Posts by a0062995

1) Message boards : News : Help in the fight against COVID-19! (Message 93317)
Posted 4 Apr 2020 by a0062995
Post:
So how do you find out it is running less efficent, if the WU is always running 8 hours?
In your Account page, Computers, Details, Application details, Average Processing Rate (APR).
The number of Floating Operations performed each second. The higher, the better (ie, the more numbers crunched per hour, day, week etc).
So far it's looking less efficient, but only by a small amount- about 0.6% at this stage (at least on my system- those with more cores & threads the impacts will probably be greater). However, we would need to run the same Tasks that were processed using the other Applications on this Application for a true comparison- the present Tasks may be "chewier" than the previous ones, and may have run even slower on the previous Application than they do on this new one.


Those values do not quite make much sense.
Is the value stated per Core?
Ryzen 7 3700X 2.87 GFLOPs (8x @3.6 GHz; 30 MB L3; 51.2 GB/s RAM) (okay, old program with 32-bit programm in parallel)
E5-2690 2.74 GFLOPs (8x @3.3 GHz; 20 MB L3; 51.2 GB/s RAM)
i5-4570S 2.75 GFLOPs (4x @3.3 GHz; 6 MB L3; 25.6 GB/s RAM) (only 3 values so far)
E5-2670 v3 2.79 GFlops (12x @2.6 GHz; 30 MB L3; 59.7 GB/s RAM)

If the L3-cache is a bottleneck especially the i5-4570S should be a lot slower it the value is per core.
And the Ryzen should be a lot faster regarding the fact it is newer and should be way more efficient.

I've sent an email to that Doug Refren, maybe he could elaborate a bit more.

I am also open for testing, if you could tell me which project in WCG uses fixed work, not fixed runtime I am happy to run tests for comparison.
2) Message boards : News : Help in the fight against COVID-19! (Message 93305)
Posted 3 Apr 2020 by a0062995
Post:
So how do you find out it is running less efficent, if the WU is always running 8 hours?
How do you know how much "work" you have done for your WU?

I have made these 15 second intervall measurements (Idle Desktop, 1 COVID-19 rosetta, 8 PPSE LLR primegrid WUs, one of those finding a prime i take for comparison measurements of energy consumption, speed and achievable turbo, got quite a lot data for several CPUs by now)





Can't see a lot more misses, on either situation, but am starved for now regarding rosetta WUs.
3) Message boards : News : Help in the fight against COVID-19! (Message 93293)
Posted 3 Apr 2020 by a0062995
Post:
The Rosetta/L3 cache issue has been discussed quite a lot over at the Microbiome Immunity Project (MIP) forums on World Community Grid, a project which also uses Rosetta.
Running more than 1 WU per every 5 MB of L3 cache resulted in increase runtime for all tasks running (non-Rosetta too), which results in loss of efficiency considering the increase in runtime and constant electricity usage. One of the MIP scientists said they were aware of the issue and were looking into it, but this was years ago by now.


I don't think so.
I have one E5-2690 (8 Cores @ 3.3 GHz with 20 MB L3 Cache), one E5-2670 v3 (12 Cores @ 2.6 GHz with 30 MB L3 Cache), one i5-4570S (4 Cores @ 3.2 GHz with 6 MB L3 Cache) and all are crunching WUs with the same amount of time (~29000 seconds).
And even if i have for a time only a fraction of cores running WUs, it would not get faster (if correction for turbo is factored in).
The last 6 WUs on my E5-2670 v3 are run simultanously, they are not faster than the other ones that ran 12-parallel.
So could you give some examples?






©2022 University of Washington
https://www.bakerlab.org