GPU issues/suggestions for possible replacement?

[Malcolm Payne]

I have an Nvidia Quadro P4000 GPU which has been running perfectly satisfactorily and without any apparent issues since it was installed in my previous Windows 10 system in 2017.

I recently built a new Windows 11 PC with an i9-13900K CPU and 64GB RAM on an Asus Z790 Proart Creator board and transferred the P4000 to that machine. Everything else runs perfectly on the new machine, but since then I have been plagued with random GPU crashes on file export in both C1 and PTGui, although no other software appears to be affected. Both applications may run without issue for several days, or may crash either immediately or part way through a batch export (C1) with various OpenCL and nvlddmkm errors. Following the initial crash, and even if one application or the other was previously running without any problems, both then continue to crash every time until the system is restarted, which appears temporarily to reset the card, driver or whatever is causing the crash. Both C1 and PTGui also run perfectly when exporting via either the CPU or the iGPU. As far as I can tell, the Topaz AI applications, which use the GPU heavily but run under OpenGL, not OpenCL, also all run without any problems under the P4000.

Neither C1 support nor PTGui have been able to reproduce the issue or to offer any further suggestions, and I have so far tried just about everything obvious apart from the more intrusive measures such as a complete teardown and rebuild of the machine to check for physical damage (bent LGA socket pins etc) or a full reinstall of Windows 11 from scratch.

The crashes appear to be directly connected with the GPU and/or the Nvidia driver, but it is not at all clear whether that might be due to a faulty GPU or motherboard, the driver, a GPU/motherboard/Windows 11/bios incompatibility, or anything else. I have already tried a full driver uninstall using DDU and a clean installation of various drivers going back several major versions, with no improvement.

In the event that no contributory factors can be located in the main system and I am forced as a last resort to replace the P4000, can anyone offer any suggestions for a suitable alternative? Use would be mainly C1, PTGui, Photoshop and the Topaz AI suite. I'm rather out of the loop with recent GPUs, and everything seems to be primarily gaming-oriented (and overpriced) these days. The simple answer would be another Quadro, but the nearest direct equivalent is some £980 and I don't want to go there. I'm also not confident that another Nvidia card might not have similar issues, and I know nothing about AMD.

Has anyone tried Intel Arc A750/770? It looks very interesting on paper as well as reasonably priced in comparison to Nvidia and AMD, and there have been some positive reviews about its performance with Photoshop and other creative apps, but I can find nothing about C1. There is also the question of it being a first generation product as well as the drivers very much still under development, and it is also unclear as to whether it supports 10-bit output. The latter is perhaps not an absolute deal-killer, but I do highly-accurate art repro professionally and it would be nice to have (and the original reason for the Quadro).

Any comments or suggestions for suitable GPUs would be gratefully received; and, if anyone has any ideas as to what might be causing the crashes in the first place, that would be a very welcome bonus!

Many thanks.

[Paul2660]

I switched a few years ago to Radeon. Using a 580 with 8Gb of ram. Win10. No issues. Not on
Win11 so not sure if that would a contributing factor.

Curious. In C1 have you turned off open CL and tried to see if that helps? 

Due to the current state of affairs in the world video cards are crazy expensive at at in the US. 2x now for same range of card that the 580 was 3 or so years ago.

Paul

[Malcolm Payne]

Thank you for that suggestion, Paul. Out of curiosity, why did you switch? The difficulty with researching suitable alternatives to the P4000 has been that almost every review is solely based around gaming benchmarks which are largely or wholly irrelevant to business use. It's very difficult to find anything online that even mentions creative applications and the like, and the current grossly inflated GPU prices are a nightmare.

Yes, if I turn off 'Hardware Acceleration (Use OpenCL)' in C1 it works fine. As it also does if I force the use of the iGPU instead of the Quadro. Enabling OpenCL for display only in C1 also seems to work normally; it's only when processing the output files that it fails.

Interestingly, running the Geekbench 6 OpenCL GPU benchmark when the Quadro is apparently working normally returns a benchmark figure of ca. 36000, which seems to be typical for this card. Following a crash, that figure drops to slightly over 10000 for the same benchmark, with the Face Recognition routine falling to zero.

As noted previously, the only other application affected seems to be PTGui, also OpenCL based. That would seem to indicate an OpenCL error specifically; but everything else, including the OpenCL-based benchmark in Affinity Photo 2, appears to run normally as far as I can tell.

The randomness of the failure makes it hugely difficult to diagnose a possible cause other than by swapping out components. A different GPU would at least rule out that entire part of the chain, and may in the end be the only option, but I don't have a spare one available and I'm reluctant to fork out several hundred UKP on a new one without at least a fair idea where the problem lies in case that doesn't fix it.

Anyway, thanks again for taking the trouble to respond.

Malcolm

[Paul2660]

I had to go back a ways to remember, but I ended up switching for a similar reason, I was having blue screens of death, and other issues with C1 21 with the Nvidia card I was using (I can't remember the type number, but it only had 4GB of ram).  I then started having issues with Photoshop CC not giving me the option to use open CL, CC saw I had a Video card and told me what it was, but the selection areas to pick to use open CL were greyed out.  Lightroom seemed to have no issues.

I also agree that trying to reach support from any Video card company is pretty worthless, as they only seem to be focused on gaming/Video creation. 

C1 support has dropped way down recently also IMO.  I still prefer C1 for raw conversion mainly due to their handling of layer, and ability to work with a color within a layer, something LR/Camraw still can't offer (at least I have never figured out a way, color selection only works globally). 

C1 and open CL are important when working with larger files, 45MP and up especially with multiple layers and using the masking tools. 

I switched to Radeon, as it Apple used it in my MacPro (I believe I have that correct it is a 2016 Mac Pro).  I figured it was going to be easier to purchase a different card than spend the hours with tech support from C1 and or Nvidia and get nowhere.

Paul

[Arlen]

You might take a look at Puget Systems computers and hardware recommendations. They are generally well regarded, and one of their specialties is building computers for image processing, Photoshop in particular. Besides their computers themselves, they offer a lot of information about the components and what they recommend. Whenever I build a new computer (I'm about due now), I always at least check the information there. It helps in putting together hardware that they have tested and know which components play well together.

 https://www.pugetsystems.com/solutions/photo-editing-workstations/adobe-photoshop/hardware-recommendations/

[Malcolm Payne]

@Paul: Ain’t that the truth. Nvidia’s community support forum appears to be focussed wholly on their GeForce gaming cards, and there seems to be no way as an end user to submit a bug report directly to their Developer forum nor any other way to get in touch at all. You would think they would want to know about potential issues with their products. It's doubly frustrating that a Google search on the specific error messages throws up a host of identical error reports on Nvidia cards stretching back a number of years, with no apparent resolution or feedback from the manufacturer in that time.

In fairness, C1 support did at least investigate the issue further before eventually closing the ticket, saying they were unable to commit further resources to continue working on such a random edge case but would keep the ticket on file. PTGui basically just said they were unable to reproduce the crash and gave up, apart than making a few generic comments and suggestions. I did request both companies at least to submit a bug report to Nvidia on my behalf, but there is no indication that either has done so.

There is most likely an Nvidia driver update due imminently so I'm inclined to see if that produces any resolution before I start tearing the machine apart again, although I'm not optimistic. Otherwise it may just come down to a change of GPU in the end and hoping that does the trick, though now is far from the ideal time to source a replacement.

Malcolm

[Malcolm Payne]

@Arlen: Good call, thank you. I hadn't thought of that. I'll have a look and see what they suggest; it might at least put me in the right ballpark if I do need to replace the GPU.

Malcolm

[Doug Peterson]

There are a lot of advantages to PC.

But this is an example of one of the weaknesses.

There are only a handful of Mac models at a given time, and a numerable number across a several-year-span, and the adoption rate of new MacOS versions is quite high quite quickly.

All this means a developer can thoroughly test their software on most recent Mac+OS combinations with only a small room's worth of machines and people. And a bug with any of those combinations is worth fixing since a meaningful total number of people have that specific combination.

But on PC there are SO many combinations of OS, Hardware, and Driver that it would take a large room's worth of machines and people to scratch the surface.  And the number of people with any given specific combination is quite small. It's plausible you are the only C1 user in the world with this specific OS, driver, GPU.

None of this is an excuse or a banal "switch to mac" conclusion – just an explanation.

[Paul2660]

Point well taken. One of main reasons I switched to Mac for C1 to reduce confusion and simplify when working with C1 support. Plus since the M1 MacBook Pro I don’t feel an intel machine currently can run C1 anywhere near as fast especially when working with 100mp or larger files and many layers combined with masks.

Paul

[Roscolo]

Suggestions, some of which you may have tried:

Swap out the GPU for another one and see if the problem replicates.

Check the power cable connections to the GPU and/or change the power cables.

Are you updated to latest BIOS?

Disable the built in Intel GPU.

[Malcolm Payne]

Update:

@Roscolo: Thank you for your suggestions. I had already tried all of those (and many more) other than the GPU swap, as I didn’t have anything available at the time that was compatible with the new PC.

This weekend I swapped out the Nvidia P4000 for an Intel Arc A770 and that seems to have sorted the problem, at the cost of an incompatibility with PTGui (see below), so it seems that the Quadro itself is most likely faulty.

I replaced the Quadro in its original Windows 10 machine, where its behaviour has since been a bit odd; it initially crashed immediately with C1, then settled down and apparently worked normally with C1 over a number of extended test runs that would have reliably crashed it every time in the new machine, but still failed intermittently with PTGui. Tried again this morning, and it is crashing again with C1. It had previously worked seemingly faultlessly in the Win 10 PC, so the failure appears either to have been entirely coincidental with its transfer to the new machine or a direct result thereof, though I can’t imagine what might have caused the latter.

For anyone interested, the choice of the Intel Arc GPU over a new Nvidia/AMD card was a calculated risk; I was wary of Nvidia after the Quadro problems, and the current Nvidia and to some extent AMD GPU rip-off pricing was also a disincentive. There had been good (albeit limited) reports about the Arc compute performance and potential in productivity software, which is all I need it for, and thus far it mostly seems to be performing well. The only issues I have found have been that it is not officially supported in PTGui and corrupts the stitch and output file, though oddly the UHD770 iGPU on the i9-13900K (also not officially supported) appears to stitch perfectly. And also that Topaz Gigapixel doesn’t seem to use the Arc when that is selected as the processor. Both issues are currently with Intel & Topaz for comment; at least both companies have accessible and responsive support departments, unlike Nvidia.

I note the earlier comments about the predictability of the Mac system; unfortunately the corollary of that is the lack of (and very expensive at the point of purchase) upgradability and repairability, particularly with the M-series machines, and also Apple's habit of breaking/obsoleting software with OS updates. Not to start a Mac/PC war, as I have had a MacBook Air M1 for the past year for site work, but I much prefer Windows overall, having used it in its various incarnations since the mid-80s. Not do I have much choice for a studio machine, as my print RIP (ColorGATE Productionserver) is PC only and I have far too much money and time invested in that even to think of changing.

I shall update further in due course with any feedback from Intel & Topaz in case anyone else might also be interested in the Arc GPU.

[Doug Peterson]

I note the earlier comments about the predictability of the Mac system; unfortunately the corollary of that is the lack of (and very expensive at the point of purchase) upgradability and repairability, particularly with the M-series machines, and also Apple's habit of breaking/obsoleting software with OS updates. Not to start a Mac/PC war, as I have had a MacBook Air M1 for the past year for site work, but I much prefer Windows overall, having used it in its various incarnations since the mid-80s. Not do I have much choice for a studio machine, as my print RIP (ColorGATE Productionserver) is PC only and I have far too much money and time invested in that even to think of changing.

Completely fair!

Unfortunately I have no suggestions re your current issue, but wish you luck. Maybe it will be one of those problems you exhaust all efforts on and then a random Tuesday a driver update, OS update, or C1 update sweeps it away.

[Malcolm Payne]

Updates as promised on the couple of issues identified with the Intel Arc GPU:

(1) The problem with Topaz Gigapixel AI not using the Intel GPU has been confirmed by Topaz as a bug at their end. It is scheduled for a fix by their developer, but with no timescale currently available.

(2) The corruption in PTGui panoramas stitched on the Arc is basically down to this card not being supported in the software. There were a couple of requests on the PTGui community forum in December 2022 for the developer to look into providing support for the Intel GPU(s), to which he agreed, but I have not seen any reports of further progress.

Also noted, the Arc series of GPUs currently only supports 8-bit output. Intel’s suggested workaround for anyone needing 10-bit output is to use the Arc for compute and route the output via the iGPU, which does support 10-bit although unfortunately currently only via HDMI. This is also subject to limitations where only HDMI 1.4 is available. Supposedly, 10-bit support on the USB-C and DisplayPort outputs is being worked on, but again with no available timescale, and it is not clear whether this will apply to both the UHD770 (or similar?) iGPUs and the A7xx Arc dGPUs.

Otherwise the card appears to be working very well and with no issues with everything else I have tried.

Hope that might be of some interest to anyone considering one of these new Intel GPUs instead of the more usual alternatives.