Bigadv points changes

[Dead Things]

Originally Posted by 3.0charlie

re. the bolded section. In order to support this statement, I wonder on a weekly basis how many WUs are completed, per Client. Then we would find out if that ratio of completed WUs per Client is really an issue for Stanford.

This separates GPUs from CPUs and PS3s. But the CPU numbers combine bigadv and regular SMP and uniproc.

As you can see, even when all CPU-bound clients are combined, they still come up way short of the TFLOPS from GPU and about on a par with PS3 folding. Plus, when you consider that the vast majority of the 100 million points per week posted by Default is from uniproc clients, it really goes to show just how insignificant SMP and bigadv folding is in the big picture.

[10e]

Originally Posted by lowfat

What kind of PPD are you getting w/ your 48-core machine now w/ the standard bigadv's?

A 6901 for example used to give me (4x6168 Opterons@1.9Ghz) around 305,000 PPD, and now it's down to something like 245,000 PPD.

The 6903s used to give over 500,000 PPD and are now down to 396,000 PPD

4DoorGTZ has a faster set of procs so he would do a fair bit better.

Originally Posted by 3.0charlie

I wrote this to our Captain earlier this week.



We are Stanford's clients. Talk to us. Discuss. Benchmark. Then take action. But don't change the point attributed to WUs after the fact..

Unfortunately it may happen again, and very shortly too. If indeed BD becomes the reality that is everyone's Dream, those same rigs that are responsible for this points change will (probably, hopefully) crush the record tpfs we've been seeing lately.

1M ppd rig? It ain't too far.

On the 6901s in Linux 4x16-core BD processors of unknown speed were getting a 3:50 TPF which was over 800,000 PPD.

So I'd say 1m PPD is still easily a reality even with the latest points nerf.

Originally Posted by chrisk

Stanford is always changing things up which is why I try to be careful when buying parts. A new WU, new client, etc often comes along and shakes things up.

But as DT said, this time it is a bit different as we are not talking about $100-$200 cards, but folks who have dropped $2000-$3000 on relatively efficient folding rigs.

I think Stanford put themselves in a hard spot. They still need the science done by GPU and other clients, and folks were migrating quickly to Bigadv and Bigbeta rigs. But if Stanford wants that science done, I think it would be wise to now give more points to GPU folding than in the past. Lots of heat, noise, and mostly, electricity. Electricity costs are skyrocketing, so there will be additional anger from folks who have Bigadv setups and like them for the lower power and heat.

Like I mentioned before, I think that instead of killing PPD for the 'Big' clients, they should just have setup the bonus system for GPU clients too. That way, folks spending big money on 1 million PPD rigs would still reach that goal, while the GPU clients would get some additional attention.

I'd agree with that. I'm not happy about it nor am I really unhappy about it because I've had the 48 core rig for some time now, but for those that just took the leap and got a 48 core encouraged by the egregious PPD output I could see they would be miffed.

My one exception where I am truly NOT happy is that they also nerfed the 2684s which they just had upgraded. This was stupid. I don't get why, as these units take WAY too long as it is. WTF?

So Stanford giveth and taketh away, and their donors might do the same.

[3.0charlie]

Originally Posted by Dead Things

This separates GPUs from CPUs and PS3s. But the CPU numbers combine bigadv and regular SMP and uniproc.

As you can see, even when all CPU-bound clients are combined, they still come up way short of the TFLOPS from GPU and about on a par with PS3 folding. Plus, when you consider that the vast majority of the 100 million points per week posted by Default is from uniproc clients, it really goes to show just how insignificant SMP and bigadv folding is in the big picture.

If that is the case (and I saw that page prior to replying), why did Pr. Kasson bother with his statement:

but we don't want demand for bigadv to overwhelm the rest of the project or imbalance the points system.

If indeed SMP + Bigadv + Bigbeta is overwhelmed in term of WU quantity vs. GPGPU + Uniproc, then his comment holds no ground. Stanford decided to stop the ppd race, but unfortunately BD will change all this again.

[Dead Things]

Originally Posted by 3.0charlie

then his comment holds no ground.

I'm afraid that all the evidence supports this notion. Which may lead someone to suspect that perhaps foundation-level changes to the PPD system are in development or perhaps - with this move - in the early stages of implementation. What I means is...

Is it possible that PG would do something to risk upsetting many of its most ardent contributors for what equates to a minor issue in the grand scheme of things? Hell yes.

But... is it possible that PG has a reason for making this risky move that far exceeds the official "bringing things into alignment" word? Makes you go hmm, doesn't it? Considering how little PG cared about the P2684 points discrepancy for so long, why do they suddenly care so much about the bigadv points discrepancy vs. other projects writ-large?

I'm not saying this is going to happen, but might I suggest one extreme possibility: PG has begun to prepare us for the eventual elimination of bigadv units entirely. Hey - anything's possible with these guys, right?

[3.0charlie]

And I believe you are quite correct. I've learned from all my years of folding that PG sometimes makes no sense, and will again surprise us with a major change. Instead sticking to WU folding whatever the hardware used (heck, not everyone wants to babysit a quad, just like not everyone wants the heat dump from GPUs) and to Hell with the points, Stanford puts its nose into what most folders eventually come back to: ppd.

It's part of Human Nature - always become better, faster, bigger. And take the necessary steps to reach those new goals. Building those monster rigs did indeed increase the gap between GPU and bigadv, but really: who cares?

The Work Units are still being folded.

[Dead Things]

Who cares indeed! It sure isn't stopping me. I won nice contract with a new client and am using some of the funds for a bunch of upgrades. In the past couple of weeks I've bought 2 x Xeon E5649's, 4 x Opteron 8393SE's and 4 x Opteron 8439SE's.

[Rison]

Pissed? naw.. I just bought 3k worth of SR2s, and don't really mind. Changing bigadv does stop me from looking at getting a quad opteron box at work though.
I do like the SR2's for the ram channels and massive clocks for VMs.. they weren't specifically bought for folding@home though.

[geokilla]

This is why you don't build dedicated folding machines.

[badatcards]

I would have rather seen them up the bonus points for the smp WU's and start giving bonuses to GPU WU's then see them take points away from any WU. But with that said the crazy point's the Bigadv WU's were getting killed most of the big F@H Farms, Making GPU's virtually point less to run because they inter fear with the big bonus, and the extra heat the GPU's make and bad PPD/Watt and summer heat does not make them desirable at all this time of year.

GPU and CPU client's are different so I dont see how taking points away from 1 client is going to help the other 1.

Who know's they might see a error in their change and return it to the way it was, Giving you guys your lost point's back in 1 catch up day.

This Summer was looking good for the team. This points change could change some things. just going to have to wait and see how things end up. keep it up guys

[10e]

BigAdv is not going anywhere.

IMHO PG created this &()*&)*$ mess in the first place, by releasing new units based on an old "benchmark" to BigAdv that were completely overwhelmed/dominated by the latest set of overclocked, and stock multi-cpu systems. Had they made an effort to include/involve the "multi CPU maniacs" in their testing they probably could have prevented this, at least to some extent.

Basically it boils down to the fact that a number of non-bigadv folders argued against the giant BigAdv points bonus, and I would theorize even against the 12 and higher thread units. It may have even been a small number of vocal non BigAdv folders that PG believes represented the masses.

It doesn't affect me one way or another. I'll continue to fold, but I may put off some upgrades this year and next in order to take a slightly more "wait and see" attitude.