GPU folding is here!

[Clocker]

Now if we can just get Clocker to reinstall 6.11 and restart GPU folding and then theres Handruin ....

I only have an 128MB X300 right now. But hey, that 350Mhz core overclocks to ~500Mhz. LOL

I have a 256MB 7300GT coming that I got for like $40, but that won't help much with folding. I'm not foling anymore because I don't leave my computer on 24/7 anymore.

 

[P5-133XL]

I only have an 128MB X300 right now. But hey, that 350Mhz core overclocks to ~500Mhz. LOL

I have a 256MB 7300GT coming that I got for like $40, but that won't help much with folding. I'm not foling anymore because I don't leave my computer on 24/7 anymore.

The X300 is not capable of GPU folding and for that matter no Nvidia cards work either, including the 7300GT.

 

[Pradeep]

Looks like the PS 3 client will be released shortly. Will be interesting to see how many points they give for the units. Certainly a good excuse to get one.

 

[Clocker]

The X300 is not capable of GPU folding and for that matter no Nvidia cards work either, including the 7300GT.

Thanks Captain Obvious.

 

[CougTek]

Using the ATI Tool 0.26, I managed to shave the time it takes to compute a frame down to 8m30s, which means that every 14 hours and 10 minutes, I should complete a 330 points unit. That's around 550ppd, considering the small amount of time not folding while the client is sending/receiving data from Stanford's server. It's still short of the 590ppd I got from the (60$CAN cheaper) X1950 Pro, but it's as good as I'll get from that X1950XT.

3D Mark 2006 and every game benchmarks say it's faster than the X1950 Pro (5690 3D Marks vs ~4540 3D Marks for the Pro), but the FAH client says otherwise.

 

[Fushigi]

I've lost some GPU folding time as the GPU client has been waiting for work for a few hours now. :frowner:

 

[P5-133XL]

My GPU client had been failing to contact a server too. Stanford had a network issue, that has been fixed and my GPU client is now working ...

 

[Fushigi]

Mine resumed as well.

 

[Fushigi]

So what's the current best GPU Folding card? Still the X1960 512MB or is the (more expensive) HD3000 series worth a look?

 

[P5-133XL]

There is no current folding client for the x2900 or the x3000 series: It is still in development. So there has been no recent change with the x1900 series as the best option. The top performing is the 512 MB x1900XTX which has always been horribly expensive, and very hot. The 512 MB x1950 pro series is at a good cost-effective point. Unless one likes, regular pauses in your normal day-to-day computer activities, always get a 512MB version.

 

[Fushigi]

Beta Folding clients for Radeon 2xxx & 3xxx have been released and can be downloaded here. No Crossfire support but multiple GPU support is there.

 

[P5-133XL]

Also, the original ATI Radeon 1xxx GPU clients have been pulled and Stanford will no longer be sending out WU for those clients in a months time time frame.

 

[Fushigi]

That's rather annoying. I wonder how many contributors made specific investments in 1xxx cards to support the cause. I'm running 2 1950Pros myself. Both are less than one year old & were picked for the express purpose of Folding.

 

[P5-133XL]

Oh, I'm not disagreeing. I feel the same pain and am I'm quite annoyed.

On the other hand, now that the GPU2 client is out, Stanford does want to transition to it since it is more flexible in the type of work that can be done: One can't fight progress. Unfortunately it does not support the 1xxx cards. It is my understanding, the one month continuation is there so that some papers, already in the works, can be finished. After that, everything will be going to the second generation GPU clients.

 

[Fushigi]

So, since I may be rebuilding my wife's PC in the next few months, taking it from an Athlon 2800 to a C2Q, what would be the recommended GPU based on currently shipping hardware? 3870X2?

At this time I don't think I can justify replacing my 1950s with newer cards but I can certainly put them in newer builds.

 

[P5-133XL]

At some point, the X2 may work with both GPU's but at the moment only one runs folding. So currently, there is no need to spend the extra money and a simple 3870 is the best GPU folder.

Just as a side note, currently the Q9450 is significantly better than the q9300 which is significatly better than the Q6600 for folding. From the numbers I'm seeing at stock speeds the Q9300 is getting an extra 20% and the Q9450 is getting an extra 40% over the base Q6600. In addition, Everything gets even better if you OC (another 40-80%). Currently, both the Q9300 and the Q9450 are in short supply, but hopefully that will resolve itself over the next couple of months.

 

[Fushigi]

Oh yeah, the 9450 is definitely the CPU to get, although I haven't bothered to OC much lately, prefering stability since I don't have the time to devote to tweaking. Microcenter had the 9450 for $350 with a $50 gift card; it was the cheapest source I could find at a net $330 or so (CPU + tax -$50). I should have taken the plunge.

I'll waffle on the 3870 v. 3870X2 for the moment; maybe by the time I'm ready to build Stanford will have X2 support. Seems like it would be more power-efficient than getting 2 3870s.

 

[Fushigi]

I'm not finding any details on the VisionTek 3870 OC edition, including on VisionTek's site. I'll stop by BestBuy & check out the packaging for details to see how it compares to this card from HIS.

 

[Fushigi]

I couldn't find out anything specific about the VisionTek card but I bought one anyway. :cyclopsani: It has an 800MHz clock (as reported by Catalyst Control Center) so it isn't quite as OCed as the HIS card, which runs at 850MHz.

With the new GPU client, you have to dedicate one of your host CPU cores to the GPU client so one of my Folders now has 3xCPU client + 1xGPU client. It'll be interesting to see what the point difference is between a CPU client + the old GPU client and the new GPU client. It is fast at whatever kind of WUs they're throwing at it, completing approx. 1% of the WU per minute with no errors.

 

[Fushigi]

It is fast at whatever kind of WUs they're throwing at it, completing approx. 1% of the WU per minute with no errors.

I should note Stanford is only sending test WUs at the moment; they are small so are completing quickly. Larger WUs will come in a week or two.

 

[Mercutio]

Apparently, one can fold on nVidia GPUs now.

 

[CougTek]

My GeForce 8800GT has been folding for two days now. It yields about 1000 points/day, but it doesn't clutter one of the CPU cores like the old client used to.

 

[P5-133XL]

You should be getting much higher PPD than that: Try around 4000-4400 PPD, stock. There is something wrong, if you are only getting 1000.

The main problem the the Nvidia GPU2 client, well two actually, is that the client isn't very stable and it slows down standard 2D graphics making your machine feel like a slug. However, for that PPD level, most avid folders seem to be willing to make the sacrifice.

No, I am not a good person to help debug the problem, having no Nvidia graphics cards. All I'm doing is reporting what other have said in the folding forums. Just as an observation, the folders seem to prefer the 177.35 drivers that have had their inf modified to allow installation on older video cards.

 

[CougTek]

I resumed GPU folding with the beta 177.40 driver from Nvidia. I hope the results will improve. Trying to install the 177.35 with the modified inf file never worked. I always ended up with a standard VGA graphic adapter.

 

[P5-133XL]

I thought that anyone interested in GPU folding might find this useful:

Nvidia GPU2 folding PPD scores and configurations.
ATI GPU2 folding PPD scores and configurations.

Some important things to note: The very upper end cards still have some leg-room because the WU's aren't yet big enough to actually stretch the capabilities of the cards [GTX 260, GTX 280, ATI 4850 and ATI 4870]. Performance very's alot depending upon CPU capability but to some extent you can compensate by running multiple instances of the GPU client if you have spare CPU cores to feed the GPU client. Yes, for both Nvidia and ATI the most recent GPU client/core combo is capable of running multiple instances but is outright difficult to configure properly. Vista has problems with running both the SMP clients with the GPU2 clients, so you may be forced to choose but with the most recent client/core there are ways to get arround it but again it is somewhat difficult to configure: Some trial and error may be necessary. At some point the Stanford officials are committed to revisiting the scoring system and I have no idea what's going to happen but it may be dramatic.

 

[Mercutio]

I'd hate to think what the power bill for a quad core SLI rig that manages to keep all its processors at or around 100% utilization.

 

[P5-133XL]

I'd hate to think what the power bill for a quad core SLI rig that manages to keep all its processors at or around 100% utilization.

Do note SLI is not currently supported, but four individual cards is.


I don't think it would be nearly as bad as you'd think. It will be relatively high but folding tends to be significantly less stressfull than actual 3d Gaming. Also, it is now common for the system power usages to be given with video card reviews. So looking at Video card power usage chart The system power usage for Nvidia 9800 GX2 SLI'd only needs 460W while under load which is theoretically equivilent to four 9800GTX's.

Four 9800GTX's on a stock Q6600, assuming that folding scales with additional boards should produce somewhere around (4 cards at 6000PPD per card) 24,000 PPD total for that 460W. However, a stock Q6600 using 150W folding an SMP client can only produce around 3000PPD (when OC'ed). So you are getting 8x the results for 3x the power. I would also give odds, that it is alot cheaper to buy one expensive Q6660 system with four $200 video cards and a $200 power supply then buying eight cheap Q6600 systems.

Now if they were four GTX 280's the numbers will change. Rather than $200 cards we are now using $600 cards. And the power usage differental between a single and a SLI'd 280 is around 200W so four of them should be somewhere around 908W. So you've doubled the wattage, tripled the card costs and gained basicly no folding productivity as compared to the 9800 GTX (for now).

 

[P5-133XL]

Just as a side note, in my local power company charges 7.23 cents per KWH which for 460W 24x7 is $24/month. and 910W would cost $47 per month

 

[Mercutio]

The power bill for my apartment full of computers, HDTVs, electronics and an extra freezer still manages to come out less than $100 a month, so that would be an awfully expensive computer.

 

[P5-133XL]

So I took the chart and found the lowest price for each card (inc rebates) on newegg. Then I divided the avg PPD by the cost. The card with the best ratio is the 8800GS with 52.63PPD/$. The next highest is the 8800GT 256MB @ 46.07PPD/$.

Updated. The numbers previously listed were based on older data.

I'm going to try to incorporate energy costs. I scoured the internet to get aprox system energy usage on the various cards while loaded. I'll also assume a $0.10/KWh cost (your local cost may be different) and then extropolate all PPD and costs over a two year lifetime.

                             Orig   Energy   Points
Card          PPD       W    Cost    Cost      Per
                              $       $/y        $
  
8800GTS        4223    283    150    248      4770
8800GTX        4538    228    170    200      5810
8800GT 256     4607    236    100    207      6540
8800GT 512     4762    269    125    236      5820
8800 Ultra     4950    330    236    289      4440
9800 GX2       9996    289    330    253      8730
8800 GTS 92    5512    264    150    231      6570
9800 GTX       5937    212    170    185      8030
GTX 260        6476    263    245    230      6710
GTX 280        6901    313    400    274      5310
GTX 280 SLI   13002    508    800    445      5620
9800GX2 SLI   19992    461    660    404      9940
8800 GT SLI    9524    271    250    237      9602

Most of the loaded System Wattage comes from AnandTech from a variety of video card reviews.
The 9800 GX2 has the PPD doubled on the assumption of running two clients.
The SLI versions have had their PPD doubled assuming twice the clients. so the x2 SLI'd version has 4 clients.
The Points per dollar does not include any CPU/SMP clients that may also be running
The GPU2 PPD is assuming a CPU capable of running the GPU client at its maximum.


The Energy cost/year calculation = System Wattage/1000 (to convert to KW) x 365days x 24Hr x .1($ 0.10 per KWh)
Example GTX 280: $274/year = 313W x 365 days x 24 hours x .1 ($/KWh) / 1000

So to get the points/$ over two years = PPD x 365 days x 2 years / (Original Cost + Energy Cost/year x 2 years)
Example GTX 280: 5310 = 6901 x 365 x2 / ($400 + 2 x 274).

I'm making this so that others can judge a more realistic cost for GPU2 folding. The Wattage numbers are at best an estimate since they have not been measured while actually folding and they incorporate the entire system but different computer are more or less efficient. So take the numbers with a certain grain of salt. That being said, I think they may have a value to some.

 

[CougTek]

For those who fold using their Nvidia GPU, upgrading to the fahcore version 1.15 and using the 178.08 CUDA driver results in at least a 10% bump (probably closer to 15%) in ppd production. I might hit 5000ppd thanks to this and I spent exactly 0$ for it.

 

[Fushigi]

My wife's XP machine, as documented in another thread, has been unstable. I've reinstalled it a couple of times. This makes her upset, which isn't good. The last time I caught the BSoD screen, it pointed to the ATI driver so yesterday I upgraded to Catalyst 8.10 (from 8.9). If that fails to fix the problem I will throw hardware at it and replace the card with an nVidia unit. The mobo has a single PCI Express slot. From the chart it seems the 9800GTX or GTX+ offers the best Folding bang per buck at the moment. I'm looking at this one from EVGA. Thoughts?

BTW, what's with the brands? Zotac? Sounds like an allergy medication.

 

[P5-133XL]

Try looking at the 9800GXT X2 as a single card that runs two GPU clients each at the same PPD as your 9800GTX. The card costs more but not twice as much as the single and uses less than twice the electricity too. So as to folding cost effectiveness it is better than either the 9800GTX or the GTX+.

One thing of note, and i'm being cautious here: The Nvidia chip manufacturing defect may or may not apply to desktop cards. To be safe, you probably need to buy a card that the chip uses 55nm technology and was manufactured in the past 3 months or so. Otherwise, I would be buying into a card with a lifetime waranty from a well known reputable card manufacturer (like EVGA)

 

[Fushigi]

I'm not seeing any 'X2' cards in the 9000 series unless you count GX2. ATI's product names aren't good, but nVidia's are downright horrible. GT, GTX, GTS, GX2 .. which is sucky/good/better/best? And now there's the GTX 200 series; how do they compare? From the chart they seem to be on par with the 8800 series & behind the 9800 series.

OK, went to nvidia's site & the GX2 is the dual-GPU offering. Back to Newegg to look up GX2 cards ..

After reading the comments I'm not impressed. The PC is in a bedroom and while it's over 20 feet from where our pillows are, I dont' think I want obtrusive fan noise in the room. I don't want a card that runs hot at idle; that's crazy in this day and age. Also not impressed that it appears to vent into the case instead of out the back like the ATI cards. For $400 I'd like a better design.

 

[P5-133XL]

They will run hot and are noisey (The basic nature of very-high performance video cards), but they all exhaust externally. I offered it as a solution to your best folding bang for the buck question. If it is not for you, then feel free to go back to your original choice.

 

[Fushigi]

One commenter mentioned the card exhausted into the case.

Anyway, I think I'd favor an OCed GTX+ like this one. But hopefully the updated Catalyst drivers will work & I won't have to drop another $200 on the PC.

 

[P5-133XL]

I think that is a fine choice.

Nvidia cards don't use Catalyst drivers. Unless you have both Nvidia and ATI cards simultanously on the same machine, I suggest you uninstall the ATI drivers before installing Nvidia's.

 

[P5-133XL]

You will also probably have to uninstall and reinstall GPU folding -- Nvidia and ATI fold different WU's and trying to fold an ATI WU using Nvidia drivers probably won't work so well.

 

[P5-133XL]

It's starting to get cold. I just ordered a couple of cheap Nvidia 9600GSO's for $40 each. I'm going to experiment over the next two weeks to see which machines work best with the two cards and then power-up a couple of 24x7 folders: Computers, used as heaters!

 

[Fushigi]

So, what's the 1. best video card for GPU Folding and 2. best back/buck video card for Folding?

I'm currently running Radeon 3870s in a couple of PCs. I know they are a couple of generations old. I prefer ATI but understand that nVidia may be the performance king right now.