Page 1 of 1
My contribution in GFLOPS??
Posted: Tue Feb 03, 2009 10:41 pm
by ewal
Hi
Is there any way to find out how much I, or my team, contributes with in GFLOPS?
To me that is far more interresting than points.
Best regards
Erik Walfridsson, Sweden.
Re: My contribution in GFLOPS??
Posted: Tue Feb 03, 2009 10:51 pm
by codysluder
Not directly.
A certain number of GFLOPS does not produce a constant amount of scientific value, particularly when you compare the results produced by one FahCore to those produced by a different FahCore.
Re: My contribution in GFLOPS??
Posted: Wed Feb 04, 2009 7:11 pm
by ewal
No, of course not, but I´m kind of competing with myself. I´m trying to maximize the practically usabele computing power in my LAN. How valuable the particular WU:s processed by my computers become to the project is a little bit over my head.
Anyway, if anyone have any good ideas I would really appreciate it.
best regards
Erik Walfridsson, Sweden.
Re: My contribution in GFLOPS??
Posted: Wed Feb 04, 2009 7:58 pm
by jaak ennuste
There is indirect method and I believe, that PPD and GFLOPS are in linear relationship. Card manufactures give GFLOPS numbers to their cards plus we know average PPD performance from experience (varies by WU's indeed)
NVIDIA GFLOPS PPD
GTX 295 1788 6500 x2
GTX 280 933 6900
9800 GX2 768 5900 x2
9800 GTX 431 6200
8800 GS 396 4305
8800 GT 336 4700
9600 GT 312 3795
8400 GS 43 580
So, You get rough estimate that You get 8 PPD for 1 GFLOPS.
Re: My contribution in GFLOPS??
Posted: Thu Feb 05, 2009 5:51 pm
by codysluder
jaak ennuste wrote:There is indirect method and I believe, that PPD and GFLOPS are in linear relationship.
So, You get rough estimate that You get 8 PPD for 1 GFLOPS.
I don't believe your first statement, but the 8 PPD ~~ 1 GFLOPS is a reasonable estimate.
The assumption that GFLOPS and FAH points or (FAH processing time) are linearly related requires you to assume that each of the FAH cores uses
only Floating Point OPerations (or, more accurately, that all
other processing can be overlapped in such a way that your FPU is
always 100% utilized). On what do you base that assumption?
There is a reasonable correlation but as hardware gains better FP processing speed, other factors become more and more important, including Integer performance, branch performance, memory latency, etc. and I believe we've reached the point where those factors are important to several of the FAH cores.
Re: My contribution in GFLOPS??
Posted: Sat Feb 07, 2009 3:43 am
by ewal
So far I only have CPU:s working for F@H
My fastest GPU is a GeForce 7950 GT and as far as I understand it I need at least GeForce 8xxx.
I have an GeF 8400 GS + RAM and mainboard laying around, but no CPU to go with it yet.
For the moment I only have two machines running F@H on their CPU:s; one Athlon XP 2800+ (approx. 2.6-2.7 GFLOPS theoretically)
and one Pentium 4 630 (3.0 GHz, 2 MB L2 cache) (approx. 3.1-3.2 GFLOPS ..theoretically).
During the night when the computers are running virus scans the avaiable resources for F@H drops down to maybe 30% on averge for an unknown period of time, so it is very har to estimate how much I contribute with ..in GFLOPS, wich is what I´m interested in.
PS I only use nVIDIA GPU:s to be sure they work well with
x-plane
DS