Blog post: "Unified GPU/SMP benchmarking scheme ..."

[Jesse_V]

KWSN_PToT, speed is everything. It increases WU throughput of the project overall. That's why the QRB was introduced by Dr. Kasson years ago and that's why it has been supported ever since. There's this incentive to complete WUs as quickly as possible, which is important because there are serial aspects of F@h's simulations (see the Simulation FAQ). Researchers don't want to start projects if they know it will take a very long time to complete. They want the results back ASAP, so that they can write the paper, publish it, and get on with tackling the next problem in molecular medicine.

This makes GPUs a more attractive client. Everyone with fast GPUs will start getting more points as a reward. Since it's a bonus, those with lower-end GPUs (I own one btw) should, in theory, get no less than they currently are. If you do a search, you'll discover that in the early days of the project 1 WU = 1 point. That's crazy looking back. It's the science performed that really matters in the end. Those that have been staying with the project for so many years have done a lot of science.

[7im]

No, speed is not everything. Science is everything. Speed is only one component.

As you noted, the addition of the QRB was a suppliment to the existing points system at the time. In the past, science = points, and it was linear. But the quickness of the return also increased the scientific value, and that added value went mostly uncredited. Before the QRB, that time factor was not accounted for by the points system.

The addition of the QRB was an attempt to assign points to represent the added scientific value of faster returns of work units. And the QRB, overall, has been helpful to the project. But in hindsight, some sort of loyalty bonus should have been given to existing folders before the QRB was started to normalize the previous contributions with the exponential nature of future contributions. A sort of previous points "conversion" to the new points values. That way 10 years of loyal folding couldn't be outstripped in only 10 days on the new points system. But like they say, hindsight is 20/20, foresight is not.

In my opinion, all existing GPU points should be "QRB" adjusted (1 time bonus) so that past contributions more closely match future contributions, going back to where SMP QRB started. That's only fair, right?! Unfortunately, Pande Group doesn't keep enough WU data to make that happen any more. Oh well. Fold on.

[P5-133XL]

The fairness comes in where you have a single point system rather than two separate point systems. When the GPU point system was created GPU's could not run the same WU's as CPU's, so a number was created saying a specific GPU is going to produce X PPD. We could not really compare the two: It was an apples to oranges comparison. All GPU points since have been roughly based on that X. Now that GPU's can run the same WU's as CPU's they can be benchmarked together against the same WU's and that X can go away. We can now compare apples to apples and that is a good thing.

I really don't see how making GPU's more valuable with bonus points interferes with the upgrade process, even at the low end of the spectrum. I am assuming that GPU's are currently undervalued compared to SMP. We will see when the new project gets released...

I found a good way to upgrade old equipment. As long as you can install a modern video card you can effectively re-purpose old computers for folding. I have a P4 machine that was gathering dust. A P4 gets a couple of hundred PPD running uniprocessor WU's and is really not worth the electricity to fold with. The machine has a PCI-E 1.0 slot so I put in an OC'ed GTX 460 and attached it all to my TV wirelessly and now I have an HTPC machine that can get 14,900 PPD right now folding p8018's with no problem and whenever I want to I can watch Netflicks or Hulu. There is no PPD drop running that P4 as compared to a modern machine with the same video card. That GTX 460 was a $130 upgrade but then the main cost for GPU folding has always been electricity and that has not changed.

[7im]

No, it was not some random arbitrary number. PG would never do anything so reckless.

While they couldn't exactly match up Apple to Apple, they certainly made a very scientific study of all the features of apples, and all the features of oranges, and decided on a very close approximation of equivalency. Now, they can finally do apples to apples.

[rhavern]

Where has it ever been stated that the points system was going to be "fair"? PG use the points system to optimize the science. PG says that faster is better for the science and the QRB represents that statement. The QRB was always intended to be used across the board, it just took this long to implement it on GPU. For dedicated 24/7 folding hardware, this may not make much difference but it will certainly penalize part-time folders, as well it should. Without a doubt, those partially completed GPU WU in the wild are slowing down the science. Why wouldn't PG want to reward faster WU return??

As to rewarding long-term folders, it has long been PG policy to not alter past scores and I fully agree with that. What's done is done. Now if they want to implement a loyalty bonus in addition to the QRB, let's call it the Loyal Folder Bonus (LFB), I'd be okay with that. What is certainly the wrong thing to do is leave the current semi-implemented QRB in place.

[jimerickson]

i for one am very happy with this development. it makes me feel that my investment in gtx480's was not in vain. for me it is a relief knowing there will be full implementation of qrb.

[7im]

Where has it ever been stated that the points system was going to be "fair"? PG use the points system to optimize the science. PG says that faster is better for the science and the QRB represents that statement. The QRB was always intended to be used across the board, it just took this long to implement it on GPU. For dedicated 24/7 folding hardware, this may not make much difference but it will certainly penalize part-time folders, as well it should. Without a doubt, those partially completed GPU WU in the wild are slowing down the science. Why wouldn't PG want to reward faster WU return??

As to rewarding long-term folders, it has long been PG policy to not alter past scores and I fully agree with that. What's done is done. Now if they want to implement a loyalty bonus in addition to the QRB, let's call it the Loyal Folder Bonus (LFB), I'd be okay with that. What is certainly the wrong thing to do is leave the current semi-implemented QRB in place.

Loyalty wouldn't be an adjustment of existing points, but a bonus in addition to those points. So it doesn't break the no adjustment policy. And as you said, they can add any bonus that they feel is in the best interests of the project.

I also have to wonder why you would argue against something that I already said couldn't possibly happen. Paranoid about [h]olding that #1 spot?

[Meh_Lay_Lay]

Oh! I didn't know that the GPUs can now do what the CPUs calculate. Or is that only for GPU3 and Nvidia, and do some restrictions apply?

[P5-133XL]

Oh! I didn't know that the GPUs can now do what the CPUs calculate. Or is that only for GPU3 and Nvidia, and do some restrictions apply?

They are just announcing the testing of that capability. As to the details, we don't know yet since they haven't started public testing. They will undoubtedly start out with either AMD or Nvida with restrictions (V7?) and expand the program over time to include the other brand. If you read the threads, you will find the pro-ATI camp has already decided that Nvidia will get first shot and that this is just another example of AMD discrimination by Stanford. But really no one knows even who gets first shot at these WU's.

[bruce]

Oh! I didn't know that the GPUs can now do what the CPUs calculate. Or is that only for GPU3 and Nvidia, and do some restrictions apply?

No details have been released so it's unclear what, if any, restrictions may apply. In the past, GPUs and the PS3 could only process implicit solvent models which tended to be smaller, faster, and involved a higher degree of approximation. SMP processed explicit solvent models which involved lots more atoms (protein plus solvent atoms) and consequently were slower but more accurate, assuming the same protein needed to be run for the same fraction of a second.

Apparently this is tied to a new version of Gromacs which can be incorporated into FahCores for both SMP and GPUs. With some additional development, this apparently will provide additional capabilities for both platforms. http://folding.typepad.com/news/2012/09 ... w-you.html

I would interpret the "substantial development work over the next few months" to suggest that we may see initial restrictions which are gradually removed over time so any answer to your question would be premature.

[mdk777]

you will find the pro-ATI camp has already decided that Nvidia will get first shot and that this is just another example of AMD discrimination by Stanford. But really no one knows even who gets first shot at these WU's.

I've made $10,000 bets on much less accurate information.

PS, discrimination is not always a pejorative. The last AMD GPU beta launch was in February...and no posts followed up on it.
since many Nvidia betas have launched since, it doesn't take a great prognosticator to predict the future.

"Stanford" will use discrimination in choosing the programing that they have invested the most time and have acquired the most facility in using.
No one said that an established software ecosystem doesn't matter.

Its just a shame to let the hardware become obsolete.

[alancabler]

If you read the threads, you will find the pro-ATI camp has already decided that Nvidia will get first shot and that this is just another example of AMD discrimination by Stanford. But really no one knows even who gets first shot at these WU's.

I take it that you are commenting on the AMD/ATI camp deciding that they aren't getting a fair shake- otherwise, I'd have to say- you should know better, by now.
edit: I know that P5-133XL knows that I know that he gets it and isn't whining...
I always figured that people who got upset that their AMD cpu didn't make as many points as an Intel cpu were just bad at math and/or logic... witness the pretzel logic employed in the "use AMD for benchmark" arguments.

I think this will have a chilling effect on those with older equipment.
There are many dedicated folders (veterans of many years) who are not able to afford to upgrade their hardware every few months/years.
Now, there is an added dis-incentive: Early Return Bonus (for GPUs).

Some members of my team have expressed dismay at this move, and are seriously considering dropping F@H for BOINC/WCG.

As an alternative to Dr. Pande's suggestion of unification (which actually gives unfair advantage to newer/faster hdwe), why not give bonus points for "consistent" contribution (over some period of months or years)???

I happen to be the #1 point-holder on my team (at the time of this post) and I have made a significant investment (recently) in newer hardware.
And still, I think this move is manifestly unfair to others who have provided many years of contribution (but their wallets and/or life-priorities will not allow them to upgrade).

Sorry to be such a Negative-Ned, but...

Unfair advantage to new, fast equipment?

[k1wi]

If you read the threads, you will find the pro-ATI camp has already decided that Nvidia will get first shot and that this is just another example of AMD discrimination by Stanford. But really no one knows even who gets first shot at these WU's.

I take it that you are commenting on the AMD/ATI camp deciding that they aren't getting a fair shake

That's exactly what he stated.

[alancabler]

If you read the threads, you will find the pro-ATI camp has already decided that Nvidia will get first shot and that this is just another example of AMD discrimination by Stanford. But really no one knows even who gets first shot at these WU's.

I take it that you are commenting on the AMD/ATI camp deciding that they aren't getting a fair shake

That's exactly what he stated.

Dang.
you beat me to it... I was gonna edit that post, also tone down what I'd said about lameness and whineyness... don't want to hurt anyone's feelings.

[Brazos]

Well my two cents is that I will be switching from a AMD 5770 to a Nvidia GTX 660 Ti . I've grown tired of waiting for AMD to help out F@H. The race has been won as far as I'm concerned. It's all about the science.