32GB RAM -- beneficial or just money sinker?

[blueskull]

For the members who run VMs, is there a necessity to go with 32GB RAM rather than 16GB?

I'm talking about casual stuff, not heavy video editing or large FEA simulation.

Windows will be the host, running Altium, Edge (Chrome-based), VS2019 and maybe some FPGA tools, while VM will run Linux, mostly running test code or at most doing some compiling, nothing more than 8GB RAM consumption (which is the worst case, Yocto compiling, requires 12GB RAM or 8GB RAM+swap, tested with iMX6ULL BSP).

For such applications, should I consider 32GB upgrade or stick with 16GB if the price difference is $300 (soldered, not user upgradable)? I plan to keep this computer for at least 2 years, and I expect to score a good price when selling it, so it must still be usable (not essentially blazing fast) in 2022.

Thanks for any advices.

[james_s]

How many VMs? I mean I use VM's on my 8GB laptop, I used them on my 1GB PC years ago, normally I'm only running one or two machines at a time though. I have 16GB in my desktop and work laptop and that has been totally adequate for what I'm doing. Then again, $300 is not really a huge amount of money and if you can't upgrade it later it might make sense to just max it out now.

[james_s]

Sounds like today you'd be perfectly fine with 16GB. Whether that will be true in a year or 4 is anyone's guess.

I was about to upgrade my laptop from 8 to 16GB but I went to a SSD instead and that made such a big speed improvement that I don't really care anymore. Virtual memory is painless now, and supposedly SSD endurance for better ones is high enough that no normal person will wear one out.

[hamster_nz]

16GB fine, 8GB too little, 32GB overkill.

WSL is actually quite usable for a lot of things, so install it as well, if only for  experience. Be waned that AV tends to make WSL slower than native WIndows.

Oh, and don't go for small SSD + huge HDD. The price of midsized SSDs is pretty sharp now.

I would do for 500GB NVMe or M.2 SSD, and archive off old data and backups to an external USB HDD. When an NVMe can do 500+ MB/s at sub-millisecond latencies it makes an "extra RAM for disk caching" argument very weak.

I really like my new-to-me Intel NUC with an older i5, 16GB RAM and 500GB NVMe, even if on-CPU 3D is pretty average. So much smaller!

[maginnovision]

Sort of depends what kind of RAM you want/need for the VM. I don't think 32GB is needed if you stick to 4GB. If you go to 8GB and disable the windows swap file(like you should) then you can start seeing a reason to break 16GB. Regardless of an SSD making a swap file less annoying it still takes up time that it shouldn't but for my customers PC's I don't disable swap without 16GB. They do a lot of media processing so it does eat up a bit.

In the end I personally would go for the 32GB even though 300 is a little higher than I'd want to pay. You'll be running enough software on both the PC and VM that it may be necessary for peak performance. It also allows you to throw 16GB to the VM if you need it in the future. For my PC I have 32GB and a couple of TB NVMe drives running windows and ubuntu VM's(VirtualBox). Both windows and the VM run well with an average of 54% memory used. That's without running any memory intensive stuff.

[daqq]

Given your application range I'd also say that 16GB is fine. 32GB is only needed very rarely and if you need it you know it. I have 32GB on my machine, but the only time I actually used them was for a larger FDTD simulation. They were actually purpose bought for that reason :-)

[ebclr]

I have 64 GB, And still, need more, 

[I wanted a rude username]

As primarily an IT consultant, I feel I should give a contrary opinion.

Buying computers with insufficient RAM is one of the most common mistakes I see in the industry. And it happens again and again. Makes it easy for me to resolve performance issues by telling clients to buy more, but you can't do that with most sub-14" laptops.

When you see n GB of RAM used, that doesn't take into account the further amount the OS uses for cache ... or the amount already swapped because the OS wants to maintain at least some cache. SSDs are quick but the latency is still orders of magnitude worse than RAM. If you are seeing in the order of 16 GB used now, and you want to run VMs, and you want to use your computer for another three years ... and the RAM can't be upgraded ... do yourself a favour and spend that $300.

[wraper]

If you go to 8GB and disable the windows swap file(like you should) then you can start seeing a reason to break 16GB.

Like you shouldn't  . It won't do any good but possibility of causing problems is quite real. Those who are concerned about SSD wear are just ill informed.

[madires]

300 bucks for +16GB? Even with windows as host 16GB are sufficient to run a VM. 32GB would make sense for a small dedicated VM host system. And caching is a complex topic, it isn't "the more the better".

[olkipukki]

For such applications, should I consider 32GB upgrade or stick with 16GB if the price difference is $300 (soldered, not user upgradable)? I plan to keep this computer for at least 2 years, and I expect to score a good price when selling it, so it must still be usable (not essentially blazing fast) in 2022.

This is a game changer.
If a resale value is your primary objective vs a bit of compromise and discomfort in case when 16GB will be not enough (such as close an app to "free" memory, "loss" of time to make things faster etc. ) lead to a simple question - can you return your $300 (directly or indirectly)?

As example, 6 years ago I have maximised MacBook Pro 2013 Retina with 16GB (max) and 512GB SSD (max) when standard was 4GB/8GB and 128/256.
Today, I still use it and can run Altium in VMware in case of emergency 

[LeonR]

32GB makes sense if you are working with 4+ VMs simultaneously or doing nested VMs (clusters, etc) labs. I for one don't like being constrained to setting my lab VMs to 1-2GB since performance would go to a crawl on some occasions.

As other suggested, also invest on a large SSD (500GB+) especially if you are into virtualization.

[ebclr]

Sooner we all will have terabytes  of RAM, 16 GB will be a joke

https://www.intel.com/content/www/us/en/architecture-and-technology/optane-dc-persistent-memory.html

[BravoV]

For such applications, should I consider 32GB upgrade or stick with 16GB if the price difference is $300 (soldered, not user upgradable)?

What computer is that ? Apple ?

[LeonR]

Optane DC isn't a DRAM replacement but rather a ultra-low latency storage that pushes the boundary of 2nd tier memory.

https://www.storagereview.com/node/7416

It isn't even a new tech. Sandisk and DiabloTeK (or something like that) have done it some years before with the ULLtraDIMM:

https://www.tweaktown.com/reviews/6815/sandisk-ulltradimm-ddr3-400gb-ssd-enterprise-review/index.html

Since LGA2011 dual socket made its debut it's feasible to get a 1TB+ of RAM on a single system. But it's a bit of a moot point since desktop-class systems standard usages are more CPU than RAM oriented.

[Mechatrommer]

i've been living with 3GB for many years fine (not a moment virtual HDD/SSD memory need to be allocated edit: ok its paging, but i cant feel it thats why the previous (striked) comment). but then otoh i'm going to fill all empty slots of my newly acquired used HP Z800 to 72GB all because i can think a way to deplete all those and the 12 cores. so for me, "it depends" on what applications you run on it. but then i'm not at the liberty to talk about this since i dont VM. but i suppose VM doesnt use much RAM and i suppose the real answer lies on the task manager's memory usage report.

[Jeroen3]

For the members who run VMs, is there a necessity to go with 32GB RAM rather than 16GB?

If you want to. Now is probably the last time DDR3 is affordable. (see price history here)
If you can upgrade by adding modules, because your CPU allows extending life a couple of years, go for it.
If you have to replace modules, meh.

A new workstation with VM purposes definitely can benefit from 32 GB.
Recently build an Ryzen 3600 pc with 32 GB as replacement for my old E3-1230 with 16 GB. Now I can run two windows VM's at once  No more swapping.

[tszaboo]

I have 16GB in my main desktop. I had 24 when I was compiling Yocto, plus a 6 core xeon. I think it could've used more. My next machine for sure will have 32GB, because RAM is cheap. 80 EUR or so for the extra 16 GB. I even wrote pythn computation stuff that could've used more. Had games which went above 16GB. And I'm quite sure that chrome with a single webpage open will use 16GB memory in a few years.

With VMs, for sure. Although a VM can be simple 512MB machine that barely runs. Or something that runs OpenCV, processing multiple simultaneous video streams.

The other thing I'm getting rid of in my home desktop is the spinning rust.

[Bud]

I run Altium routingly at 12 Gb RAM usage. I have 32Gb so that is about right considering i might fire 2-3 VMs sometime in addition to that so the total usage comes to 20-ish Gb. I also use part of the RAM for a RAM drive and i point the Temp folders and scratch folders to that RAM drive to speed things up. I d say consider using the extra RAM for a RAM drive, you may find that helping with performance.

[mariush]

When you're doing that compiling that uses lots of memory, does it also do a lot of disk access?
If that's the case, it may be worth getting more ram in order to give you the ability to create a temporary drive and copy the files into the ram drive before compiling.

[coppice]

I have 64 GB, And still, need more, 

I HAD 64GB, and I rarely used more than 20GB rest of rendering videos, which I now have another desktop to do the task.

How often did you use between 16 and 20GB? If it was frequently, then having only 16GB might be rather annoying. If it was rare, then you might just live with some swapping.

[Jeroen3]

If you compile a lot using -j (multithreaded) then more ram is very useful.

[senso]

Using a couple VM's(with Win7/10) on a base Win 10, all with updates on, 16GB wasn't cutting it, and the laptop had a free slot, so I went with an extra 8GB(because they where on sale, got 2666Mhz 8Gb stick for 30€), but regret it a bit, having a nice round 32GB would be better, but with 24GB I can run 3 VM's with 6GB each dedicated RAM and feel no slowdowns due to lack of RAM, only due to running the VM's still in an HDD.. 

Waiting for some decent prices on 1TB NVMe drive to upgrade the crappy 240GB SN 520 WD that the laptop came with.

[james_s]

If you go to 8GB and disable the windows swap file(like you should) then you can start seeing a reason to break 16GB.

Like you shouldn't  . It won't do any good but possibility of causing problems is quite real. Those who are concerned about SSD wear are just ill informed.

I had mine disabled for a while with 8GB and it was mostly fine but occasionally after being up for a few months it would start running out of memory. I reenabled the swapfile and noticed no negative effects whatsoever and it never gets screwy from running out of memory.

In a nutshell I agree, no reason whatsoever to disable swap.

[rstofer]

I just did a quick price check of 32GB memory cards at Amazon and, without considering features, the price was around $139.  That isn't a number worth thinking about.  If a 16GB card is half that then the difference is only about $65.  Why even think about it?

That's pretty much how I wound up with 32GB of 4133 MHz DDR4. 

[james_s]

When the RAM is soldered to the motherboard as it is in most small laptops now, the price of memory modules is irrelevant. Also the largest single modules are normally much more expensive than the same amount of memory apread axross smaller modules.

[nctnico]

For the members who run VMs, is there a necessity to go with 32GB RAM rather than 16GB?

It depends on the number of VMs you run and what you run inside them. For some of the tasks I run the VM needs at least 12GB so having 32GB or more is necessary if you want to run another memory hungry application. And I do recommend to turn of swap files especially for VMs because the translation layer makes disk access incredibly slow already.

[maginnovision]

If you go to 8GB and disable the windows swap file(like you should) then you can start seeing a reason to break 16GB.

Like you shouldn't  . It won't do any good but possibility of causing problems is quite real. Those who are concerned about SSD wear are just ill informed.

What problems will it cause? Is there a number of years it takes so I can prepare for a callback? My customers have computers from as far back as 2011 that are still in use.

[nctnico]

If you go to 8GB and disable the windows swap file(like you should) then you can start seeing a reason to break 16GB.

Like you shouldn't  . It won't do any good but possibility of causing problems is quite real. Those who are concerned about SSD wear are just ill informed.

What problems will it cause? Is there a number of years it takes so I can prepare for a callback? My customers have computers from as far back as 2011 that are still in use.

I've been running with the swap disabled on all my machines for years without any problems. Especially on Windows this gives a good performance boost because Windows likes to stuff things in swap even though there is enough memory. Also swap is so much slower that a desktop machine will slow down to a crawl when a signficant amount of memory is in the swap file. On a modern PC there really isn't any reason to have a swap file especially since RAM is pretty cheap anyway. Windows should disable swap by default on PCs with 8GB or more.

[wraper]

Also swap is so much slower that a desktop machine will slow down to a crawl when a signficant amount of memory is in the swap file. On a modern PC there really isn't any reason to have a swap file especially since RAM is pretty cheap anyway. Windows should disable swap by default on PCs with 8GB or more.

It will crawl only if all of RAM is used, and swap is on slow mechanical drive. If swap was disabled in such situation, you will have applications crashing as minimum. As of during normal operation, use of swap file does not degrade performance, even when used on HDD, not SSD.
http://www.tweakhound.com/2011/10/10/the-windows-7-pagefile-and-running-without-one/

[nctnico]

Also swap is so much slower that a desktop machine will slow down to a crawl when a signficant amount of memory is in the swap file. On a modern PC there really isn't any reason to have a swap file especially since RAM is pretty cheap anyway. Windows should disable swap by default on PCs with 8GB or more.

It will crawl only if all of RAM is used, and swap is on slow mechanical drive. If swap was disabled in such situation, you will have applications crashing as minimum.

No and no. Windows will always push stuff to the swap because it tries to keep as much ram unused in case a new application is started. Leave an application unused for a while and then switch to it -with swap enabled- and you'll see the application won't pop-up immediately. In case of low memory you'll get a warning saying memory runs low. It is not like everything will crash and burn.

[wraper]

Also swap is so much slower that a desktop machine will slow down to a crawl when a signficant amount of memory is in the swap file. On a modern PC there really isn't any reason to have a swap file especially since RAM is pretty cheap anyway. Windows should disable swap by default on PCs with 8GB or more.

It will crawl only if all of RAM is used, and swap is on slow mechanical drive. If swap was disabled in such situation, you will have applications crashing as minimum.

No and no. Windows will always push stuff to the swap because it tries to keep as much ram unused in case a new application is started. Leave an application unused for a while and then switch to it -with swap enabled- and you'll see the application won't pop-up immediately.

BS, windows actually preloads stuff it thinks might be needed into RAM. It uses more RAM than is actually needed by running apps to improve performance. Claim that it tries to keep RAM free is nonsense. https://en.wikipedia.org/wiki/Prefetcher

In case of low memory you'll get a warning saying memory runs low. It is not like everything will crash and burn.

You'll get warning and soon after that some app will be killed by windows to free up RAM as RAM usage increases. I've done some experiments in the past.

[maginnovision]

Also swap is so much slower that a desktop machine will slow down to a crawl when a signficant amount of memory is in the swap file. On a modern PC there really isn't any reason to have a swap file especially since RAM is pretty cheap anyway. Windows should disable swap by default on PCs with 8GB or more.

It will crawl only if all of RAM is used, and swap is on slow mechanical drive. If swap was disabled in such situation, you will have applications crashing as minimum. As of during normal operation, use of swap file does not degrade performance, even when used on HDD, not SSD.
http://www.tweakhound.com/2011/10/10/the-windows-7-pagefile-and-running-without-one/

Windows has always been sloppy with the swap file. Just as a test I enabled the swap file and restarted this PC. It had 3GB allocated and 2.5GB IN USE. that's not a good thing. I have plenty of RAM to keep everything there where I get instantaneous access to everything. Let's not even pretend I'm trying to compile some version of Linux or render any videos on an HDD. I literally just booted the PC. You think I want windows to use this same logic when I'm trying to run one or more VM's to do other work? Absolutely not. If you want to keep your swap file go for it but blueskull should absolutely disable the swap file for doing what he plans. Anyone running a VM should disable the swap file, and have an appropriate amount of RAM to run however many PC's they plan on running. It's that easy.

By the way all your data is old. Especially with a modern SSD you don't need any sort of prefetching because it's already available and you're wasting drive space and time. If you don't have the memory to run without a swap file you shouldn't run without one, that's obvious.

[james_s]

That's what was happening to me. I'd suddenly have software start to fail in strange ways, the whole machine would feel sluggish and Windows would pop up a warning that it was out of memory and things were going to be closed. It's just one of those annoying situations that is completely alleviated by leaving swap alone. It's virtually transparent and with a decent SSD it should not cause any perceptible performance degradation.

If you have a lot of RAM you are less likely to have problems running out but if swap is enabled your butt is covered in case it does happen. If having it enabled actually felt slower then I'd feel differently but I have not found that to be the case. It would be interesting to see some actual data though.

[SiliconWizard]

32GB is certainly beneficial for some heavy uses.

The problem here is the laptop. Unless its RAM can be expanded with SODIMM modules, getting one with 32GB is currently pretty expensive.

That said, if I need 32GB RAM, I'll currently consider that I need a proper workstation computer, and not a laptop. Just my opinion.

[legacy]

I have 64 GB, And still, need more, 

I have 2Gbyte, I cannot have more, but I am fine.
What do you have to do with 64GB of ram?

(edit: kidding, but ... my router has 64Mbyte of ram, and it was able to recompile cmake, and then erlang, both locally in something about three weeks no-stop moving things on a swap over NFS

real men can do it )

[SiliconWizard]

I have 64 GB, And still, need more, 

I have 2Gbyte, I cannot have more, but I am fine.
What do you have to do with 64GB of ram?

What are you doing with 2GB? 640KB ought to be enough!!

[wraper]

By the way all your data is old. Especially with a modern SSD you don't need any sort of prefetching because it's already available and you're wasting drive space and time.

Dunno what you mean by that and it's not obsolete. Windows preloads stuff into unused RAM, it's a fact. https://www.techcrises.com/windows-10/superfetch/

By default, SuperFetch uses all free memory of the machine when pre-loading apps. But as soon as the system will need more memory, the SuperFetch will free up the needed RAM.

[legacy]

What are you doing with 2GB? 640KB ought to be enough!!

IBM/Lenovo X61S is my daily laptop. Xilinx ISE v10.1 is more than perfect for spartan 3E, which is my preferred FPGA. It eats up to 0.9Gbyte of ram when it synthesizes stuff for my Arise-v2 core, while  Mabaxterm eats something like 50Mbyte, and Codewarriors + Avocet each something like 200Mbyte, and for the worst scenario, well, the rest is used by Chrome and Openoffice on Windows XP.

These two are heavy elephants but not as big and fat as Eclipse inside a VM (Yes, Windriver only offers this way to develop for VxWorks  ).  After all, the laptop only has two slots, and you can only install 1Gbyte per slot, therefore 2Gbyte is the max you can do, but I have never experimented a single problem with this setup.

OK, a lot of stuff runs remotely on other machines, since mabaXterm is just a ssh + X11 interface 

p.s.
for Windriver, RDP saves my laptop. The big heavy VM runs remotely and I only have to import the session.

[Mr. Scram]

If this thread isn't a can of worms. 

[nctnico]

What are you doing with 2GB? 640KB ought to be enough!!

IBM/Lenovo X61S is my daily laptop. Xilinx ISE v10.1 is more than perfect for spartan 3E, which is my preferred FPGA. It eats up to 0.9Gbyte of ram when it synthesizes stuff for my Arise-v2 core, while  Mabaxterm eats something like 50Mbyte, and Codewarriors + Avocet each something like 200Mbyte, and for the worst scenario, well, the rest is used by Chrome and Openoffice on Windows XP.

for Windriver, RDP saves my laptop. The big heavy VM runs remotely and I only have to import the session.

Wait until you are going to route a larger Virtex FPGA or compile the UnrealEngine. The first needs a machine with 8GB the latter needs 16GB.

[james_s]

I've compiled for a Virtex4 (or 5?) on a 2GB laptop. Wasn't fast but it did work. Sold that board though when I realized none of my projects were big enough to benefit from it.

[KaneTW]

I run 128GB and it's just about fine. 16GB is stifling, 32GB is low but acceptable.

[legacy]

Wait until you are going to route a larger Virtex FPGA or compile the UnrealEngine. The first needs a machine with 8GB the latter needs 16GB.

Yeah, and the last smartphone OnePlus 6T McLaren 2019 Edition comes with 10 GB RAM, and it's just a smartphone, or what is it supposed to be? a spaceship?

do you need 10Gbyte of ram on a smartphone? NO! do you need more than 4Gbyte in a laptop? usually NO, but it depends ... why do you need Virtex FPGA? And for doing what?

I feel seriously embarrassed, *all* my business runs fine with spartan3/500, and my phone only has 8Mbyte of NVRAM (used mostly to store contacts).


edit:
oh, not to mention the *brilliant idea* to force Linux's applications (like "Fireflop") to have more than 512Mbyte of stack, just because WE have a lot of ram 

[KaneTW]

As a software engineer not involved with any of the latest web bloatware, 16GB on a laptop is under the bare minimum for a new device. It works, but it's not comfortable.

[wraper]

As a software engineer not involved with any of the latest web bloatware, 16GB on a laptop is under the bare minimum for a new device. It works, but it's not comfortable.

16 GB is more than enough for ~99% of people. Anything more than that is beneficial only for special use cases.

[james_s]

That seems bizarre to me. My daily driver laptop has 8GB and it's fine, my work laptop has 16GB and it is spacious. I don't think any of the developers there have more than 16GB, a few new hires might have 32 by now. More than that is far outside of the norm for a workstation, and would suggest bloat or mismanagement of resources to an extreme degree.

[legacy]

I've compiled for a Virtex4 (or 5?) on a 2GB laptop.

We have a sort of mainframe for doing this. I say "mainframe" but it's rather a big PeeeeCeeee with a lot of RAM. It's in the computer room, anyway, with a lot of 10Gbit/sec network, then locally downconverted into 1Gbit, in order to support dummy terminals, which are laptops, used as RDP "terminals".

WmWare comes with a lot of interesting support for this, and Windows is OK with this.

Big stuff runs only there, on the big-iron. Anyway, it comes with 64Gbyte of ram, but it has to serve sixteen desks (including mine) via "hypervisoring" resources.

64/16 means you get 4Gbyte per desk, and the IT guys think it is enough for each desk.

OK, two desks can be stacked into one "special" request on-demand, but hey? it's no more than 8Gbyte as extra.

[Mechatrommer]

Wait until...

wait until you run 10x VM, 10x surreal engine, 10x sims etc etc all work together. no, 32GB 24 cores never been enough... there's windows management, there's human management, and then there's a person's reality... as shown so far... it depends... ymmv.

[Halcyon]

For the members who run VMs, is there a necessity to go with 32GB RAM rather than 16GB?

It depends what you're doing and how much memory you're using/require.

I personally find 16GB borderline (for running 4 VMs) and sometimes it begins over-provisioning to the SSD drive. I'm adding another 16GB next week.

[brucehoult]

I have 64 GB, And still, need more, 

I have 128 GB, but then I have 32 cores, and often keep them all busy. A previous machine with 18 cores and 64 GB often went into swapping, which was awful.

But the OP has much different needs.

I have a NUC with an i7-8650U and 32 GB RAM which I use for general use (lots of web browser windows) and light development. It is always fast, never goes into swap.

I also have a ThinkPad X1 Carbon with the same i7-8650U and 16 GB RAM. As I type this it is 1 GB into swap and it reasonably frequently has unexplained pauses when I for example start to resize a window.

Both are running Ubuntu 18.04.

16 GB is enough for general use in 2019 most of the time, but if the motherboard will take it I'd definitely go to 32 GB as the cost is not large.

[nctnico]

I've compiled for a Virtex4 (or 5?) on a 2GB laptop. Wasn't fast but it did work. Sold that board though when I realized none of my projects were big enough to benefit from it.

Well... the case is ofcourse with a design which is big enough to make the large FPGA make sense 

[NiHaoMike]

I have 48GB, about 23GB in use right now not including disk cache. A large amount of that is running VMs.

With a quad channel X79 machine, adding 32GB (4x8GB) doesn't cost much more than adding 16GB (4x4GB), so I got 32GB to add to the existing 16GB. That was back in 2016 when there was a sale on DDR3 and earnhoney was very well worth mining. Between the 3 earnhoney mining VMs, other VMs, and regular desktop usage, RAM usage peaked at almost 40GB.

[legacy]

I guess, this topic is actually talking about personal EGO 

[beanflying]

More like some can 'get away' with less than others. I stuck it out with a clunker on 8Gb for longer than I should have. Freezing working on CAD from time to time and generally being a slug on complex files. Couldn't work with video above 1080P in spite of having 4K capable cameras. I couldn't run a browser with a few tabs opened and run CAD which was just a PITA. 8GB is all but dead.

16 on my current box is 95% of the time a nice level of overkill and the 32 I fitted to it is likely a 5 year future proofing of it. And when I play with 4K video it really is worth it NOW.

[Howardlong]

For general day to day development and unit testing, without VMs, 16GB I still find to be fine. 8GB is very marginal IME.

However I do also sometimes have to debug and system test resilient distributed systems, which can mean I need eight or ten VMs. Certainly it’s not unusual to need half a dozen, in which case 32GB is really the absolute minimum nowadays, something you can even do well on a 13” laptop nowadays. 64GB for more but I’ve never found a need for more than 64GB: you can do that on a modern high end 15” laptop.

I do also have a 192GB 1U dual xeon 24 core monster that I use remotely if I’m in the field because it has IPMI so I have complete control over it, and can simulate pretty much any production environment, although I’ve never needed anywhere near the 192GB it has.

[NiHaoMike]

I do also have a 192GB 1U dual xeon 24 core monster that I use remotely if I’m in the field because it has IPMI so I have complete control over it, and can simulate pretty much any production environment, although I’ve never needed anywhere near the 192GB it has.

Make good use of tmpfs.

[KaneTW]

I've not found tmpfs to be a significant increase in performance over NVMe SSD+caching.

[NiHaoMike]

I've not found tmpfs to be a significant increase in performance over NVMe SSD+caching.

But unlimited endurance, so best for temporary data that's constantly changing.

[wraper]

I've not found tmpfs to be a significant increase in performance over NVMe SSD+caching.

But unlimited endurance, so best for temporary data that's constantly changing.

Unless you write like half of total size of SSD every day, endurance is a non issue.

[coppice]

I've not found tmpfs to be a significant increase in performance over NVMe SSD+caching.

But unlimited endurance, so best for temporary data that's constantly changing.

Unless you write like half of total size of SSD every day, endurance is a non issue.

Most SSDs contain a huge proportion of long term files, so what they are able to rewrite every day is a small free area of the disc, and they can wear that out quite quickly. I don't know how well Tesla manages its flash memory, but there are now reports of early Model S cars failing because they write such a huge amount to the log in that memory they have worn some of it out.

[wraper]

Most SSDs contain a huge proportion of long term files, so what they are able to rewrite every day is a small free area of the disc, and they can wear that out quite quickly. I don't know how well Tesla manages its flash memory, but there are now reports of early Model S cars failing because they write such a huge amount to the log in that memory they have worn some of it out.

SSDs have overprovisioned memory and they shuffle data to avoid what you wrote from happening. If there are old files, it does not mean they stay written in one place. Your point is ill informed.

[mariush]

Most SSDs contain a huge proportion of long term files, so what they are able to rewrite every day is a small free area of the disc, and they can wear that out quite quickly. I don't know how well Tesla manages its flash memory, but there are now reports of early Model S cars failing because they write such a huge amount to the log in that memory they have worn some of it out.

Most SSD controllers will shuffle the data around during idle times, so that memory cells will wear out evenly.
Basically, those long term files are not fixed in one location. If the erase counter of some pages (where a page is usually 512KB or higher of data)  goes too high, those pages will no longer be erased and data from other pages with lower erase count value is moved somewhere else to get those pages in use.

Also, keep in mind SSDs reserve a lot of space for endurance, ex a 500 GB TLC drive may have somewhere around 520-550 GiB of actual memory... those 20-50 GiB are spare.

Some SSDs also use some part of the SSD in SLC mode, as write cache... that also helps reduce the wear and tear.

[nctnico]

Most SSDs contain a huge proportion of long term files, so what they are able to rewrite every day is a small free area of the disc, and they can wear that out quite quickly. I don't know how well Tesla manages its flash memory, but there are now reports of early Model S cars failing because they write such a huge amount to the log in that memory they have worn some of it out.

SSDs have overprovisioned memory and they shuffle data to avoid what you wrote from happening. If there are old files, it does not mean they stay written in one place. Your point is ill informed.

What you forget here is that the data retention time depends on the number of times a sector is written. So if you leave an SSD on 24/7 it will seemingly last forever. However, if you turn it off the clock starts ticking and a heavily used SSD may have a much shorter data retention time then you'd like. Under certain conditions the retention time can deteriorate down to several months.

[Mr. Scram]

What you forget here is that the data retention time depends on the number of times a sector is written. So if you leave an SSD on 24/7 it will seemingly last forever. However, if you turn it off the clock starts ticking and a heavily used SSD may have a much shorter data retention time then you'd like. Under certain conditions the retention time can deteriorate down to several months.

Few people will leave their Tesla unused for months and it has a beautiful inbuilt UPS so that doesn't appear to be an issue.

[james_s]

Commodity SSDs have been around long enough that we should be seeing some meaningful data on their reliability and longevity by now. I have not worn one out yet, even a very cheap $20 SSD I put in an older laptop as much as an experiment as anything else. I don't think I know anyone offhand who worn one out, anyone here have direct experience with a failed SSD? Obviously they do occasionally fail, but I'm reasonably convinced at this point that the decent modern ones are holding up at least as well as spinning drives for typical use.

Goes without saying that no matter what sort of drive you're using it's important to keep backups of your data. Drives fail, computers get lost/stolen/flooded/burned in fires, etc, the hardware can be replaced, not always the case for your data.

[mariush]

What you forget here is that the data retention time depends on the number of times a sector is written. So if you leave an SSD on 24/7 it will seemingly last forever. However, if you turn it off the clock starts ticking and a heavily used SSD may have a much shorter data retention time then you'd like. Under certain conditions the retention time can deteriorate down to several months.

See JEDEC SSD Specifications Explained (Alvin Cox, Seagate) : https://www.jedec.org/sites/default/files/Alvin_Cox%20%5bCompatibility%20Mode%5d_0.pdf

ex... ambient temp 25c (y axis), active temp 40c (x axis) : 105 weeks retention time or 26 years.
Even if you store the SSD in a box in a hot room (say 40c) you're looking at ~50 weeks.

[Mechatrommer]

Commodity SSDs have been around long enough that we should be seeing some meaningful data on their reliability and longevity by now. I have not worn one out yet, even a very cheap $20 SSD I put in an older laptop as much as an experiment as anything else. I don't think I know anyone offhand who worn one out, anyone here have direct experience with a failed SSD? Obviously they do occasionally fail, but I'm reasonably convinced at this point that the decent modern ones are holding up at least as well as spinning drives for typical use.

my earliest SSD is intel brand, not the cheapest but not the most expensive, iirc around $100 or less 120GB. it experienced (almost) daily usage for 6 years now in my main computer, Windows XP (no TRIM), 3GB RAM that i just found out occasionally got swapped to the SSD, not 4096 bytes aligned, because i was too ignorance to care about the importance until recently. several time got reformatted or complete OS/SWs restore etc. its still working on the other computer after i changed my main PC SSD to 250GB. that intel SSD imho on par or surpasses conventional HDD that i usually encountered. fwiw.

[olkipukki]

I don't think I know anyone offhand who worn one out, anyone here have direct experience with a failed SSD? Obviously they do occasionally fail, but I'm reasonably convinced at this point that the decent modern ones are holding up at least as well as spinning drives for typical use.

A few months back, the primary SSD (OCZ Vertex 4 256GB) has 'passed away' after almost 7 years in service since I put these into a new Mac Mini 2011 Server. There are two drives in the system, but I didn't check 2nd wear factor yet, just replaced with Samsungs.

[wraper]

I had one drive which failed. However certainly not because of wear as I don't think entire drive size (1TB) was written on it even once. That drive actually let the data to be read but slowly. But writes failed. SSD fail not due to wear. But due to electrical faults or lock due to data/firmware corruption which may happen due to firmware bugs or power loss at unlucky moment.

[iMo]

Better SSDs achieve 2-9PB writes..

[wraper]

OCZ Vertex 4 256GB) has 'passed away' after almost 7 years in service

When OCZ was still alive they stated that 9 out of 10 dead RMA drives actually were physically fine, only firmware got in panic state and locked the drive. IIRC especially true fro Sandforce drives like Vertex 2 and few other series. People often actually manage to unbrick those. https://buger.dread.cz/ocz-vertex2-unbrick.html

[nctnico]

What you forget here is that the data retention time depends on the number of times a sector is written. So if you leave an SSD on 24/7 it will seemingly last forever. However, if you turn it off the clock starts ticking and a heavily used SSD may have a much shorter data retention time then you'd like. Under certain conditions the retention time can deteriorate down to several months.

See JEDEC SSD Specifications Explained (Alvin Cox, Seagate) : https://www.jedec.org/sites/default/files/Alvin_Cox%20%5bCompatibility%20Mode%5d_0.pdf

ex... ambient temp 25c (y axis), active temp 40c (x axis) : 105 weeks retention time or 26 years.
Even if you store the SSD in a box in a hot room (say 40c) you're looking at ~50 weeks.

105 weeks= little over 2 years. Not 26!

[tkamiya]

The only way you can know that is by testing it in your particular use case.  Just open VM as you usually do and keep watching memory usage and swap activities. 

Windows do act weirdly though.  No matter how much RAM you have, you always see some swap activities.  I have a Dual Xeon box with 48GB of memory.  It still swaps, although memory usage at the time may be only 8GB. 

At one scenario, I had only 8GB on this box.  It was constantly swapping.  Added 40GB to it, and now, it hardly ever go beyond 10GB or so.  It often stays below 8GB.  Swap is on SSD by the way.

Memories are reasonably inexpensive.  I'd just add 32GB if you want that and be happy. 

Linux is more reasonable.  It won't swap until it has to.

[tkamiya]

Try setting swap size to ZERO on Windows box.  It will still swap, according to the performance monitor.  As to linux, of course, you have to size the swap area accordingly.  But it will do what you told it to do.

I work at a database company.  Swapping of any kind at any time while database is running is considered a sin. 

[tkamiya]

Last time I did that, Windows automatically created a temporary one.  Maybe things has changed.  I haven't messed with it for quite some time, as anything that sort of thing matters, I choose Linux or Unix.

[legacy]

, anyone?

[Mr. Scram]

Last time I did that, Windows automatically created a temporary one.  Maybe things has changed.  I haven't messed with it for quite some time, as anything that sort of thing matters, I choose Linux or Unix.

There shouldn't be a pagefile if you set it correctly. Not now and not 8 years ago.

[Black Phoenix]

, anyone?

Too expensive for the storage size compared with other kind of mediums, not really that fast, compared with PCIe4 NVME drives and with a low scope of uses where the performance difference will be noted (simulations, scientific research, mathematics calculations)

It's innovative yes, the endurance of the memory itself is great compared with the NAND memory, the fast access almost like RAM is great and the amount of IOPS but it doesn't justify the price for most of the common users or even most of the companies.

Data Centers, Scientific Research Departments/Schools/Universities, Medicine Companies, Oil Prospectors and probably Space related companies are where this technology will be really at their best. For the normal consumer the price must come down or the size up a lot.

[Mechatrommer]

the lovely part is "maybe.. if you must have the best of the best" i like the guy, he is giving "practical" advice instead of some pushers who only like to show appealing stats alas heavy sided. where did i saw someone with signature.. "sphere of influence... linustechtip"?   i tried to find reason if i need to add nvme to my pc (old) system? no i dont find any appealing reason benefit/cost wise, instead it will give me extra work, probably headache to deal with extra drive letter that has no apparent purpose other than as a swap drive. maybe.. if i buy a new motherboard that support booting from nvme, then i'll get nvme and kiss the SATA SSD goodbye.

[NiHaoMike]

if i buy a new motherboard that support booting from nvme, then i'll get nvme and kiss the SATA SSD goodbye.

You don't really need BIOS support for NVMe boot, you can just put the /boot partition on some other device like a USB drive.

[Mechatrommer]

btw, i disabled swap/pagefile on my 24GB RAM Win7x64 machine, mostly 4GB is used running some CAD and game, so swap drive efficiency is solved there. i think nctnico is right about this, Win7/XP still swaps memories (few MB) even on highly oversized RAM, but for my low 3GB RAM WinXP_x86 running 10s of apps at a time, i dont think disabling pagefile will be a good idea.

[mariush]

I can't give you guys links to documents, but I remember reading tech blogs written by Microsoft programmers that were saying it's best to leave the pagefile / swap enabled, even if set to something as low as 100 MB.
The reason for that was because inside the memory management code there are some "shortcuts" or tricks made to improve memory allocation and handling and those tricks or shortcuts work because the code knows if the attempt fails, there's always pagefile to fall back to and save the day, in those rare worst case scenarios .... with pagefile disabled, some of those algorithms default to "guaranteed always works" but they're much slower (use more cpu cycles to complete).

So basically, a program that allocates and destroys lots of memory even if it's small chunks of memory, may work a bit better (perform more operations per second, stuff like that), even if they never hit the page file, even if they only work with way less than the maximum memory installed in the system.

Maybe things have changes in the latest versions of Windows, and maybe with the modern processors that have 8 cores, 16 threads, and 4 Ghz+ frequencies, those extra cpu cycles don't matter that much anymore.

 Still, it doesn't affect me in any way to leave some page file...  Windows is often too eager to kill applications if it seems it's running out of memory.
For example, I had at some point pagefile locked to 256MB (fixed size) and 8 GB of RAM  ... I was playing a game and alt-tabbed to a browser to read a walkthrough for the game.
The page was full of animated GIFs showing the route through the map of the game, and animated avatars for characters and so on ... at that point Firefox still kept the uncompressed GIFs in memory, so the page went up to 5 GB of memory usage from all the GIFs on the page and that made Windows ask to kill the game to make more room for memory.
 

[RoGeorge]

Some years ago I came up to the conclusion that, when I build a new PC, it makes sense to buy as much memory as the board allows.  Now I have 32 GB of RAM, the maximum allowed by this motherboard, first bought 16 then another 16 in a month or so, just for the peace of mind.  I did that because it happened to me in the past to upgrade very late, when the type of memory my PC has was almost deprecated, so hard to find, and thus more expensive.

Virtual machines were one reason, and I can testify is better to have too much memory rather then lack it, then it is that I have a ZFS file system on some disks, and ZFS could take advantage of the extra RAM.

My advice, if you can afford the cost, buy as much memory as permitted by motherboard 's chipset.  If not, buy as much as your budget allows.

For my daily usage on a workstation, it doesn't matter much if it's CL3 or CL2, so I won't spend much for high speed memories, but it does make a difference if it's 16 vs 32 GB.

[legacy]

Data Centers, Scientific Research Departments/Schools/Universities, Medicine Companies, Oil Prospectors and probably Space related companies are where this technology will be really at their best. For the normal consumer the price must come down or the size up a lot.

Precisely the reason why I asked: anyone from those areas is using it with profit?

[madires]

Some years ago I came up to the conclusion that, when I build a new PC, it makes sense to buy as much memory as the board allows. 

The last few decades taught us that you should buy what you need and not what you might need in a few years, since hardware prices will drop or you'll get more bang for the buck. If you're the family's repair cafe old RAM modules and so on will pile up anyway, in case you want a RAM upgrade for an old PC.

[nctnico]

Some years ago I came up to the conclusion that, when I build a new PC, it makes sense to buy as much memory as the board allows.

I usually leave slots open on a new PC. Later on I upgrade.