I tried running it on my AMD AMD Radeon RX 9060 XT, I have 2 of them in the same pc and they both crunch most projects with no problems, but neither gpu returned valid tasks!!

Task 55787973
mikey · log out
Name LL-dc_85M_4_wu_171_8
Workunit 50225733
Created 1 Jun 2026, 21:55:53 UTC
Sent 2 Jun 2026, 2:26:27 UTC
Report deadline 9 Jun 2026, 2:26:27 UTC
Received 2 Jun 2026, 3:03:32 UTC
Server state Over
Outcome Computation error
Client state Compute error
Exit status 0 (0x0)
Computer ID 222552
Run time 4 sec
CPU time
Validate state Invalid
Credit 0.00
Device peak FLOPS 786.66 GFLOPS
Application version LL - dc v0.09 (opencl_ati_200)
Peak working set size 59.11 MB
Peak swap size 276.07 MB
Peak disk usage 2.65 MB
Stderr output
<core_client_version>8.2.9</core_client_version>
<![CDATA[
<stderr_txt>
2026-06-01 22:58:43 (11508): wrapper: running prpll.exe (-d 0 -v -use NO_ASM)
2026-06-01 22:58:43 (11508): wrapper: created child process 1200
20260601 22:58:43 PRPLL a3e4c7e starting
20260601 22:58:43 config: -d 0 -v -use NO_ASM
20260601 22:58:43 device 0, OpenCL 3661.0 (PAL,LC), unique id ''
20260601 22:58:43 85869361 No FFTs found in tune.txt that can handle 85869361. Consider tuning with -tune
20260601 22:58:44 85869361 config: -DNO_ASM=1
20260601 22:58:44 85869361 Using CARRY64
20260601 22:58:44 85869361 OpenCL: 3661.0 (PAL,LC):gfx1200, args -cl-finite-math-only -cl-std=CL2.0 -DNO_ASM=1 -DEXP=85869361u -DWIDTH=1024u -DSMALL_HEIGHT=256u -DMIDDLE=4u -DCARRY_LEN=8u -DNW=4u -DNH=4u -DAMDGPU=1 -DCARRY64=1 -DFFT_VARIANT=101u -DMAXBPW=4750u -DWEIGHT_STEP=0.038353673847609501 -DIWEIGHT_STEP=-0.036937004041686809 -DTAILT=U2(-7.52981578e-05f,0.0122715384f) -DTRIG_SCALE=9 -DTRIG_SIN={6.6579027251980952e-07,3.7209369054580932e-23,-4.9188217704570848e-20,1.0901995091303198e-33,-1.1506191102407305e-47,7.0839240359575376e-62,-2.8545227803597818e-76,8.0307778151820938e-91,} -DTRIG_COS={1,-2.2163834349100114e-13,8.1872592175725296e-27,-1.209740380477418e-40,9.5758876440886212e-55,-4.7164085661095887e-69,1.5838111582820031e-83,-3.830342691138796e-98,} -DTAILTGF31=U2(509684486u,293249438u) -DTAILTGF61=U2(249938719029223731ull,1245372627562045535ull) -DFFT_TYPE=4 -DWordSize=8u -DDISTGF31=819200u -DDISTWTRIGGF31=2560u -DDISTMTRIGGF31=768u -DDISTHTRIGGF31=131776u -DDISTGF61=1638400ull -DDISTWTRIGGF61=3072ull -DDISTMTRIGGF61=1792ull -DDISTHTRIGGF61=263040ull -DFRAC_BPW_HI=4061759487u -DFRAC_BPW_LO=4294967295u
20260601 22:58:44 85869361 FFT: 2M 4:1K:4:256:101 (40.95 bpw)
20260601 22:58:44 85869361 Loaded testTime : 328ms
20260601 22:58:44 85869361 V_NOP : 384307168202282304.00 cycles latency; time min: -1; avg 0
20260601 22:58:44 85869361 V_ADD_I32 : 384307168202282304.00 cycles latency; time min: -1; avg 0
20260601 22:58:44 85869361 V_FMA_F32 : 384307168202282304.00 cycles latency; time min: -1; avg 0
20260601 22:58:44 85869361 V_ADD_F64 : 384307168202282304.00 cycles latency; time min: -1; avg 0
20260601 22:58:44 85869361 V_FMA_F64 : 384307168202282304.00 cycles latency; time min: -1; avg 0
20260601 22:58:44 85869361 V_MUL_F64 : 384307168202282304.00 cycles latency; time min: -1; avg 0
20260601 22:58:44 85869361 V_MAD_U64_U32 : 384307168202282304.00 cycles latency; time min: -1; avg 0
20260601 22:58:44 85869361 Loaded transposeIn : 323ms
20260601 22:58:45 85869361 Loaded readResidue -DREADRESIDUE=1: 305ms
20260601 22:58:45 85869361 LL loaded @ 0 : 0000000000000004
20260601 22:58:45 85869361 In file included from C:\Users\mike\AppData\Local\Temp\comgr-c7ff2d\input\CompileSource:1:
In file included from C:\Users\mike\AppData\Local\Temp\comgr-c7ff2d\include\fftp.cl:5:
In file included from C:\Users\mike\AppData\Local\Temp\comgr-c7ff2d\include\fftwidth.cl:18:
C:\Users\mike\AppData\Local\Temp\comgr-c7ff2d\include\fftbase.cl:1136:8: error: assigning to '__private F2' (aka '__private float2') from incompatible type 'ulong2' (vector of 2 'ulong' values)
1136 | base = U2(fma(a, -w.y, w.x), fma(a, w.x, -w.y));
| ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1 error generated.
Error: Failed to compile source (from CL or HIP source to LLVM IR).

20260601 22:58:45 85869361 Compiling 'fftp.cl' error COMPILE_PROGRAM_FAILURE (-15) (args -cl-finite-math-only -cl-std=CL2.0 -DNO_ASM=1 -DEXP=85869361u -DWIDTH=1024u -DSMALL_HEIGHT=256u -DMIDDLE=4u -DCARRY_LEN=8u -DNW=4u -DNH=4u -DAMDGPU=1 -DCARRY64=1 -DFFT_VARIANT=101u -DMAXBPW=4750u -DWEIGHT_STEP=0.038353673847609501 -DIWEIGHT_STEP=-0.036937004041686809 -DTAILT=U2(-7.52981578e-05f,0.0122715384f) -DTRIG_SCALE=9 -DTRIG_SIN={6.6579027251980952e-07,3.7209369054580932e-23,-4.9188217704570848e-20,1.0901995091303198e-33,-1.1506191102407305e-47,7.0839240359575376e-62,-2.8545227803597818e-76,8.0307778151820938e-91,} -DTRIG_COS={1,-2.2163834349100114e-13,8.1872592175725296e-27,-1.209740380477418e-40,9.5758876440886212e-55,-4.7164085661095887e-69,1.5838111582820031e-83,-3.830342691138796e-98,} -DTAILTGF31=U2(509684486u,293249438u) -DTAILTGF61=U2(249938719029223731ull,1245372627562045535ull) -DFFT_TYPE=4 -DWordSize=8u -DDISTGF31=819200u -DDISTWTRIGGF31=2560u -DDISTMTRIGGF31=768u -DDISTHTRIGGF31=131776u -DDISTGF61=1638400ull -DDISTWTRIGGF61=3072ull -DDISTMTRIGGF61=1792ull -DDISTHTRIGGF61=263040ull -DFRAC_BPW_HI=4061759487u -DFRAC_BPW_LO=4294967295u -DFFT_FP64=0 -DFFT_FP32=1 -DNTT_GF31=1 -DNTT_GF61=1 )
20260601 22:58:45 85869361 Can't compile fftp.cl
20260601 22:58:45 Exception "Can't compile fftp.cl"
20260601 22:58:45 Bye
2026-06-01 22:58:45 (11508): prpll.exe exited; CPU time 0.000000
2026-06-01 22:58:45 (11508): called boinc_finish(0)

</stderr_txt>
<message>
upload failure: <file_xfer_error>
<file_name>LL-dc_85M_4_wu_171_8_0</file_name>
<error_code>-240 (stat() failed)</error_code>
</file_xfer_error>
</message>
]]>

After running multiple jobs on my RTX 5080, the task
https://srbase.my-firewall.org/sr5/result.php?resultid=56125592 errored out

last part of the log:

The task that I mentioned earlier with the number 55785824 finally completed after around 15 hours run time on my RTX 3070-ti but strangely logged only 1h28. I had to pause it twice for computer shutdown. The Stderr file started logging the last 3,5 hours and also shows the last interuption but everything before that is gone. So the official runtime is 1h28 instead of 15h.

Host https://srbase.my-firewall.org/sr5/show_host_detail.php?hostid=251426
1080ti returns only errors. For now I switch back to TF

I have the same issue as mikey on linux & windows

In file included from /tmp/comgr-9915-19-4657d4/include/fftp.cl:5:
In file included from /tmp/comgr-9915-19-4657d4/include/fftwidth.cl:18:
/tmp/comgr-9915-19-4657d4/include/fftbase.cl:1136:8: error: assigning to '__private F2' (aka '__private float2') from incompatible type 'ulong2' (vector of 2 'ulong' values)
1136 | base = U2(fma(a, -w.y, w.x), fma(a, w.x, -w.y));

Running on a Radeon R9 290, 100% according to Boinc after only 15 hours, stderr.txt as below, but still producing heat. I assume it's on another section? I shall leave it running. Unfortunately the incorrect progress means Boinc scheduler left it for a bit until it panicked.

Hi Rebirther! Is there an option (somehow) to unpack the app (prpll-linux64-v3n), run a tune session (creates a config.txt + tune.txt file) and repack?

Or can I use app_config.xml to point to a cuda version of prpll?
I have no idea whether the project accepts this or not.

Why do I want this you might ask?! On ll-dc runs (85M), a tune session that takes just ~5-10 minutes improves the speed by +30%. Runtime decreases from 12:40 (HH:MM) to 08:00.

You might not be aware of all functions of prpll, but generally a tune can improve the speed of a run by a lot.

Curious what you thoughts are!

EDIT: also the hardcoded "NO_ASM" is a major slowdown on nvidia-cards. (like -20%). I can see why it might help older cards (or AMD?) but modern nvidia-cards suffer quite a lot from this switch.

I figured an easier way. Copied the package from /var/lib/boinc/, ran tune and pasted the line from config.txt to a app_config.xml file.

That works well, already improved by 16-20%. The only thing in the way is the hardcoded "NO_ASM", which I cannot get rid off. I can understand why its there, but its a major slowdown on modern nvidia cards (4xxx,5xxx).

Did some more checking, actually its a lot less than I thought. About 4.5% lost due to "NO_ASM".

In a standalone version (which has config.txt and tune.txt) the iteration time is 346 and ETA is roughly 08:00 (HH:MM)

Vanilla boinc app (opencl_nvidia_200) ETA is roughly 12:40 (HH:MM)

Using app_config (copy line from config.txt) I can get the iteration time down to 450 and ETA is roughly 10:26.

Pausing boinc and pasting tune.txt in slot 0 and restarting again, I can get the iteration time down to 364 and ETA is roughly 08:23.

Ofcourse this is not useful to do every time, but it demonstrates how much time we leave on the table. My situation: 4070 Ti NVIDIA card.

GPU max WUs in progress limit = 1

I am trying a work unit on a NVidia GTX 750 Ti, 3 days already and it shows ETA 9 days. That will be a little bit far from its deadline, isn't it? Still worth to keep?

Just wondering that is running but I have extended the deadline for this WU.

no, 1 is enough for better scaling.