PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Generalized Fermat Prime Search : Because I am on a weird "let's try this" mode (Message 153435)
Posted 162 days ago by SeanHsu
https://www.primegrid.com/forum_thread.php?id=9821&nowrap=true#153138

Yves Gallot said:
Multithreading is not needed for GFN-17. The data size is 2 MB. On Intel processors the L3 cache size is #cores * 2MB and on AMD CPU it is #cores * 2MB or * 4MB or more.


I guess it's meant to utilize as much cache available on your CPU to improve your throughput.
2) Message boards : Problems and Help : Radeon RX 6700XT Error on GFN-15 trough GFN-19 (Message 152494)
Posted 207 days ago by SeanHsu
Good news! Update 21.12.1 fixed the error.
https://www.amd.com/en/support/kb/release-notes/rn-rad-win-21-12-1

The new OpenCL driver (3354.13) that AMD released in December is doing OCL2 transform properly!
Say yes to previously blacklisted GFN tasks again!

... Or after the Geminids Shower Challenge. (XD)
3) Message boards : Problems and Help : Radeon RX 6700XT Error on GFN-15 trough GFN-19 (Message 152273)
Posted 227 days ago by SeanHsu
I run a residue test for OCL2, and get random residues every time.
(6600XT, driver 21.10.2)

.\geneferocl_windows_3.3.3-2.exe -r -x ocl2

Output:
geneferocl 3.3.3-2 (Windows/OpenCL/32-bit) Copyright 2001-2018, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke Copyright 2010-2012, Shoichiro Yamada, Ken Brazier Copyright 2011-2014, Michael Goetz, Ronald Schneider Copyright 2011-2018, Iain Bethune Genefer is free source code, under the MIT license. Command line: C:\boinc\projects\www.primegrid.com\geneferocl_windows_3.3.3-2.exe -r -x ocl2 Normal priority change succeeded. Running tests for transform implementation "OCL2" Testing 10000^32+1... Using OCL2 transform Running on platform 'AMD Accelerated Parallel Processing', device 'gfx1032', vendor 'Advanced Micro Devices, Inc.', version 'OpenCL 1.2 AMD-APP (3302.6)' and driver '3302.6 (PAL,LC)'. 16 computeUnits @ 2359MHz, memSize=3072MB, cacheSize=16kB, cacheLineSize=64B, localMemSize=64kB, maxWorkGroupSize=256. Starting initialization... Initialization complete (0.001 seconds). Expected residue [8d38b1fa99437373] does not match actual residue [118b70e8f7b28b0f] maxErr exceeded for 10000^32+1, 1.0000 > 0.4500 during final check

Second time:
Expected residue [8d38b1fa99437373] does not match actual residue [0005b8ef3b09d4a9]

Third time:
Expected residue [8d38b1fa99437373] does not match actual residue [31e2a6ac297e6c14]

What else can I test? OCL, OCL3, OCL4, and OCL5 transforms all pass the residue tests for small Ns, didn't wait for larger Ns.
4) Message boards : General discussion : Windows 11 discussion (Message 152078)
Posted 241 days ago by SeanHsu
Prime95 benchmark throughputs are about the same on Windows 11 + update KB5006746 + AMD Chipset Driver 3.10.08.506, which solved the L3 latency issue on Ryzen processors. Earlier in this post I wrote they were 5% lower, but I forgot to turn off some background tasks before benchmarking. Now it performs as good as before and I am happy.
5) Message boards : Number crunching : best wu for thread ripper? (Message 151985)
Posted 246 days ago by SeanHsu
hmm very odd. i try a 16 limit. and it stop giving me wu on any pc.
will further look into this.
just wanted to let you two know about it(bug i think)


You can reset the job limit in PrimeGrid preferences.

According to WikiChip, Threadripper 2990WX has 8x8MiB L3 cache. Each 8 MiB is shared between 4 cores within a CCX. You can use 4 cores (4T or 8T) to crunch up to 1024K=1M FFT within the L3 cache. 1024K per 4 cores.
6) Message boards : Problems and Help : Temperature problems. (Message 151868)
Posted 253 days ago by SeanHsu
@Dennis Menace That is impressive! For around 115W, your temperature delta is 33 degrees (78-45), but my delta is 56 degrees (88-32). I think it's partly because of air vs. water cooling, and partly because of the power density? My CPU consumes 12-13W per core under manual overclocking. (Another CPU, 3300X, can easily do 14W per core.) In this situation my margin is too thin to run 16 vs. 8 PPS-MEGA tasks (or SGS.)

I did find that bigger LLR tasks that prefer multithreading, such as GCW (Martin Gardner's Birthday Challenge!!), do quite a bit better on SMT like other people said, based on this Prime95 benchmark result.

Timings for 1920K all-complex FFT length (8 cores, 1 worker): 1.00 ms. Throughput: 997.51 iter/sec. Timings for 1920K all-complex FFT length (8 cores, 2 workers): 1.65, 1.67 ms. Throughput: 1203.07 iter/sec. Timings for 1920K all-complex FFT length (8 cores hyperthreaded, 1 worker): 0.92 ms. Throughput: 1086.86 iter/sec. Timings for 1920K all-complex FFT length (8 cores hyperthreaded, 2 workers): 1.60, 1.60 ms. Throughput: 1249.40 iter/sec. Timings for 2240K FFT length (8 cores, 1 worker): 1.19 ms. Throughput: 840.72 iter/sec. Timings for 2240K FFT length (8 cores, 2 workers): 2.52, 2.51 ms. Throughput: 795.10 iter/sec. Timings for 2240K FFT length (8 cores hyperthreaded, 1 worker): 1.09 ms. Throughput: 915.97 iter/sec. Timings for 2240K FFT length (8 cores hyperthreaded, 2 workers): 2.52, 2.63 ms. Throughput: 776.19 iter/sec.


And my CPU running 1-2 of these tasks using all 16 threads is less hot (100W) than running 8 PPS-MEGA tasks, which helps a lot.

In my last post I recommended Bill2's Process Manager for single-threaded tasks, but now I can recommend Process Lasso again, for its Instance Balancer works with spreading 2 or more multithreaded tasks across the cores, as long as I don't "disable SMT" for them, which did break the Instance Balancer function... But as I said, MT tasks benefit from SMT and run less hot, so I will not worry about temperatures now.
7) Message boards : Problems and Help : Radeon RX 6700XT Error on GFN-15 trough GFN-19 (Message 151780)
Posted 260 days ago by SeanHsu
Ubuntu 20.04.3 with "amdgpu-pro-21.30-1290604-ubuntu-20.04" driver,
Works very well.
https://www.primegrid.com/result.php?resultid=1263537497
<core_client_version>7.16.6</core_client_version> <![CDATA[ <stderr_txt> geneferocl 3.3.3-2 (Linux/OpenCL/64-bit) Copyright 2001-2018, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke Copyright 2010-2012, Shoichiro Yamada, Ken Brazier Copyright 2011-2014, Michael Goetz, Ronald Schneider Copyright 2011-2018, Iain Bethune Genefer is free source code, under the MIT license. Running on platform 'AMD Accelerated Parallel Processing', device 'gfx1032', vendor 'Advanced Micro Devices, Inc.', version 'OpenCL 2.0 ' and driver '3314.0 (HSA1.1,LC)'. 16 computeUnits @ 2900MHz, memSize=8176MB, cacheSize=16kB, cacheLineSize=64B, localMemSize=64kB, maxWorkGroupSize=256. Supported transform implementations: ocl ocl2 ocl3 ocl4 ocl5 Command line: ../../projects/www.primegrid.com/geneferocl_linux64_3.3.3-2 -boinc -q 134438730^65536+1 --device 0 Normal priority change failed (needs superuser privileges. Checking available transform implementations... OCL transform is past its b limit. OCL3 transform is past its b limit. OCL4 transform is past its b limit. OCL5 transform is past its b limit. Using OCL2 transform Starting initialization... Initialization complete (0.060 seconds). Testing 134438730^65536+1... Estimated time for 134438730^65536+1 is 0:01:46 134438730^65536+1 is complete. (532711 digits) (err = 0.0000) (time = 0:01:48) 17:14:47 17:14:47 (39062): called boinc_finish </stderr_txt> ]]>
8) Message boards : Problems and Help : Radeon RX 6700XT Error on GFN-15 trough GFN-19 (Message 151773)
Posted 261 days ago by SeanHsu
you say that extracting the amdocl.dll from a working package is not gonna work?

Not for my GPU (that didn't exist then). Chances are it will work for a 6700XT and higher. But the process involved is a little messy. Extract the dll. Find the working copy under C:\Windows\System32\DriverStore and replace it (need to change ownership and permissions). If it's locked because some program is using it, go to Device Manager and disable the Graphics device. It will fall back to Microsoft Basic Adapter, and now you can replace the dll and re-enable the device and do the testing.
9) Message boards : Problems and Help : Radeon RX 6700XT Error on GFN-15 trough GFN-19 (Message 151769)
Posted 261 days ago by SeanHsu
I can confirm that the 2020-21.4.1-win10-64bit-apr20 driver works
This was the one that worked for me when i ran successful tasks

I will test 21.5.2 soon as i get home :)


Seeing your testimony is enough. No need to revert to 21.5.2, then.
    OpenCL 1.2 AMD-APP 3240.6 (PAL,LC) from Radeon Software 21.4.1, 21.5.1, and 21.5.2 -- works



We find the culprit -- it's the June update 21.6.1 and the OpenCL 3276.6 (PAL,LC) driver for the RX 6000 series cards. RX 5000 series are fine.

PAL=Platform Abstraction Library
HSAIL=HSA (Heterogeneous System Architecture) Intermediate Language
LC=(Legacy?)

I should also test Linux...

10) Message boards : Problems and Help : Radeon RX 6700XT Error on GFN-15 trough GFN-19 (Message 151763)
Posted 261 days ago by SeanHsu
OpenCL 1.2 AMD-APP (3276.6), but on other GPUs, returned good results.

https://www.primegrid.com/result.php?resultid=1262643841 RX 5700 XT (3276.6 (PAL,LC))
https://www.primegrid.com/result.php?resultid=1262763778 RX 590 (3276.6 (GSL))
https://www.primegrid.com/result.php?resultid=1261446620 RX Vega (3276.6 (PAL,HSAIL))
https://www.primegrid.com/result.php?resultid=1263259145 Radeon VII (3276.6 (PAL,HSAIL))

OpenCL 1.2 AMD-APP (3240.6), but on the R9 290X and the R9 Fury, also worked.

https://www.primegrid.com/result.php?resultid=1262867574 R9 290X (3240.6)
https://www.primegrid.com/result.php?resultid=1263101266 R9 Fury (3240.6)

Now it narrows down to the RX 6000 series, on Windows 10, with the 3276.6 (PAL,LC) OpenCL driver, and maybe the 3240.6 (PAL,LC) driver.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2022 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 2.45, 2.39, 2.53
Generated 30 Jun 2022 | 19:04:44 UTC