GPU vs CPU!!

[Evilsizer]

hmm i wonder what a voodoo 5 5500 would do....

 

[Macaholic]

soooo, should we call it pause and resume in the future Mac??

Right now lets keep on going. I've just finished a couple of units in question, which appears in line with Dr. Pande's findings;

ok, I think we're on the trail of this. We'll test our fix internally before rolling it out. This could still take a couple of days to fix and then do the Q&A.

For now, it looks like this isn't a disaster (i.e. all GPU clients failing to do calculations) as much as a pretty major "cosmetic" bug.


_________________
Professor Vijay S. Pande, PhD
Director, Folding@Home Distributed Computing Project, and
Associate Professor of Chemistry and of Structural Biology, Stanford University

 

[pscout]

I thought both you and Mac were returning completed GPU wu's o.k. Anyway, later is better than never.


I just got my "GPU rig" going last night; a budget Pent. D rig with the x1900xt. I started a 2711 before going to bed, but this morning before I went to work, I checked it and it had hung...message stating "starting GUI server..."
Why GUI server? Maybe that's a GPU-thing.

IH ... you are just seeing some symptoms of the gpu beta ... your gpu is probably folding fine, just not reporting ... everyone else is seeing the same i think, me included ... seems to be a problem with the core on gen 2 or later wus'

 

[ihrsetrdr]

IH ... you are just seeing some symptoms of the gpu beta ... your gpu is probably folding fine, just not reporting ... everyone else is seeing the same i think, me included ... seems to be a problem with the core on gen 2 or later wus'

Very strange, these GPU wu's. Rather than clutter this thread with a blatent hi-jack I'll start another, perhaps at FC.org.

 

[Joe Camel]

damn this is SCARY!! BOTH of you are in the top 20 list

all in all, i say this is VERY impressive for 5x "rigs" regardless of who wins

way to FOLD guys !!!

 

[LandShark]

bump

 

[Macaholic]

Headed for the home stretch with Sharky in the lead by a quarter length;

16 The_LandShark 3,540pt 24hr avg. 34,332pt Total
17 Macaholic 3,347pt 24hr avg. 30,702pt Total

It just might be a photo finish.

 

[ChasR]

What this goes to show you is that ,with OC'd E6600s and OC'd GPUs, you could make 6000 ppd with 5 rigs on currently available WUs. That would make you the team's number 8 producer.

 

[Joe Camel]

does that take into account the fact that 1/2 the CPU cores would need to be dedicated to feeding the GPUs?

The_LandShark .....3,540pt 24hr avg. .....34,332pt Total
17 Macaholic ......... 3,347pt 24hr avg. .....30,702pt Total

so...

3540 /2 = 1770
+..............3347
=..............5117 PPD on 5x OCed 6600s with 5x OCed X1900XT (X ?)


---


ive been kicking around an upgrade or two to the home farm and have been leaning toward either 805's with X1900's or conroe's alone. i hate to buy a dual core conroe and only use 1x core to FOLD, id feel much better "sacrificing" a 805 core to feed the GPU.

 

[leelegend]

that is ssooooo mad, i have 50 cores (admittedly none of them clocked) to get into the top 10!!!!!

lee

 

[LandShark]

What this goes to show you is that ,with OC'd E6600s and OC'd GPUs, you could make 6000 ppd with 5 rigs on currently available WUs. That would make you the team's number 8 producer.

but don't forgot I'm not allow to shuffle WU around! and with nowaday's WU, I'll have no problem feeding all 10 CPU core w/ 600pointer all day long (or let's say 90+% of time) which will net me 5300+ppd. therefore, my whole point is, by adding 5 GPU, it will "only" get me no more than 700ppd!! and how much is the cost of 5 1900xtx again?? I'd still rather build another 3 more barebone conroe for the $$ of 5 x1900xtx and get another 3000+ppd imho.

does that take into account the fact that 1/2 the CPU cores would need to be dedicated to feeding the GPUs?

so...

3540 /2 = 1770
+..............3347
=..............5117 PPD on 5x OCed 6600s with 5x OCed X1900XT (X ?)
---
ive been kicking around an upgrade or two to the home farm and have been leaning toward either 805's with X1900's or conroe's alone. i hate to buy a dual core conroe and only use 1x core to FOLD, id feel much better "sacrificing" a 805 core to feed the GPU.

yes, I agree, if anyone has anything less than a Conroe, it'll worth to get a x1900xtx and do GPU folding as you don't have much to lose to start with! I too thought about getting some cheap used barebone rig (Prescott/Northwood) and just let it fold on GPU too!

 

[ihrsetrdr]

yes, I agree, if anyone has anything less than a Conroe, it'll worth to get a x1900xtx and do GPU folding as you don't have much to lose to start with! I too thought about getting some cheap used barebone rig (Prescott/Northwood) and just let it fold on GPU too!

That's what I'm doing with my "aging" p945 rigs, rather than sell them for pennies-on-the-dollar.


Regardless of the outcome, congrats to both contestant's for the Top Drawer competitive Folding event! <-<-Guiness, I'll catch the bar tab

 

[ChasR]

Mac is doing 3347 on stock 2D clocks. The GPUs will make 4200 ppd on OC'd 3D clocks. Adding 1770 gives you 5970 which I rounded to 6000.

 

[Macaholic]

but don't forgot I'm not allow to shuffle WU around! and with nowaday's WU, I'll have no problem feeding all 10 CPU core w/ 600pointer all day long (or let's say 90+% of time) which will net me 5300+ppd. therefore, my whole point is, by adding 5 GPU, it will "only" get me no more than 700ppd!! and how much is the cost of 5 1900xtx again?? I'd still rather build another 3 more barebone conroe for the $$ of 5 x1900xtx and get another 3000+ppd imho.

Those 600pt Gros won't last forever. <cough> QMD <cough> In my opinion, time is money, and I don't want to have to waste time shuffling units. Besides, not that many donors have 5+ machines at their immediate disposal that they can use for shuffling units around. I really don't see a problem sacrificing one core for GPU folding at the current points level. And I would not be surprised to see it nudged up a little more in the future. If I were to do a bit of OCing and fold on one core of these machines they would definitely come out ahead with minimal effort on my part.

 

[Joe Camel]

Mac is doing 3347 on stock 2D clocks. The GPUs will make 4200 ppd on OC'd 3D clocks. Adding 1770 gives you 5970 which I rounded to 6000.

ahhhh i follow now. (sorry)

huh...

when you say "OC'd 3D" do you mean >stock 3D clocks or just 3D clocks?

 

[LandShark]

Those 600pt Gros won't last forever. <cough> QMD <cough> In my opinion, time is money, and I don't want to have to waste time shuffling units. Besides, not that many donors have 5+ machines at their immediate disposal that they can use for shuffling units around. I really don't see a problem sacrificing one core for GPU folding at the current points level. And I would not be surprised to see it nudged up a little more in the future. If I were to do a bit of OCing and fold on one core of these machines they would definitely come out ahead with minimal effort on my part.

I hear you. yes, for those set it and forget it, GPU does give you the most stable, and pretty high, ppd!! and especially true for those have less than a high clocked Conroe too! but for people/farmer that has a few rig to play with and have some highend folder, adding a GPU to fold just might not be the best option imho.

when you say "OC'd 3D" do you mean >stock 3D clocks or just 3D clocks?

> stock 3D clock!

 

[7im]

I may just be talking pie in the sky, but those 600 point WUs, and 149x WUs, are probably the *best* points the Conroe will ever get. However, the current work units for the GPUs are likely the *worst* points we will see. There is no additional upside to the Conroe, but there is a potential for bonus WUs on the GPUs. And the efficiency on the CPU client is maxed already, while the GPU client is only 3 weeks old. Lot's of room for upside there too.

If Mac and LS ended in a dead heat for points, would it still be considered a close race?

 

[Silver]

Bought an XT in anticipation of the GPU units and am presently running it given that the points were raised to 330 per unit. Frame times are at 8.1 so it really means that I have pulled one cpu down to get pretty much the same points but added electrical cost plus the purchase of the card. Will most likely take it back off line till there is a net gain with something less than a $400 dollar card and start converting over to boards capable of running pci-x 16. Was not the best investment for my farm at this time.

 

[harlam357]

Regardless of the outcome, congrats to both contestant's for the Top Drawer competitive Folding event! <-<-Guiness, I'll catch the bar tab

QFT

Yes! Great entertainment and demonstration of the new GPU client. Thanks guys!

Red Stripe for me tonight!

 

[LandShark]

I may just be talking pie in the sky, but those 600 point WUs, and 149x WUs, are probably the *best* points the Conroe will ever get. However, the current work units for the GPUs are likely the *worst* points we will see. There is no additional upside to the Conroe, but there is a potential for bonus WUs on the GPUs. And the efficiency on the CPU client is maxed already, while the GPU client is only 3 weeks old. Lot's of room for upside there too.

If Mac and LS ended in a dead heat for points, would it still be considered a close race?

don't forget every now and then, we all treated with some nice bonus WUs, QMD, DGromacs, 1495, 600 Ribo, etc.

I agree 1495s probably will be the "best" a conroe could do! but who had guess a single conroe rig could net 2000+ppd even just 3-4 months ago?! and look at some latest WUs, 240+MB of memory usage!! therefore, there's still some good hugh bonus WUs for the cpu for sure. the only question is how often only?! and that's why I understand and agree from Mac's point of view, set it and forget it.

dead heat? don't seem like it! unless I stop at least a day earlier and Mac will then have a chance to make up the points. oh, I've never mentioned, I had 1 conroe down for 2 days during the course too!

Bought an XT in anticipation of the GPU units and am presently running it given that the points were raised to 330 per unit. Frame times are at 8.1 so it really means that I have pulled one cpu down to get pretty much the same points but added electrical cost plus the purchase of the card. Will most likely take it back off line till there is a net gain with something less than a $400 dollar card and start converting over to boards capable of running pci-x 16. Was not the best investment for my farm at this time.

exactly my thought Silver!!!

gpu folding has it's appreal for some folder, but not all for sure!! it's entry fee and net gain might not be as well to some people. but will make some nice (huge) different for others too!

that's why I say, IF for a conroe owner looking to upgrade for more ppd, spending the $300+ for video card and extra eletricity usage doesn't sound too appreal to most people/famer!! HOWEVER, if for anyone who has anything less than a high clocked conroe or running out of space or has a small farm/folders, going to gpu folding might be a better option for them atm.

I guess, it's mainly the high entry fee vs the not too high reward (atm!) that keeps me from gpu folding right now. but again, I think I will do gpu folding one day myself too! it's simply just "not right now for me".