TR Forums

So here's the stats: i9 7900x all core OC @4.6, 2x 1gb NVME drives in RAID 0 for boot, 32gb RAM @4000, Titan Pascal (the one before the XP)

I frequently get out of RAM messages, or my screens just go black and I have to reboot if I'm doing too many things at once. By too many things I mean Photoshop, Illustrator, MS Office, a bunch of chrome tabs, slack and a few other small programs like a music player. I got 32gb of RAM specifically so I would NOT get those sorts of messages. On top of that every time the system goes to sleep, my windows are all rearranged, frequently ending up on my large wall mounted display screen that's usually off unless I'm presenting something or watching a movie.

It's just freaking annoying that I have to stop what I'm doing to reboot, recover from a crash, or just trying to find where the hell my windows are (this is especially annoying with the graphics programs where the windows and tools end up all over the place).

I'm designing for web, it's not like I'm even editing huge print images.

Anyone else having these kinds of aggravations with Windows 10?

How many displays are connected? I've had issues with multiple displays on Windows in the past, but not on W10. Kinda shocking about the RAM issue. I currently only have 16GB sitting in my motherboard and haven't hit that message and I do a lot of the same things on my home PC. Surely you've checked what is taking up so much space?

Have you disabled your pagefile?

Screwy window rearrangement behavior when there are multiple displays that aren't always connected seems to be par for the course on pretty much all desktop OSes these days.

When an overclocked machine is getting screwy, just run it at stock for a week, especially during troubleshooting. If I had a dollar for every time someone had said "but my overclock is stable" and it was found the be contributing to the problem, I'd be able to buy a very small helicopter.

It might not be your overclock, but there's no point trying random troubleshooting steps whilst overclocked because it's REALLY hard to prove the difference between 100% stable and 99.999% stable. That .001% could be your corrupt files or flipped bit or whatever....

1. Memtest x86, confirm it's stable overnight
2. Run sfc /scannow and then Windows updates. Stop the update service, take ownership of and the delete the c:\windows\softwaredistribution folder if you have any errors.
3. Perform a clean install of the GPU drivers.
4. Simultaneous Furmark and CPUBurner/Prime Torture to check the PSU isn't flaking out.
5. Check your RST drivers are up to date, write-cache is enabled and that the array isn't giving you any warnings.
6. Sacrifice a goat.

If at stock speeds it passes all those tests and behaves itself for a week, put your overclocks back and see if it stays good.
If you still have problems, then maybe you didn't use the right type of goat - at this point you probably need to start digging around in the event log for further clues. Could be anything from pagefile to intermittent display connectivity :|

DancinJack wrote:

How many displays are connected? I've had issues with multiple displays on Windows in the past, but not on W10. Kinda shocking about the RAM issue. I currently only have 16GB sitting in my motherboard and haven't hit that message and I do a lot of the same things on my home PC. Surely you've checked what is taking up so much space?

4 displays, one is a 46 inch tv. When I fire up the task manager it shows 52% memory in use (which seems like a lot), CHrome is taking 4.2gb, Photoshop 3.9, Illustrator about a gig, and the rest are smaller processes.

@JBI, I haven't disabled the page file. Sometimes it seems like the GPU driver has crashed rather than getting the RAM message and my window title bard look weird when the screens come back on. I don't remember having the multiple display problems in Win 7 or 8 (although it might have happened on 8 I don't really recall). I've been on 10 since it came out so lots of this stuff isn't new, I just felt the need to gripe after yet another crash.

Chrispy_ wrote:

When an overclocked machine is getting screwy, just run it at stock for a week, especially during troubleshooting. If I had a dollar for every time someone had said "but my overclock is stable" and it was found the be contributing to the problem, I'd be able to buy a very small helicopter.

It might not be your overclock, but there's no point trying random troubleshooting steps whilst overclocked because it's REALLY hard to prove the difference between 100% stable and 99.999% stable. That .001% could be your corrupt files or flipped bit or whatever....

1. Memtest x86, confirm it's stable overnight
2. Run sfc /scannow and then Windows updates. Stop the update service, take ownership of and the delete the c:\windows\softwaredistribution folder if you have any errors.
3. Perform a clean install of the GPU drivers.
4. Simultaneous Furmark and CPUBurner/Prime Torture to check the PSU isn't flaking out.
5. Check your RST drivers are up to date, write-cache is enabled and that the array isn't giving you any warnings.
6. Sacrifice a goat.

If at stock speeds it passes all those tests and behaves itself for a week, put your overclocks back and see if it stays good.
If you still have problems, then maybe you didn't use the right type of goat - at this point you probably need to start digging around in the event log for further clues. Could be anything from pagefile to intermittent display connectivity :|

Thanks for the advice, I'll give some of those things a shot. My PSU is a Seasonic 1200 Platinum so would be really surprised if it was doing dodgy stuff.

The RAM is overclocked too, I take it? Can you put it back into DDR4-2666W as well as dialling the CPU cores back to 3.3 GHz base and spec'd turbo clocks?

RAM usage went through the roof this year, I got 32 GB in summer 2016 thinking I was good for a few years, but turns out not so and I could do with going to 64 GB.

As for the windows rearranging, I get that too on my main machine with only 2 screens (plus the virtual one from the LOM) which Windows 'sees' as a third monitor.

Topinio wrote:

The RAM is overclocked too, I take it? Can you put it back into DDR4-2666W as well as dialling the CPU cores back to 3.3 GHz base and spec'd turbo clocks?

RAM usage went through the roof this year, I got 32 GB in summer 2016 thinking I was good for a few years, but turns out not so and I could do with going to 64 GB.

As for the windows rearranging, I get that too on my main machine with only 2 screens (plus the virtual one from the LOM) which Windows 'sees' as a third monitor.

It's actually under-clocked, it's rated for 4266, but my system didn't like the XMP profile.

It doesn't sound like you should be anywhere near out of memory, and that one is fairly easy to look into.

* Have you confirmed that the pagefile is set up correctly, even if you never touched it yourself?
* What's typical reported RAM usage? Does it rise with increasing uptime when it shouldn't?
* Are any programs obviously using more RAM than they should? If not, does the sum of them match up with total reported usage?

Topinio wrote:

RAM usage went through the roof this year, I got 32 GB in summer 2016 thinking I was good for a few years, but turns out not so and I could do with going to 64 GB.

I'm still fitting pretty comfortably in 32GB, but I haven't been using VMs as much lately (which was the reason I got 32GB in the first place).

synthtel2 wrote:

It doesn't sound like you should be anywhere near out of memory, and that one is fairly easy to look into.

* Have you confirmed that the pagefile is set up correctly, even if you never touched it yourself?
* What's typical reported RAM usage? Does it rise with increasing uptime when it shouldn't?
* Are any programs obviously using more RAM than they should? If not, does the sum of them match up with total reported usage?

I forgot I set the pagefile to 16gb for mining, think that's screwing something up?
RAM usage does seem to rise with increased uptime.

leor wrote:

I forgot I set the pagefile to 16gb for mining, think that's screwing something up?

Maybe... but with 32GB of RAM I am still surprised it is happening.

leor wrote:

RAM usage does seem to rise with increased uptime.

Can you tell which process(es) are using the additional RAM?

just brew it! wrote:

Can you tell which process(es) are using the additional RAM?

I can in a few days since I just crashed and rebooted

leor wrote:

I forgot I set the pagefile to 16gb for mining, think that's screwing something up?

Not impossible, but I doubt it.

leor wrote:

RAM usage does seem to rise with increased uptime.

leor wrote:

I can in a few days since I just crashed and rebooted

That definitely sounds like a memory leak.

If no particular process seems to be using the extra, it could be a driver. I keep seeing Killer ethernet drivers do that (fixable by using the vanilla Qualcomm drivers).

leor wrote:

It's actually under-clocked, it's rated for 4266, but my system didn't like the XMP profile.

Yeah, maybe (though those ratings are always IMO a little sketchy) but the iMC is rated for 2666.

Topinio wrote:

leor wrote:

It's actually under-clocked, it's rated for 4266, but my system didn't like the XMP profile.

Yeah, maybe (though those ratings are always IMO a little sketchy) but the iMC is rated for 2666.

I concur, IMC determines whether the RAM is overclocked or not, not the XMP profile. It's as clear as mud but the IMC's rated speed/voltage/timings are usually the lowest common denominator

There are still various reasons an IMC's rating might be where it is.

On the AMD side, at least one factor seems to be the SoC voltage AMD considers safe. SoC voltage behaves as you would expect while at or below the RAM speed spec, but goes straight to a high, out-of-spec value (1075mV on mine) when RAM speed goes out of spec. The VSoC AMD considers safe is 1V, which just about any OG Zeppelin die should be able to do 2666 at. 2933 could be a 100% rock-solid sure thing even with 4 sticks of 2R and so on, and AMD still wouldn't want to cover it under warranty. I assume that Zen+ can manage 2933 at 1V.

The apparent gap between the rating and reality is a lot bigger for Intel (a 9900K is still only rated for 2666), which makes me think they've got something similar going on.

I've been running Win 10 on 32GB since it came out and never ran out of memory that caused the display to go black, even when I dedicated 28GB to running VMs and I was thrashing the pagefile.

Sounds like either the system isn't stable or some piece of software or driver running is bad. Definitely follow what Chrispy suggested. If you can't find anything with a couple runs of Memtest then an overnight run of Prime95 blends would be in order, though I've found that it can take a long while to proc calculation errors with 32GB.

One other idea is to verify all 32GB is in fact usable. If you open up Resource Monitor, go to the Memory tab and look for the color bar and make sure the gray "Hardware Reserved" is <100MB. If the hardware is reserving gigabytes of RAM then that indicates pin damage in the socket, a bad startup Windows configuration setting, or a UEFI setting issue.

Hardware reserved is listed at 322 MB. 17gb is listed as standby with 682mb free. 14gb in use.

Rules that out then, that's good. My bet is still on Memtest or Prime95, I forget how long Memtest takes with 32GB so ya might as well leave it going overnight, it will track how many full cycles it has completed.

It's definitely not real usage driving this - it's either a driver leaking memory or something unstable. I do more with my 16 GB and never have issues (pagefile at 32 GB on the big SSD array).