Have you thought about doing a version of E@H for GPUs? (Is that even possible with the current generation, or will we have to wait for DirectX 10 before this becomes realistic?)
Well, the GPU version sounds well.
But I have some problem with this:
- GPU programming needs the sourcecode
- ATI and nVidia aren't generous with documentation
- i have never used GPU for programming
- my best videocard is an old TNT2
Have you thought about doing a version of E@H for GPUs? (Is that even possible with the current generation, or will we have to wait for DirectX 10 before this becomes realistic?)
I think the current generation is theoretically capable of it. Akos's tnt2 is manifestly not though.
PS have any DC projects been sucessfully ported to GPU yet?
I decided to dust off my ancient desktop (Pentium III 667 Coppermine) and use S41.07 on it just to see what kind of performance my dinosaur would get. Turns out an average WU will complete in just over 3 hours (11371 s estimated).
I wanted to thank you, Akos, for this amazing work. I never dreamt my 6 year old computer with an optimized client would outperform my Athlon 64 3700+ with a stock client.
AthlonXP 2400+
S41.06 and S41.07 are within 2 minutes of each other on the (now medium) results (50 minute runs).
I'll let it run overnight to get some more work done.
you could try setting it to another "location" on the computerpage under your Einstein account and set the general preferences for that location to use only 1 processor on multi-cpu computers.
That way HyperThreading will still be enabled, but there will only be running a single Einstein process, which should get (nearly) all available cpu-resources (assuming you're only running Einstein on that computer).
Thx for the advice m8, that certainly helped. The best results I got in HT mode was with S40.04, where a "standard" wu took approx. 2.30 hours, and the newer optimized bins was MUCH slower.
Now I set # of cpu's to 1 in the einstein prefs, upgraded to S41.06 and a wu now ran in approx. 55 mins.
So setting the prefs to 1 cpu have now improved performance significantly, and I'm able to use the new and faster optimized bins.
S41.07 is significantly slower than S41.06 on my PIII 801 mhz, waiting for first true S41.07 result, but so far this one crunching atm, also seems MUCH slower with S41.07
RE: Have you thought about
)
Well, the GPU version sounds well.
But I have some problem with this:
- GPU programming needs the sourcecode
- ATI and nVidia aren't generous with documentation
- i have never used GPU for programming
- my best videocard is an old TNT2
RE: Have you thought about
)
I think the current generation is theoretically capable of it. Akos's tnt2 is manifestly not though.
PS have any DC projects been sucessfully ported to GPU yet?
RE: PS have any DC projects
)
Folding@Home are working on it, but they haven't released a client app yet though.
I decided to dust off my
)
I decided to dust off my ancient desktop (Pentium III 667 Coppermine) and use S41.07 on it just to see what kind of performance my dinosaur would get. Turns out an average WU will complete in just over 3 hours (11371 s estimated).
I wanted to thank you, Akos, for this amazing work. I never dreamt my 6 year old computer with an optimized client would outperform my Athlon 64 3700+ with a stock client.
S41.06 got one Access
)
S41.06 got one Access Violation (0xc0000005) : resultid=27808140
One out of a few hundred results though.
AthlonXP 2400+ S41.06 and
)
AthlonXP 2400+
S41.06 and S41.07 are within 2 minutes of each other on the (now medium) results (50 minute runs).
I'll let it run overnight to get some more work done.
SpeedUP -> S41.06 vs original
)
SpeedUP -> S41.06 vs original -> 5.433x
short WU
aplication time
------------------
original 4553s
S41.06 838s
------------------
on ATHLON XP 3200+ (Barton)
report on
)
report on S41.06/07:
valid:77
invalid:0
RE: you could try setting
)
Thx for the advice m8, that certainly helped. The best results I got in HT mode was with S40.04, where a "standard" wu took approx. 2.30 hours, and the newer optimized bins was MUCH slower.
Now I set # of cpu's to 1 in the einstein prefs, upgraded to S41.06 and a wu now ran in approx. 55 mins.
So setting the prefs to 1 cpu have now improved performance significantly, and I'm able to use the new and faster optimized bins.
Once again MANY MANY thx for your advice.
First mixed
)
First mixed S41.06/S41.07.
S41.07 is significantly slower than S41.06 on my PIII 801 mhz, waiting for first true S41.07 result, but so far this one crunching atm, also seems MUCH slower with S41.07