Author Topic: AMD Ryzen - New CPU Series that is cheaper but better then Intel Core I Series  (Read 22349 times)

0 Members and 1 Guest are viewing this topic.

Offline Terrius

  • Contributor
  • Posts: 49
  • Country: ca
    • Tinkering with Terrius
I run windows 10 with no issues.

Funny how people claim to run Win 10 without issues then explain they needed the pro/enterprise version and how and what they disabled and turned off and it is still a privacy leak and of course no one knows what the next update which most people can't refuse will bring.

There is a reason why almost twice as many people are still running Win 7 compared with Win 10 despite Win 10 being free and rammed down their throats by Microsoft. Sadly if enough sheeple accept the polished turds Microsoft produce that is what all of us are going to get, but, hey I'm glad you have no issues, apart from the ones you had, and have, and that updates will bring.

On Ryzen Win 7 support opinion is divided. AMD said they would and then said they wouldn't and yes some Ryzen hardware seems to list Win7 drivers.

I run Windows 10 home on a windows 7 era laptop with no issues as well (the laptop didn't even have windows 8 support). So you certainly don't "need" nor did I say you "need" the pro or enterprise version.  I simply stated I run windows 10 with no issues though I use Win 10 Pro and disabled the bloat and dodgy stuff.

My PC ran perfectly fine with the bloat and dodgy privacy stuff , but anyone who knows anything at all about IT security immediately disables all the key-loggers and other things baked into Win 10, or at least I would hope they do.

I find it particularly amusing when people freak out about windows privacy issues and then immediately go on facebook/social media on their phone... Never realizing or caring that all of the things they are scared about windows 10 are already baked into Android, IOS and the majority of social media apps.

Windows 10 has 25.6% of the market share now, so it has already done far better than windows 8 or 8.1 combined. While there are people who prefer windows 7 still, that market shrinks each month especially in the enthusiast and business markets.

It is important to note however that windows 10 falls in a period where most enterprise and business customers are looking to migrate to a newer version of Windows due to 7's extended support cutoff in 2020. Windows 8 and 8.1 both fell outside of this upgrade period so most enterprise/business customers ignored them.


I happily acknowledge that there are issues with windows 10, but it is far from the fiasco that was Windows 8. There seems to be some sort of rose coloured glasses issue going on where people remember Windows 7 as being perfect and having no privacy/security/usability issues... and yet at launch out the majority refused to upgrade from XP and many had the exact same issues that people have with windows 10. Fact is no matter how good an OS is at launch it will never be as stable and secure as the OS that has been patched for the last 5-10 years, change is always difficult.

Microsoft is actually fairly open about what their updates will bring, it is easy enough to find the patch notes if you actually care.

The biggest pet peeve I have, is the automatic install of all service packs and driver updates. In my opinion it is far from a "Polished Turd" but it all depends on how open to adaptation you are. The fact remains, this is the direction Windows is headed, and while it may not be ideal, it is certainly better than Windows 8 and 8.1 which both failed horrendously due to innumerable compatibility issues that still plague the OS.



As for the Ryzen support, it is listed on the official AMD drivers page as supporting Windows 7, so it is supported. Whether Motherboard and device manufacturers wish to use engineering time developing for Win 7 which is E.O.L.(End Of Life) and has already been dropped from Mainstream support and will be dropping extended support in 2020, well that is a completely different issue.




Anyways back on topic,

I am excited to see what Ryzen 5 and Ryzen 3 will bring to the table. If they can manage the same single threaded performance as Ryzen 7 but in cheaper more mainstream machines, it could become a decent rival to the I5s and I3s.

Offline Hensingler

  • Regular Contributor
  • *
  • Posts: 144
  • Country: gb
I find it particularly amusing when people freak out about windows privacy issues and then immediately go on facebook/social media on their phone... Never realizing or caring that all of the things they are scared about windows 10 are already baked into Android, IOS and the majority of social media apps.

I find it particularly amusing when people try to use the privacy invasion of google and facebook as an excuse for the new privacy invasions perpetrated by Microsoft.
 

Offline BradC

  • Super Contributor
  • ***
  • Posts: 1661
  • Country: au
I've noticed that intel has reacted to the release of the Ryzen by immediately lowering the cost of the competing cpu's to lower then amd's retail costs.

Can you point to any evidence of that? There have been a couple of very poorly researched clickbait articles that have said that was the case, but as yet they've proven to be hot air. If you have any credible evidence I'd appreciate a link.

Frankly I don't think Intel can *afford* to do an immediate price drop. A) that would draw immediate attention to the AMD products and announce loudly that they are a viable option to everyone who might not have noticed that yet, and B) that would also be admitting they have been taking the absolute piss for the last 5 years and reaming their clients with a pineapple. Nether are particularly a good look.
 

Offline Terrius

  • Contributor
  • Posts: 49
  • Country: ca
    • Tinkering with Terrius
I find it particularly amusing when people freak out about windows privacy issues and then immediately go on facebook/social media on their phone... Never realizing or caring that all of the things they are scared about windows 10 are already baked into Android, IOS and the majority of social media apps.

I find it particularly amusing when people try to use the privacy invasion of google and facebook as an excuse for the new privacy invasions perpetrated by Microsoft.


I wrote a fairly lengthy reply to that quote but it seems like a discussion for a separate topic, this thread was about the new Ryzen CPUs and whether they will put up a fight against Intel. So if you want to discuss that further I'd happily join another thread about it! It is quite an interesting topic to be honest. 



Back on point, so far the benchmarks are mixed, but mostly agree on the fact that in Multi threaded application these new chips are killers, and in single threaded they are respectable. It's been many years since AMD has hit even a slight parity in the CPU markets!  I am really looking forward to the R5 and R3 chips, their potential is quite high if they can maintain this performance but in a cheaper more mainstream package.



I've noticed that intel has reacted to the release of the Ryzen by immediately lowering the cost of the competing cpu's to lower then amd's retail costs.

Can you point to any evidence of that? There have been a couple of very poorly researched clickbait articles that have said that was the case, but as yet they've proven to be hot air. If you have any credible evidence I'd appreciate a link.

Frankly I don't think Intel can *afford* to do an immediate price drop. A) that would draw immediate attention to the AMD products and announce loudly that they are a viable option to everyone who might not have noticed that yet, and B) that would also be admitting they have been taking the absolute piss for the last 5 years and reaming their clients with a pineapple. Nether are particularly a good look.


I don't know about anywhere else but there were about $20-$40 price drops on the Intel chips here in Canada that started on March 1st. Uncharacteristic for this time of the year, normally those drops don't happen until closer to big sale days/months. Their socket 2011 chips haven't had any movement price wise though, just the 7000 series chips.

Offline Deridex

  • Regular Contributor
  • *
  • Posts: 162
  • Country: de
  • IMHO
Ryzen performs as i expected.
And it even seem to work well for a completly new cpu-architecture. Even if there are a few issues left, i gotta say: Well done AMD  :clap:
But i will wait some more months, till all fixes for the performance issues are done. I don't like being a beta tester.
 

Offline Howardlong

  • Super Contributor
  • ***
  • Posts: 4821
  • Country: gb
What seems clear now is that the overclocking potential is minimal unless you go to impractical exotic methods. Partially this may be to do with their use of an indium based thermal interface to the heat spreader, providing excellent thermal conductivity, and therefore being able to run the chips close to their limit even with air based coolers. Even the extreme de-lidders don't appear to be able to do much to better it.

The one question seems to be around the 1080p games performance anomaly, which isn't anywhere near as pronounced in 1440p or 4k when compared to Intel Broadwell-E. Anecdotally there also seems to be a marginally improvement by disabling SMT.

Irrespective, for creators and productivity, there is now a compelling alternative to Intel in the single socket market.
 

Offline FireFlower

  • Regular Contributor
  • *
  • Posts: 129
  • Country: fi
  • Wish had a cabin in Lappland
Poor OC potential is probably due architecture noise amount on signal paths. Better cooling (liquid) improves a bit heat noise and you get a little bit more overclocking but that's about it with conventional methods.  :horse:

Sub-temps the heat noise is not truly not an issue anymore and you can go much higher overclock.  Seems like same problem experienced with Phenom II series where temps were fine but you hit a wall around 4GHz and nothing you can do about it above room temps.

Basically AMD engineers are probably at this moment improving 14nm process and tweaking signal paths on architecture to go for higher clocks.


Game performance problems is pretty obvious when Intel easily has more than 75% of the market share on gaming platforms so you an easily think game developers optimizing their games for just Intel processors and ignored AMD's optimization guides? Dunno I am not game developer, just  my personal opinion why game performance has so much gap compared to similar clock speed Intel counter parts.
 

Offline aandrew

  • Regular Contributor
  • *
  • Posts: 235
  • Country: ca
Personally I still run 4 i7-920's overclocked which are circa 2008/9 because i've seen no need to upgrade as there has been no significant performance increases at the price level at around 199 retail at the time.

I'm in the same boat. I have a 1st-gen i7 (i7-950) that I bought new what must be 6 or 7 years ago now. When I bought it, I bought a good power supply, motherboard and maxxed out the RAM (24G) as well. I have had *no* need to update this machine, with the possible exception of power consumption. I have an ebayed Dell C6100 blade server whose Xeon L5520s run lower power than the 950.

I have to say, it's quite nice to be in this position after so many years of wanting newer hardware for the speed increases.
 

Offline Homer J Simpson

  • Super Contributor
  • ***
  • Posts: 1092
  • Country: us

OC3D TV

Asus Crosshair VI AM4 Ryzen Overclocking Motherboard Review

 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 10665
  • Country: us
  • DavidH
Game performance problems is pretty obvious when Intel easily has more than 75% of the market share on gaming platforms so you an easily think game developers optimizing their games for just Intel processors and ignored AMD's optimization guides? Dunno I am not game developer, just  my personal opinion why game performance has so much gap compared to similar clock speed Intel counter parts.

Intel disables the use of instruction set extensions in the libraries they provide if the processor does not identify as an Intel processor.  Last time I checked, they still did this on their compiler despite  the lawsuit over it but I do not know how many game companies use their compiler.
 

Offline brucehoult

  • Super Contributor
  • ***
  • Posts: 1402
  • Country: us
  • Formerly SiFive, Samsung R&D
Personally I still run 4 i7-920's overclocked which are circa 2008/9 because i've seen no need to upgrade as there has been no significant performance increases at the price level at around 199 retail at the time.

I'm in the same boat. I have a 1st-gen i7 (i7-950) that I bought new what must be 6 or 7 years ago now. When I bought it, I bought a good power supply, motherboard and maxxed out the RAM (24G) as well. I have had *no* need to update this machine, with the possible exception of power consumption. I have an ebayed Dell C6100 blade server whose Xeon L5520s run lower power than the 950.

I have to say, it's quite nice to be in this position after so many years of wanting newer hardware for the speed increases.

I had an i7 860, which is similar generation to your 950. I overclocked it quite a bit and got (stock, then overclocked):

2388/8365 https://browser.primatelabs.com/geekbench3/7509
2721/9389 https://browser.primatelabs.com/geekbench3/332502

Now I have an i7 6700K which I have left stock:

5200/20290 https://browser.primatelabs.com/geekbench3/5266295

Over twice faster was definitely worth spending US$1200 for new motherboard, CPU, RAM, M.2 SSD. In fact the SSD is 3x faster write, 5x faster read.
 

Offline Muxr

  • Super Contributor
  • ***
  • Posts: 1345
  • Country: us
I've been running my Ryzen build now for over a week. I must say I love it so far. 8 cores and 16 threads really make a difference when it comes to compiling stuff and running bunch of VMs. I don't really game much but I tried it and everything just feels very smooth. I upgraded from my i7 4770K Intel Haswell machine quad core. Ryzen feels much smoother overall.

The beast in its naked state:


Noctua 15D is just massive.. I swear I could probably just remove the fans and let my case's airflow do the job:


Lights up like a Christmas tree (all these gaming components now days have RGB LEDs it seems), maybe it's a bit tacky but I think it looks cool personally:



One thing I have yet to test is how much faster I can synthesize my FPGA projects now. Maybe I do some benchmarks compared to the old system.

edit: also it's been a while since I had bought a new case. These new cases man, they make cable management so much easier. I can just hide everything behind the motherboard, it's a piece of cake to work on these and make it all look nice.
« Last Edit: March 19, 2017, 12:43:12 am by Muxr »
 

Offline Fsck

  • Super Contributor
  • ***
  • Posts: 1157
  • Country: ca
  • sleep deprived
needs more ram. "better" is such a relative term.
I went from my overclocked 920 (24GB of ram) to a 5820k *2* years ago because I wanted more ram. 96GB of ram is really shiny to have.

kind of disappointed with AMD still, since my last AMD rigs are S939.
"This is a one line proof...if we start sufficiently far to the left."
 

Offline Muxr

  • Super Contributor
  • ***
  • Posts: 1345
  • Country: us
needs more ram. "better" is such a relative term.
I went from my overclocked 920 (24GB of ram) to a 5820k *2* years ago because I wanted more ram. 96GB of ram is really shiny to have.

kind of disappointed with AMD still, since my last AMD rigs are S939.
I have 32Gb in there (2x16Gb), and that's plenty for now. Apparently Ryzen also supports ECC RAM, so if I were going with that much RAM would probably invest in ECC. AM4 doesn't support more than 64Gb though. They are supposed to release Naples next. Which should support truck loads, but it's also going to cost an arm and a leg I would imagine. Although it would be nice to have a workstation with 32 cores and 64 threads, also 64 PCIe lanes lol.
 

Online wraper

  • Supporter
  • ****
  • Posts: 11040
  • Country: lv
I have 32Gb in there (2x16Gb), and that's plenty for now. Apparently Ryzen also supports ECC RAM, so if I were going with that much RAM would probably invest in ECC. AM4 doesn't support more than 64Gb though. They are supposed to release Naples next. Which should support truck loads, but it's also going to cost an arm and a leg I would imagine. Although it would be nice to have a workstation with 32 cores and 64 threads, also 64 PCIe lanes lol.
Nope, there are 128 PCI-E lanes. If two CPUs config is used, half of those lanes goes to interconnect between CPUs and you have 64 PCI-E left on each, still 128 total.
EDIT, each CPU also have 8 RAM channels.
« Last Edit: March 19, 2017, 01:38:57 am by wraper »
 

Offline Fsck

  • Super Contributor
  • ***
  • Posts: 1157
  • Country: ca
  • sleep deprived
needs more ram. "better" is such a relative term.
I went from my overclocked 920 (24GB of ram) to a 5820k *2* years ago because I wanted more ram. 96GB of ram is really shiny to have.

kind of disappointed with AMD still, since my last AMD rigs are S939.
I have 32Gb in there (2x16Gb), and that's plenty for now. Apparently Ryzen also supports ECC RAM, so if I were going with that much RAM would probably invest in ECC. AM4 doesn't support more than 64Gb though. They are supposed to release Naples next. Which should support truck loads, but it's also going to cost an arm and a leg I would imagine. Although it would be nice to have a workstation with 32 cores and 64 threads, also 64 PCIe lanes lol.

haswell-e consumer chips don't support ecc. only the xeons do and they don't overclock. I would've bought ecc if it was supported. Anyways, ECC DDR4 UDIMMs cost an arm and a leg. 180$ per 16GB stick here, and very few options.

a theoretical overclockable naples would be amusing. I still have an overclocked opteron 165 rig around.
"This is a one line proof...if we start sufficiently far to the left."
 

Offline blueskull

  • Supporter
  • ****
  • Posts: 13243
  • Country: cn
  • Power Electronics Guy
haswell-e consumer chips don't support ecc. only the xeons do and they don't overclock. I would've bought ecc if it was supported. Anyways, ECC DDR4 UDIMMs cost an arm and a leg. 180$ per 16GB stick here, and very few options.

Xeon E5 supports RDIMM, which is cheaper than UDIMM because there are used sticks available from decommissioned servers (yes, there are rich companies phasing out DDR4 2133 now!). The same is for ECC DDR3. It is just a matter of scale. Not too many people can find a use of UDIMM, therefore LRDIMM and RDIMM are cheaper, just because there are much more of them on second hand market.

Xeon E5 is also kinda overclock-able. E5-16xx v3/v4 are just server versions of i7x, and have unlocked frequency multipliers. E5-26xx v3/v4 have locked multipliers, but BCLK is overclock-able, you can get ~3% performance boost without affecting stability, though reliability may suffer if you run them 24/7.

A recent crack on Intel BIOS allows all Broadwell Xeon cores to stay at their maximum single core turbo boost frequency, which can also boost performance by a lot. However, this hack doesn't allow CPUs to breach TDP-wall, so don't expect any AVX/SSE performance increasing. Integer performance, on the other hand, can improve by a lot.
 

Offline Muxr

  • Super Contributor
  • ***
  • Posts: 1345
  • Country: us
I have 32Gb in there (2x16Gb), and that's plenty for now. Apparently Ryzen also supports ECC RAM, so if I were going with that much RAM would probably invest in ECC. AM4 doesn't support more than 64Gb though. They are supposed to release Naples next. Which should support truck loads, but it's also going to cost an arm and a leg I would imagine. Although it would be nice to have a workstation with 32 cores and 64 threads, also 64 PCIe lanes lol.
Nope, there are 128 PCI-E lanes. If two CPUs config is used, half of those lanes goes to interconnect between CPUs and you have 64 PCI-E left on each, still 128 total.
EDIT, each CPU also have 8 RAM channels.
Ahh, interesting, that's even crazier. heh

I do think those 32 core chips will probably be $2k+. There is no reason for AMD to undercut Intel as they perform better for their intended use. Better perf/watt which is king in the datacenter.
« Last Edit: March 19, 2017, 02:05:06 am by Muxr »
 

Offline Fsck

  • Super Contributor
  • ***
  • Posts: 1157
  • Country: ca
  • sleep deprived
haswell-e consumer chips don't support ecc. only the xeons do and they don't overclock. I would've bought ecc if it was supported. Anyways, ECC DDR4 UDIMMs cost an arm and a leg. 180$ per 16GB stick here, and very few options.

Xeon E5 supports RDIMM, which is cheaper than UDIMM because there are used sticks available from decommissioned servers (yes, there are rich companies phasing out DDR4 2133 now!). The same is for ECC DDR3. It is just a matter of scale. Not too many people can find a use of UDIMM, therefore LRDIMM and RDIMM are cheaper, just because there are much more of them on second hand market.

Xeon E5 is also kinda overclock-able. E5-16xx v3/v4 are just server versions of i7x, and have unlocked frequency multipliers. E5-26xx v3/v4 have locked multipliers, but BCLK is overclock-able, you can get ~3% performance boost without affecting stability, though reliability may suffer if you run them 24/7.

A recent crack on Intel BIOS allows all Broadwell Xeon cores to stay at their maximum single core turbo boost frequency, which can also boost performance by a lot. However, this hack doesn't allow CPUs to breach TDP-wall, so don't expect any AVX/SSE performance increasing. Integer performance, on the other hand, can improve by a lot.

haswell-e was a q3 '14 release, I purchased in q1 '15, and nobody was decomissioning <2 quarter old gear yet.

it will be interesting for me in a year or so to see if there will be ryzen servers in the <200$ server/desktop market which is currently occupied with old SB/IVB systems. the power efficiency improvements are nice but I doubt the 10W or so you'd save would be worth it to pay more.

it's still early (and I haven't checked very hard), but I haven't seen anything announced by AMD to fight it out with the xeon-d or C2000/C3000 markets, or old SB/IVB which is also used in this low-range market.

I know that companies which don't require high performance are still deploying SB-E, IVB-E, haswell, haswell-e and broadwell-e but there are also those caveman companies with S940 and LGA775 systems still in play on DDR1 and DDR2 respectively even though spending a few bucks for a new system would save them so much on power that would result in a lower TCO.
« Last Edit: March 19, 2017, 02:53:44 am by Fsck »
"This is a one line proof...if we start sufficiently far to the left."
 

Offline Howardlong

  • Super Contributor
  • ***
  • Posts: 4821
  • Country: gb
needs more ram. "better" is such a relative term.
I went from my overclocked 920 (24GB of ram) to a 5820k *2* years ago because I wanted more ram. 96GB of ram is really shiny to have.

Does that work on a 5820k? I thought they were limited to 64GB.

I recently upgraded to 32GB for my daily driver from a 16GB machine, but I also have 64GB and 192GB dual Xeon rendering and VM boxes, and I can't say I've ever really needed more than 24GB, but never say never. What do you do that uses all that RAM?
 

Offline blueskull

  • Supporter
  • ****
  • Posts: 13243
  • Country: cn
  • Power Electronics Guy
Does that work on a 5820k? I thought they were limited to 64GB.

As long as you can plug more without degrading signal integrity, it will take it. i7-5820K is identical to E5-1650v3 (same die design, different eFUSE bits), just with 0.1GHz higher boost frequency. E5 supports ECC, while i7 supports it, but BIOS code prevents it from being used. i7 will boot with ECC, but ECC feature will not be turned on. Both i7 and E5-16xx support overclocking if you use an X99.

I recently upgraded to 32GB for my daily driver from a 16GB machine, but I also have 64GB and 192GB dual Xeon rendering and VM boxes, and I can't say I've ever really needed more than 24GB, but never say never. What do you do that uses all that RAM?

I recently experienced a weird Yocto compiling issue caused by insufficient memory. It demands more than 10GB of RAM when 16 threads of make are running concurrently.
It is safe to say if I run a VM or something else, it can easily consume more than 24GB of RAM.
Also, numeric simulation eats a lot of memory. It's normal to see COMSOL or ANSYS to eat more than 50GB of RAM.
Finally, I occasionally have to run multiple VMs at a time. I'm not a fan of Linux container, so whenever I need a new node in my software test bench, I add a new VM. For the extreme case, I run 4 VMs at a time, 1 Ubuntu Desktop for my development environment, 2 Ubuntu Servers for testing and a Qemu inside the Ubuntu Desktop for ARM simulation. PITA exclusively for embedded developers :palm:.
My daily driver has 64GB of RAM, though for most of the time, I use no more than 8GB of them. However, had I have less, it would be really painful when I really need them, such as running ANSYS or a bunch of VMs.
 
The following users thanked this post: Iwanushka

Offline Muxr

  • Super Contributor
  • ***
  • Posts: 1345
  • Country: us
If you need to run bunch of VMs which require a lot of RAM, you can try giving Docker containers a try instead. Basically you can have multiple Docker containers run inside one big VM and instead of needing dedicated RAM for each VM they would share it (paravirtualization).
 
The following users thanked this post: blueskull

Offline blueskull

  • Supporter
  • ****
  • Posts: 13243
  • Country: cn
  • Power Electronics Guy
If you need to run bunch of VMs which require a lot of RAM, you can try giving Docker containers a try instead. Basically you can have multiple Docker containers run inside one big VM and instead of needing dedicated RAM for each VM they would share it (paravirtualization).

Yep, that is a cool idea, and I will find a time to play with it to virtualize some of my test platforms.
Some other test platforms, however, cannot be virtualized by a container, such as a driver test bed or anything that needs to directly talk to kernel.
The most important use of my test benches is to allow me to validate designs for ARM on X86, which mostly involves x11 and related drivers, ALSA driver modules and some character driver modules.
For apparent reason, no containers AFAIK allows me to insmod at this time, until Linux kernel itself has built in kernel level paravirtualization itself.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf