Crashes with OpenCL enabled

[MrBeep]

Hi, I've just tried Affinity Photo and it crashes everytime when Opening or Creating new file. With OpenCL disabled it doesn't crash. I'm on Windows 11, RTX 4090 and latest drivers version 546.01

[Callum]

Hi Mrbeep,

This is a known issue, for the time being I'd recommend working with Open CL disabled.

[BR22]

Can you please say more about this issue? I'm not able to run with OpenCL acceleration with either my RTX 4090 or my RTX 3080 on a fully patched Windows 11 Pro using the Nvidia 546.33 driver. I'm using the AP beta 2.3.0.2165 build. This same build with Windows 11 Pro runs fine on my laptop which has an RTX 3060 laptop GPU.

The crash is happening inside nvvm64.dll at the bolded line below: 

Not Flagged    >    15244    0    Worker Thread    Win64 Thread    nvvm64.dll!00007ffeafb81149
                              nvvm64.dll!00007ffeafb81149()
                              nvvm64.dll!00007ffeafb7c23e()
                              nvvm64.dll!00007ffeafb884b5()
                              nvopencl64.dll!00007ffeb554cbb4()
                              nvopencl64.dll!00007ffeb554ddea()
                              nvopencl64.dll!00007ffeb552f788()
                              libraster.dll!00007ffecc47a2a1()
                              libraster.dll!00007ffecc47cde6()
                              libraster.dll!00007ffecd6a3415()
                              libraster.dll!00007ffecd6889e2()
                              libraster.dll!00007ffecd774ee9()
                              libpersona.dll!00007ffec8196408()
                              libkernel.dll!00007ffedb1335b1()
                              libkernel.dll!00007ffedb423231()
                              libkernel.dll!00007ffedb42573b()
                              ntdll.dll!TppWorkpExecuteCallback()
                              ntdll.dll!TppWorkerThread()
                              KERNEL32.dll!BaseThreadInitThunk()
                              ntdll.dll!RtlUserThreadStart()

The exception is: Unhandled exception at 0x00007FFEAFB81149 (nvvm64.dll) in 78a9f3cd-83b7-413a-b0bb-c117631567c2.dmp: 0xC0000005: Access violation reading location 0x0000000000000008. Pretty much anything I do in AP with an image will cause it to crash, sometimes loading, but always when I try to save an image or change a layer's opacity, for example.

If I disable CUDA on both my cards, then AP will use the motherboard Intel GPU for acceleration without crashing. I am able to run the AP benchmark with CUDA on all cards turned on without it crashing, although the results are sometimes odd for the raster score.

Any insight or workaround you can provide on this issue would be greatly appreciated.
 

 

 

[MrBeep]

On 11/28/2023 at 11:53 AM, Callum said:

Hi Mrbeep,

This is a known issue, for the time being I'd recommend working with Open CL disabled.

Is that issue going to be resolved?

[MikeTO]

On 1/1/2024 at 8:55 PM, BR22 said:

I can't help with the issue, but what CPU was this test run on? We're compiling a table of benchmarks and it would be nice to add your benchmark report. Thank you.

 

[MrBeep]

What's the point of using this software when the serious bug is still there?

[walt.farrell]

13 hours ago, MrBeep said:

What's the point of using this software when the serious bug is still there?

You can avoid the bug by not using OpenCL, and we (users) have no indication that the bug is in the Affinity code. Looking at the info posted by BR22 above, for example, the failure looks like it was in Nvidia's code, not part of Affinity.

Turning off the OpenCL support should let you use the application, and get your work done, for now. Just, perhaps, a bit slower.

 

[BR22]

2 hours ago, MikeTO said:

I can't help with the issue, but what CPU was this test run on? We're compiling a table of benchmarks and it would be nice to add your benchmark report. Thank you.

 

This was with an Intel i9-14900K.

[MikeTO]

1 hour ago, BR22 said:

This was with an Intel i9-14900K.

Thank you

[Johnson4]

Reopening this post.  New PC here (RTX 5080 and AMD 9800).  With Open CL enabled anytime I try to "export" a file to PNG or JPG the image will tile itself (and save the same way).  Also, when I jump back to the original image it's just a series of small lines -- completely unintelligible.

I would very much like to have hardware acceleration enabled when using a new whiz-bang GPU. Please fix!

[AiDon]

It seems that there are many issues with the RTX5080 and RTX5090 cards. It you do a search on the internet there are many posts on both the NVIDIA forums and Reddit forums.

The current solution is to disable the integrated graphics card. You can either disable via the Windows Device Manager or via the BIOS, I would acceas the BIOS and disable from there as it won't be loaded at startup before Windows disables via the device manager.

Then you shouldn't have to disable OpenCL.

[photobomb]

Same problems as described above on Asus Scar 18" Intel Ultra 9 275HX RTX 5080 Win 11 Pro.  If I use Topaz Photo AI instead of Affinity, there are no issues saving as TIFF or DNG or exporting as JPEG.  Also, in Affinity, adding a JPEG as a layer to the active photo, also "scrambles" both.  Sorry but I'm old and when I booted to the BIOS I did not see the integrated intel GPU and, hence, any way to disable it.   If one of you young'n pros could briefly explain, or the other option how to disable OpenCL (yes, I know how to disable the Intel Graphics in Device Manager), I would be most appreciative.  Otherwise, until the Affinity fix, my workflow will be Camera RAW to Topaz (cropping, sharpening, denoise, at least) dng to Affinity then back to Topaz to save as a jpg.  Ugh

[Komatös]

On 5/8/2025 at 4:21 AM, Johnson4 said:

I would very much like to have hardware acceleration enabled when using a new whiz-bang GPU. Please fix!

It's a problem with NVIDIA, not Serif. And NVIDIA isn't paying as much attention to consumer cards as it is to its AI accelerators.

 

@photobomb

In Affinity Photo, go to the Menu Edit -> Settings -> Performance and disable hardware acceleration there.