Folding Forum

Adam A. Wanderer wrote:Snip

... is Stanford moving away from CPU generated units?

No!

Foxbat wrote:especially if the GP104 doesn't make it on to the GPUs.txt Whitelist for a few months.

I wouldn't worry too much, 7im is normally quite responsive maintaining the GPUs.txt. I would worry more that they are sold out in few seconds.

Adam A. Wanderer wrote: The newer CPUs should also speed up the input of completed units of folding. Or, is Stanford moving away from CPU generated units?

They are not moving away from CPU units but they are currently out of favour, they have a pair of new cores in very early testing but I wouldn't expect to see them anytime soon

Nathan_P wrote:snip

They are not moving away from CPU units but they are currently out of favour, they have a pair of new cores in very early testing but I wouldn't expect to see them anytime soon

"Out of favor" seems a little strong in light of the fact they are developing new CPU cores. Since PG doesn't comment on development cycles, we can expect to see them when they are announced.

Foxbat wrote:After the new Pascal cards become available, I'm assuming Stanford needs to put them through some sort of validation testing procedure to ensure good science. They might make awesome gaming cards, but does that mean that they will necessarily make just as awesome Folding devices?

The word "validation" and your "good science" aren't accurate because that suggests that there's a chance that they'd produce inaccurate results. That possibility is so remote that it's fair to call it impossible. NV wouldn't let that happen.

It's highly likely that there will be a new driver version (or several which may gradually improve performance). It's also possible that the OpenMM code that interfaces with those drivers can be re-optimized somewhat. All in all, the initial benchmarks may not be quite as good as the final performance, but that'll probably be a small improvement.

"Out of favor" might just be a donor attitude.

For those with top-of-the-line GPUs, running CPUs too may seem like too small an increase in science (or in PPD) to be bothered with. For others, GPU+CPU is always more than GPU only.

7im wrote:

Nathan_P wrote:snip

They are not moving away from CPU units but they are currently out of favour, they have a pair of new cores in very early testing but I wouldn't expect to see them anytime soon

"Out of favor" seems a little strong in light of the fact they are developing new CPU cores. Since PG doesn't comment on development cycles, we can expect to see them when they are announced.

The current push is for GPU computational power, most of the recent projects are for GPU's running on core 21, AVX has been in cpu's since 2011 and we still do not have a working core, machines with more than 24 threads have little to no work available for them unless you start running multiple slots. CPU's working on a 1 million atom+ WU get less points than a upper mid range GPU running a 100k atom WU. If that's not out of favour then I don't know what is

I know, everything will be sorted soon

@7im, I am looking forward to the new A7/A8 cores, just a pity the AVX one wasn't out sooner

https://folding.stanford.edu/home/2015-achievements/

From the 2016 Vision section...

New Gromacs core. The Gromacs core continues to be a workhorse core for CPU clients and we are working on releasing an updated core with more broad hardware support, including support for AVX, which will give some donors an automatic performance boost.

Obviously GPU folding was the lowest hanging fruit, but Pande Group doesn't like to leave fruit on the tree if it can help it. But sometimes it takes a while to build the ladder to reach the higher fruits like AVX.

Adam A. Wanderer wrote:... we each must do what we can do.

Well said.

Nvidia released GTX 1080 for benchmarks and here are some FAHbench 1.2 explicit numbers.
Titan X => 84 ns
GTX 1080 => 106 ns
And this card likes overclocking so 2 GHz may be possible.
https://i0.wp.com/cdn.arstechnica.net/w ... 80x720.png

The k-factor might produce insane PPD figures on the new GPU ...

toTOW wrote:The k-factor might produce insane PPD figures on the new GPU ...

Well, the bonus system is designed to reward quicker return of WU, since pascal looks a lot quicker it should get more points. They can't alter the k factor for projects already released - that devalues the donations of existing folders, and if they nerf the points too much on the current cards by lowering the K factor people will either cherry pick WU or slow/shutdown.

Certain WUs could be limited to pascal if its too out of whack. Not that I'm suggesting or want it.

mmonnin wrote:Certain WUs could be limited to pascal if its too out of whack. Not that I'm suggesting or want it.

How could they be "out of whack?" faster results are rewarded as they are considered 'better' by the project, so we are told, so faster is better. Bigger GPU units are coming out so if these cards deliver on their promise , which remains to be seen, then if there is to be any limiting it would seem sensible to direct them to whatever are the more powerful cards. Still all the promise remains that at present, hard data will hopefully not be too far off.

These Cards are more expensive than I thought. Around 840$ here in Switzerland. Not sure if I going to buy one.

Love to see some results for Folding first, before I decide.