Radeon RX 6700XT Error on GFN-15 trough GFN-19

[Dennis]

For some reason i get error when trying to run GFN-15 trough GFN-19
GFN-20 (that have same requirements, run with no problems)
GFN-21 and GFN-22 that have same requirements plus additional double precision requirement run with no problems

Why can't i run the smaller tasks when the big ones run fine?

the errors GFN-15 trough GFN-19 is giving me is as follows:

<core_client_version>7.16.11</core_client_version>
<![CDATA[
<message>
The system cannot find the device specified.
(0x14) - exit code 20 (0x14)</message>
<stderr_txt>
geneferocl 3.3.3-2 (Windows/OpenCL/32-bit)

Then it starts running anyway and about a few minutes in to the task i get this error:

maxErr exceeded for 134332300^65536+1, 1.0000 > 0.4500
Errors occurred for all available transform implementations
Waiting 10 minutes before attempting to continue from last checkpoint...

Then it starts over and over and over and...........

I'm running Windows 10 Enterprice
Asus Tuf gaming X570 mother
Ryzen 9 3900X
32 GB ram
Sapphire NITRO+ AMD Radeon™ RX 6700 XT

as i said... all other Primegrid tasks run fine. Even "Do You Feel Lucky? (GFN World Record)" run with no problem on my system... but the smaller Generalized Fermat Prime Search from 15 to 19 wont run....
I believe the problem is that those tasks it want to use OCL2 transform for those smaller tasks... but as i said i don't know?
How to fix the problem?

Thanks in advance

[Prime95]

I'm not sure anyone here can help. Have you tried the forums over at PrimeGrid?

[moebius]

Unfortunately I cannot help you with your problem. But I have one request for you. I think many users are interested in the performance of your quite expensive 6700 XT with gpuowl.

I need the software timings (iteration times in µs/it) for for the first 300000 iterations for a specific exponent.
I am currently creating a benchmark list in (.ods) format, which better reflects the performance for gppuowl than mersenne.ca which is better suited for CudaLucas You can use Linux or Windows-version, the values ​​should only differ by a few percent. The Linux version is around 5-10% faster. The windows version works without any further installation effort.

gpuowl -iters 300000 -prp 77936867

Fastest Windows Version (with Proofs) for most FFT sizes
https://mersenneforum.org/showpost.php?p=551594&postcount=8

Linux binary version Ubuntu 18.04
https://mersenneforum.org/showpost.php?p=556311&postcount=40

Thread with hints for fine tuning (probably like 6900XT)
https://mersenneforum.org/showthread.php?p=590193#post590193

Downloadlink for .ods sheet (with no macros) LibreOffice Calc 7
https://drive.google.com/file/d/10fC...enkBdAaRP/view

[Dennis]

@moebius I ran the command on GPUOWL you requested.... I have no idea on how to finetune GPUOWL.... but i hope you can use the results :)
The card is running factory OC tuning.....

C:\Users\Menace>e:\Tools\gpuowl-v6.11-364-g36f4e2a\gpuowl-win.exe -iters 300000 -prp 77936867 -maxAlloc 10000 -nospin
2021-10-13 09:09:51 gpuowl v6.11-364-g36f4e2a
2021-10-13 09:09:51 Note: not found 'config.txt'
2021-10-13 09:09:51 config: -iters 300000 -prp 77936867 -maxAlloc 10000 -nospin
2021-10-13 09:09:51 device 0, unique id ''
2021-10-13 09:09:51 gfx1031-0 77936867 FFT: 4M 1K:8:256 (18.58 bpw)
2021-10-13 09:09:51 gfx1031-0 Expected maximum carry32: 583B0000
2021-10-13 09:09:51 gfx1031-0 OpenCL args "-DEXP=77936867u -DWIDTH=1024u -DSMALL_HEIGHT=256u -DMIDDLE=8u -DPM1=0 -DAMDGPU=1 -DMM_CHAIN=1u -DMM2_CHAIN=2u -DMAX_ACCURACY=1 -DWEIGHT_STEP_MINUS_1=0xa.c42d0d7cec038p-5 -DIWEIGHT_STEP_MINUS_1=-0x8.0e50c8817ddf8p-5 -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2021-10-13 09:09:54 gfx1031-0 OpenCL compilation in 2.62 s
2021-10-13 09:09:55 gfx1031-0 77936867 OK 957600 loaded: blockSize 400, 4787ae3610fd9338
2021-10-13 09:09:55 gfx1031-0 validating proof residues for power 8
2021-10-13 09:09:55 gfx1031-0 Proof using power 8
2021-10-13 09:09:56 gfx1031-0 77936867 OK 958400 1.23%; 1127 us/it; ETA 1d 00:06; c6a2c99a13d20cea (check 0.49s)
2021-10-13 09:10:44 gfx1031-0 77936867 OK 1000000 1.28%; 1123 us/it; ETA 1d 00:01; 9bf9d9e6bff4286e (check 0.49s)
2021-10-13 09:14:29 gfx1031-0 77936867 OK 1200000 1.54%; 1124 us/it; ETA 0d 23:58; da42731132f00140 (check 0.49s)
2021-10-13 09:15:33 gfx1031-0 Stopping, please wait..
2021-10-13 09:15:34 gfx1031-0 77936867 OK 1257600 1.61%; 1125 us/it; ETA 0d 23:58; 129a3b2a78942cc0 (check 0.49s)
2021-10-13 09:15:34 gfx1031-0 Exiting because "stop requested"
2021-10-13 09:15:34 gfx1031-0 Bye

[moebius]

Thank you very much, a good replacement for the Vega64, which is no longer available.