New bigadv project 6904

Yea... I picked up 2x 5830's today and am picking up 2x 6950's tomorrow...
 
This definitely sucks, I agree the system needed balancing but I would prefer they have applied higher points or bonus factors to other classes of units rather than cut the rug out from under us.

I hope no one gets too discouraged by this and leaves, at the end of the day we're still doing this for the science and if we are to stand a chance on points we need everyone to continue rallying and not be defeated by this.
 
MIBW is NOT going to be pleased about this.

he's probably about to start his saturday morning cup of coffee and sit down to check hardforum.com and his HFM status, and..........

Amazingly I am not as peeved as I should be about it - there were hints that this would happen.

PPD is down almost exactly 20%. But this might end up the same on average as it was a month ago, depending on how many super-bigadv flow.

Dont get me wrong, Stanford are simply useless re: consultation, but now I expect it.

The biggest casualty is that we had just become the #1 PPD team, and would probably of held on even with EVGAs start of the month ramp. But now I think the tables have been turned. And potentially more pain to come if they boost GPU points. We will see.
 
You get the impression they might boost gpu points?

I wasn't sure what kasson meant when he referenced normalizing points across the board.
 
I heard they might introduce a quick return bonus for gpus, which I take it to mean would elevate the average.

But I find my mental state much easier to maintain if I assume that the most dedicated folders will get screwed by unintended consequences of Stanford. The last month was too good to be true.

But given that these are the same tone-deaf people that did not see the obvious consequences of setting in place the bigadv program, did anyone think that their "correction" would be handled any better?

They are playing laser tag with a kitten, and we are the kitty.
 
It was fun the time it lasted. What I do find pityful is that Kasson in his explanation mixed the points given per WU and the usefullness of said WU. I taught that Stanford did not give a crap about points... also,

we don't want demand for bigadv to overwhelm the rest of the project or imbalance the points system.
I wonder how many SMP, uniproc WU and GPU WUs are folded for each bigadv? Is Kasson's statement true?
 
After thinking about it, you're probably right that any further tweaks to the points system would only boost GPU and possibly lower end CPU points more.

I will commence keeping my mouth shut.
 
Ok here's some stats to cheer people up:
06.30.11:
EVGA 16,393,510
[H] 24,632,220

yup that's 8.2 million more points :D

Yup thats the day before they nuked the points...

So Stanford is saying my contribution is worth 20% less to them today than it was yesterday (according to HFM with the lastest base point values - I am looking at 2x6904s, a 6903, a 2685, and a 2684)....that really makes me want to keep supporting them...

Thats what I am reading...
 
Amazingly I am not as peeved as I should be about it - there were hints that this would happen.

PPD is down almost exactly 20%. But this might end up the same on average as it was a month ago, depending on how many super-bigadv flow.

Dont get me wrong, Stanford are simply useless re: consultation, but now I expect it.

Yeah but i bet that this puts 2684 back where it was before they upgraded the credit a couple of months back:mad:
 
Ok here's some stats to cheer people up:
06.30.11:
EVGA 16,393,510
[H] 24,632,220

yup that's 8.2 million more points :D
7/1/2011:
*Ring Ring*
EVGA Business Manager: Kasson, What's going on here? You haven't forgotten our arrangement have you? We've poured a lot of resources into this relationship and it feels like we aren't aligned here. EVGA sells GPU's, not CPU's....

Kasson: Yes sir, right on it!

:p

In seriousness I am confident they must have been putting pressure on Stanford for sometime now and they must have hit the fever pitch needed. Sure the rest of the community has complained about the point system for over a year but nothing changed until team rankings started shifting...
 
Last edited:
The biggest casualty is that we had just become the #1 PPD team, and would probably of held on even with EVGAs start of the month ramp. But now I think the tables have been turned. And potentially more pain to come if they boost GPU points. We will see.

The biggest casualty is that a fair few people will now hold off upgrade plans to see how this all pans out, I for one, will not be adding a bulldozer rig to my garden for a while now.

It will probably also cause a fair few people to fold less or maybe even quit altogether which won't help the project
 
7/1/2011:
*Ring Ring*
EVGA Business Manager: Kasson, What's going on here? You haven't forgotten our arrangement have you? We've poured a lot of resources into this relationship and it feels like we aren't aligned here. EVGA sells GPU's, not CPU's....

Kasson: Yes sir, right on it!

:p

In seriousness I am confident they must have been putting pressure on Stanford for sometime now and they must have hit the fever pitch needed. Sure the rest of the community has complained about the point system for over a year but nothing changed until team rankings started shifting...

*sigh*

For the sake of the project, I fear that this parody might actually be accurate. I'm not going to get hit as hard as you guys who have spend magnitudes more or your hard earned cash to better science, but it doesn't mean that i'm not going to feel a bit down about this. The timing is what makes it most suspicious.
 
This news truly does hurt teams like us, especially the top 20-50 folders here. We've been able to keep pace with EVGA even though we have 300ish fewer active folders because of the bigadv WU and the 2P and 4P folders.

Given the change to K-factor and base points, will bigadv still be better PPD than regular SMP for those of us with 8 cores? I will try and find a calculator around somewhere, but I suspect some of our more experienced folders would know off-hand.

It's better, but not tons better depending on work unit. I have some i7 2600 machines running both bigadv and regular smp and the difference between them can be as small as ~4k ppd. Some of the smp units are around 21k ppd. No guarantees though on the ones you'll get.
 
The last month was too good to be true.

I think this gets to the root of the problem. Some people have been complaining since the beginning that bigadv points were too inflated, and I agreed. The problem is that point inflation just kept getting bigger and bigger, with the new Linux client and then the super-bigadv, and some people accepted that as the "new normal" where a single box could generate more PPD than some whole teams.

I think the "discussion" mentioned by PG refers to all the bitching about point inflation that's been going on for months. Think about the people out there still running uniproc clients and getting a couple hundred PPD or the PS3 folks locked to about 900 PPD. If they look at their points (and thankfully many don't) they must be extremely discouraged when they've been running F@H for years and now someone with a fast computer and a new beta client (remember you have to hunt for even the SMP or GPU client on the F@H homepage) blows by them in the stats after a few days or weeks.

Like everything else, PG has handled this point adjustment badly, but I think it had to be done. I hope the next step will be to adjust the uniproc and PS3 points to make those clients worth running. I don't think GPU needs any help since graphics cards continue to get faster and more power-efficient. Ideally someone running four uniproc clients should have about the same PPD as someone running SMP on a quadcore, maybe with a small (small!) bonus for SMP since it will complete WUs faster.

BTW, I own two SR-2s, but I never bought them with the expectation that that the F@H point gravy train would continue forever. I run lots of other DC projects too, and I just like having lots of cores in a single box vs. lots of boxen. Currently one SR-2 is running GIMPS and the other is running 24 instances of Prime Sierpinski PRP. :) Granted, SR-2s are cheaper than the AMD quad-proc monsters now being built.
 
so how do i get one of these (6904 WU), all i get is 6903's.
do i have to change to bigbeta flag?
 
so how do i get one of these (6904 WU), all i get is 6903's.
do i have to change to bigbeta flag?

I thought these were released into the general bigadv pool but I run bigbeta anyway so that I don't have to reconfigure machines if another big unit hits the beta pool.
 
I hope the next step will be to adjust the uniproc and PS3 points to make those clients worth running.
I wonder if perhaps the apathy with regard to boosting these units is because Stanford really isn't particularly interested in these anymore... now that the power is available to do more, maybe they've moved on to bigger and better things and just leave these clients laying around because they don't want to shut anyone out.
Ideally someone running four uniproc clients should have about the same PPD as someone running SMP on a quadcore, maybe with a small (small!) bonus for SMP since it will complete WUs faster..

I actually thought the point of the bonus for SMP, and even bigadv, was to eliminate that behavior. When dual-cores broke people used to run 2 uniproc clients, then they came out with SMP.. then people started getting quads and running 2 SMP's, so they released bigadv. Now regular people have 8 thread rigs so they've come up with 12-thread units. Each step of the way theres been a hefty bonus associated with moving to the new client.. which is the same as offering an incentive to discontinue old behavior for new behavior.. so to return the uniproc to the value of SMP would be implying that uniproc is just as important as SMP.. Its being strongly implied that thats not the case
 
anybody gettin better then 18 min TPF on the 6904 other than sfield?

his is arnd 15min because of better ram chips.
 
Getting about 20m tpf on 6904's with 6176's and CL9 ram. Dont have the guts to do the memory profile re-writing for my chips, they're 7-8-7-24 @1600 stuck at CL9 on my SM board @1333.
 
Getting about 20m tpf on 6904's with 6176's and CL9 ram. Dont have the guts to do the memory profile re-writing for my chips, they're 7-8-7-24 @1600 stuck at CL9 on my SM board @1333.
really? what's your setup like?

I get about 21:00 average on my 6172s, also with 1333 cas 9 RAM using ubuntu 10.10 + Kraken

edit. after running the numbers I guess that sounds right.
 
16:00 TPF with 6180s running at 2.7ghz, 6-7-5-20-1T memory running at DDR3-1440. NB @ 1944. ~760W power draw

16:12 TPF with different chips running at 2.7ghz, 6-7-5-20-1T memory running at DDR3-1333. NB @ 1800. 788W folding power draw (1.175v cpu0, 1.15v cpu1-3)

15:45 TPF with different chips running at 2.8ghz, 6-7-5-20-1T memory running at DDR3-1333. NB @ 1800. 850W folding power draw. Folding temps 55-58C (single 360 rad) Will try DDR3-1440, NB 1944 later... 8 instances of IntelBurnTest -- 1 instance per NUMA node, 6 threads per node -- at this clock pulls 1100W. At pstate 4 (800mhz) set to 0.825v, idle power draw is a meager 145W.

At 3ghz, it should be possible to get around 14:30TPF. Folding power draw will likely be around 1050W. Will do PPD/W math later -- this may be past the threshold of meaningful gains (the point system changes don't provide as much an incentive to push at these extremes now...).

Migrating a 4P into the primary SR-2 chassis this week (which has 3x 360 rads) -- much more cooling potential.

anybody gettin better then 18 min TPF on the 6904 other than sfield?

his is arnd 15min because of better ram chips.
 
Last edited:
great info sfield, thanks.

you helped answer a question that has been bothering me a bit....I went with cheap DDR3-1333 Cas 9 for my 4x6172 system, but I have been wondering if I should have bought somthing a bit faster. It looks like your ~8% increase in memory bandwidth moving from 1333 to 1440 resulted in a TPF reduction of around 1%.
 
oh i see bonus points are calculated based on the final deadline.

well then, I'm OK with this change :)
 
Last edited:
I just picked up a 6904 the other day and noticed that PPD data isn't showing up in HFM. I've downloaded projects from stanford but its still not showing total credit or PPD. what gives?
 
Gents,

What is the "right" tag for smp to get bigadv WUs (6903, 6904)?
I've been using -bigbeta on my i7 970 rigs, but could not get any of those 2 for last 2 weeks... keep getting 2689 (suckie), 2684
 
With -bigadv, you may get 6903. You have to use -bigbeta to have a chance at getting 6904 since it is a beta project still. When using -bigbeta, you may get other projects that are in beta testing.

I'm personally running -bigbeta and have pulled one 6904 over the last two weeks. As kendrak said, there are not that many floating around right now. I'm still waiting for my 3p system to pick up a 6904 :)
 
Thanks all for your reply.

I've been using -bigbeta from day1 .. and pretty much would get 6904/6903 either back-to-back or 1/2 normal bigadv WUs in between.

But for last almost 2 weeks, got NONE .... :D .. weird huh ? :D
 
I havent seen crap for bigbeta units either. A ton of 2864s 2865s and 6900s...
 
Stupid question, but you are trying to pick up a 6903/6904 while running Linux right?
Not to jinx myself, but they are flowing rather well for me lately using -bigbeta.
 
Stupid question, but you are trying to pick up a 6903/6904 while running Linux right?
Not to jinx myself, but they are flowing rather well for me lately using -bigbeta.

I went a week or so without 6903/4
 
Oh thanks God .. got 2x 6904 after the complaining .. lolz ...Maybe SF sends someone here to read all the complaints :D
 
Back
Top