News from PG re GPU benchmarking

Better find those Sr-2's and asus Z9 boards with 7 pcie slots, Thats north of 100k ppd on smp and then a shit ton more on 4 or 5 gpu's. Theres your 1st 1m PPD rig
 
Nope the first 1,000,000 PPD rig would be a 4P with 1 or 2 GPU's an 8102 and 1- GTX 580 running a 8047 would get you there. ;)
 
Grandpa_01 said:
Nope the first 1,000,000 PPD rig would be a 4P with 1 or 2 GPU's an 8102 and 1- GTX 580 running a 8047 would get you there. ;)
Click to expand...

Nope, because running a 4p with an 8102 would require the gpu's to run under wine - that has always been a dodgy endeavour at best;)
 
Nathan_P said:
Better find those Sr-2's and asus Z9 boards with 7 pcie slots, Thats north of 100k ppd on smp and then a shit ton more on 4 or 5 gpu's. Theres your 1st 1m PPD rig
Click to expand...

4 x GTX 690's on a single motherboard?

Hohoho... first 2 million ppd machine! (or more... if these units prefer 1536 newer 600 series cuda cores vs. 512 of the 500 series cores)

But seriously... almost 250k ppd from a 580??
 
colinstu said:
4 x GTX 690's on a single motherboard?

Hohoho... first 2 million ppd machine! (or more... if these units prefer 1536 newer 600 series cuda cores vs. 512 of the 500 series cores)

But seriously... almost 250k ppd from a 580??
Click to expand...

Sell your multi-proc while you can and invest in GPUs.
 
Nathan_P said:
Why not, Atlas used to run 4 x GTX 295 on multiple mobo's
Click to expand...

By the questionmark I meant "heck you can get 8 GPUs in a single machine and make 2m ppd possibly"

:p

If I had a trigger finger I'd buy some cheap-ass used MSI k9a2 platinums while ya still can and start filling them up with beefy video cards.
 
Maybe I'm missing something - but is nVidia GPU folding now back in a huge way? If I want folding on a tight budget and like playing games on the side, would a GTX 670 do more folding than a 3770k would?
 
Lets not get into too much of a panic or start dropping big bucks on multiple GPUs just quite yet.

Fairly certain these numbers will be adjusted.
Isn't the whole point of this that with GPU now being able to fold the same work as SMP, if a GPU completes that same work in the same time as SMP, it should be awarded the same points?

Now, unless GPUs are that much more efficient at folding than SMP, there is no way they can leave the points as they are. Though this is Stanford, so who knows........:p
 
BlueMax said:
Maybe I'm missing something - but is nVidia GPU folding now back in a huge way? If I want folding on a tight budget and like playing games on the side, would a GTX 670 do more folding than a 3770k would?
Click to expand...

If they leave the current beta testing GPU points system as is when they release to the wild, the yes a 670 will do 2-4 times the PPD of a 3770k depending on the WU.

these are the first results of the beta testing though. Unless Stanford can prove that GPUs are that much more efficient at folding than SMP, they will have to adjust the points unless they want over a years work in developing QRB for GPU to blow up in their face from angry CPU/multi-proc Folders.

Lets give it some time. Though this is Stanford, so who knows what they will do.
 
I think the problem is that the GPUs can be this much more efficient than SMP, they're just harder to program for. When considering DP FLOPS numbers I think a 4p system is fairly similar to GPU. Their molecular dynamics type of problem works very well in a highly parallel environment and GPU was really their big game changing technology a couple years ago. Bigadv kind of screwed that. I can't wait until GPU folding with big multi card systems is back, it will just suck if they screw with the points system for everyone who built around bigadv.
 
You know - points be damned... I just want to fold the important stuff for the sake of science and killing the evil cancerous beast!

If GPU folding does a million tiny, nearly-useless jobs and CPU folding does the "important stuff" I'll still aim in that direction.
 
Simba123 said:
If they leave the current beta testing GPU points system as is when they release to the wild, the yes a 670 will do 2-4 times the PPD of a 3770k depending on the WU.

these are the first results of the beta testing though. Unless Stanford can prove that GPUs are that much more efficient at folding than SMP, they will have to adjust the points unless they want over a years work in developing QRB for GPU to blow up in their face from angry CPU/multi-proc Folders.

Lets give it some time. Though this is Stanford, so who knows what they will do.
Click to expand...

Catch being...current nvidia GPUs were already on par with current gen smp CPUs in ppd even before applying a new gpu QRB, if not higher at stock clocks....if anything smp's points should be cranked to make it remotely worthwhile in the points competition.
 
I think the problem is that the GPUs can be this much more efficient than SMP, they're just harder to program for. When considering DP FLOPS numbers I think a 4p system is fairly similar to GPU. Their molecular dynamics type of problem works very well in a highly parallel environment and GPU was really their big game changing technology a couple years ago. Bigadv kind of screwed that. I can't wait until GPU folding with big multi card systems is back, it will just suck if they screw with the points system for everyone who built around bigadv.

- that's the problem with the points system and enthusiast, competitive people. These people invest big bucks and lots of time in chasing the most points and pushing the envelope of what their hardware can do.
when a game changing thing like GPGPU comes along, those people are left feeling out in the cold.

You know - points be damned... I just want to fold the important stuff for the sake of science and killing the evil cancerous beast!

I fold for a cure, the points are a nice little earner for my competitive side, but I really don't care for them. If I had investing $1000s in hardware like many have, no doubt I would care more.

If GPU folding does a million tiny, nearly-useless jobs and CPU folding does the "important stuff" I'll still aim in that direction.

The whole point is that now we can do the same work on GPUs as on SMP. If GPU is that much more efficient at doing the work than SMP, then points should reflect that.
Stanford should be encouraging people to Fold in the most efficient manner possible.
 
Simba123 said:
I think the problem is that the GPUs can be this much more efficient than SMP, they're just harder to program for. When considering DP FLOPS numbers I think a 4p system is fairly similar to GPU. Their molecular dynamics type of problem works very well in a highly parallel environment and GPU was really their big game changing technology a couple years ago. Bigadv kind of screwed that. I can't wait until GPU folding with big multi card systems is back, it will just suck if they screw with the points system for everyone who built around bigadv.

- that's the problem with the points system and enthusiast, competitive people. These people invest big bucks and lots of time in chasing the most points and pushing the envelope of what their hardware can do.
when a game changing thing like GPGPU comes along, those people are left feeling out in the cold.

You know - points be damned... I just want to fold the important stuff for the sake of science and killing the evil cancerous beast!

I fold for a cure, the points are a nice little earner for my competitive side, but I really don't care for them. If I had investing $1000s in hardware like many have, no doubt I would care more.

If GPU folding does a million tiny, nearly-useless jobs and CPU folding does the "important stuff" I'll still aim in that direction.

The whole point is that now we can do the same work on GPUs as on SMP. If GPU is that much more efficient at doing the work than SMP, then points should reflect that.
Stanford should be encouraging people to Fold in the most efficient manner possible.
Click to expand...

Except the CPU client is platform agnostic. So long as you have an x86 processor you can fold on anything so long as it is adequately cooled and you don't mind poor returns.

The new QRB bonus is Nvidia only and insofar as the points competition goes automatically renders useless anyone who has any ATI GPU...Even then, it may only be certain Nvidia GPUs. If they want people folding mono-proc CPU in smp to give up in the points race, they should just come out and say it than dicking around about it-as let us face it this is the nail in the coffin for making SMP utterly pointless in the points competition....because this strikes me as little more than an X-mas present to a certain team that recently got passed that coincidentally gets most of their points from GPUs. It also strikes me as Stanford practically subsidizing Nvidia.

I doubt that Stanford will tone down the WU QRB on these WUs. I really really doubt it. Before releasing them they talked and debated internally what to make the QRB, I'll be incredibly surprised if they change their minds.
 
Last edited:
funny how things are on the other shoe

look I wasn't happy when I lost Bigadv (140k PPD down to 35k)...science moves forward.. so does stanfords plans

if a gpu can do the same work in 1/2 or 1/4 the time then the bonus is much higher (non-linear QRB..so the faster the more points exponentially)

As other have noted QRB has to change - and a rebalance over the entire project will have to be done from GPU to SMP to Bigadv

Look I don't even know if these points will stick..maybe they get adjusted - that's the norm
if they do stay for GPU then there is going to be a shift again between hardware and between teams(most likely)

A rebalance is fair - Bigadv and 4p users can't dictate any more than GPU folders

And if your saying just because you have 4p system that cost some dough you get to say what happens over those that don't have those systems?
kybosh

what about those who spent dough on 3930k's or 3960x's that got whacked by the changes?
was it fair? $2-3k per machine could have been spent on 4P's

I myself was only planning to run SMP (and bigadv) at first befor the changes and now was planning on doing SMP only (with only the odd GPU) and eventually get back to bigadv
with these changes I might have to rethink things again

as well as all teams will have to wait and see the outcome of all the changes before deciding what the plans going forward will be.

All I can say is don't ever count on your hardware being the "cat ass" forever or being top dog.
Stanford has a way of changing things and upsetting all our plans. get use to it

what this says is every team will have to keep on it's toes and refocus with hardware and software changes with stanford and not to get overly ladden with one type of hardware
 
Last edited:
This is crazy. A 10x overnight increase in the points given for one particular type of work is stupidity.

Kendrak, I would have hoped the beta team or DAB or somebody would have been aware of this pending change before they hit the streets. This type of drastic change can only cause disillusionment, disappointment, confusion, and not to mention, supply/demand laws will kick in and we will all flock to GPU folding in a big way.

Speaking of which, are these units still in beta or are they in production? How do I get them on my GTX470s? I still am getting nothing but 8054.
 
jebo_4jc said:
This is crazy. A 10x overnight increase in the points given for one particular type of work is stupidity.
Click to expand...

[Skripka=Biting Sarcasm Voice]

But "equal pay for equal work", to quote Stanford

[End Biting Sarcasm Voice]
 
those units are beta. I haven't snagged one yet... steady diet of 8018's

No one knows if the points will stay that high...but expect a big increase in GPU points in the future

this is a benchmark test unit
all other work units will be based after they figure things out and after testing each WU with the new baseline
 
Lets not turn this into a p!ssing match between [H] and eVga.

We know that the QRB for GPU has been in the works for over a year. It's not like
Stanford woke up yesterday and decided, lets put a QRB on GPU today.
They certainly did not do it to aid a particular team, so lets just leave that alone.


jebo_4jc

The units are in beta, not general pool.
You can set up your client to get them by adding the 'beta' tag.(as far as I'm aware that is all you need to do)

As for the DAB. All DAB representatives are under a non-disclosure restriction from Stanford, they are not allowed to talk about what is discussed in the DAB until Stanford makes an 'official' statement first.
 
Except the CPU client is platform agnostic. So long as you have an x86 processor you can fold on anything so long as it is adequately cooled and you don't mind poor returns.

The new QRB bonus is Nvidia only and insofar as the points competition goes automatically renders useless anyone who has any ATI GPU...Even then, it may only be certain Nvidia GPUs. If they want people folding mono-proc CPU in smp to give up in the points race, they should just come out and say it than dicking around about it-as let us face it this is the nail in the coffin for making SMP utterly pointless in the points competition....because this strikes me as little more than an X-mas present to a certain team that recently got passed that coincidentally gets most of their points from GPUs. It also strikes me as Stanford practically subsidizing Nvidia.

I doubt that Stanford will tone down the WU QRB on these WUs. I really really doubt it. Before releasing them they talked and debated internally what to make the QRB, I'll be incredibly surprised if they change their minds.


Ignoring the favoring a certain team bit, I agree.
If folding on nVidia GPUs is 10x more efficient than SMP/Multi-proc, then Stanford should just come out and state that that is the future for folding.

It hurts and sucks big-time for the people who have invested thousands in multi-proc rigs.For the project though, if that is where the most efficient returns can be generated, then that is what they should be encouraging.




__________________
 
Chances are the points will change, if they don't they i will go down to 1 rig instead of the 3 that i have/will have. What it will mean is that i will chew both SMP and GPU wu.

If they do change i have choices that may mean i fold all 3, bigadv, SMP and GPU
 
Macaholic has just raised an important point over at EVGA, the PG GPU benchmark is a GTX460, does someone have numbers for one of them on the new WU......
 
ppd.png


taken from here
http://www.overclockers.com/forums/showthread.php?p=7324327#post7324327

3770k & 2 670's (I have no idea what the frequencies are - not my rig)

I uploaded it to imageshack because you can't see images on that forum unless you are a member.

Depending on how well V7 is calculating that PPD, (and we all know how we it does that :p)
 
It would be nice to get one of these work units captured. Please send it to Tear if you happen to get one! We need to benchmark these units in a controlled environment, and try to see how they scale across different hardware.
 
Maybe GPU folding will be viable in the future then?
Reminds me of the old days:






I still have most of the gear laying around.
 
Jebo, the first number I saw were posted in this thread.

DAB is more policy than raw numbers..... that is where the beta team comes in.

That said I'm hoping for a re-adjust back down to earth.
 
I have 6 460s spinning on beta but haven't seen a single one of these yet. I expect this is down to the masses using the beta flag to get WU for their Kepler GPUs. As PG now use a whitelist for the GPUs, I expect this should be how the beta WU distribution ought to work as well.

Dear DAB rep (Kendrak?), please pass this on, as it really kills the usefulness of beta testing if actual registered beta testers can't get beta WU to, you know, test.
 
rhavern said:
I have 6 460s spinning on beta but haven't seen a single one of these yet. I expect this is down to the masses using the beta flag to get WU for their Kepler GPUs. As PG now use a whitelist for the GPUs, I expect this should be how the beta WU distribution ought to work as well.

Dear DAB rep (Kendrak?), please pass this on, as it really kills the usefulness of beta testing if actual registered beta testers can't get beta WU to, you know, test.
Click to expand...

have to agree with this. beta units should be restricted to beta folders. Controlling the group makes it easier to get feedback as well.
 
How do you set the beta flag on v7 for a GPU? I have two 460s and a 560Ti folding but have never set the beta flag for GPUs.
 
http://foldingforum.org/viewtopic.php?f=66&t=18031#p180270

To request Beta Project and FahCores, you will need a special configuration that SHOULD NOT BE used by Non-Beta Testers. Non-Beta Testers do so at their own risk. The special configuration varies with the F@h Client that is running:
v6 F@h Clients
A) -betateam -> Works for all v6.xx F@h Clients
B) -bigbeta -> Works with v6.xx SMP2 Clients only and gives access to Beta WUs for bigadv trials
Using the -verbosity 9 flag is highly recommended while Beta Testing with all v6.xx F@h Clients. The flag increases the detail in the FAHlog.txt, thus helping with troubleshooting if a problem occurs. If there are no Beta WUs that match your F@h Client's configuration, the client will be assigned Public WUs (normal, advmethods and bigadv). However, you can change the WU Size settings or flags to get Beta WUs if they are available.
Note: Please do not mix -betateam flag with -advmethods flag. Do not mix -bigbeta with -bigadv and -advmethods flag. Use either the beta flags or the public flags, but not both at the same time.
V7 F@h Client
A) client-type = beta -> Works for V7 Client on all Slots
B) client-type = bigbeta -> Works for V7 Client on SMP Slot only and gives access to Beta WUs for bigadv trials
We recommend the verbose setting of 3 in FAHControl. If there are no Beta WUs that match your FAHClient's configuration, you will be assigned either normal/advanced/bigadv client-type WUs. However, you can change the WU Size settings and client-type to get Beta WUs if they are available.
Note: Please do not mix beta client-type with advanced client-type or bigbeta client-type with bigadv/advanced client-type. Use either the beta/bigbeta or the normal/advanced client-type, but not both at the same time. Do note that you can individually set each Slot's client-type.
 
Even though I have expert selected and get to this screen, when I try to add the Extra-slot options, the window opens but it will not except anything I type.
Nothing even shows when I type and I have to hit Escape just to close the window.
 
Last edited:
equalworkb.JPG


K factor the same
base points 10x
atom count 1/24th
preferred deadline double
final deadline 20% longer

I somehow doubt that this will be the final numbers as all this does is boost gpu wu's by 10x instead of actually giving them equal work.
 
Last edited:
I know the significantly higher points for these WUs are exciting/frustrating but let's not forget that this is just the first step in PG's effort to give "equal points for equal work". I'm all for it!

I can only hope they finalize this initiative as soon as possible, without causing too much confusion and chaos.
 
Geez, I set the "-betateam" flag on my three v6 GPU clients, and I keep getting these 762X units and their PPD is terrible.

I just added the beta client type to my v7 GPU client, so we will see if v7 is more likely to get the omgwtfppd units.
 
theGryphon said:
I know the significantly higher points for these WUs are exciting/frustrating but let's not forget that this is just the first step in PG's effort to give "equal points for equal work". I'm all for it!

I can only hope they finalize this initiative as soon as possible, without causing too much confusion and chaos.
Click to expand...

before they can give equal points for equal work...they have to port an smp unit over...
Otherwise you are just doing more points for same old work...
All they did was boost the value of an old wu...
 
You must log in or register to reply here.
Back
Top