foldinghomealone wrote:The database is currently under maintanance.
There is an issue that that WU name and path don't match.
I don't know what messed this up, probably it was me.
The PPD and HW stats seem to be ok, though.
Currently I don't have time to set up everything again.
There are too many uploaded single files I have to process manually.
skydivingcatfan wrote:That is interesting. The opposite of what was reported on GamersNexus and the database on TechPowerup.
How did you determine that you had a TU106?
I can tell you from my end that GPUs.txt does not differentiate between TU106 2060s and TU104 versions and reports them all as TU106-200A. I suspect that's what markdotgooley is looking at. GPU-Z should be used to check the actual GPU.
skydivingcatfan wrote:That is interesting. The opposite of what was reported on GamersNexus and the database on TechPowerup.
How did you determine that you had a TU106?
I can tell you from my end that GPUs.txt does not differentiate between TU106 2060s and TU104 versions and reports them all as TU106-200A. I suspect that's what markdotgooley is looking at. GPU-Z should be used to check the actual GPU.
All I can find on this Ubuntu Linux machine is what's on the FAHControl application, and the 2060 KO is called a TU106-200A and the vanilla 2060 is called a TU-106. I don't know where else to look. The NVIDIA X Server Settings calls both merely RTX 2060. Clueless about where else to look.
I would keep it simple and just label it 2060, but change it to TU104. I can tell the current format is the same as that used by devicehunt.com. The format is pretty good, but I would love to know what "GL" & "GLM" stands for if you know. Also, my OCD is being triggered by the inconsistent use of upper and lower case Super/SUPER (the brand uses all upper case FYI), and that some entries don't have any names at all. I notice you are adding the GFLOPS to the end of some of the entries; would you like me to find that info for the rest?
Radeon VII performance, as observed on my side and as mentioned by user NormalDiffussion, of around 1.2MPPD compared to 1.635MPPD mentioned in the table raises questions and makes me consider where the difference might come from. So it would be nice to have information how the 1.635MPPD have been achieved. That is about 35% higher than achieved here and even considering the (double effect of speed on) bonus points for faster return indicates a serious performance difference.
I wonder whether I could improve the performance of the four Radeon VII contributed to the project by me. Information might be helpful for others, too. My setup is stock clocks and voltages for the GPUs, Linux Debian Buster with CPU Ryzen 3700X on one, Ryzen 3900X on the other machine. Graphics driver is upstream kernel amdgpu (kernel 5.5/5.7) with ROCM opencl implementation.
What is performance on Linux with AMD proprietary driver, any experience? What is performance on Windows? Are the results in the table coming from overclocked Radeon VIIs or stock?
Would be appreciated if the contributors of data on Radeon VII for the table could share information about their systems. Thanks!
On another issue the bonus PPD are a decision by the project to increase incentive, but are misleading regarding the true performance and contribution of different GPUs to the project. Thus information on PPD for base points would be nice to have in the table - but that's probably not so easy to achieve?
Thanks for the great contribution with this work. It helps to judge whether the own system performs appropriate.
Two, one Ryzen 3700X, one Ryzen 3900X. The Ryzen 3900X received the second GPU only today, since a power supply upgrade was required to support two GPUs, so for the last days was working with one GPU only. But the PPD for each GPU is approximately the same, independant of how many GPUs are in the computer, so PPD is not limited by CPU overload.