EEVblog Electronics Community Forum
General => General Technical Chat => Topic started by: Simon on November 09, 2018, 05:36:08 pm
-
So my not so trusty and way too young PC seems to be slowly dying. So time for a new motherboard (the faulty bit) some RAM (upgrade to DDR4) and well processor as that will need replacing for the DDR4 RAM.
As always not being a computer buff anymore I am baffled by which processor from which all other selections will transpire.
My laptop is i7 and works great, my tablet (smaller touchscreen laptop) is an i5 and i am quite happy with it but have not tried anything too grunty on it. I type this on it as it drives a 4K screen so can't complain.
I see that as always AMD have an infinite number of models on offer and of course cost less. My current PC is an FX-4350 but I am not sure if the more expensive models are worth while as they simplf seem to be more cores not speed. I am sure video editors that every CPU is benchmarked on will love the 12 core 24 thread CPU but erm I'm not sure circuit studio will know what to do with the remaining 11 cores and 23 threads.
I want a good machine and am only really going for a PC still because separate graphics is the only way to get the real RAM and CPU performance and laptops with dedicated graphics are expensive. I have a perfectly adequate graphics card other bits.
I won't do anything too heavy but want a fast one but multicore does not mean fast and unless they independently clock down every time one core fires up all the others will too using more power.
-
Anything from the last 3 years will do (Core 6xxx generation and up, AMD Ryzen), max out the memory (16GB at least), get a decent SSD and you are done.
-
What is your budget ? what form factor ? eg ATX, Micro Atx
speeds have been fairly stagnant for a few years but core counts have gone up
A SSD and 64bit OS are a must these days, graphics cards are a big factor in a system too
Im still running my X99 (5820K hex core) with ddr4 and that still handles what I throw at it
-
Seriously consider one of the Ryzens. Some of the mid range models have excellent performance and low power consumption to the point that some power supplies won't handle how little power they draw at idle. That makes it easy to make a nice quiet machine with low operating costs with an upgrade path if you decide that you need a faster processor later. Of course, if you need the performance, you can go straight for one of the higher-end grunty ones right now. :)
FreeBSD runs great on them (though the Vega graphics in chips like the 2400G still isn't quite supported yet but you already have a graphics card, right?) and I've been able to still run Windows 7 on the chipsets I've been using, though you do have to mess with it a bit. Most customers I have gone to running FreeBSD on the bare metal and then Windows or MacOS or whatever in a VM depending on their needs.
Definitely use an SSD if you're not already, and since modern software is bloated to the extreme and RAM is relatively inexpensive right now, don't skimp on the size of your first two DIMMs so you'll have space to upgrade later unless you go straight for the MAX that the chipset & cpu supports.
Try not to skimp on the motherboard, either. It pays off in the long run. I almost always use ASUS (occasionally 2nd choice is Gigabyte depending on feature set) and loved the Sabretooth line for general purpose to workstation class but they've discontinued that "line." Just don't cheap out on the board. :)
-
I would be running windows although i have to admit I'd love to leave it. The main thing tying me down at the moment is circuit studio. RAM cheap? not the last time i looked.
-
What is the difference between the Ryzen numbers ?
-
2200G: 4c/4t embedded GPU
2400G= 4c/8t embedded GPU
Ryzen '7' (2700,1800,1700) 8c/16t
Ryzen '5' (2600,1600) 6c/12t
Ryzen '5' (2500,1500,1400) 4c/8t
Ryzen '3' (2300,1300,1200) 4c/4t
1700x vs 1700 per example is just a bit higher clocked, usually can be achieved by a mild overclock yourself (but also comes with different cooler)
2700 vs 1700 per example is a generation bump with a few 100MHz higher clock, smaller process node, a few minor improvement.
-
Looks like it has to be a 7 for DDR4 support.
So what are i5's to i7's ?
-
oh and as i have a GPU already why buy a Rayzen 3 or 5 as i don't need their included GPU
-
Before replacing the whole system, do you have any of the following:
* A Intel quad-core processor from within this decade, an AMD Ryzen processor or better
* At least 16GB of RAM,
* A recent mid-tier graphics card or better
* A SSD, hopefully a SSD that uses PCIe interface.
If your system is less than 5 years old, some spot upgrades like an NVMe SSD, maxing our the RAM, or a new graphics card can do wonders while not breaking the bank. Depending on the system, it can be helpful to comb eBay or AliExpress for a compatible CPU upgrade too.
Also don’t put the old parts in trash, donate it or put it in your parts bin instead. That is better for the environment, and you might find one day the parts accumulated in that bin can be put together as a potato PC for YouTube and Facebook or a basic home server running some lightweight Linux OS like xubuntu or Raspbian-x86.
-
The motherboard is fucked! if I am going to buy another motherboard i may as well get some shiny new DDR4 and a new CPU has to come with that. The CPU will also use 50% the power. As for "at least" 16GB of RAM 16GB is plenty for me.
I do so hate it when people come on like experts and recommend far more than is required for it to be brilliant. of course it will be at 200% the price!
-
So what are i5's to i7's ?
What I would watch out for some of the newer boards put the tracks over the heatsink bracket hole areas and they can get crushed if they are tightened too much.
If you compare the I5's and I7's by generation
https://en.wikipedia.org/wiki/List_of_Intel_Core_i5_microprocessors
https://en.wikipedia.org/wiki/List_of_Intel_Core_i7_microprocessors
You will see the difference in cache size.
I5 6MB L2 cache
I7 8MB L2 cache
-
Yea, AMD it is, 16MB cache 2/3 the price
-
https://www.ebuyer.com/830814-amd-ryzen-7-2700-am4-processor-with-led-wraith-spire-cooler-yd2700bbafbox (https://www.ebuyer.com/830814-amd-ryzen-7-2700-am4-processor-with-led-wraith-spire-cooler-yd2700bbafbox)
-
The motherboard is fucked! if I am going to buy another motherboard i may as well get some shiny new DDR4 and a new CPU has to come with that. The CPU will also use 50% the power. As for "at least" 16GB of RAM 16GB is plenty for me.
I do so hate it when people come on like experts and recommend far more than is required for it to be brilliant. of course it will be at 200% the price!
If the motherboard died the spot upgrade method goes nowhere then. I was assuming that the machine still works.
As of now the bang for the buck looks like something like this to me:
* Ryzen 5 2nd gen
* 16GB DDR4-2666
* B350 motherboard
* RX 570 or GTX 1060 6GB graphics card
* 256GB NVMe SSD for boot drive
* 2TB spinning rust as data drive.
Since applications are getting bigger and bigger, having 16GB RAM will help a lot. At least here on my machine even 16GB barely makes it with ~30 tabs in Google Chrome and I went to 32GB RAM a yamear ago.
-
https://www.ebuyer.com/868954-exdisplay-msi-amd-x370-gaming-pro-carbon-am4-atx-motherboard-ebr2-x370-gaming-pro-carbon (https://www.ebuyer.com/868954-exdisplay-msi-amd-x370-gaming-pro-carbon-am4-atx-motherboard-ebr2-x370-gaming-pro-carbon)
-
For features with the Intels, they've been making i5 and i7 for a while so you have to go by the "generation", such as 5xxx, 4xxx, and so on. As an example, I have a ancient 3 series i5, an i5-3570K. A newer one would be i5-8600. I think DDR4 support started with 6 series maybe?
-
ok forget that motherboard, it won't support decent ram speed without overclocking.
-
Looks like it has to be a 7 for DDR4 support.
So what are i5's to i7's ?
i5 and i7 are marketing terms. Traditionally you could infer a few things from them, but since AMD got Intel in a panic they've started messing with the numbers in all kinds of ways. The terms have also been used for ages, so ageing chips go by the same name.
It's good to note that IPC is almost identical for both current generations AMD and Intel chips. Comparing clockspeeds is still not an apples to apples thing, but it's not as unreasonable as it was in the days of the FX chips. Right now the sweetspot seems to be somewhere around the Ryzen 5 or 7 series, depending on how much power you can actually use. The 5 series is a very nice and potent range for a comparatively modest budget.
-
https://www.ebuyer.com/868954-exdisplay-msi-amd-x370-gaming-pro-carbon-am4-atx-motherboard-ebr2-x370-gaming-pro-carbon (https://www.ebuyer.com/868954-exdisplay-msi-amd-x370-gaming-pro-carbon-am4-atx-motherboard-ebr2-x370-gaming-pro-carbon)
X370 is overkill for a mid-tier machine, unless you overclock it.
Also usually the motherboard don’t fail that frequently. Maybe it is because me being so used to server grade gear that is designed to last for a long time, but from my experience a problematic system can often be fixed by a deep cleaning to the hardware and a fresh OS install on the software.
Speaking of, you can actually look into building your next machine using decommissioned server hardware commonly available on eBay and AliExpress for fairly low prices. The only drawback AFAIK is the power draw since they are older technology with worse energy management, but since servers are often ahead of the performance curve than consumer grade stuff older generation server grade gear often perform as good as, if not better than current generation mid-tier consumer grade stuff.
-
If the motherboard died the spot upgrade method goes nowhere then. I was assuming that the machine still works.
As of now the bang for the buck looks like something like this to me:
* Ryzen 5 2nd gen
* 16GB DDR4-2666
* B350 motherboard
* RX 570 or GTX 1060 6GB graphics card
* 256GB NVMe SSD for boot drive
* 2TB spinning rust as data drive.
Since applications are getting bigger and bigger, having 16GB RAM will help a lot. At least here on my machine even 16GB barely makes it with ~30 tabs in Google Chrome and I went to 32GB RAM a yamear ago.
If the budget's available, something like this is a very nice system without overdoing things. Just make sure to buy a proper PSU and not some cheapie.
-
well a 7 costs not a lot more as it drops the GPU I'd be paying for but not using. the £259 CPU is fine, still far cheaper than the entry level i7.
Now I just need a motherboard that does not tell lies about the RAM speed which mean the lying cunts at MSI can sod off! their motherboards can do wonderful things until you read the specs and find that the 2 low end speeds are the only real ones. Had an Asus motherboard like that once and it was a nightmare.
-
what a load of codswallop:
Supports AMD 2nd Generation Ryzen™/ Ryzen™ with Radeon™ Vega Graphics/ 1st Generation Ryzen™ Processors
Dual Channel Non-ECC Unbuffered DDR4, 4 DIMMs
HDMI, DVI-D Ports for Multiple Display
Integrated I/O Shield of Ultra Durable™ Design
Dual Ultra-Fast NVMe PCIe Gen3 M.2 (x4, x2) with One Thermal Guard
High Quality Audio Capacitors and Audio Noise Guard with LED Trace Path Lighting
RGB FUSION with Multi-Zone LED Light Show Design, Supports Digital LED & RGB LED Strips
GIGABYTE Exclusive 8118 Gaming LAN with Bandwidth Management
Smart Fan 5 Features 6 Temperature Sensors and 4 Hybrid Fan Headers with FAN STOP
APP Center Including EasyTune™ and Cloud Station™ Utilities
CEC 2019 Ready, Save Power with a Single Click
-
what a load of codswallop:
Supports AMD 2nd Generation Ryzen™/ Ryzen™ with Radeon™ Vega Graphics/ 1st Generation Ryzen™ Processors
Dual Channel Non-ECC Unbuffered DDR4, 4 DIMMs
HDMI, DVI-D Ports for Multiple Display
Integrated I/O Shield of Ultra Durable™ Design
Dual Ultra-Fast NVMe PCIe Gen3 M.2 (x4, x2) with One Thermal Guard
High Quality Audio Capacitors and Audio Noise Guard with LED Trace Path Lighting
RGB FUSION with Multi-Zone LED Light Show Design, Supports Digital LED & RGB LED Strips
GIGABYTE Exclusive 8118 Gaming LAN with Bandwidth Management
Smart Fan 5 Features 6 Temperature Sensors and 4 Hybrid Fan Headers with FAN STOP
APP Center Including EasyTune™ and Cloud Station™ Utilities
CEC 2019 Ready, Save Power with a Single Click
Where are all the important metrics and part numbers?
-
what a load of codswallop:
Supports AMD 2nd Generation Ryzen™/ Ryzen™ with Radeon™ Vega Graphics/ 1st Generation Ryzen™ Processors
Dual Channel Non-ECC Unbuffered DDR4, 4 DIMMs
HDMI, DVI-D Ports for Multiple Display
Integrated I/O Shield of Ultra Durable™ Design
Dual Ultra-Fast NVMe PCIe Gen3 M.2 (x4, x2) with One Thermal Guard
High Quality Audio Capacitors and Audio Noise Guard with LED Trace Path Lighting
RGB FUSION with Multi-Zone LED Light Show Design, Supports Digital LED & RGB LED Strips
GIGABYTE Exclusive 8118 Gaming LAN with Bandwidth Management
Smart Fan 5 Features 6 Temperature Sensors and 4 Hybrid Fan Headers with FAN STOP
APP Center Including EasyTune™ and Cloud Station™ Utilities
CEC 2019 Ready, Save Power with a Single Click
Where are all the important metrics and part numbers?
Exactly! what are Audio capacitors? didn't anyone tell them that digital signals are far faster than audio ones and if audio capacitors is all the assholes put in then they ain't clever!
-
what a load of codswallop:
Supports AMD 2nd Generation Ryzen™/ Ryzen™ with Radeon™ Vega Graphics/ 1st Generation Ryzen™ Processors
Dual Channel Non-ECC Unbuffered DDR4, 4 DIMMs
HDMI, DVI-D Ports for Multiple Display
Integrated I/O Shield of Ultra Durable™ Design
Dual Ultra-Fast NVMe PCIe Gen3 M.2 (x4, x2) with One Thermal Guard
High Quality Audio Capacitors and Audio Noise Guard with LED Trace Path Lighting
RGB FUSION with Multi-Zone LED Light Show Design, Supports Digital LED & RGB LED Strips
GIGABYTE Exclusive 8118 Gaming LAN with Bandwidth Management
Smart Fan 5 Features 6 Temperature Sensors and 4 Hybrid Fan Headers with FAN STOP
APP Center Including EasyTune™ and Cloud Station™ Utilities
CEC 2019 Ready, Save Power with a Single Click
Where are all the important metrics and part numbers?
Exactly! what are Audio capacitors? didn't anyone tell them that digital signals are far faster than audio ones and if audio capacitors is all the assholes put in then they ain't clever!
Classic consumer hardware marketing wank.
That is another reason I have almost completely switched to workstation/server hardware: no marketing wank and cut directly to the technology details, after their longevity optimized validation process and their much broader hardware compatibility.
-
Intel still is king in single core performance, but AMD is not far behind. But Intel has production shortages right now, so their premium priced products are even more premium. If you don't need 8 cores, get one with less. Simple as that. AMD is looking very competitive in price/performance across the complete spectrum now.
Personally I would recommended getting an AMD CPU with fast RAM. The RAM speed spec for AMD Ryzen 2nd gen is 2933MHz, so any slower (2666MHz) will run below factory speeds. Any board should support factory speeds. If not, return it.
You do have much more faster memory like 3200, 3600, etc. But basically you're overclocking at that point. If you want a stable system, don't touch any OC feature.
The necessity of 16GB or 32GB is a personal thing. 5 years ago, I built an i5 system with 16GB DDR3 (costed me 100 euro's back then). 16GB serves me well in both my Linux workstation (which I use to run 2 Windows VMs + Jetbrains IDEs at the same time) and my Windows desktop (CAD + entertainment). , With recent DDR4 shortages the 16GB DDR4 prices haven't even settled down to that level.
Unused RAM won't hurt anything, so if you can afford sure go ahead..
If fast RAM is at the cost of a SSD (NVME if possible), then :-//
-
Looks like it has to be a 7 for DDR4 support.
If you mean the Ryzen chips, no, anything with the AM4 socket is DDR4.
oh and as i have a GPU already why buy a Rayzen 3 or 5 as i don't need their included GPU
Most of the earlier version Ryzen chips do not have integrated graphics, only the newer ones do, like anything with a G suffix and some models of X. You need to check the processor specifications for the particular models you're interested in. While it doesn't add much to the cost to have it in there, if you're never going to use it then it is kind of silly to get one. :)
As for memory speed, the memory controller is in the CPU so as long as the motherboard is physically designed properly you maximum memory speed is limited by the CPU, not the board. Most of the current chips officially support up to 2933 MHz memory speed. Most of the systems I've been building lately, though, I get proper 3200 MHz rated RAM and run it at that. Have not had a single issue yet. You system will probably default to 2666 or 2933 depending on the CPU installed and the BIOS on the board but "overclocking" the memory controller part of the CPU by 10% isn't going to hurt anything and on a decent board stability is not an issue. if you buy a cheap-ass motherboard that had the PCB layout done by a chimpanzee, then probably stick with the official 2933 maximum clock. :)
3400 MHz would probably be fine also in virtually all cases (on a good MB) but trying to run in the 3500-4500 MHz range is less likely to work properly without careful tuning, even with RAM that is intended to run at those speeds.
-
It's not purely layout with the RAM, most people never seem to realise that the actual distance between RAM and CPU is an issue due this inconvenient thing called the speed of light and even slower speed of electrons in a PCB trace......
-
It's not purely layout with the RAM, most people never seem to realise that the actual distance between RAM and CPU is an issue due this inconvenient thing called the speed of light and even slower speed of electrons in a PCB trace......
Uhhh... How is that not a layout issue?
The DIMM sockets need to be right by the CPU with careful attention to trace length matching, impedance of the traces on the finished board, what kind of crazy stuff you're running around there on other layers, etc. That is "board layout" in my book.
-
True
-
I have set my eyes on some 3.2GHz RAM and the motherboard supports up to 3.45GHz so I might just push the CPU a tiny bit.
-
I have set my eyes on some 3.2GHz RAM and the motherboard supports up to 3.45GHz so I might just push the CPU a tiny bit.
Ryzen performance is helped by fast RAM more than that of Intel chips.
-
I always wonder about RAM controller clock to CPU clock. I found this on the FX4250 I have that just adding 200MHz made a difference even though it was just a 10% jump. I assume that just 10% increase in RAM speed might mean that the data the CPU wants is ready one clock cycle before it would have been before, if the data comes 5% after the last CPU tick it's no good and will take a further 0.95 of a tick before usable, but if i go 10% faster then it is ready at 0.05 before the next CPU tick that is effectively 1 whole CPU cycle early.
-
I always wonder about RAM controller clock to CPU clock. I found this on the FX4250 I have that just adding 200MHz made a difference even though it was just a 10% jump. I assume that just 10% increase in RAM speed might mean that the data the CPU wants is ready one clock cycle before it would have been before, if the data comes 5% after the last CPU tick it's no good and will take a further 0.95 of a tick before usable, but if i go 10% faster then it is ready at 0.05 before the next CPU tick that is effectively 1 whole CPU cycle early.
Note that the FX has an entirely different architecture which cannot be compared other than through benchmarks.
-
Have a look at mindfactory.de for cheap PC HW for good prices.
https://www.mindfactory.de/search_result.php?select_search=0&search_query=ryzen+2700 (https://www.mindfactory.de/search_result.php?select_search=0&search_query=ryzen+2700)
-
I am in the UK, considering current exchange rates they are about the same as the supplier i am already registered with.
-
Warranty could be a pain too.
-
I am buying the parts now in the UK, £651 for the ryzen 7 2700, 16GB of 3200MHz RAM, a 512GB M.2 drive and a motherboard
-
Out of curiosity, what motherboard have you decided on.
-
I am in the UK, considering current exchange rates they are about the same as the supplier i am already registered with.
Things like CPUs are a commodity item so in most locales the prices should be similar unless there are significant tarrifs or something. After currency conversion, that listed German website is right about $20 CAD more than the typical high-volume retail or online price around here, so about right overall. Small quantity wholesale pricing knocks another 10-15% or so off of that at my suppliers currently, so that's in the ballpark, not getting ripped off or anything. :)
-
Out of curiosity, what motherboard have you decided on.
https://www.ebuyer.com/835547-gigabyte-x470-aorus-ultra-gaming-am4-atx-motherboard-x470-aorus-ultra-gaming (https://www.ebuyer.com/835547-gigabyte-x470-aorus-ultra-gaming-am4-atx-motherboard-x470-aorus-ultra-gaming)
-
Simon, just curious what settings are you using that work for you?
I have an fx6300.
I always wonder about RAM controller clock to CPU clock. I found this on the FX4250 I have that just adding 200MHz made a difference even though it was just a 10% jump. I assume that just 10% increase in RAM speed might mean that the data the CPU wants is ready one clock cycle before it would have been before, if the data comes 5% after the last CPU tick it's no good and will take a further 0.95 of a tick before usable, but if i go 10% faster then it is ready at 0.05 before the next CPU tick that is effectively 1 whole CPU cycle early.
Also, if you are upgrading, and want to use an AMD CPU, with Linux in the future to pursue a number of kinds of scientific computing using your GPU, just in case the technology is something you want to get into in the future, make sure you get hardware that is known to be compatible with the RoCM software project. Most newer HW is likely compatible but since you are upgrading all at once, you should double check.
Basically this means PCIe gen 3 atomics and there may be some other particulars about the way the main bus is laid out.
See https://rocm.github.io/hardware.html
Would I get an Nvidia graphics card rather than an AMD? thats a difficult question. Id be lying if said that I haven't had moments regretting picking AMD recently. This project is supposed to make it possible to run cuda software on AMD graphics and CPU HW. But it may still make much more sense for some kinds of work (neural networks, deep learning and many kinds of computer graphics) to have a cuda card, even though they are more than 50% again expensive for an equivalent performance.
-
i just bumpet it up by one increment which i think is 200MHz, it runs at 2000MHz, 2200 is fine, 2400 could be unstable
-
That Gigabyte X470 AORUS ULTRA GAMING is a good board, it should serve you well.
It looks like you've picked a good combo, I think you'll be pleased with the performance. Total processing power-wise it is about 3 times what your FX-4350 can muster... More than enough to notice the difference, especially with a decent PCIe SSD. :)
-
Heh. Got the same gear for 100£ less at mindfactory. But your funeral :D
-
Simon, you should run some tests with the memory.
The thing with very high frequency memories is that they also have very loose timings.
Ryzen likes very fast memory but from some point (around 3000-3200 Mhz) the benefits are minimal and in fact it could actually be a bad idea (infinity fabric inside cpu may not cope and may get corrupted transfers and waste time repeating tranfers etc)
You should do some tests with that memory running at 3200 and 3000, in both cases with the smallest timings that are stable.
-
i just bumpet it up by one increment which i think is 200MHz, it runs at 2000MHz, 2200 is fine, 2400 could be unstable
I think you were probably changing the CPU clock multiplier then. On most decent motherboards should be able to change the actual master clock speed in 1 or 5 MHz increments, then fiddle with multipliers (on chips that support it and aren't locked, of course,) RAM speed, etc.
I usually do all my burn in-testing for customer machines about 5% higher than I intend to run the combo at, then back it off that little bit. Often the machine is actually stable at another 10+% above that but I don't normally severely overclock machines going out to customers (and other than the memory being above "rated" I normally run the processor clock at stock (or slightly below for things like silent media PCs, etc.)
-
I always wonder about RAM controller clock to CPU clock. I found this on the FX4250 I have that just adding 200MHz made a difference even though it was just a 10% jump.
I don't know about AMD, but different Intel processors have Memory buses of different width. The wide-bus Intel chips have over 2000 pins and the memory bus is twice as wide. In theory, this should matter more than 500 MHz difference in DDR4 speed. Although I haven't seen the benchmarks. Of course, these bigger chips are more expensive and require a different motherboard, which is also more expensive.
-
Ryzen has Infinity Fabric which connects the two CCX (core complexes, up to 4cores 4 threads each) and the memory controller and everything else... pci-e , the SOIC part (usb3 and sata ports on cpu)
It's tied to the memory frequency, see https://en.wikichip.org/wiki/amd/infinity_fabric (https://en.wikichip.org/wiki/amd/infinity_fabric) and https://www.tomshardware.com/reviews/amd-ryzen-threadripper-1950x-game-performance,5207-2.html (https://www.tomshardware.com/reviews/amd-ryzen-threadripper-1950x-game-performance,5207-2.html)
-
I believe a lot of people here have mentioned, AMD's Infinity Fabric is dependent on RAM clock speed, and since Infinity Fabric works almost as if the front side bus, it has direct and significant impact on overall system performance. Even if you don't plan on overclocking your system too much, it would be worth it to crank that part up using RAM sticks rated at higher speed like 2933MHz or 3200MHz.
On the other hand for a mid tier machine there is no point going with a X470 motherboard if you don't plan to overclock the CPU sky high. B450 will work just fine and allow some overclocking, and it will cost less.
Also there is a parity issue with CPU and GPU. They should be around the same tier, otherwise one of them will bottleneck and you are not going to get your money's worth in computing performance.
-
I don't know about AMD, but different Intel processors have Memory buses of different width. The wide-bus Intel chips have over 2000 pins and the memory bus is twice as wide. In theory, this should matter more than 500 MHz difference in DDR4 speed. Although I haven't seen the benchmarks. Of course, these bigger chips are more expensive and require a different motherboard, which is also more expensive.
The "desktop" quad-channel memory Intel processors use the 2011 and later 2066 pin LGA "sockets" originally borrowed from the server Xeons. To take advantage of that configuration you obviously need to populate four DIMM sockets (most motherboards have 8 DIMM sockets.)
On AMD you currently have to go to the server level chips to get 4 memory channels.
-
On AMD you currently have to go to the server level chips to get 4 memory channels.
The Threadrippers are not *that* expensive.
-
The Threadrippers are not *that* expensive.
Nope. Certainly a valid way of getting 4-channel performance.
The Intel 4-channel "desktop" chips are certainly not cheap! :)
Also, the "entry level" Intel CPUs that use LGA2066 only have dual-channel memory support.
-
The Threadrippers are not *that* expensive.
Nope. Certainly a valid way of getting 4-channel performance.
The Intel 4-channel "desktop" chips are certainly not cheap! :)
Also, the "entry level" Intel CPUs that use LGA2066 only have dual-channel memory support.
If you need 4-channel on the cheap, it might be a better idea to use second hand Intel LGA2011 based systems. Those were decommissioned en masse a few years ago and cleap LGA2011 (not the -1 or -3 variety) ans matching DDR3 memory flooded the market.
-
It's nice to have many cores and more threads, but make sure your software can take advantage of them!
I have a dual CPU box with TWO 2.somethingGHz Xeon 6 core 12 thread CPU.
Even with software package like Photoshop and FlightSimulator, most cores/threads will stay idle while just a few would reach 100%. It would have been better with less core but fast CPU.
Also, I have 56GB memory but again, 12GB peak at the most gets used.
Total waste of money!
BenchMarking CPU and chipset is one thing but reality is something totally different.
-
It's nice to have many cores and more threads, but make sure your software can take advantage of them!
I have a dual CPU box with TWO 2.somethingGHz Xeon 6 core 12 thread CPU.
Even with software package like Photoshop and FlightSimulator, most cores/threads will stay idle while just a few would reach 100%. It would have been better with less core but fast CPU.
Also, I have 56GB memory but again, 12GB peak at the most gets used.
Total waste of money!
BenchMarking CPU and chipset is one thing but reality is something totally different.
You must be one of those people who close their browser tabs, aren't you? ;D
I think Photoshop and Flight Simulator both hail from another period. Programmers are catching on how to properly make use of more cores. Games have been using them for a while now and things like renderers of course have almost always done it. Most programs benefit from more cores, but developers simply didn't know how to or want to spend the effort making use of them. That being said, having fast cores is still a safe bet for consumer tasks.
-
On AMD you currently have to go to the server level chips to get 4 memory channels.
The Threadrippers are not *that* expensive.
No, they are the price of a decent i7 but I don't need to go that far. The threadripper was clearly aimed at the server market but may well have desktop applications for some particular usage cases including over night increase in the size of you penis by 50% when installed on machines that will never get to use them properly.
-
I believe a lot of people here have mentioned, AMD's Infinity Fabric is dependent on RAM clock speed, and since Infinity Fabric works almost as if the front side bus, it has direct and significant impact on overall system performance. Even if you don't plan on overclocking your system too much, it would be worth it to crank that part up using RAM sticks rated at higher speed like 2933MHz or 3200MHz.
On the other hand for a mid tier machine there is no point going with a X470 motherboard if you don't plan to overclock the CPU sky high. B450 will work just fine and allow some overclocking, and it will cost less.
Also there is a parity issue with CPU and GPU. They should be around the same tier, otherwise one of them will bottleneck and you are not going to get your money's worth in computing performance.
i am hoping this motherboard lasts a little longer, yes it may have been a little cheaper than the last one but 3 years and conked out? not happy.
I was hoping that the higher end one would last longer, i don't bother to change machines that often these days. I don't need the performance really although i do like to run BOINC.
The CPU is rated at 2933 so the motherboard and RAM at 3200 will already be able to exceed that. If i can crank the RAM channel to 3200 I will be perfectly happy.
Indeed on the last machine it was the memory controller I increased the speed of not the RAM per se as i had more or less worked out that it was going to be a hub of some kind joining several things together like the external memory controllers/North bridges used to do.
FSB? oh yes I remember my first PC with a 900MHz AMD spitfire, it had a DDR FSB at 100MHz and 133MHz RAM, the asshole that I bout it from set the RAM speed to 100MHz to err.... match the speed of the FSB. My that was a speed boost of 33% that was a long time coming.
-
Heh. Got the same gear for 100£ less at mindfactory. But your funeral :D
http://www.ebuyer.com/product/830814 (http://www.ebuyer.com/product/830814)
http://www.ebuyer.com/product/835547 (http://www.ebuyer.com/product/835547)
http://www.ebuyer.com/product/856118 (http://www.ebuyer.com/product/856118)
http://www.ebuyer.com/product/743022 (http://www.ebuyer.com/product/743022)
I'm not too bothered, I get the VAT back anyway and shipping is free and it is easier to deal with them in the even of problems plus some local people get employed even if they are cheap eastern European migrant warehouse workers |O
-
Just built a new PC for myself, as follows:
- Corsair 200R case
- Noctua NH-U12S Ultra Quiet Single Tower CPU Cooler
- 32GB (2x16GB) Corsair DDR4 Vengeance LPX Black, PC4-21300 (2666), Non-ECC Unbuffered
- Intel Core i7 8700K, S 1151, Coffee Lake, 6 Core, 12 Thread, 3.7GHz, 4.7GHz Turbo, 12MB Cache, 1200MHz GPU, 95W
- ASUS PRIME H370-PLUS, Intel H370, S 1151, DDR4, SATA3, Dual M.2, 2-Way CrossFire, Realtek GbE, USB 3.1 Gen2 A, ATX
- 500GB Samsung 970 EVO, M.2 (2280) PCIe 3.0 (x4) NVMe SSD, Phoenix, MLC V-NAND, 3400MB/s Read, 2300MB/s Write, 370k/450k
- 750W Corsair RMx Series RM750x, Full Modular, 80PLUS Gold, SLI/CrossFire, Single Rail, 62.5A, 135mm Fan, ATX PSU
- GTX1050Ti (pulled from old PC)
Motherboard was chosen because I wanted a couple of PCI slots in addition to PCIe. A lot of new board only have one, if any, PCI slots. I need them for a couple of legacy cards I use.
PC is used for software dev and video editing/rendering mostly.........and of course browsing these forums...:-)
Ian.
-
I dearly hope that this is the last PC i build for a while. with 8 cores and about the max speed any processor will do the real performance developments are now left to the software writers to make their code multithreading. The software industry is lagging badly and selling money for old rope. We use solid edge at work and for years it did not multithread (not sure if it does now) and despite having 4 core i7 processors (which worked faster if hyperthreading was turned off) 3/4 of the proccessor was idle. And this really annoyed me because large assemblies would take a long time to move on the screen yet were a perfect multithread use case as once you establish the absolute position of all components each core can independently move a part on it's own making full use of the processor. But no, Siemens don't give a shit and turn out barely usable code and when you find a bug pretend it's you fault.
-
If you need 4-channel on the cheap, it might be a better idea to use second hand Intel LGA2011 based systems. Those were decommissioned en masse a few years ago and cleap LGA2011 (not the -1 or -3 variety) ans matching DDR3 memory flooded the market.
Having owned a 2011 system from new, they're fantastic machines albeit much less energy efficient than newer machines of comparable specs. And keep in mind a standard desktop Ryzen 7 has 8 cores while the 2011s (Sandy Bridge E and Ivy Bridge E) only have 6 for the single socket versions.
It's nice to have many cores and more threads, but make sure your software can take advantage of them!
I have a dual CPU box with TWO 2.somethingGHz Xeon 6 core 12 thread CPU.
Even with software package like Photoshop and FlightSimulator, most cores/threads will stay idle while just a few would reach 100%. It would have been better with less core but fast CPU.
Also, I have 56GB memory but again, 12GB peak at the most gets used.
Total waste of money!
BenchMarking CPU and chipset is one thing but reality is something totally different.
All the extra RAM and cores seems perfect to install Virtualbox and spin up a few earnhoney mining VMs. Especially in the winter when the heat generated can be put to good use. Have 48GB in my PC with about 20GB in use by all apps including some VMs and up to 24GB allocated for use by tmpfs.
-
i run 16GB without a pagefile (ruins SSD's) and never had a problem. If i need more i have 2 more slots that i can fill.
-
I dearly hope that this is the last PC i build for a while. with 8 cores and about the max speed any processor will do the real performance developments are now left to the software writers to make their code multithreading. The software industry is lagging badly and selling money for old rope. We use solid edge at work and for years it did not multithread (not sure if it does now) and despite having 4 core i7 processors (which worked faster if hyperthreading was turned off) 3/4 of the proccessor was idle. And this really annoyed me because large assemblies would take a long time to move on the screen yet were a perfect multithread use case as once you establish the absolute position of all components each core can independently move a part on it's own making full use of the processor. But no, Siemens don't give a shit and turn out barely usable code and when you find a bug pretend it's you fault.
Once again, decommissioned server hardware often last longer than brand new consumer hardware, simply because they are built to last. If you can find a solid used overclockable Intel X79 motherboard, that plus a Xeon E5-1650v2 can work better than a Ryzen: it has a ring bus architecture instead of Infinity Fabric (a 2-way NUMA, similar to a dual Xeon E5-2643v2 system.) it has 4-channel memory support, it has 44 PCIe lanes for multi-GPU or a lot of random PCIe cards, and while being a Xeon it has an unlocked clock multiplier and can be overclocked sky high, and thanks to it being a Xeon the base voltage and temperature are lower allowing for more overclocking headroom.
If you need 4-channel on the cheap, it might be a better idea to use second hand Intel LGA2011 based systems. Those were decommissioned en masse a few years ago and cleap LGA2011 (not the -1 or -3 variety) ans matching DDR3 memory flooded the market.
Having owned a 2011 system from new, they're fantastic machines albeit much less energy efficient than newer machines of comparable specs. And keep in mind a standard desktop Ryzen 7 has 8 cores while the 2011s (Sandy Bridge E and Ivy Bridge E) only have 6 for the single socket versions.
Like above, that Ryzen is more like dual E5-2637v2 in the same package: a NUMA system. NUMA puts a lot of pressure on the memory management system of the OS since the memory access latency will be irregular. The ring bus of Ivy Bridge Xeon has a much regular memory access pattern.
Also specifically for the E5-16xx series, maybe with the exception of the lowest end E5-1620 and E5-1620v2, they have an unlocked multiplier allowing for sky high overclocks on a solid X79 motherboard. Being a Xeon those chips are binned from the get go for lower temperature and voltage, allowing even more headroom at overclocking.
It's nice to have many cores and more threads, but make sure your software can take advantage of them!
I have a dual CPU box with TWO 2.something GHz Xeon 6 core 12 thread CPU.
Even with software package like Photoshop and FlightSimulator, most cores/threads will stay idle while just a few would reach 100%. It would have been better with less core but fast CPU.
Also, I have 56GB memory but again, 12GB peak at the most gets used.
Total waste of money!
BenchMarking CPU and chipset is one thing but reality is something totally different.
All the extra RAM and cores seems perfect to install Virtualbox and spin up a few earnhoney mining VMs. Especially in the winter when the heat generated can be put to good use. Have 48GB in my PC with about 20GB in use by all apps including some VMs and up to 24GB allocated for use by tmpfs.
I have a workstation PC with 2x Xeon E5-2680 (8-cores each) and 128GB RAM. I was using that as an ESXi VM host, until I finally stopped using it as so and now it is my high core count Windows workstation.
-
You must be one of those people who close their browser tabs, aren't you? ;D
How did you know?
-
Maybe consider Intel? You said single thread performance has priority, but do not want to not give-up multi-core (no low end chips).
From passmark:
cpu - price - total cpu mark - cpu mark value - single thread mark - single thread value
AMD FX-4350 Quad-Core $129.99 5333 41.03 1532 11.79
AMD Ryzen 7 2700 $265.99 15063 56.63 2009 7.56
Intel Core i5-8600K @ 3.60GHz $239.99 12791 53.30 2518 10.49
Intel Core i5-9600K @ 3.70GHz $279.99 13452 48.05 2674 9.55
For similar price of Ryzen 7, i5 has large single thread performance gain over Ryzen (130%[Ryzen] vs 160%[i5-9600k] over fx chip) while still having decent multithread. Also the 8600k is supposed to be good for overclocking. Once you get above these ($300+), you pay something like $100 more for 5% increase in single thread performance and then $500 more for the next 5%...
Also, E-2126G ($254) and E-2136 ($284) JUST came out, they are basically i7-8700 series ($439.98 15152 34.44 2628 5.97) xeon equivalents with slightly lower turbo but much lower price than the i7.... if you dont plan on major overclocking (these are multiplier locked, but you can likely set the turbo power limits)..these might be worth looking into.
-
What about AMD's "epyc" or whatever its called? Its a server class chip.
A Chinese company also sells them, but AMD makes them, don't know if they are available elsewhere (probably not) or any cheaper that way.
-
What about AMD's "epyc" or whatever its called? Its a server class chip.
A Chinese company also sells them, but AMD makes them, don't know if they are available elsewhere (probably not) or any cheaper that way.
AFAIK those Chinese Epyc chips are sold only in China - I have a feeling that AMD licensed it out as a way to circumvent potential trade restrictions. Also I was talking about decommissioned server gear like Intel Xeon E5-1650v2 from 2013, not new ones.
Maybe consider Intel? You said single thread performance has priority, but do not want to not give-up multi-core (no low end chips).
I am still backing an overclocked Xeon E5-1650v2. Spec for spec it is almost the same as the highest end Ivy Bridge desktop chip i7-4960X Extreme Edition, including the unlocked multiplier. 12716 PassMarks at stock speed, it is on par with i5-8600K which is also a 6-core processor. It can be overclocked sky high like that Extreme Edition i7 and that Skylake i5 chips too, and it supports DDR3 Registered ECC memory which is much cheaper on the used market. Compared to that Skylake i5, E5-1650v2 has the benefit of 28 additional PCIe lanes on the CPU, allowing direct CPU connection of NVMe SSD, additional graphics cards, or even one of those decommissioned Xeon Phi coprocessor cards. If you round up the overall cost, the E5-1650v2 might be the best bang for the buck for a high end system.
-
Well I looked at the chepest i7 and it was more than the Ryzen, for what I do it won't make any difference and it's same money for flops at worse. Parts are on order so no changing that.
-
At the end the day its down to budget and your own personal requirements
My X99 5820k (2011-3) is still as good as the day I built it 4 years ago it still keeps up with the more modern cpus (passmark of 12990, single core 2021)
-
These are servers that are being removed from service? Sometimes they sell for very little. I dont run my computer all the time so that might make sense for me as a development platform, as I want to get into machine learning.
If it supports PCIv3 atomics + is very cheap.
Could you tell me a bit more? I am on a very tight budget.
What about AMD's "epyc" or whatever its called? Its a server class chip.
A Chinese company also sells them, but AMD makes them, don't know if they are available elsewhere (probably not) or any cheaper that way.
AFAIK those Chinese Epyc chips are sold only in China - I have a feeling that AMD licensed it out as a way to circumvent potential trade restrictions. Also I was talking about decommissioned server gear like Intel Xeon E5-1650v2 from 2013, not new ones.
Maybe consider Intel? You said single thread performance has priority, but do not want to not give-up multi-core (no low end chips).
I am still backing an overclocked Xeon E5-1650v2. Spec for spec it is almost the same as the highest end Ivy Bridge desktop chip i7-4960X Extreme Edition, including the unlocked multiplier. 12716 PassMarks at stock speed, it is on par with i5-8600K which is also a 6-core processor. It can be overclocked sky high like that Extreme Edition i7 and that Skylake i5 chips too, and it supports DDR3 Registered ECC memory which is much cheaper on the used market. Compared to that Skylake i5, E5-1650v2 has the benefit of 28 additional PCIe lanes on the CPU, allowing direct CPU connection of NVMe SSD, additional graphics cards, or even one of those decommissioned Xeon Phi coprocessor cards. If you round up the overall cost, the E5-1650v2 might be the best bang for the buck for a high end system.
-
At the end the day its down to budget and your own personal requirements
My X99 5820k (2011-3) is still as good as the day I built it 4 years ago it still keeps up with the more modern cpus (passmark of 12990, single core 2021)
Intel has been stagnating for quite a while since Nehalem really. Here is a comparison with the Xeon E3-1230 series across generations: (I have that Xeon E3-1231v3, and this series was fairly popular in China before Skylake as a high end mainstream desktop processor over corresponding high end mainstream i7, since it is usually 0.1-0.2GHz slower than that i7, lacked overclocking support, while cost almost as low as the high end i5)
Pulled out a lot of data and let's see...
(https://www.eevblog.com/forum/chat/time-for-a-new-computer-what-processor/?action=dlattach;attach=568924)
Passmark doesn't have a score for E-2134 which is the best approximation of what E3-1230v7 would be should the old naming scheme is kept. The E-2146G really is a better approximation of E3-1245v7.
-
AFAIK those Chinese Epyc chips are sold only in China - I have a feeling that AMD licensed it out as a way to circumvent potential trade restrictions. Also I was talking about decommissioned server gear like Intel Xeon E5-1650v2 from 2013, not new ones.
I am still backing an overclocked Xeon E5-1650v2. Spec for spec it is almost the same as the highest end Ivy Bridge desktop chip i7-4960X Extreme Edition, including the unlocked multiplier. 12716 PassMarks at stock speed, it is on par with i5-8600K which is also a 6-core processor. It can be overclocked sky high like that Extreme Edition i7 and that Skylake i5 chips too, and it supports DDR3 Registered ECC memory which is much cheaper on the used market. Compared to that Skylake i5, E5-1650v2 has the benefit of 28 additional PCIe lanes on the CPU, allowing direct CPU connection of NVMe SSD, additional graphics cards, or even one of those decommissioned Xeon Phi coprocessor cards. If you round up the overall cost, the E5-1650v2 might be the best bang for the buck for a high end system.
Ivy Bridge technology is both woefully inefficient and insecure compared to modern chips. Not something to invest in at this point.
-
These are servers that are being removed from service? Sometimes they sell for very little. I dont run by computer all the time so that might make sense for me as a development platform, as I want to get into machine learning.
Could you tell me a bit more? I am on a very tight budget.
These are individual components pulled from decommissioned servers. In some cases rach-mounted servers have proprietary motherboards, but processor, RAM modules and expansion cards are usually standardized components. You can buy used, decommissioned server processor, RAM and peripherals for very cheap from various online sellers, and put those into a matching motherboard and build a cheap but powerful system.
In their official documentation Intel maintained a separation of server chipsets and desktop chipsets as separate market segments while processors themselves used the same socket and are often electrically compatible, before Skylake Intel never enforced this separation and motherboard vendors often added support to both desktop and server processors to the motherboard regardless of the market segment of the motherboard based on that electrical compatibility.
This lead to my suggesting the E5-1650v2, which is spec for spec effectively an i7-3970X (top-tier Ivy Bridge desktop chip) with 3MB less cache, 0.1GHz lower top turbo clock speed, but with support for ECC memory (cheap due to limited support not allowing the supply on the used market drain up fast enough,) faster DDR3-1866 RAM over DDR3-1600, close to 3x more memory (allowing for 8x 16GB memory modules for 128GB total,) supports PCIe 3.0 over the desktop chip's PCIe 2.0 (almost doubles NVMe performance,) has 20W less TDP (either a save on electric bill or a good amount of overclocking headroom, since both chip has unlocked clock multiplier) than the desktop chip, and costs a little over half as much even back when they are sold at MSRP.
Ivy Bridge technology is both woefully inefficient and insecure compared to modern chips. Not something to invest in at this point.
If you are building a mission critical server, buy the latest stuff with security patches applied. If you are just using it as a workstation behind a firewall, you might be able to get away with the older components for the cheap. If your motherboard has a socketed BIOS chip (like my Asus Z9PE-D16C/2L) and you have a programmer suitable for it (can be the CH341-based cheapies, universal programmer like TL866, or even just an Arduino with appropriate programming) you can even find BIOS modding tools to add those patches yourself (and new features like NVMe.)
-
The problem is that all of the performance tests are usually done with multithreading video encoding so not a fair comparison for most software.
-
With the multicore war just heating up investing in old server hardware doesn't seem sensible. It's all inefficient, hot and insecure compared to newer stuff. Patching can only get you so far, it's still a workaround for hardware issues and impacts performance even more. You don't need to do mission critical stuff to be badly hurt by a breach.
Besides, what was exotic server hardware not long ago is fairly mundane desktop territory today. With the added benefit of higher single core clocks on top.
-
The problem is that all of the performance tests are usually done with multithreading video encoding so not a fair comparison for most software.
Depending on what you do really. If you have a single thread exclusive workload, it might be cheaper to buy an overclockable Pentium or Core i3, count on the fact that there are only 30% or less of the chip active, and crank that clock sky high on standard cooling solutions. If it is heavily multithreaded you will need a high end Ryzen Threadripper or even a dual Broadwell-EP machine (dual-socket processors since Skylake are crazy.)
I pulled up that list noting Passmark per core per GHz, boiling down to almost just cache per core and instruction per clock difference across generations of Intel processors, which is barely.
With the multicore war just heating up investing in old server hardware doesn't seem sensible. It's all inefficient, hot and insecure compared to newer stuff. Patching can only get you so far, it's still a workaround for hardware issues and impacts performance even more. You don't need to do mission critical stuff to be badly hurt by a breach.
I am not using decommissioned server hardware as a server, I am using it as a workstation or even a home desktop. Market segment is being crossed here.
Besides, what was exotic server hardware not long ago is fairly mundane desktop territory today. With the added benefit of higher single core clocks on top.
It is natural for server technology trickle down to the mainstream.
Newer things often has fancies like warranty and R&D cost attached to it and it is on the high point of the tub curve. Those used server grade stuff doesn't have those fancies attached to drive up their prices, and being a few years down it is well in the bottom of the tub curve too. You do miss out on newer technology like energy savings and new instruction sets, but for a budget-limited machine build or a machine you intended to use privately for the decade to come, used server hardware cna make sense.
As of the single core clocks, overclockable Xeons can be pushed sky high just like their desktop counterparts and their lower stock temperature and lower power draw than the desktop chips of the same architecture can even affords them more overclocking overhead. As of the instruction per clock per core efficiency, the benchmark says it have barely improved for the recent decade.
-
Assertion: with modern processors the key factors are:
- the size of the L1/2/3 caches and latency accessing it
- the bandwidth to main memory and latency accessing it
Discuss.
-
Well the Ryzen is 16MB versus the intel 8MB. The Ryzen is running 2933MHz RAM which I think about the fastest I see available. I don't know about cache access time.
-
Assertion: with modern processors the key factors are:
- the size of the L1/2/3 caches and latency accessing it
- the bandwidth to main memory and latency accessing it
Discuss.
Most generic assessments fall apart on there being distinct architectures.
-
- the size of the L1/2/3 caches and latency accessing it
Both brands have settled in a 2MB L3 per core formula. Given the fact that few software can run the same code across multiple cores (this is something optimizing compilers often use SIMD on, which is faster than multi-core processing) the total cache size makes less sense than cache per core.
- the bandwidth to main memory and latency accessing it
This is simply tied to the memory speed specs. And for an overclock-capable motherboard and processor (e.g. Ryzen on any X470, or E5-1650v2 on X79) those are manually tunable. Unless you are looking at non-OC chips (most of them Intel) there is little point looking into this spec on the CPU.
Well the Ryzen is 16MB versus the intel 8MB. The Ryzen is running 2933MHz RAM which I think about the fastest I see available. I don't know about cache access time.
Ryzen and Intel both have that 2MB per core ratio here. And with overclocking being so prevalent now that RAM speed spec holds little weight. Especially memory overclocking which is doable even on Intel's multiplier-locked chips.
-
...
As of now the bang for the buck looks like something like this to me:
* Ryzen 5 2nd gen
* 16GB DDR4-2666
* B350 motherboard
* RX 570 or GTX 1060 6GB graphics card
* 256GB NVMe SSD for boot drive
* 2TB spinning rust as data drive.
Since applications are getting bigger and bigger, having 16GB RAM will help a lot. At least here on my machine even 16GB barely makes it with ~30 tabs in Google Chrome and I went to 32GB RAM a yamear ago.
...
Ryzen 7 2700x
Cheapest B450 mobo ($77)
DDR4 3200C16 8G*2
512GB NVMe (M8PeG, not recommended anymore, go with PM981 instead)
GTX1050Ti
SeaSonic 80plus Gold 550W
Cheapest chassis
Total: $1000
Or:
Ryzen 5 2600.
Corsair Vengeance (2x8GB, 3000MHz, CL15).
B450 Mainboard.
And now, depending on how much you want to spend:
GTX 1050 or GTX 1060.
256GB or 512GB (NVMe)
etc..
-
Or:
Ryzen 5 2600.
Corsair Vengeance (2x8GB, 3000MHz, CL15).
B450 Mainboard.
And now, depending on how much you want to spend:
GTX 1050 or GTX 1060.
256GB or 512GB (NVMe)
etc..
The R7 was chosen because it eliminated the need for a discrete video card.
-
And just to throw in a curveball, there's an Intel CPU out there that uses AMD IP for the GPU. I guess that's Intel admitting their own GPUs are not very good?
-
And just to throw in a curveball, there's an Intel CPU out there that uses AMD IP for the GPU. I guess that's Intel admitting their own GPUs are not very good?
1) Intel never admitted that their GPU is any good anyway. Except that I’ll-fated Larabee which seems to be having a comeback as of late.
2) That chip still contains the Intel on-die GPU, and the display can be switched between the Intel and AMD graphics cards as if the AMD one is a separate chip.
3) It is not AMD GPU IP, it takes a whole separate chip. Those mobile chips are already CPU + PCH MCM to begin with, now this is CPU + PCH + GPU 3-way MCM.
-
Or:
Ryzen 5 2600.
Corsair Vengeance (2x8GB, 3000MHz, CL15).
B450 Mainboard.
And now, depending on how much you want to spend:
GTX 1050 or GTX 1060.
256GB or 512GB (NVMe)
etc..
The R7 was chosen because it eliminated the need for a discrete video card.
Mine does not have a GPU? I really don't get this putting GPU's in CPU's. it totally kills the performance, that is why laptops are always slow, as it is they tend to use a slower RAM and then you make that RAM work for the graphics as well. I prefer to keep them seperate unless an integrated GPU will be working on the same job as the CPU as another co-processor.
-
Mine does not have a GPU? I really don't get this putting GPU's in CPU's. it totally kills the performance, that is why laptops are always slow, as it is they tend to use a slower RAM and then you make that RAM work for the graphics as well. I prefer to keep them seperate unless an integrated GPU will be working on the same job as the CPU as another co-processor.
My mistake, I thought I read this in the thread. The benefit is that you don't need a discrete card, which means considerable power and financial savings, less noise and potentially a smaller case or motherboard. If you don't run graphics intensive tasks, it's a very attractive option. Just plug a monitor into the motherboard and be done with it. Modern IGPs are quite potent too.
-
R3s and R5s have them but they get ditched in the R7. I don't do anything intensive but a 4K monitor plus another HD can be an issue although Intel's HD520 in my laptop copes.
-
R3s and R5s have them but they get ditched in the R7. I don't do anything intensive but a 4K monitor plus another HD can be an issue although Intel's HD520 in my laptop copes.
I believe you're correct that the only Ryzen 7 CPUs that currently have GPUs in them are the mobile 2700U, Pro 2700U and 2800H.
I'm not sure exactly which desktop 2700 series version you ordered but none of the 2700E, 2700, Pro 2700, 2700X or Pro 2700X have GPUs in them. I assumed you ordered the just plain "2700" model. Regardless, it won't have a GPU in it. :)
-
I believe you're correct that the only Ryzen 7 CPUs that currently have GPUs in them are the mobile 2700U, Pro 2700U and 2800H.
I'm not sure exactly which desktop 2700 series version you ordered but none of the 2700E, 2700, Pro 2700, 2700X or Pro 2700X have GPUs in them. I assumed you ordered the just plain "2700" model. Regardless, it won't have a GPU in it. :)
Given the power and thermal envelope there is next to no way for AMD or Intel to put both a powerful CPU and a powerful GPU in one chip package that makes sense for a mainstream desktop processor socket without it letting the blue smoke out. I am excluding high end server sockets which do have the thermal and power capacity but no way for the graphics signals to get out, or laptop chips which are large for the sake of it really being a SIP and has purpose built thermal solution on them.
The only Ryzen with integrated GPU are the lower end ones, and they still has a very inferior GPU in it. Intel never had a decent GPU to start with.
AMD actually learned something from the system builders preferring E3-1230's over i7's for high end mainstream desktop: E3-1230's are always a close approximation of the top tier mainstream desktop i7 from the same generation but lacked the integrated GPU, and decided not to even bother with built in graphics on high end desktop chips since nobody would bother even using it to begin with.
-
I am currently in the process of specifying a system to replace my current AMD Phenom 940 workstation which runs Windows and Intel is not competitive on price because I want ECC just like on all of my workstations going back 10+ years which all still work. That leave my decision between:
Ryzen 5 2400G Zen 4(8) Cores 3.6(3.9)GHz 65W $170 Includes Graphics
Ryzen 7 2700 Zen+ 8(16) Cores 3.2(4.1)GHz 65W $290 +PCIe Graphics Card
And a likely Asus motherboard costing about $100 although there may be some other options. It is not always clear which motherboards support ECC and which do not but the Asus ones do.
I will probably go with the 2700 because I can make use of the extra cores despite the increased cost of the processor and having to buy a graphics card also. The Ryzen 7 2700X is marginally faster for only $40 more but also draws almost twice the power.
The comparable Intel Xeon processors all cost as much or more than the Ryzen 7 2700 but have half the performance or less, draw more power, and if they have graphics it is terrible. And their motherboards cost at least twice as much because Intel ties ECC support to the south bridge; you have to pay extra for it. I suppose that is an improvement though. When I built my Phenom 940 system, the Intel premium was even greater because all of their ECC systems required FB-DIMMs; just the cost of the memory for an Intel processor was as much as the CPU, motherboard, RAM, and GPU that I bought.
-
Search eBay. There are a lot of cheap Xeon E5 chips (illegal, OEM ones, never supposed to be resold as part of purchase agreement, not listed in ARK) from China and Japan.
I got my E5-2696v4 (OEM version of E5-2699v4) for a bit short of half the price.
Don't get confused by the word OEM. They are not ES/QS. ES/QS are samples which are not fully qualified, while OEM chips are fully qualified and production-ready.
That is an idea but then I still have to put up with a more expensive motherboard, lower performance, and higher power.
Earlier in this project the E3-1246v3 and E3-1220v5 were on my short list of candidates but I took so long that AMD's Zen became available.
-
Intel's older generation higher end consumer boards (X79, X99) support ECC out of the box, with no hacking needed. A $100 used X99 is all you need.
I considered something similar when I built my Phenom 940 but at the time the latest consumer boards from Intel which supported ECC were out of production and obsolete.
It is difficult to justify a used and out of production or soon to be obsolete system when a new one with better performance is of comparable cost.
FB-DIMMs are cheaper than regular DIMMs due to the large quantity from used market, and the incompatibility with regular mobos.
They are not but they sure were not then. I have an Intel OR840 board which takes RAMBUS and supports ECC but I am sure not going to refurbish that no matter how cheap used memory is for it now.
-
... I want ECC just like on all of my workstations going back 10+ years which all still work. That leave my decision between:
Ryzen 5 2400G Zen 4(8) Cores 3.6(3.9)GHz 65W $170 Includes Graphics
Ryzen 7 2700 Zen+ 8(16) Cores 3.2(4.1)GHz 65W $290 +PCIe Graphics Card
And a likely Asus motherboard costing about $100 although there may be some other options. It is not always clear which motherboards support ECC and which do not but the Asus ones do.
Asrock X470 board, officially supports ECC memory too, and ECC is only supported with PRO CPUs.
Edit : Official memory supported list -> https://www.asrock.com/mb/AMD/X470%20Taichi%20Ultimate/index.asp#MemoryPR (https://www.asrock.com/mb/AMD/X470%20Taichi%20Ultimate/index.asp#MemoryPR)
-
...ECC is only supported with PRO CPUs.
Uhhh... Are you sure about that? I thought it was only the Raven Ridge series (like 2400G) where it is only supported on the PROs. I thought Pinnacle and Summit both all supported ECC (though it may not be officially listed anywhere in AMD's specs.)
I haven't tried it myself personally and I don't have any unbuffered ECC DDR4 anywhere here to actually try it and then attempt to force an ECC error to watch it and see if it actually works and reports (I would expect 1 bit correction and 2 bit notification and interrupt) but there are reports on various forums of it working on various processors including a plain 2700, being detected as running in ECC mode by various OSes and supposedly actually correcting and reporting detected errors.
Supposedly the biggest issue is finding unbuffered reasonably high speed DDR4 ECC DIMMs at a reasonable price.
-
...ECC is only supported with PRO CPUs.
Uhhh... Are you sure about that? I thought it was only the Raven Ridge series (like 2400G) where it is only supported on the PROs. I thought Pinnacle and Summit both all supported ECC (though it may not be officially listed anywhere in AMD's specs.)
I haven't tried it myself personally and I don't have any unbuffered ECC DDR4 anywhere here to actually try it and try to force an ECC error to watch it and see if it actually works and reports (I would expect 1 bit correction and 2+ bit notification and interrupt) but there are reports on various forums of it working on various processors including a plain 2700, being detected as running in ECC mode by various OSes and supposedly actually correcting and reporting detected errors.
Supposedly the biggest issue is finding unbuffered reasonably high speed DDR4 ECC DIMMs at a reasonable price.
I don't use ECC for my desktop machine, but currently I use Asrock X370 board (X470 predecessor). quoted from the Asrock board spec web site.
-> https://www.asrock.com/mb/AMD/X470%20Taichi%20Ultimate/index.asp#Specification (https://www.asrock.com/mb/AMD/X470%20Taichi%20Ultimate/index.asp#Specification)
See attached below.
-
The ASRock B450 Pro4 specification page says:
- AMD Ryzen series CPUs (Pinnacle Ridge) support DDR4 3200+(OC) / 2933/2667/2400/2133 ECC & non-ECC, un-buffered memory*
- AMD Ryzen series CPUs (Summit Ridge) support DDR4 3200+(OC) / 2933(OC) / 2667/2400/2133 ECC & non-ECC, un-buffered memory*
- AMD Ryzen series CPUs (Raven Ridge) support DDR4 3200+(OC) / 2933/2667/2400/2133 non-ECC, un-buffered memory*
- Max. capacity of system memory: 64GB**
- 15μ Gold Contact in DIMM Slots
*For Ryzen Series CPUs (Raven Ridge), ECC is only supported with PRO CPUs.
... seems to indicate that Summit and Pinnacle support ECC and only talks about the Raven Ridge needing to be PRO series to do ECC, otherwise not supported.
That aligns with what I have read elsewhere.
-
I don't use ECC for my desktop machine, but currently I use Asrock X370 board (X470 predecessor). quoted from the Asrock board spec web site.
Right. That spec says the same as the 450... According to your graphic above, on Raven Ridge, it has to be a Pro. Pinnacle and Summit say support ECC.
-
Here is a report from someone using a Pinnacle Ridge, supposedly non-PRO, regular 2700 in ECC mode on an ASUS X470 Prime PRO:
BTW. I believe ECC is enabled and working.
Here is a little info of the CPU and os / kermel.
jmd1 ~/shell-scripts # uname -a
Linux jmd1.comcast.net 4.16.13-gentoo-20180603-1145-jmd1.comcast.net #3 SMP Sun Jun 3 11:52:55 EDT 2018 x86_64 AMD Ryzen 7 2700 Eight-Core Processor AuthenticAMD GNU/Linux
This tells me its enabled.
jmd1 ~/shell-scripts # dmesg | grep ECC
[ 8.557846] systemd[1]: systemd 238 running in system mode. (+PAM -AUDIT -SELINUX +IMA -APPARMOR +SMACK -SYSVINIT +UTMP -LIBCRYPTSETUP +GCRYPT -GNUTLS +ACL -XZ +LZ4 +SECCOMP +BLKID -ELFUTILS +KMOD -IDN2 -IDN +PCRE2 default-hierarchy=hybrid)
[ 9.132922] EDAC amd64: Node 0: DRAM ECC enabled.
This tells me there have been 0 errors (I expect that from server experience ECC errors should be rare)
jmd1 ~/shell-scripts # edac-util -v
mc0: 0 Uncorrected Errors with no DIMM info
mc0: 0 Corrected Errors with no DIMM info
mc0: csrow0: 0 Uncorrected Errors
mc0: csrow0: mc#0csrow#0channel#0: 0 Corrected Errors
mc0: csrow0: mc#0csrow#0channel#1: 0 Corrected Errors
edac-util: No errors to report.
This tells me the mode of error correction for the first rank
jmd1 ~/shell-scripts # cat /sys/devices/system/edac/mc/mc0/rank0/dimm_edac_mode
SECDED
Same goes for the 2nd rank
jmd1 ~/shell-scripts # cat /sys/devices/system/edac/mc/mc0/rank1/dimm_edac_mode
SECDED
Here is what SECDED means
EDAC_SECDED
Single bit error correction, Double detection
-
The Threadrippers are not *that* expensive.
Nope. Certainly a valid way of getting 4-channel performance.
The Intel 4-channel "desktop" chips are certainly not cheap! :)
Also, the "entry level" Intel CPUs that use LGA2066 only have dual-channel memory support.
If you need 4-channel on the cheap, it might be a better idea to use second hand Intel LGA2011 based systems. Those were decommissioned en masse a few years ago and cleap LGA2011 (not the -1 or -3 variety) ans matching DDR3 memory flooded the market.
THIS
I have a workstation built from two E5-2690 XEONs, which are 8 cores (16 threads) each, and have quad channel memory (each) at 1866 MHz, using dirt cheap registered ECC DDR3 (I paid $3/GB). The memory bandwidth per CPU is nearly 60 GB/s, and the CPUs have 20 MB of L3 cache each. Yes, the pure MHz speed is slower than the fastest DDR4, but more channels does make up the difference. The CPUs are 2.9 GHz nominal, but run at 3.3 GHz continuously with all cores loaded or up to 3.8 GHz one core loaded, which is right on par with modern typical high performance desktop CPUs. It is very impressive compute power for the money. Power draw of the system is rather high under full load, but well under 100 W at idle (including several HDDs). The power consumption per performance is the biggest drawback of these older CPUs, and that is the main reason that these otherwise fast and viable server CPUs are cheap and plentiful on the 2nd hand market. For a server farm under constant high load, the energy cost outweighs cost of more efficient newer hardware. For a home system which is lightly loaded most of the time, but needs the performance on occassion, these older CPUs are a fantastic bargain. You are unlikely to spend as much on extra energy vs. extra cost for newer more efficient hardware of equal performance.
-
Considering how long I "hope" this machine will last I will take the modern hardware with lower power consumption.
-
Considering how long I "hope" this machine will last I will take the modern hardware with lower power consumption.
Indeed. I'm seriously considering starting to replace more of my old hardware that is doing server duty with Ryzen setups just for the power savings. (Especially if I can find a good supply of some reasonably priced unbuffered ECC DDR4.) Even with our power cost usually down in the 3-6 cent per kWh my monthly power for the servers here is over $200 / mo, and if power spikes up to 10-12+ cents that's a significant chunk of change, even just for a month or two long spike in the rate. (I just pay the market rate and take my chances... :) Overall it is cheaper than power plan options in the long run with an essentially fixed load.)
The power consumption at idle of the last desktop Ryzen setup I tested it on (I think that was one of the 2400Gs with a single SSD and a DVD burner, using only internal GPU and a good PSU) was only a few watts. Not even worth trying to set the thing to use the "sleep" modes to save power. The fans stop spinning at idle anyway and the CPU is so power efficient at idle that it's just not necessary, they throttle themselves down to virtually nothing anyway.
I actually had problems with the first 2400G system I built since it ran FreeBSD. While you would never notice it if it were a Windows box since it would never get that idle, it would crash if you left it idle for long enough, like overnight at a prompt after recompiling the whole OS, it would just be frozen in the morning because the PSU didn't keep the power clean enough at no load. ASUS actually had to add an option in the BIOS to tell it to NOT go into the deepest powersave state to intentionally make it draw more juice. If they hadn't added that option (I had pre-ordered that first 2400G before they even went on sale) by the time I wanted to deliver it to a customer I would have had to leave a resistor rigged up in there just to burn a couple extra watts of power all the time.
What a change in efficiency compared to traditional setups!!
Simon, I think you'll be pleased with your setup and it will probably pay for itself anyway vs. a used server chip. Excellent for machines anyone wants to leave on all the time, for sure!
-
Considering how long I "hope" this machine will last I will take the modern hardware with lower power consumption.
The increase in processor power may not have been very impressive in recent years, but efficiency has definitely been improved quite a bit. Especially the powering down of unused subsystems has massively changed the past half decade or so.
-
The increase in processor power may not have been very impressive in recent years, but efficiency has definitely been improved quite a bit. Especially the powering down of unused subsystems has massively changed the past half decade or so.
For sure! These latest generation desktop chips have all the computing power you expect in a high performance CPU with the typical power consumption more traditionally associated with mobile-only platforms.