Message boards :
Number crunching :
Put in a RX 550. What's the optimal cmdline and where can I read in depth about cmdline options for testing?
Message board moderation
Author | Message |
---|---|
marmot Send message Joined: 15 May 99 Posts: 144 Credit: 1,220,664 RAC: 0 |
Got this card for decent price and it behaves like a WX 3100 (1219/1500 mhz, 4GB RAM and max 33W) even though it's identified as an RX 550 by AMD drivers. It's for the server's PCIe slot but the cooling fan is 3mm from the case (no grill) and it runs at 80C and about 35% slower than when in my overclocked HT Link Gigabyte MB. So it's back in the Gigabyte MB and it's nice to see I made a great purchase (RX 550 is/was top efficiency GPU on Seti@Home). It's running the 0.45b optimized anonymous platform now but I'd like to mess with the cmdline options, vary the inputs and get some data on the RX 550's performance. Would like to get to know the cmdline parameters. Heard that Mike is the go to guy. PS Also picked up a 1060 3gb with highest natural OC made by EVGA for good price on eBay. But it's doing GPUGrid atm. |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
You shouldn`t expect much improvement on a RX 550. You can try this cmdline -sbs 786 -spike_fft_thresh 2048 -tune 1 64 1 4 With each crime and every kindness we birth our future. |
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
PS Also picked up a 1060 3gb with highest natural OC made by EVGA for good price on eBay. But it's doing GPUGrid atm.For that Mike supplied me with, -sbs 1024 -period_iterations_num 10 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64, for SoG and, -unroll 10 -oclFFT_plan 256 16 512 -ffa_block 12288 -ffa_block_fetch 6144, for Astropulse. Cheers. |
marmot Send message Joined: 15 May 99 Posts: 144 Credit: 1,220,664 RAC: 0 |
You shouldn`t expect much improvement on a RX 550. Thankyou. Now where do I read about what each and every cmdline option means, how to vary them and is there a spread sheet of option variances with return results? |
marmot Send message Joined: 15 May 99 Posts: 144 Credit: 1,220,664 RAC: 0 |
PS Also picked up a 1060 3gb with highest natural OC made by EVGA for good price on eBay. But it's doing GPUGrid atm.For that Mike supplied me with, Thanks. It has a lot of WUs to get through to fill out my WUProps DB, not sure if it will get to SETI@Home before next winter. |
marmot Send message Joined: 15 May 99 Posts: 144 Credit: 1,220,664 RAC: 0 |
Assuming the first number is supposed to be a multiple of 256, I changed it to 768 and placed the line in the Lunatics file named: mb_cmdline_win_x86_SSE2_OpenCL_ATi_HD5.txt This rx 550 has 4 GB RAM, not 2GB, so maybe the thresh needs to be 4096? I want to read a guide, understand what these settings mean, and make my own changes. I can't find the guide for the command line settings in the forums. |
Siran d'Vel'nahr Send message Joined: 23 May 99 Posts: 7379 Credit: 44,181,323 RAC: 238 |
Hi Marmot, This page gives information on what you are looking for. That page is Greek to me. ;) Have a great day! :) Siran CAPT Siran d'Vel'nahr - L L & P _\\// Winders 11 OS? "What a piece of junk!" - L. Skywalker "Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath |
marmot Send message Joined: 15 May 99 Posts: 144 Credit: 1,220,664 RAC: 0 |
Thanks, but those are cmdline options to the BOINC client and not to the SETI WU application. Those I've known of for years and aren't applicable in this setting. |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
No, -spike_fft_thresh 4096 is for high end GPU`s Check for a file called ReadMe_MultiBeam_OpenCL_ATI.txt, its located in you projects folder sub docs. Its not a guide but you will find some useful information about each cmdline switch. At the buttom i wrote some examples in the tuning section. With each crime and every kindness we birth our future. |
marmot Send message Joined: 15 May 99 Posts: 144 Credit: 1,220,664 RAC: 0 |
Cool, thanks. Looks like I can create the MultiBeam_ATI_config.xml file which I'd prefer as an aid to my learning curve but it doesn't appear to cover all options. This test machine only has 8GB, ran it out of RAM into a swap crisis, so don't trust the last 12 hours of results. Starting over and will let you know how things go. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
Now where do I read about what each and every cmdline option means, how to vary them and is there a spread sheet of option variances with return results? The explanation for the SoG application parameters and what they mean and do was written by the developer Raistmer and is available here Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
marmot Send message Joined: 15 May 99 Posts: 144 Credit: 1,220,664 RAC: 0 |
Now where do I read about what each and every cmdline option means, how to vary them and is there a spread sheet of option variances with return results? Cool, thanks Keith |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
The one person that I know of that did a real good spread sheet type of post for variances with the SoG options was RueiKe. But he only uses AMD hardware so there could be some differences in the parameter sets because of the architecture differences of AMD and Nvidia. I would look at these posts first to start with. https://setiathome.berkeley.edu/forum_thread.php?id=82043&postid=1895584 https://setiathome.berkeley.edu/forum_thread.php?id=82043&postid=1895483 https://setiathome.berkeley.edu/forum_thread.php?id=82043&postid=1895620 Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
marmot Send message Joined: 15 May 99 Posts: 144 Credit: 1,220,664 RAC: 0 |
You shouldn`t expect much improvement on a RX 550. Sorry it took so long to get back to this. Been working on it every day, actually. Crimson 17.11.4 driver was causing crashes and now on the 19.3 driver. (From years of experience, I tend to keep drivers for GPU to the first year after release of the GPU but the strategy failed this time). This setting works sometimes and other times leads to a driver stall. Actually had 2 valid guppy results that completed in average 31 minutes from: -hp -sbs 256 -period_iterations_num 1 -spike_fft_thresh 2048 -tune 1 64 1 4 memory load 358mb, GPU utilization was 75%, memory controller load 51% and ran 1219 mhz (max) but the card only drew 22.1 watts of potential 33 watts and ran at a cool 60C. 2 more valid completed in 29.5 mins with better credit with: -hp -sbs 320 -period_iterations_num 4 -spike_fft_thresh 2048 -tune 1 64 1 4 memory load 450mb, GPU utilization 55%, memory controller load 31% and 19.8 watts (same temp and frequencies). Couldn't get any -sbs 768 WU's to run at default -period_iterations_num (20) but 2 WU's ran at -period_iterations_num 32. Not sure is this is still an unresolved driver problem or that this card can't handle -sbs 768, It's certainly under using the 4gb RAM and the power potential. |
marmot Send message Joined: 15 May 99 Posts: 144 Credit: 1,220,664 RAC: 0 |
The one person that I know of that did a real good spread sheet type of post for variances with the SoG options was RueiKe. But he only uses AMD hardware so there could be some differences in the parameter sets because of the architecture differences of AMD and Nvidia. I would look at these posts first to start with. Those are very interesting reads. Spent an hour perusing them and learned a lot about how these WU's are being implemented. His card has 8x the CU's of a WX 3100 / RX 550 and 2x the onboard VRAM. Once I get a better understanding of how the data is being divided among the compute units; I should be able to make some use of his conclusions. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.