New Gpu Stats

[nada391]

Hello ,

i a, receveing a new GPU;

is there a way to track the performance of the GPU vis-a-vis the currently in use?

have seen that there is a Stats Page in v.8-5; but it seems not listing stats on respective GPU in use;



thanks.


(GPU is W6400)

[muziqaz]

Hello ,

i a, receveing a new GPU;

is there a way to track the performance of the GPU vis-a-vis the currently in use?

have seen that there is a Stats Page in v.8-5; but it seems not listing stats on respective GPU in use;



thanks.


(GPU is W6400)

FAHclient only shows historical data, what you already folded, and it will not show you performance of that GPU if you have not folded at all.

[appepi]

Is there a way to track the performance of the GPU vis-a-vis the currently in use? ... (GPU is W6400)

I know nothing about ATI/AMD GPU's later than the ATI Firepro V3700 of 2008 that I bought on eBay in a batch of 8 cheap GPUs in 2016 and which died just the other day. However, at Techpowerup the entries for https://www.techpowerup.com/gpu-specs/r ... 6400.c3873 and https://www.techpowerup.com/gpu-specs/r ... 6400.c3813 (both NAVI 24) both have general performance similar to that of a NVIDIA GTX 1050.

At https://folding.lar.systems/gpu_ppd/overall_ranks you won't find Folding performance stats for any GPU with "W6400" in its title, but a search on "6400" will give ranks of 233 (PPD 365K, N=281) and 263 (PPD 216K PPD, N=500 most recent samples out of 1,232) in the GPU database for a RX6400/6500XT. The same database shows ranks of 240, 248, 252 for NVIDIA GTX1050, exluding 1050Ti entries) around 300K PPD.

LAR SYSTEMS data covers a wide range of "real world" folding performance, which will depend not only on WU's but also on on user settings and operating procedures. For example I limit temperatures on my GPUs to a target of 70 Deg C and otherwise the settings are factory standard, they generally do nothing but fold and without concurrent non-folding CPU work, they generally run for at least 9 hours per day and WU's are rarely paused, they are usually running Windows. Performance generally falls within the fairly wide ranges I find in the LAR database and I explore further only if it is consistently higher or lower for a particular GPU or WU.

In summary: My best guess is that you might expect maybe 300K Points (assuming that FaH's Quick Return Bonus applies) on a good day (24 hours, otherwise in proportion to run time ) but don't be too surprised by occasional variations down to 200K PPD. [For example, one of my devices (RTX 2060 running in a HP Z440 under W10, 8pm-7am but allowed to finish whatever is still running at 7am) has turned in "daily" performance between 950-1350 KPoints over the last month see https://folding.extremeoverclocking.com ... &u=1301962. [Note: each of my folding devices is defined as a separate "user" within my "Team" since I did not know any better back there in 2020.]

[nada391]

alright,

it seem can indicate along the "PPD" number;

if the actual grpahics card gets number "194.000" and the newer card gets higher number, it should say that the new card is more apt at folding?

[appepi]

if the actual grpahics card gets number "194.000" and the newer card gets higher number

Other things being equal, more PPD is better.

If that "194.000" means 194 thousand Points Per day (PPD) that has been estimated and reported by the FaH app for a particular job (WU) while it is running, and the "higher number" is for another job (WU) while it is running on the other card, then you have one observation of the difference, and a higher PPD is better. Even so, unless it is a huge difference you would need to repeat the observations because WUs vary in difficulty and you might just be looking at a WU that suited the newer card better.

BUT, if that "194.000" means 194 thousand Points awarded for a particular job (WU) after it finished, and the newer card was awarded more for another job (WU) after it finished then it means nothing, because (for example) one might be for a really complex WU that took 1 hour on the old card and the other for a really simple WU that took 5 hours on the new card.

[nada391]

am referring to the PPD number that is shown in webview-homeclient

https://ibb.co/gM1WKj2N

in this example , the "server" gets PPD of around 2.200; which is comparable to the PPD (not shown here) of 980 that another SBC received; which is around 1/3 cpu strength, which account probably to real-world-comparision;
but previously, the same "server" has shown an PPD of aound 36.000, no probably this number is not so significant;

have not fully understood yet, how this sistem works;

[appepi]

It is a while since I ran a trial of the Version 8 Folding app, because I decided to stay with version 7.6.21 because it better suits the way I run things. Thus it would be better for someone who used version 8 to explain this.

But if I remember rightly, then referring to the image you posted, the short answer is that the 224,409 PPD for project 18267 (presumably your new W6400) is 14% better performance than the 196,913 PPD for project 18266 running on your NVIDIA (Quadro) T400, but there isn't much between them and they were running WUs from different projects and while it seems they were both running CPU WUs concurrently, these were allocated 15 CPUs vs 3 CPUs so that's another variable. You would need to monitor this for a while if the difference is important. From a practical point of view I would treat them as equivalent.

[nada391]

thanks appepi, the gfx1102 should be the radeon 7600, it should be quite more capable than t400, so this whole theory PPD doesnt seem to make much sense for the moment.

will keep an eye to it and see if can make sense of oit after making the gpu switch;

[appepi]

Yes, a Radeon 7600 should certainly generate heaps more PPD than a Quadro T400.

I happened to be up in the attic starting the Asus Turbo RTX 2060 in Z443 on its evening run (8pm - 7am), and I logged onto the forum while here. The RTX 2060 picked up a WU from project 18262, which is poorly rewarded on the 2060 so it is only getting an estimate of 1.76 Million PPD on this one, as against its usual 2.2 Million PPD. However, the LAR systems Fah 7.6.21 Chrome extension also says that the average of 20 returns for 2060s on WU for this project is about 2.0 M PPD, so mine is underperforming by 14%. This might be because I limit temperature to 70 Deg C with ASUS GPU Tweak 3, so it is averaging 145W power as against 160 max. But normally I only take a performance hit of a few percent with these settings.

Anyway, Techpowerup says that the RTX 2060 only has 72% of the performance of your Radeon 7600, and while this ratio of relative performance in their graphic card benchmarking may not apply to Folding performance, if it did you should be getting 2.2M x 100/72 = about 3 M PPD. Unfortunately the LAR systems GPU rankings at https://folding.lar.systems/gpu_ppd/overall_ranks don't have a Radeon 7600, just the (S) variant which is a mobile version and significantly less powerful, but even that says 1.5 M PPD.

My overall point is just that if you follow the average FAH performance of a device as measured by PPD and compare it with others you will soon know what to expect and be able to tell if anything unexpected is occurring - where "unexpected" means "outside the substantial variation".