Are smp atoms equal to gpu atoms? The numbers Nathan posted show that the different cores are awarded different points, regardless of the atom count.
Isn't it well known that a gpu can out-simulate a cpu? Perhaps Pande is moving
towards harnessing the power of the gpu with further optimization. I'm hoping for a day
in the near future where a gpu, with the right core, can out perform a 4p system.

To me the big issue is that Pande is giving any bonus at all. Even a small adjustment is huge
considering how many gpu wus are completed everyday.

texinga
For Rjbelans:

It would appear that [H]'s DAB rep is already posting in the DAB to take issue with this beta WU's point structure.  I'm not saying that [H] is necessarily wrong in their position, but since we are a GPU heavy team, what are you going to be saying about our position on the matter?  Unless we agree with [H]'s DAB rep, then we need to offer our position so that it is also heard and considered.

[H]'s post:
http://hardforum.com//sho...74080&postcount=97

and I think the post that their DAB rep is referring to:
http://hardforum.com//sho...71968&postcount=77

I saw the post yesterday already, but thanks for the heads up. I also posted with the same comment I made here that GPUs may just be that much more efficient and any change to the QBR should be looked at across the board, not just a change for one type of hardware.

Now this is FINALLY going in a great direction. With the 4P's getting all the recent love for points, why not offer similar calculations for GPU's if the work completed is relative.
 
All I hope for... Fairness/equality that makes those excited about CPU or GPU farms have a choice of weapons. And not have the huge deficit we see today. In other words... It is time for both types to have something to be excited about and build around as we help humanity...

Afterburner

Now this is FINALLY going in a great direction. With the 4P's getting all the recent love for points, why not offer similar calculations for GPU's if the work completed is relative.

All I hope for... Fairness/equality that makes those excited about CPU or GPU farms have a choice of weapons. And not have the huge deficit we see today. In other words... It is time for both types to have something to be excited about and build around as we help humanity...

Agreed. Time to add some GPU to both my 2P Asus Z8NA-D6C machines. Need to find a couple of GTX 590 to maximize space usage.

It's not only the number of atoms, but also the simulation length (number of steps) that counts.
From what I've seen on some of my SMP WUs (P6943, 6987, 8055) the number of steps is 500,000.
I'm running regular GPU WUs, so I don't know the number of steps for a P8057 WU, but on my GPU a P8054 WU the number of steps is 50,000,000.
 
So that's 100x more than these CPU WUs (Edit: due to the parallel computing architecture of the GPU).
And maybe there are other factors that count.

-ZS-Carpenter
I would not suggest buying any hardware until well after this is out of beta. Remember the first fermi test WUs? 450s getting 15k+ ppd? Wait until this goes live and we see real WUs come out. Fermi cards, with less cuda cores compared to kepler, could fall on their face just like the 450s did. Might be great points for a couple months then the regret will set in.

Yeah, but if you get it now, it comes with Borderlands2 and $20 rebate!
 
 

I don't know if kendrak is a beta tester or not. There are plenty of people running the beta flag now that aren't a part of the beta team. Once the flags became public knowledge, there was no way to control who runs it.

I feel bad for some of the other teams. People put alot of money into P2 and P4 systems. But then again their NOT losing any points like when they pulled the rug out on BA for 8 core and 12core I7's
 
 

I noticed that another Folder (over in the FAH Forum) referenced the post that I'm copying below from Dr. Pande last April.  There are some important clues in that message as to why we could be seeing PG needing to emphasize GPU Folding.  It could also explain the shift we are seeing with the GPU QRB implementation.  It may not be a comfortable shift for some people, but the Folding project itself has to be kept in mind here.  In effect, what are the best moves to grow the overall project, not necessarily what is best to maintain someone's individual points needs.  That is a truth/lesson that we all have to face as Folders.
 
Dr. Pande's post from the FAH forum:

Re: Discussion: what is holding F@h back?

by VijayPande » Wed Apr 25, 2012 5:24 pm
It has been interesting to read this thread. We have had different priorities at different times. Years ago, we had a big push to get to new platforms such as GPUs and true SMP (many other d.c. projects just run multiple WUs on each core, which uses cores, but at least for us is a big loss scientifically compared to true SMP). Now, our main goal is to grow the number of clients so we can get more work done. This includes items like (in no special order):
* Make GPU computing easier. This includes install issues and heat issues, but especially usability issues (GPUs don't have process priority, so efficient GPU code can really monopolize a machine). I've been thinking about bringing the screen saver back for GPUs since that would be a natural way around this. If we could get 100,000 GPUs, we would be at the ~100petaflop level which would be an amazing feat computationally, but more importantly would be a huge resource for FAH's scientific calculations. Towards this end, we have been experimenting with a linux GPU client (no promises or an ETA, but it is something we've been looking into).
* Make running FAH in general easier for newbies/non-experts. The V7 client I think has helped with this, but there's still a lot we can do there.
* Provide more feedback in the client. We've made a push to do more with the blog and I think that's had positive reactions. I think the client could do a lot more as well. The v7 client rewrite took a lot of effort since we started from scratch. Adding in new features won't take nearly as long. I'd also like to improve its GUI to be slicker, more modern, and more visually appealing.

I should stress that this isn't to say that I don't think these are the only issues but that these are the ones on the forefront of my mind these days in terms of the donor experience (there are also tons of other things, in terms of the server backend and the science, but I think that wasn't the main point of this thread).

Also, I should add that I'm not so concerned about the long term prospects along the lines of d.c. going away. There will always be plenty of machines out there, but just not an exponential growth. If we can make FAH much easier to use and less intrusive for donors and make its benefits more clear, I think we can do very well. There are 100M GPUs out there. It would be great to get even 1% of them running FAH!
Prof. Vijay Pande, PhD

jkefalas

Ironically, I am now pulling those nasty 76xx units at 40% drop (as well as 101xx smps), lol. Come on QRB! Maybe this is some sort of initiation?

Same here, they also forced update to ver 2.25 core on my Fermi's with these 76xx's.

Fold On

Seriously wondering if these are some kind of stress test units, or preliminary whatnots. We will see what the rest of the week brings. Hang in there boys!

i got a few 8057's but now I am back on 7625's :(

TheWolf

Same WU I been working all along, so I dough its either.
Its just some badly hacked up core to try to make Kepler look good over others.

I'm running a Kepler with the same borking of points, so these are DEFINITELY not Kepler optimized.  Considering the 8018's had me at 20-21k ppd, and these 76xx's have me at 14k ppd, they are just a poorly rated WU.  Hopefully this is a small project...

jkefalas

Sleinous

No betas yet for me :) And I have quite a big fleet to get one from

Those that were getting them seem not to be any longer, as well.  Wonder what's to come...

EDIT:  My 660ti is scheduled to arrive on Friday.  A weekend release of hearty QRB's would make my weekend! 

+1 Fingers crossed