PC airflow, parts and performance

[uzbuzz]

Hi all, I hope I can get some advice from yourselves about what I might be doing wrong with my build and where I might get some improved performance. I guess it might be useful to give you all a summary of my build to get things started:

Mobo - Asus x570 tuf gaming +WiFi
CPU - 5950X (water cooled with Arctic Liquid Freezer ii 420 aio)
GPU - PNY 3090 XLR8 (water cooled with alpha cool Eiswolf 360 aio)
RAM - 64GB Corsair 3600mhz
Power supply - Corsair RM1000e
Case - phantecs G500A

Right, so I've had ongoing system instability issues since before I decided to watercool my GPU and even with my previous 3070 GPU I'd have frequent crashes in otherwise simple games, like fortnite. I've downloaded HWInfo and noticed that at its peak, my GPU memory junction temperature can reach up to 90°C but I'm unsure if that's the cause for the crashing. Also, I'd be interested to know if my card fan configuration is ok:

I've got 3 120mm intakes at the front
3 140mm (slim) exhausts at the top
1 140mm intake rear.

Is this enough positive air pressure?

Any advice would be amazing, and thank you for those of you that got this far, appreciate your time!

 

[uzbuzz]

Just seen your edited post.

Your case fan configuration is fine. The GPU memory junction is ok too.

Is your CPU and/or RAM overclocked? What are your CPU and GPU core/hotspot temps?

Yeah, typing on my mobile and accidentally hit post before finishing! I've not overclocked anything myself at all.

Idle CPU core: Average: 28°c - minimum: 26°c - maximum: 58.6°c
CPU hotspot: all hovering around 33-35°C when idle

Idle GPU core: Average: 27°C minimum: 26°C maximum 29°C.
Gpu hotspot: all hovering around 45°C when idle


At load (and for this I've run heaven benchmark)

GPU Core: around 55°C on average throughout the test
GPU hotspot(memory junction is actually the highest rating): reaches 100°C


CPU core: 64.1°C max but hovers around 42°C mostly
CPU hotspot: 64.1°C again, as above.

Also, my benchmark score on heaven seems to suck for the kit I've got: it managed to wrangle out 3310 on ultra settings, with extreme tessellation, 8xAA in windowed mode (2560x1440).
It's worth noting, I've had my fans maxed out to avoid any damage to components by potential overheating, so... Yeah. Any help with that info?

 

[uzbuzz]

An average of +10/20c hotspot over your core temp is normal. The fact it's almost double tells me that's why you're getting random crashes. If you've got it in a WB make sure it's making good contact with the core etc, and repaste etc the pads may be the cause of this if it's not making good contact

Yeah, that's what I suspected. With the memory chips on the 3090 on both sides of the card, are there any tell-tell signs which thermal pads aren't properly making contact with which memory modules? I'm unsure which set of dimms would need re-padding. Got a back plate from alphacool that was a bit generic and I'm concerned that it's not quite fitting the PNY card correctly; the instruction manual was very generic and not designed specifically for my card. For clarity, they definitely said it was compatible with my PNY GPU, but clearly the thickness for the memory thermal pads appears to have been incorrect, causing misalignment.

 

[uzbuzz]

Have you tried updating the motherboard bios to the latest version? There have been several updates to increase system stability and compatibility so it's worth doing if you haven't already done it.

Not in a while, so I'll try to have a look at their website when I get a chance. Armoury Crate is really... Buggy for me and I really want to minimise all the random bloat I'm having to install. I'm considering doing a fresh windows install just to ensure everything is up to scratch.

 

[uzbuzz]

Systems should not crash due overtemp in CPU or GPU because both will react to high temps by clock throttling.
Try to find out the symptoms of the crash. When it crashes check the time immediately and note it. Then examine the event log and look for entries preceding that time. They might be a clue. If its a system crash then run WinDbg on the memory dump. Have you already checked for WHEA?

Yeah, this is what I had thought too, but my GPU memory temps climb and climb, then I either get a crashed game or a system reboot. Another thing I've noticed, besides huge frame drops, is that when I'm playing online, it often takes me ages to enter a game. For instance, with fortnite, I'll still be loading into a game after everyone has jumped off, started playing, the bus has already dropped me off and I don't enter the game until my character is halfway down! This usually happens for the first game, every time I log in to play. Subsequent games I load in normally. Unlikely to be internet connection - 1gb fibre, Ethernet connection, no high bandwidth activity on the network and it's always the first game and not subsequent games.

I've attempted to look at windows logs after crashes and tried googling what the dump files mean, but had very little luck making sense of any of the information. Perhaps next time it happens, I'll copy and paste the text into this thread for a little more information.

Will look into WHEA as I've not heard of it before and attempt to address it if it's the root.

I've not run windbg on a dump file - any advice on how to do this?

 

[uzbuzz]

I take it as a symptom of case temperature climbing. Do you have system temp readings idle and after a longer GPU load?

My first suspect in thermal related crashing would be RAMm then maybe CPU. I.e. stable at low temp but starts crashing when case temperature is up.
If your crashes are easily repeatable, could try to eliminate these. Run couple cycles with ram downclocked to say 3200, timings and voltages kept the same.

As for GPU memory, well again, could try to downclock it. Can't improve cooling much as you already have rear 140 as intake.

Slow loading time points at slow storage, but you must be running NVME and there is no way to make any nvme throttle that hard on reads.

Yeah, you're right, NVME drive, not the best, but still, not anything near full capacity or any other reason to slow down load times.

Tempted to down clock GPU, but feel a bit saddened that I paid for the 3090 and the aio and I'm having to compromise on performance.

I have investigated ram as a possible culprit for the crashes and run multiple tests and all the sticks have come back fine. Also, regarding case temp... It's possible, but I think it's unlikely, as I've moved all my components into an open case (thermaltake core p3) and had the same temps climbing etc.

 

[uzbuzz]

I think I have a reasonable set of next steps. I will re-pad my GPU VRAM, just because that's the ONLY reading that goes high, even GPU hotspot on HWInfo reads lower than memory junction. Then I think just to play it safe, I'll do a fresh windows install (gah, I have so many files to back up!!!) Update drivers, and see if there's any improvement to stability.

Just wanted to say a quick thank you so much to everyone that's helped and is continuing to help troubleshoot. Will keep you posted with the process and any updates.

 

[uzbuzz]

Don't use it. Download the bios to a usb stick, reboot the pc and enter the bios and flash the latest bios using the in built EZ-Flash 3 tool. Nobody should be using windows based software to flash a bios. It would have been even easier if your board has usb bios flashback but Asus seemed to have made the ridiculous choice of only having that on their higher end boards.

Oh my goodness. Update my BIOS and decided to nosey around on case I messed about with any settings in the past. Enabled D.O.C.P to get memory working better (this went back to default because of the bios update) and then stumbled across "Above 4G decoding" and "resizeable bar" and realised I hadn't enabled either of these in the past. Reading through a few sites, it seems like I really should have had these two settings enabled a long time ago!

Going to run a benchmark to see if there's any noticeable difference in performance.

Also came across a blog from someone staying the thermal pads recommended by alphacool are a little thin for the memory units on reference cards. Will crack open my GPU to see if there's any indentation on the pads, and if not, replace with slightly thicker ones.

 

[uzbuzz]

Easier way to check if back memory is doing good contact is simply touching backplate. If hot, not just warm, is as it should be. 3090 main issue was rear memory not getting enough cooling. Reason I went with an active backplate when I had a 3090.
The 3090 is also infamous for voltage spikes. Sometimes even a good PSU won’t be able to cope with them, despite the output being, sometimes, way greater than recommended.
Some games will simply crash when RAM is set to run at certain frequencies. Battlefield V, in my rig, would crash. A great 8Pack 3200 (later replaced by the 3600) would crash, unless failing back to 3000MHz or less. Also some RAMs don’t set automatically the correct voltage. For the 8Pack kits, every ASUs board I’ve tried, I had to set manually the correct voltage. MSI was fine.
For the G500A, I don’t think there’s any chance the case is lagging behind on airflow. The case is very well ventilated, almost non-existent restriction from mesh panels.
The big gap between core temperature and hotspot is 99% chance mounting pressure/thermal paste/thermalpad application.
A previous 6800XT, Pulse, was showing same issue. But because I noticed that the screws holding the backplate and the heatsink from the backplate side were quite loose, I decided to tighten them myself. Hotspot issue was solved, but because the card would crash on every tested game of Ray Tracing was enabled (I know, not the card for Ray Tracing, but just to check if such feature was worth) the warranty was voided because o tightened the screws. Didn’t take the heatsink apart, but solving their incompetence in assembling a GPU, the warranty was voided.

Wow, that's a dreadful experience! Manufacturers really ought to be held to account when shoddy products get shipped like that.

I'm really embarrassed to admit that when I took a close look at my backplate... There were actually two screws missing!!! Super newbie mistake and can't believe I've been running my computer for so long with half the backplate loose. Found the screws (I have no idea how I left two screws unaccounted for when I put the thing together - actually I do, I generally only get time to work on my PC ridiculously late at night, so probably half asleep when I put it together!!) Tightened the backplate and saw memory temps drop from 100°C to 88°C under heavy load; max settings, ray tracing etc. (When I reached 100°C I didn't have my settings anywhere near that high)

But gameplay on fortnite was still choppy - still had a really long loading time for my first game. Wondering if there's something wrong with the NVME drive, or a software issue.

I wouldn't know where to begin with manually setting voltages for ram. Think I need some time practicing fitting all the correct screws into things before I level up to playing with voltages!

Absolutely love the G500A - a big change from my Core P3 (which was also really fun). I've actually managed to fit a 420 and 360 rad in the case by using slim fans on the 420 at the top of the case.

Regarding the voltage spikes, I heard as much, which is why I bumped up to a 1000W PSU from my 750, to give myself some headroom for transient spikes. I used to get lots of random shut downs with the old PSU, but none since the upgrade, barring a few blue screens, which need investigating...

 

[uzbuzz]

I didn't read your posts well enough. I see that your GPU spikes to 100C under load. HW throttling starts at 95C for the 3090, but the boost clock should have been reduced before this. Halving the GPU clock should bring the temp down but if not then it can reach the threshold at which the GPU will protect itself by shutting down. This should be 98C. That will indeed cause an application crash or even a system crash.

Thanks for this. Weirdly, my memory junction temperature has been higher than the hotspot temp. When I hover over it on HWInfo - the note mentions that since it's reading is from inside the silicone, temps tend to read higher and that thermal throttling is expected to start at 110°C. But still, it's my preference to get them as low as possible without limiting too much performance..

 

[uzbuzz]

What NVME drive have you got?

May need a firmware update.

Download a tool kit for your brand (Samsung Magician for example) and it should tell you if one is available/needed.

Does it have/need a heatsink fitted?

Currently running off of a crucial p2 - it's only Gen3 so I think I'm well overdue an upgrade to something a lot faster. Any recommendations? Price not too much of an issue, but the most I'll be doing on it is gaming, a bit of blender - not saving huge files - but also maybe some 4k video editing, so perhaps some quite big files...

Also considering purchasing some 2.5" sata drives for storage of old files, just cheap and cheerful with a good lifespan... Any advice on that would be great - I try to back my phone files up every so often and with the size of videos and photos I take, it can be quite a task doing backups.

 

[uzbuzz]

I have a similar setup for cooling.

5800X w/ Kraken 360 rad. 3080 w/ Eisenwolf 360.

The eisenwolf is a beast! Keeps the 3080 under 60*C under the heaviest loads and for "casual" gaming like World of war ships, with VSync enabled the fan profile doesn't even spin the fans up. Silent gaming.

The Kraken is a waste of time. An air cooler would probably do better. The issue is the transfer from the CPU to the cooler. The temp delta under benchmark load STARTs at like +50*C and gets worse. So if your water temp is 25*C your CPU will hit 75*C immediately under full load. It will then slowly soak rise to 85, 95 and throttle. The water temp is still only 32*C. I don't see the point in this. If the cooler can't get the heat away from the CPU it's a complete failure.

If you can stretch, I would recommend you do what I intend to do, to fix mine. Use the Eisenbar CPU AIO. Then link the two loops and top it up. A proper, copper cooler "should" improve things over a cheap AIO. (A kraken is not cheap!).

Hey man, if I hadn't already got the arctic, then I would have definitely gone for the Eiswolf for both CPU+GPU together. Honestly, I'm really happy with the performance of the arctic; not really had any CPU cooling issues so far. I suppose the only issue is from the thickness of the rad and how much difficulty I had during two triple fan radiators in one case, but that's my fault for not planning on advance and just adding more and more stuff into my case.

Are both your radiators triple fans? What case are you using?? Edit - just saw you said both of us are 360's!!! Nice!

 

[uzbuzz]

I would email Alphacool. They will know exactly what you mean and may be able to ship you custom pads.

Yes, those pads are critical. Yes they have to accomodate dual sided cards. Mine came for 3080/3090 with 2 sets of pads. The 3090 pads where (in my case Gigabyte Eagle) thinner, but more of them as it has DIMMS on the back side.

The thing is... there is no active cooling on the backside, it's just going to be a thermal pad to a flat sheet of metal. Again I'd ask Alphacool.

Thanks for this. Seems tightening the screws on the backplate made a big difference. Haven't had a crash in a few days now, and temps have fallen in the memory junction by +10°C so I'm happy with that result. Other sensors in the graphics card aren't showing any higher than ~65°C max.

 

[uzbuzz]

Trying to troubleshoot an old nvme drive.
It didn't show up in Windows. Well, in disk management, there's a blank volume that I assume is the NVME drive, but I can't do anything with it.
It didn't show up in bios boot manager, or in my list of nvme/sata drives. However, when rummaging through menus in the bios, there's one place it does show up in bios:

Advanced\NVME an controller and drive information\NVME controller and drive information self test

Here, I can see the model number, device id, vendor I'd, but it says total size: 0.0GB (it's a 500GB drive).

Anyone come across this issue and know of a way I can get this drive running again?

Many, many thanks for any help.

 

[uzbuzz]

Where is teh NVMe drive plugged in?

Some Mobo's have 2 or 3 sockets, but often have restrictions on when and what settings need to be on to use more than 1.

Try re-seating it. Check for BIOS settinngs for it's socket.

Beyond that I would need to go into Linux to find out what is actually wrong with it.

If I recall correctly, I tried using it in both nvme slots when the problem first occurred, so I think the fault may be with the drive, rather than the port, but I'm going to play around with it tonight to see...

I heard some asus motherboards have issues with nvme port 1 being used at the same time as SATA port 1, so if that's the case, I'll switch the sata to a different port too.

Just fine it strange that the drive is kind of being recognised in bios, but not everywhere in bios, just the self-scan section.