Page MenuHome

Render crashes Blender when used with Volume Absorption
Needs Triage, NormalPublic

Description

System Information
Operating system: Windows 10
Graphics card: 5700 XT

Blender Version
Broken: 2.83
Worked: None, I am new to blender

Short description of error
Render crashes the whole application when used with OpenCL rendering with CPU+GPU or just GPU. The culprit is the Volume Absorption modifier. It does not crash when looking at the rendered mode within the viewport. Just the F12 final render. It works if I just use CPU rendering but that takes a lot longer.

Exact steps for others to reproduce the error

Event Timeline


This is the exact moment the render hangs.

Germano Cavalcante (mano-wii) changed the task status from Needs Triage to Needs Information from User.Jun 19 2020, 5:13 PM

I cannot reproduce the problem.
Does the blender close or just hang?
Make sure that it is not the compilation of the kernels that is taking time.
You can see the compilation process in Window -> Toogle System Console


Operating system: Windows-10-10.0.18941 64 Bits
Graphics card: Radeon (TM) RX 480 Graphics ATI Technologies Inc. 4.5.13596 Core Profile Context 20.10.19.02 27.20.1019.2002


It shows this for a bit, then spits out this

Read prefs: C:\Users\jrdn\AppData\Roaming\Blender Foundation\Blender\2.83\config\userpref.blend
found bundled python: C:\Program Files\Blender Foundation\Blender 2.83\2.83\python
Read blend: C:\Users\jrdn\3D Objects\Donut.blend
Cycles: compiling OpenCL program split_subsurface_scatter...
Cycles: compiling OpenCL program split_shadow_blocked_dl...
0x00007FFC03210AF5 (0x000001CF302BAE40 0x000001CF275CD8A0 0x000001CF302BAE40 0x000000000000014C), amd_comgr_symbol_lookup() + 0x1408A25 bytes(s)
0x00007FFC0370CC3F (0x000001CF29818110 0x0000006A3D7FB310 0x0000000000000001 0x0000000000000001), amd_comgr_symbol_lookup() + 0x1904B6F bytes(s)
0x00007FFC0370DDF4 (0x000001CF29818080 0x000001CF275CD8A0 0x000001CF2E817060 0x000001CF2FC2E3D8), amd_comgr_symbol_lookup() + 0x1905D24 bytes(s)
0x00007FFC0365C916 (0x0000000000000000 0x0000000000000000 0x00007FFC054B93B0 0x0000006A3D7FB720), amd_comgr_symbol_lookup() + 0x1854846 bytes(s)
0x00007FFC03EE0ACE (0x000001CF301F68E0 0x000001CF301F68E8 0x000001CF2C308501 0x000001CF2EBB2100), amd_comgr_symbol_lookup() + 0x20D89FE bytes(s)
0x00007FFC03CE502F (0x000001CF2EBB2140 0x0000000000000000 0x000001CF2EBB1F90 0x000001CF2A2753A0), amd_comgr_symbol_lookup() + 0x1EDCF5F bytes(s)
0x00007FFC03CE4CEB (0x000001CF2C30A701 0x0000006A3D7FB960 0x0000000000000000 0x000001CF301F68E0), amd_comgr_symbol_lookup() + 0x1EDCC1B bytes(s)
0x00007FFC03CE5B38 (0x000001CF27F0E900 0x0000000000000000 0x000001CF27ECE990 0x000001CF2EBB2140), amd_comgr_symbol_lookup() + 0x1EDDA68 bytes(s)
0x00007FFC03EE0F2E (0x000001CF2F0C2600 0x000001CF2F0C2700 0x000000002F0C2600 0x0000000000000012), amd_comgr_symbol_lookup() + 0x20D8E5E bytes(s)
0x00007FFC03EE07F0 (0x0000000000000000 0x0000006A3D7FBCE0 0x0000000000000000 0x0000000000000000), amd_comgr_symbol_lookup() + 0x20D8720 bytes(s)
0x00007FFC01E55598 (0x0000006A3D7FBFD0 0x00007FFC00000005 0x0000006A3D7FC140 0x0000000000000000), amd_comgr_symbol_lookup() + 0x4D4C8 bytes(s)
0x00007FFC01E57E57 (0x000001CF275C0BC0 0x000001CF29B8C070 0x000001CF311C70B0 0x000001CF275C0BC0), amd_comgr_symbol_lookup() + 0x4FD87 bytes(s)
0x00007FFC01E1D4F7 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), amd_comgr_symbol_lookup() + 0x15427 bytes(s)
0x00007FFC021BC62A (0x0000000000000000 0x0000006A3D7FC930 0x000001CF26D2F160 0x000001CF26D2F160), amd_comgr_symbol_lookup() + 0x3B455A bytes(s)
0x00007FFC021AB0B3 (0x000001CF3019EF50 0x000001CF2766F8B0 0x000001CF2766F8B0 0x000001CF29BF8D80), amd_comgr_symbol_lookup() + 0x3A2FE3 bytes(s)
0x00007FFC01E184C0 (0x000001CF2766F8B0 0x0000006A00000000 0x000001CF29E727A8 0x000001CF00000000), amd_comgr_symbol_lookup() + 0x103F0 bytes(s)
0x00007FFC01DFB432 (0x0000006A3D7FD5E8 0x000001CF26D2F160 0x000001CF29E72720 0x000001CF26D2F160)
0x00007FFC01DFE170 (0x000001CF305E7A70 0x000001CF305F63F0 0x0000006A3D7FDD30 0x00007FFC04146C44)
0x00007FFC01DFE521 (0x000001CF305E7A70 0x000001CF00000007 0x0000006A3D7FE138 0x000001CF305E7A70)
0x00007FFC01DFA8B0 (0x000001CF2FC48730 0x000001CF2FC48730 0x000001CF30662530 0x00007FFC01E06ECC)
0x00007FFC01E06AB1 (0x000001CF00000006 0x000001CF2EC2A9C0 0x000001CF30662530 0x00007FFC6BCC5D21), amd_comgr_do_action() + 0xB31 bytes(s)
0x00007FFC08428325 (0x0000000000000033 0x0000000000000000 0x0000006A3D7FEA00 0x000001CF276B2401), clGetPipeInfo() + 0x2A1C5 bytes(s)
0x00007FFC08430C70 (0x0000000000000000 0x000001CF272C5E50 0x000001CF30662530 0x0000000000000000), clGetPipeInfo() + 0x32B10 bytes(s)
0x00007FFC08426F73 (0x000001CF00000000 0x000001CF276855B0 0x000001CF29BE1140 0x000000000000014F), clGetPipeInfo() + 0x28E13 bytes(s)
0x00007FFC0841906D (0x000001CF29B9C901 0x000001CF29A98C68 0x0000000000000000 0x0000000000000000), clGetPipeInfo() + 0x1AF0D bytes(s)
0x00007FFC083F1359 (0x000001CF29C039B0 0x0000006A3D7FF209 0x0000000000000000 0x0000006A3D7FF209), clBuildProgram() + 0xC9 bytes(s)
0x00007FF6612CC793 (0x000001CF27CBCBE1 0x0000000000000000 0x0000006A3D7FF2D0 0x000001CF29BE1150)
0x00007FF66128275D (0x000001CF2912EEE8 0x0000000000000000 0x0000000000000006 0x000001CF2A9DD8B8)
0x00007FFC4CCB6BED (0x0000006A00000001 0x0000000000000006 0x000001CF2A9DD8B8 0x000001CF275013A0), _PyMethodDef_RawFastCallKeywords() + 0x22D bytes(s)
0x00007FFC4CD83252 (0x0000006A3D7FF490 0x000001CF27DEA2B8 0x000001CF2A5E1410 0x0000006A3D7FF490), PyEval_GetFuncDesc() + 0x182 bytes(s)
0x00007FFC4CD801E8 (0x000001CF2A4B63F0 0x0000000000000000 0x0000000000000000 0x000001CF291459C0), _PyEval_EvalFrameDefault() + 0x2C28 bytes(s)
0x00007FFC4CD82035 (0x000001CF275013A0 0x0000000000000000 0x000001CF291459C0 0x0000000000000000), _PyEval_EvalCodeWithName() + 0x9B5 bytes(s)
0x00007FFC4CDDCDED (0x000001CF290F9E70 0x0000000000000000 0x0000000000000000 0x000001CF00000000), PyRun_FileExFlags() + 0x48D bytes(s)
0x00007FFC4CDDC928 (0x0000000000000000 0x000001CF29112638 0x000001CF2652E901 0x0000000000000000), PyRun_StringFlags() + 0xE8 bytes(s)
0x00007FF6609D3B18 (0x0000000000000003 0x0000000000000003 0x000001CF2858F278 0x00007FF662B029FA)
0x00007FF660483DD5 (0x0000000000000003 0x000001CF27087920 0x0000000000000003 0x000001CF27087920)
0x00007FF662AF746B (0x0000000000000000 0x000001CF270878F8 0x0000000000000000 0x000001CF270878F8), PyInit_aud() + 0x5725B bytes(s)
0x00007FF6604821B7 (0x0000000000000005 0x0000000000000000 0x000000005ED7B5DE 0x0000000000000000)
0x00007FF66303AF74 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), PyInit_aud() + 0x59AD64 bytes(s)
0x00007FFC6AFA6FD4 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), BaseThreadInitThunk() + 0x14 bytes(s)
0x00007FFC6BCFCEC1 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), RtlUserThreadStart() + 0x21 bytes(s)
AL lib: (EE) alc_cleanup: 1 device not closed
Cycles: compiling OpenCL program split_subsurface_scatter...
0x00007FFC03210AF5 (0x0000014B4A0CDE50 0x0000014B4BDA8700 0x0000014B4A0CDE50 0x000000000000014C), amd_comgr_symbol_lookup() + 0x1408A25 bytes(s)
0x00007FFC0370CC3F (0x0000014B4C0ED760 0x000000333DDFB740 0x0000000000000001 0x0000000000000001), amd_comgr_symbol_lookup() + 0x1904B6F bytes(s)
0x00007FFC0370DDF4 (0x0000014B4C0ED6D0 0x0000014B4BDA8700 0x0000014B49DC1890 0x0000014B4F4768F8), amd_comgr_symbol_lookup() + 0x1905D24 bytes(s)
0x00007FFC0365C916 (0x0000000000000000 0x0000000000000000 0x00007FFC054B93B0 0x000000333DDFBB50), amd_comgr_symbol_lookup() + 0x1854846 bytes(s)
0x00007FFC03EE0ACE (0x0000014B492197F0 0x0000014B492197F8 0x0000014B5023EB01 0x0000014B49ED3700), amd_comgr_symbol_lookup() + 0x20D89FE bytes(s)
0x00007FFC03CE502F (0x0000014B49ED3740 0x0000000000000000 0x0000014B49ED33E0 0x0000014B49FEAD10), amd_comgr_symbol_lookup() + 0x1EDCF5F bytes(s)
0x00007FFC03CE4CEB (0x0000014B5023F101 0x000000333DDFBD90 0x0000000000000000 0x0000014B492197F0), amd_comgr_symbol_lookup() + 0x1EDCC1B bytes(s)
0x00007FFC03CE5B38 (0x0000014B4F25F400 0x0000000000000000 0x0000014B4ED59120 0x0000014B49ED3740), amd_comgr_symbol_lookup() + 0x1EDDA68 bytes(s)
0x00007FFC03EE0F2E (0x0000014B4BCD7500 0x0000014B4BCD6900 0x000000004BCD7500 0x0000000000000012), amd_comgr_symbol_lookup() + 0x20D8E5E bytes(s)
0x00007FFC03EE07F0 (0x0000000000000000 0x000000333DDFC110 0x0000000000000000 0x0000000000000000), amd_comgr_symbol_lookup() + 0x20D8720 bytes(s)
0x00007FFC01E55598 (0x000000333DDFC400 0x00007FFC00000005 0x000000333DDFC570 0x0000000000000000), amd_comgr_symbol_lookup() + 0x4D4C8 bytes(s)
0x00007FFC01E57E57 (0x0000014B47C88640 0x0000014B47B8B040 0x0000014B4C393B30 0x0000014B47C88640), amd_comgr_symbol_lookup() + 0x4FD87 bytes(s)
0x00007FFC01E1D4F7 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), amd_comgr_symbol_lookup() + 0x15427 bytes(s)
0x00007FFC021BC62A (0x0000000000000000 0x000000333DDFCD60 0x0000014B49D14990 0x0000014B49D14990), amd_comgr_symbol_lookup() + 0x3B455A bytes(s)
0x00007FFC021AB0B3 (0x0000014B4C6BB430 0x0000014B4741C460 0x0000014B4741C460 0x0000014B4BCF77E0), amd_comgr_symbol_lookup() + 0x3A2FE3 bytes(s)
0x00007FFC01E184C0 (0x0000014B4741C460 0x0000003300000000 0x0000014B47304DD8 0x0000014B00000000), amd_comgr_symbol_lookup() + 0x103F0 bytes(s)
0x00007FFC01DFB432 (0x000000333DDFDA18 0x0000014B49D14990 0x0000014B47304D50 0x0000014B49D14990)
0x00007FFC01DFE170 (0x0000014B4C71F360 0x0000014B4C7384E0 0x000000333DDFE160 0x00007FFC04146C44)
0x00007FFC01DFE521 (0x0000014B4C71F360 0x0000014B00000007 0x000000333DDFE568 0x0000014B4C71F360)
0x00007FFC01DFA8B0 (0x0000014B4ED90E80 0x0000014B4ED90E80 0x0000014B49AE56F0 0x00007FFC01E06ECC)
0x00007FFC01E06AB1 (0x0000014B00000006 0x0000014B4E963350 0x0000014B49AE56F0 0x00007FFC6BCC5D21), amd_comgr_do_action() + 0xB31 bytes(s)
0x00007FFC08428325 (0x0000000000000033 0x0000000000000000 0x000000333DDFEE30 0x0000014B47B51901), clGetPipeInfo() + 0x2A1C5 bytes(s)
0x00007FFC08430C70 (0x0000000000000000 0x0000014B47028C60 0x0000014B49AE56F0 0x0000000000000000), clGetPipeInfo() + 0x32B10 bytes(s)
0x00007FFC08426F73 (0x0000014B00000000 0x0000014B4731BE10 0x0000014B494D6150 0x000000000000014F), clGetPipeInfo() + 0x28E13 bytes(s)
0x00007FFC0841906D (0x0000014B4987FF01 0x0000014B496645B8 0x0000000000000000 0x0000000000000000), clGetPipeInfo() + 0x1AF0D bytes(s)
0x00007FFC083F1359 (0x0000014B47B782F0 0x000000333DDFF639 0x0000000000000000 0x000000333DDFF639), clBuildProgram() + 0xC9 bytes(s)
0x00007FF6612CC793 (0x0000014B49765AA1 0x0000000000000000 0x000000333DDFF700 0x0000014B494D6160)
0x00007FF66128275D (0x0000014B48E9CEE8 0x0000000000000000 0x0000000000000006 0x0000014B4A74D8B8)
0x00007FFC4CCB6BED (0x0000003300000001 0x0000000000000006 0x0000014B4A74D8B8 0x0000014B4729FEB0), _PyMethodDef_RawFastCallKeywords() + 0x22D bytes(s)
0x00007FFC4CD83252 (0x000000333DDFF8C0 0x0000014B47A4D668 0x0000014B4A710410 0x000000333DDFF8C0), PyEval_GetFuncDesc() + 0x182 bytes(s)
0x00007FFC4CD801E8 (0x0000014B4A2263F0 0x0000000000000000 0x0000000000000000 0x0000014B48EB89C0), _PyEval_EvalFrameDefault() + 0x2C28 bytes(s)
0x00007FFC4CD82035 (0x0000014B4729FEB0 0x0000000000000000 0x0000014B48EB89C0 0x0000000000000000), _PyEval_EvalCodeWithName() + 0x9B5 bytes(s)
0x00007FFC4CDDCDED (0x0000014B48E69E70 0x0000000000000000 0x0000000000000000 0x0000014B00000000), PyRun_FileExFlags() + 0x48D bytes(s)
0x00007FFC4CDDC928 (0x0000000000000000 0x0000014B48E83638 0x0000014B462D0E01 0x0000000000000000), PyRun_StringFlags() + 0xE8 bytes(s)
0x00007FF6609D3B18 (0x0000000000000003 0x0000000000000003 0x0000014B48B7F278 0x00007FF662B029FA)
0x00007FF660483DD5 (0x0000000000000003 0x0000014B46E1FC80 0x0000000000000003 0x0000014B46E1FC80)
0x00007FF662AF746B (0x0000000000000000 0x0000014B46E1FC58 0x0000000000000000 0x0000014B46E1FC58), PyInit_aud() + 0x5725B bytes(s)
0x00007FF6604821B7 (0x0000000000000005 0x0000000000000000 0x000000005ED7B5DE 0x0000000000000000)
0x00007FF66303AF74 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), PyInit_aud() + 0x59AD64 bytes(s)
0x00007FFC6AFA6FD4 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), BaseThreadInitThunk() + 0x14 bytes(s)
0x00007FFC6BCFCEC1 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), RtlUserThreadStart() + 0x21 bytes(s)
AL lib: (EE) alc_cleanup: 1 device not closed
Cycles: compiling OpenCL program split_shadow_blocked_dl...

it hangs for about a minute and then closes.

Here is a guide on graphics problem in blender:
https://docs.blender.org/manual/en/dev/troubleshooting/gpu/index.html

Maybe some information there might help.

Btw, could you test other versions of blender (like 2.82)?
https://www.blender.org/download/previous-versions/

Thanks for following up. Tried going through that. My drivers are all up to date. It still crashes 2.82 with OpenCL enabled. If I go to None it renders but that is much slower. It's slower than even just using my processor on OpenCL only. Which is a combination that works.

Germano Cavalcante (mano-wii) changed the task status from Needs Information from User to Needs Triage.Jun 19 2020, 8:28 PM

Hello, I just wanted to contribute my experience as I also ran into this issue. I am running an AMD 5700XT on Windows 10. I updated blender from 2.82 to 2.83 and my my driver to the 6/10/2020 driver from AMD and adding a volume absorption node to the default Principled BSDF shader and trying to render using GPU with OpenCL would cause blender to crash with no error message or useful info.

Downgrading to Blender version 2.82 and GPU driver 20.4.2 (released 5/15/2020) and running the same render, the shader compilation does not fail and blender does not crash and the render completes as expected.

I also just ran into this problem, my video card is an AMD 5700 XT on windows 10. I had no luck trying it in 2.9 neither in 2.83, and weirdly enough in 2.82 blender doesn't even want to render the default scene with a cube(it gets stuck compiling the kernels in the viewport and then crashes). GPU drivers are version 20.5.1 (5/25/2020).

2.9 gpu debug output rendering an object with principled volume:


I just found out it will unfreze if you start rendering in the viewport and then you start to render before the viewport is done, after this it didn't get stuck anymore even if I close blender and reopen the same file.

Just to contribute.

I Have a RX570. I Made a Video that shows this issue here. Just add a volume node at default cube in a fresh file created. Cycles is Unusable for me with GPU.

Another thing I noted is that if I change to cyles on this newly created file (whithout volume node), my vram utilization is highly increased.

So, I just want to add something here, when I'm working with volumetrics, after some time with blender open, If i'm with viewport render on and something is connected to the volumetrics output, If I disconnect it my blender just hangs really hard to the point that sometimes it crashes the whole windows too.

When it happens, it's usually a principled volume shader with some procedural textures on it. I tried to reproduce it only with a volume scatter, and after disconnecting, the blender recompiled the kerners for a second or two and then it freezed on this:

my gpu is an RX 5700 and I'm using the latest "stable" drivers (20.4.2)

Also I think the original bug from this thread isn't happening for me anymore in 2.90.1, tho I'd like to have some confirmation for this

So, I just want to add something here, when I'm working with volumetrics, after some time with blender open, If i'm with viewport render on and something is connected to the volumetrics output, If I disconnect it my blender just hangs really hard to the point that sometimes it crashes the whole windows too.

When it happens, it's usually a principled volume shader with some procedural textures on it. I tried to reproduce it only with a volume scatter, and after disconnecting, the blender recompiled the kerners for a second or two and then it freezed on this:

my gpu is an RX 5700 and I'm using the latest "stable" drivers (20.4.2)

Also I think the original bug from this thread isn't happening for me anymore in 2.90.1, tho I'd like to have some confirmation for this

I'm on an RX 5700XT with 20.9.2 on driver version and using Blender 2.90.1. I'm still getting crashes as in the original post but I'm also crashing even without using volumetrics. Cycles with CPU works fine but GPU still won't get past building the kernel.

@David Rodriguez (SilentMrDave) this is already a Known problem(https://developer.blender.org/T75319) with the AMD beta drivers in the current generation and is not related to this thread. you can try it with the lastest stable drivers(20.4.2)

Same problem. I am using latest driver 20.9.2 and blender 2.90.1 (this issue started to happen from 2.83). Viewport render freeze on 1st sample and I have to force stop blender.

I have the same issue in Linux. When using a volumetric shader, the GPU (5700 XT) usage goes up to 99% and blender freezes. Then even if I kill blender the GPU usage stays at 99% until I reboot the computer.
I am using blender 2.90.1 and using the latest AMDGPU-PRO version for Linux - 20.40

Hi Everyone,

I have new to blender but am having the exact same issue. The reder works fine until I add a volumetric shader and then just crashes on compiling the kernel.

I am running on a AMD Radeon RX5700 XT Red Devil, Intel i7-6700k with 32gb RAM.

Any help would be greatly appreciated.

@Aviel Warschawski (avielw) Weirdly enough I already got that 99% once, you don't need to reboot your system, puting it in sleep mode and going back is enough and faster, tho for some reason I'm not getting stuck in the kernel compile part anymore, its rendering fine, tho I still have that problem of disconnecting the volume node while rendering.

my SO is windows and i'm using v20.4.2 drivers

I've continued testing a few things regarding the crash on my Linux machine.

I thought that maybe there's a problem with the opencl amdgpu-pro (pal) driver.
I tried to test its machine learning capabilities and I found out that I that some benchmarks get calculation errors, consistently (for example when I use the "plaidbench keras mobilenet" benchmark - using the plaidml tool). Then, when I switched to the rocm opencl driver, the errors were gone.
That suggests that the problem is in the opencl amdgpu pal driver.

Then I found that someone reported an issue that sounds exactly like that on a Windows machine:
https://community.amd.com/thread/248056

But they said that the issue should be fixed...

(If someone has a Windows machine, he can test the code from this forum page on his computer to check if the problem was actually fixed)

This still is an Issue either thru Volume Absorption or even Mantaflow+principled BSDF - renders seem to stall after a certain density threshold has been reached. I noticed this again using 2.90.1 on Win 10 / RX 480 8GB using drivers 20.9.1. Also in Viewport Mode the Render does not pass the first Sample ( gets stuck there ).

Definitely an RTX 5700 XT issue. I'm getting the same issue on Blender 2.83 when doing a GPU render with volumetric shaders. Seems to work absolutely fine on a CPU render with my Ryzen 5 3600. My current graphics driver is 20.5.1 but I'm updating to 20.9.2 as I speak. Will update with findings once I've updated.


Post-update, still no dice. Still crashes while 'updating device'. I'm going to try rolling back to 2.82 as suggested above.


Aaaaaand no joy with the rollback either. Same crash at the same stage. I'll try updating to 2.9.


No luck with that either.

@Chloe Hase (ChloeHase) try to downgrade the driver to 20.4.2, its working "fine" on my 5700xt

@Chloe Hase (ChloeHase) try to downgrade the driver to 20.4.2, its working "fine" on my 5700xt

Those quote marks don't give me hope! What's the downside?

@Diogo Valadares Reis dos Santos (DiogoX2) Well I rolled back; it no longer crashes, but now my renders look like this.

@Chloe Hase (ChloeHase) are you rendering with CPU+GPU?if so, this is a known issue https://developer.blender.org/T50193. I said "fine" with quotes because its a bit unstable thing and it tends to crash, like the example I made above

@Diogo Valadares Reis dos Santos (DiogoX2) Ah, and here I was thinking ticking both of those under OpenCL would just make them both available, not use them both to render simultaneously. Weird, given I've never had that issue before with other renders! It works fine now.

I'm having the same problem. previewing and rendering perfectly fine with CPU. With the GPU, if the density isn't particularly high, items with Volume Absorption will show something completely wrong in preview mode, and sometimes render the right thing in render mode. If the density is very high, both preview and render mode will cause the system to crash. I even had problems with the blender benchmark using the GPU and couldn't run it to the end (CPU is fine).
If I hide the object in preview mode, the rest of the objects are displayed normally even with GPU, so I think it's a Volume Absorption issue.

I'm using a Radeon RX Vega 64.
Blender version 2.90.1
driver version 20.10.27.03-200826a-358464C-RadeonProEnterprise

Here are some of the tests I did. (I only made these changes: create a new file, add a volume, remove the surface)

shader


viewport shading CPU

viewport shading GPU

F12 GPU

(If using AMD ProRender)