I ended up purchasing GeForce GTX 550 Ti and the computation speed has increased twelvefold.
since your AMD is a pretty old CPU, you should try to limit boinc to use only 50% of your CPU's. GPU-tasks should run faster (maybe less than an hour per WU) then..
I've seen that thread but as both I and Dagorat wrote there is no official timeline yet.
You might have some inside info but I don't.
And I just thought I would mention that work for GPUs might be scarce around here until a new app is released.
No harm done! =)
Sorry .... jet lag.
This is the normal time line for a new app, we test before full release ... :-)
In any case don't be nervous about work ever drying up for either CPU's or GPU's, as any of the data sets looked at in GW/radio/gammas can always be re-looked at with new templates, Doppler's, dispersion estimates etc. The dev's certainly know to bring them to us 'in time' so that the E@H crunching will continue unabated. :-)
As for inside info : let's look inside the Atlas cluster which IIRC has about ~ 1500 NVidia's ( more were being added when I was there )
under the hood in lots of four per node :
which is what Bernd was referring too. As a node has a quad core ( E5504 @ 2.00GHz ) then it's CPU:GPU ( or GPGPU to be exact ) of 1:1.
Probably another five years worth IIRC was quoted recently for S5 alone - though no one seems too excited about the quality ( noise ) of S6. The BRP's particularly have a pretty limitless parameter space, meaning that any of the beam sets that have already yielded fruit can be guessed at again. The last three BRP discoveries had pretty high dispersion moduli ( dispersion means that on the way the signal spreads out so higher frequencies arrive earlier ) in effect a proxy for distance ( more electrons in the column of space b/w us and the pulsar ) so you can just go deeper and deeper looking for what's fainter or weirder or ..... :-)
Cheers, Mike.
( edit ) Oh, and these Tesla cards in particular are 'professional' versions meaning they have some measure of guarantee rather better than a 'consumer' one. Yep, you pay extra for that ...... presumably the chips would be the pick of the bins etc. No, I didn't find out a price that Atlas got them for. Actually it's pretty weird feeling walking around inside a multi-million dollar/euro room! :-)
( edit ) Don't hold me to it but : I think I recall an issue mentioned with the node's mobo PCI-express channels choking a tad when four cards are in versus three .... so that implies at least occasional contention b/w the cards and/or CPUs for access.
( edit ) A quick check shows them at ~ $3200 AUD per card !:-)
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
The only difference I see between those 2 cards is the price and the video outputs. Make sure the one you get has an output connector that matches your monitor's input connector.
Regarding availability of work, if Einstein runs out of GPU tasks you can always use it at GPUgrid.
I've yet to see an all DVI card that didn't have at least one DVI-VGA adaptor in box, so that's not a major concern. In the US I'd strongly endorse EVGA for having a superior warranty; but those policies vary by market and I've no idea what their EU branch does. EVGA has a better engineering reputation as well; but that matters more for highly customized cards than what looks like it might be a stock design.
Alright, I'm buying one of these two video cards - the price difference is small but I have no idea which one is better for crunching. Here's what I'm eyeing:
I do like the way things work around here, test before release. I'd rather go without work for a while than have new software rushed out that's not ready yet.
I have no idea how long older cards will be suitable for Einstein but it's only a matter of time before they are not suitable here either, due to insufficient RAM, inability to meet the deadline or obsolete compute capability.
My experience is that E@H strives to include as many people as possible. I doubt that many currently functioning GPU will be made obsolete any time soon. I am using a little GT220 and, while the high-end cards blow it out of the water, I don't expect it will stop being useful for quite a while. In fact, I assume I will burn it out long before E@H makes it obsolete.
I have no idea how long older cards will be suitable for Einstein but it's only a matter of time before they are not suitable here either, due to insufficient RAM, inability to meet the deadline or obsolete compute capability.
I tried running the 1.07 BRP3cuda32 app on my 32bit XP3200 with it's PCI 8400 GS the other week, it wasn't fast, but it did complete and validate O.K, and would have been a lot faster than doing those Wu's on the CPU,
Now I'm a little late with
)
Now I'm a little late with this link. But for all, who are searching for videocards, this is good ranking list.
http://www.pc-erfahrung.de/grafikkarte/vga-grafikrangliste.html
@Mike: I've seen that
)
@Mike:
I've seen that thread but as both I and Dagorat wrote there is no official timeline yet.
You might have some inside info but I don't.
And I just thought I would mention that work for GPUs might be scarce around here until a new app is released.
No harm done! =)
RE: I ended up purchasing
)
since your AMD is a pretty old CPU, you should try to limit boinc to use only 50% of your CPU's. GPU-tasks should run faster (maybe less than an hour per WU) then..
RE: @Mike: I've seen that
)
Sorry .... jet lag.
This is the normal time line for a new app, we test before full release ... :-)
In any case don't be nervous about work ever drying up for either CPU's or GPU's, as any of the data sets looked at in GW/radio/gammas can always be re-looked at with new templates, Doppler's, dispersion estimates etc. The dev's certainly know to bring them to us 'in time' so that the E@H crunching will continue unabated. :-)
As for inside info : let's look inside the Atlas cluster which IIRC has about ~ 1500 NVidia's ( more were being added when I was there )
under the hood in lots of four per node :
which is what Bernd was referring too. As a node has a quad core ( E5504 @ 2.00GHz ) then it's CPU:GPU ( or GPGPU to be exact ) of 1:1.
Probably another five years worth IIRC was quoted recently for S5 alone - though no one seems too excited about the quality ( noise ) of S6. The BRP's particularly have a pretty limitless parameter space, meaning that any of the beam sets that have already yielded fruit can be guessed at again. The last three BRP discoveries had pretty high dispersion moduli ( dispersion means that on the way the signal spreads out so higher frequencies arrive earlier ) in effect a proxy for distance ( more electrons in the column of space b/w us and the pulsar ) so you can just go deeper and deeper looking for what's fainter or weirder or ..... :-)
Cheers, Mike.
( edit ) Oh, and these Tesla cards in particular are 'professional' versions meaning they have some measure of guarantee rather better than a 'consumer' one. Yep, you pay extra for that ...... presumably the chips would be the pick of the bins etc. No, I didn't find out a price that Atlas got them for. Actually it's pretty weird feeling walking around inside a multi-million dollar/euro room! :-)
( edit ) Don't hold me to it but : I think I recall an issue mentioned with the node's mobo PCI-express channels choking a tad when four cards are in versus three .... so that implies at least occasional contention b/w the cards and/or CPUs for access.
( edit ) A quick check shows them at ~ $3200 AUD per card !:-)
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: Actually it's pretty
)
Yeah, working it one's fun too at first; but like anything else you get used to it eventually.
RE: The only difference I
)
I've yet to see an all DVI card that didn't have at least one DVI-VGA adaptor in box, so that's not a major concern. In the US I'd strongly endorse EVGA for having a superior warranty; but those policies vary by market and I've no idea what their EU branch does. EVGA has a better engineering reputation as well; but that matters more for highly customized cards than what looks like it might be a stock design.
RE: Alright, I'm buying one
)
My preference is towards EVGA as they have decent quality support and they have a neat tool called Precision for overclocking and fan control.
Thank you Mike for the info
)
Thank you Mike for the info and the pictures! =)
I do like the way things work around here, test before release. I'd rather go without work for a while than have new software rushed out that's not ready yet.
RE: I have no idea how long
)
My experience is that E@H strives to include as many people as possible. I doubt that many currently functioning GPU will be made obsolete any time soon. I am using a little GT220 and, while the high-end cards blow it out of the water, I don't expect it will stop being useful for quite a while. In fact, I assume I will burn it out long before E@H makes it obsolete.
RE: I have no idea how long
)
I tried running the 1.07 BRP3cuda32 app on my 32bit XP3200 with it's PCI 8400 GS the other week, it wasn't fast, but it did complete and validate O.K, and would have been a lot faster than doing those Wu's on the CPU,
Claggy