Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Generalized Fermat Prime Search :
Wouldn't it be nice if...
| Author |
Message |
|---|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14045
ID: 53948
Credit: 482,681,090
RAC: 578,366
                               
|
|
...there was a GeneferAVX program that could run on a single CPU core at 20% of the speed of GeneferCUDA running on a GTX-580? (For reference, GenefX64 runs at less than 5% of the speed of a GTX-580).
Just saying.
(More to come. Yves Gallot has been very busy, and there's new versions of Genefer in the pipeline. When it gets to production, this should have a significant impact on all CPU-based GFN calculations. Non-AVX 64 bit CPUs should see a smaller increase in speed, and we may even have a faster alternative to genefer80 in the pipeline. Maybe some OpenCL goodness somewhere down the road, too. Maybe. That's not certain.)
____________
My lucky number is 75898524288+1 | |
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1963
ID: 352
Credit: 6,420,431,564
RAC: 2,590,392
                                   
|
|
Well, Intel i7-3820 would be able to do as much as GTX580 - with lower power consumption...and (maybe) with higher b limits, who would be able to resist?
____________
My stats | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14045
ID: 53948
Credit: 482,681,090
RAC: 578,366
|
Well, Intel i7-3820 would be able to do as much as GTX580 - with lower power consumption...and (maybe) with higher b limits, who would be able to resist?
Yes, it looks as if the overall throughput of a fast CPU may exceed that of a GPU, although each single task will take longer.
I have no information on the new B limits, however. At the present time, GeneferCUDA has higher B limits than GenefX64. I don't know what the B limits are on the new versions.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
So the restrictions on B limits lie with the software? Or, rather, is it that the software ideally has to be coded for "specific" hardware? I use the quotation marks there because I just mean, for example, coded for AVX. Just wondering.
There are a few developers on the forums at the moment talking about writing applications - obviously the decision is theirs and we should be grateful for anything we get, but shouldn't we be looking at AVX2 apps now? It's got the fused multiply-add V.3 instruction set amongst all the other things, surely that would make a significant difference? Haswell is scheduled for... what, April or May, so if someone is starting to write an application from scratch now, then a lot of people will probably be using Haswell-based machines by the time it's out of beta stage and usable for the masses.
I'm a bit ignorant when it comes to programming, so sorry if this is irrelevant. | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14045
ID: 53948
Credit: 482,681,090
RAC: 578,366
|
So the restrictions on B limits lie with the software? Or, rather, is it that the software ideally has to be coded for "specific" hardware? I use the quotation marks there because I just mean, for example, coded for AVX. Just wondering.
It's a combination of everything, the hardware, software, and the algorithm.
I don't really have any useful comment on the AVX2 portion.
____________
My lucky number is 75898524288+1 | |
|
Message boards :
Generalized Fermat Prime Search :
Wouldn't it be nice if... |
