So who's going to buy a Titan for folding?
- Thread starter Voicon
- Start date
Well, when you do I think it'll be close for top spot - just as long as I can keep paying the electric bill (or swipe one of the nuclear reactors from work).
When/if Nvidia sort out the drivers, when the Titans become capable of folding and when Core 17 gets fully released with the new improvements I think we'll see some dramatic points increase all round
On a side note: I seem to have got Cronus back up to full speed - although it is still telling me one of the memory slots has failed - but cracking through a 8101 at under 14mins per frame.......
At one point it said there were three failed memory slots - but I did some fine fiddling (oo err) and got it back to just the one. As it's doing so well I've decided to leave it be
Many thanks to BaBa who supported me through my tinkering process
His Customer Support is second to none
When/if Nvidia sort out the drivers, when the Titans become capable of folding and when Core 17 gets fully released with the new improvements I think we'll see some dramatic points increase all round
On a side note: I seem to have got Cronus back up to full speed - although it is still telling me one of the memory slots has failed - but cracking through a 8101 at under 14mins per frame.......
At one point it said there were three failed memory slots - but I did some fine fiddling (oo err) and got it back to just the one. As it's doing so well I've decided to leave it be
Many thanks to BaBa who supported me through my tinkering process
His Customer Support is second to none
Last edited:
Thats assuming I put the titans to work on Folding, I like to support a few projects and it depends what the best fit is, in the past Seti has been a Nvidia stronghold with GPU being a better fit than CPU
But F@H is better served by 4P monsters than GPGPU from a cost/wu perspective even with the new GPU wu's
Say 4 Titans get 80K each, even if they get 100K and assuming the only thing you have to buy for the rig is the cards, thats 400K ppd for £4K or 100 points/£
A 4P system will cost you £2K and even if you don't overclock it will net you twice as many points for half the price. And the power costs will be about half as well.
The only reason I have a 7970 is to play the odd game and because I like tech, the fact its folding is because I'm a beta tester and like tech.
Weebeastie used to be a gpu folding monster because folding was primarily a gpu platform in those days, then it switched to a Boinc rig (mainly Seti) because thats where it did its best work. (And I folded on the 970X in a VM)
Things haven't changed much now, I'll probably put it on Seti if the client improves, or GPUGrid or some other project that takes advantage of the power of the Titans. I will probably run the Core 17 client to help test it for a while and benchmark etc. But will always return it to what it runs best at in the end.
However with all the change to OpenCL happening around the place it may be that things start to shift to the AMD gpu side (about time imho) then I'll have to rethink.
But F@H is better served by 4P monsters than GPGPU from a cost/wu perspective even with the new GPU wu's
Say 4 Titans get 80K each, even if they get 100K and assuming the only thing you have to buy for the rig is the cards, thats 400K ppd for £4K or 100 points/£
A 4P system will cost you £2K and even if you don't overclock it will net you twice as many points for half the price. And the power costs will be about half as well.
The only reason I have a 7970 is to play the odd game and because I like tech, the fact its folding is because I'm a beta tester and like tech.
Weebeastie used to be a gpu folding monster because folding was primarily a gpu platform in those days, then it switched to a Boinc rig (mainly Seti) because thats where it did its best work. (And I folded on the 970X in a VM)
Things haven't changed much now, I'll probably put it on Seti if the client improves, or GPUGrid or some other project that takes advantage of the power of the Titans. I will probably run the Core 17 client to help test it for a while and benchmark etc. But will always return it to what it runs best at in the end.
However with all the change to OpenCL happening around the place it may be that things start to shift to the AMD gpu side (about time imho) then I'll have to rethink.
Can you explain this a bit more please Biffa?
Stanford are shifting to a single client base model for the gpu client. This means OpenCL and an even playing field in terms of extracting performance from the hardware. Currently this means that using the OpenCL beta client turns the performance tables on its head with AMD 79xx cards being the top performers
Sure AMD has the upper hand atm with a lot of software and games being developed on around their arc but either way it appears this doesn't really matter. Not regarding Folding but check out this article anyway if adoption picks up such as the recent announcement that Adobe Premier Pro will have OpenCL support: "AMD has spent a lot of effort promoting OpenCL in the hope that developers will make use of the GPGPU in the firm's accelerated processing units (APUs). However the firm thinks most developers will shun GPGPU specific languages such as CUDA and OpenCL and stick with what they already know." Source.
Back to Folding, this shows a x2 speed boost to the HD 7970 vs GTX 680 with beta core 17 - link here.
Btw, there's a F@H benchmark out http://www.fahbench.com/
Back to Folding, this shows a x2 speed boost to the HD 7970 vs GTX 680 with beta core 17 - link here.
Btw, there's a F@H benchmark out http://www.fahbench.com/
Article is interesting and true to the extent that the vast majority of programmers are lazy and won't put the effort in to learning how to harness the full power of a gpu with either OpenCl or CUDA.
However those that do, and stanford/berkeley and other gpgpu project programmers do.. well they seem to be starting to lean towards a single model for applications, and the only single Open standard is OpenCL.
Now I fully expect and hope Nvidia pull their damn finger out and get 1.2 drivers out so that we can take full advantage of the new apps, problem is I expect they are actually doing the opposite and trying to convince everyone to stay on CUDA because that way they can lock out any competition from other manufacturers, which makes sense from a commercial perspective, its just a PITA for everyone else.
Yes we reported the potential doubling of performance with beta 17 hopefully soon, but right now Titan is not too good on the current client, you are better off running the stock client. I'll do some of my own tests once I clear out the old cards and get the Titans installed, so the benchmark is currently interesting but not reflective of real-world
performance figures
However those that do, and stanford/berkeley and other gpgpu project programmers do.. well they seem to be starting to lean towards a single model for applications, and the only single Open standard is OpenCL.
Now I fully expect and hope Nvidia pull their damn finger out and get 1.2 drivers out so that we can take full advantage of the new apps, problem is I expect they are actually doing the opposite and trying to convince everyone to stay on CUDA because that way they can lock out any competition from other manufacturers, which makes sense from a commercial perspective, its just a PITA for everyone else.
Yes we reported the potential doubling of performance with beta 17 hopefully soon, but right now Titan is not too good on the current client, you are better off running the stock client. I'll do some of my own tests once I clear out the old cards and get the Titans installed, so the benchmark is currently interesting but not reflective of real-world
performance figures
Yeah but no but...
I did a little bit of folding on one to test it works but the client and drivers are still not optimised enough.
Its even worse on Seti where they don't do any better than a 680 or so. I think some of the gpugrid stuff might be ok with it and I've heard rumours that Milkyway works well if you enable the DP switch in the drivers.
But nope I have to completely dismantle weebeasties water cooling and move the psu to the top of the case to have room for all the cards.
I did a little bit of folding on one to test it works but the client and drivers are still not optimised enough.
Its even worse on Seti where they don't do any better than a 680 or so. I think some of the gpugrid stuff might be ok with it and I've heard rumours that Milkyway works well if you enable the DP switch in the drivers.
But nope I have to completely dismantle weebeasties water cooling and move the psu to the top of the case to have room for all the cards.