Tasks are all failing.


<core_client_version>7.24.1</core_client_version>
<![CDATA[
<message>
process exited with code 1 (0x1, -255)</message>
<stderr_txt>
GPU Summary String = [CUDA|NVIDIAGeForceRTX5070Ti|1|15817MB|57099|300].
Loading GPU lookup table from file.
GPU was not found in the lookup table. Using default values:
numBlocks = 1024.
threadsPerBlock = 32.
polyBufferSize = 32768.
Setting GPU device number 0.
Cuda initialization was successful.
CHECKPOINT_FILE = wu_sf6_DS-15x11_Grp7020780of13107200_checkpoint.
Checkpoint Flag = 0.
Reading file ../../projects/numberfields.asu.edu_NumberFields/sf6_DS-15x11_Grp7020780of13107200.dat
K = x^2 - 10
S = [2, 5]
Disc Bound = 320000000000000
Skip = (P^3)*(Q^9)
Num Congruences = 32
SCALE = 1.000000
|dK| = 40
Signature = [2,0]
Opening output file ../../projects/numberfields.asu.edu_NumberFields/wu_sf6_DS-15x11_Grp7020780of13107200_0_r315192737_0
Now starting the targeted Martinet search:
Num Cvecs = 32.
Doing Cvec 1.
Error code 700: an illegal memory access was encountered file polDiscTest_gpuCuda.cu line 2213.
polDisc Test had an error. Aborting.

</stderr_txt>
]]>

I checked the database and there are successful results for 5000 series cards, but they are all with the Windows OpenCL version of the app.

It looks like you are the only one running the linux CUDA version. CUDA is supposed to be backwards compatible, unless they recently changed that with the 5000 series cards. Any chance your Nvidia driver is outdated?

It looks like you are the only one running the linux CUDA version. CUDA is supposed to be backwards compatible, unless they recently changed that with the 5000 series cards. Any chance your Nvidia driver is outdated?

I take that back. There is now a 2nd user with a 5070Ti with the same error as you. So it's not just you...

Using the 570.99 driver, Mint 22.1
Two other computers are the same except a 4070ti super in one and a 3070ti in the other. Those two a fine.

I found this in the Nvidia docs:

The NVIDIA Blackwell GPU architecture retains and extends the same CUDA programming model provided by previous NVIDIA GPU architectures such as NVIDIA Ampere and Turing architectures

So the app should be compatible with the newer 5000 series GPUs. It's possible that the code is ignoring one of the stricter rules, and the newer CUDA implementation is now enforcing it, causing the error. When I find some time in the next couple weeks I will try to rebuild the code with the latest CUDA library.