Can Einstein handle 3.5 PFlops offered by NVIDIA's K20? Cuz I'm thinking about buying one.
Language
Copyright © 2024 Einstein@Home. All rights reserved.
Copyright © 2024 Einstein@Home. All rights reserved.
RE: > Fire and brimstone
)
I had good laugh on this! :D Thank you!
> It was funnier in
)
>
It was funnier in Ghostbusters :)
RE: NVIDIA's K20 GPU has
)
Titan should be faster for DC at 1/3 the cost (but still too expensive). 2 x HD 79xx should be faster than either at even less $$$. Power efficiency should be no better than the lower cost alternatives IMO.
I see there are hosts
)
I see there are hosts sporting Titans now - and here's a host with two of them.
RE: I see there are hosts
)
It would be very interesting to know how many WUs they're running concurrently, in fact it would be nice to have this information in the results file. If it's there I'm missing it...
Quick look at those hosts
)
Quick look at those hosts last contact log's suggests 5 at a time per card.
GPU utilisation is set at 0.2 with an effective runtime just shy of 10 mins. per task.
RE: Quick look at those
)
Which means 6 wu's per hour.
My Sapphire HD7870 XT finishes 4 wu's per hour and I paid ~216.- Euro.
Power consumption < 180W when running 2 wu's at a time
Supporting CPU: i3 ivy bridge @3.3GHz, slot: pci-e3 @ x8
http://einsteinathome.org/host/6633241
A look into the statistics / best computers gives a clear picture:
places 1-5 AMD, place 6 a PC with 4 Teslas C2075.
I'm happy with my decision.
5 WUs at one time, ~10
)
5 WUs at one time, ~10 mins/WU equals 30 WUs/hour .....
RE: 5 WUs at one time, ~10
)
Maybe not so clear, it's 50 minutes or so for 5 tasks: 6 tasks/hour.
RE: RE: 5 WUs at one
)
This is how I understood it. The runtime for his wu's are reported as ~ 3000sec which points more to my calculation than to 30 wu's/h