GPU folding is here!

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
I started to run the GPU3 client on a GeForce 450GTS and the utility Bozo found tells me that it predicts 1694ppd. It better be a fucking joke. A mid-range Fermi GPU should yield ~6000ppd, not less than 1700ppd. I guess I'll know tomorrow.

I only have the card untill Monday morning. I hope I'll have the time to rank in at least 10000 points by then.

I also started the smp client on an Athlon II 635 to test the system's stability. It should also stop Monday.
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
A 450GTS gives 11938.9ppd according to the monitoring software Bozo dig. It's not bad for a 130$ investment.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,737
Location
USA
Can GPU and SMP clients be run at the same time on the same machine? I remember something from long before that both couldn't run at the same time, or maybe they could, but performance was bad?
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
I started both with the Bozo tool just before leaving the office today. I'll have the answer tomorrow morning.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,737
Location
USA
Is you heavy producing system still running Linux or did you give in and run Windows?
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,737
Location
USA
Thanks for confirming it. I started up a temp system for an artificial boost for a couple days.

Congrats on the new title and thanks for inventing it!
 

LunarMist

I can't believe I'm a Fixture
Joined
Feb 1, 2003
Messages
16,624
Location
USA
Yes, they do. Confirmed.

And on this statement, I'm Storage morphed to another Hairy Aussie. I invented the title and I now wear it.

So you are responsible for that. I feel so dirty, or is it just hairy. :queen::geek: :rr:
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,737
Location
USA
Just use that nifty folding tool you sent us, let it auto-detect it, and try starting the GPU client. That's what I did for my 8800GT. You'll see if there are any errors.
 

Bozo

Storage? I am Storage!
Joined
Feb 12, 2002
Messages
4,396
Location
Twilight Zone
Apparently the utility doesn't support ATI cards at the moment.

I downloaded the client and will see if I can get going.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
13,737
Location
USA
Then why does it have radio button selections for ATI R600 and ATI R700? Does yours have those buttons too?
 

Bozo

Storage? I am Storage!
Joined
Feb 12, 2002
Messages
4,396
Location
Twilight Zone
Yes I have the radio buttons.
When I run the 'Autodetect GPUs' I get a message that there are no NVidia compatable GPUs.
I'll keep trying.
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
I restarted GPU folding with ther new v7 beta client on my GeForce 440 and it's turning my system into a sluggish snail. I can't even watch a video on Youtube properly and just net browsing on Firefox is painfully slow. Playing Mah Jongg on my Windows 7 computer is only a faint memory now. At least it'll give me some ~4000ppd more and I have another computer I can use for my daily tasks.
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
Ok, my new system is assembled and running. Now the latest version of the FAH client (7.3.6 IIRC) only wants to enable GPU folding when the display is idled (off). WTF? How can I change it so that it folds all the time, even when i'm using the system?
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
If anyone else is doing GPU folding with Nvidia cards, you should remove the "client-type = beta" slot, if youadded it previously. I've got one of those huge unit 7661 that Mark wrote about. On a GTX 580, it would have taken more than 7 days to complete and I would have missed both the prefered and the absolute deadlines. The point per day approximation was 197 (for a card that usually makes ~30Kppd).

So I removed the beta client-type and delete the content of the work directory. Those are only good on AMD's hardware, according to Mark. It will stay that way for me.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
There is something wrong on that GTX 580 for they take about 9-11 hours on my GTX 460's. Even still they are poor PPD producers for me.
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
In another thread :
I'm seriously considering selling the 3x 580's and using that money to buy a 780 or 2x GTX 580 for a single HD 7970. The 780 stock gets 120K PPD on p8900's while the 580's get 30K. The 7970 gives 100K PPD but only costs $400. Each used 580 can be easily be sold for $200 on ebay for a total of $600+ and a 780 costs $649 new. On the short term both are a push in cost, but the single 780/7970 will give more PPD while using a ton less power.
I thought it would be more appropriate to post this here. Does the Radeon HD 7970 produces 100K ppd in Linux or in Windows too? Only with beta units or with default options too?
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
for AMD use Windows only for Core_17! The AMD proprietary drivers in Linux need work so Linux GPU folding is currently Nvidia only. The non-proprietary drivers for all manufacturer don't have the proper OpenCL support as of yet so you are stuck with proprietary drivers. So maybe eventually, but not yet.

The only core that works directly with Linux is Core_17 and that is currently only offered in beta. There is an option to run v6 under Wine. You can run beta using Wine too. It is a bear to set it up to run Core_17 under Wine though because it is not fully documented.

I was told by proteneer that Core_17/P8900 is likely to go advanced soon. That may be a problem for AMD because then those p8900 will be competing with all the other core_15/16 projects and you'll get alternating high and low PPD projects. I'm talk'n 100K PPD for p8900 followed by a 10K PPD for something else. The same will happen for low-end cards but less drastic because they don't get much QRB points for p8900. I think people are going to start demanding a way to separate core_17 from other cores so they can choose to run it or not. Currently the way is beta but that method will be going away when it gets released to the general population.
 
Last edited:

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
Just so that people know, a Radeon HD 7970 produces more like 76K ppd rather than 100K ppd. A Radeon HD 7870XT yields ~50K ppd. Both are a lot more rewarding than my older GTX 580 (~27K ppd).
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
I've just realized that at this very moment, my average 24h production is equal to awachs', one of the biggest user by out put of all time. He probably is a it, as I think it's a cloud cluster, but it's something anyway.

It's still unclear who will win the month of June. David still has a potential of ~500K points to add to his 3.8 million total (for June). Mark is in front at 4.17 million, but shouldn't add more than ~250K points, putting him barely in front of Dave. I'm a bit more than 200K points behind Mark, but I've just added a 70K ppd system and I was already producing, on average, about 20K points more than Mark. It will be a close finish (that Mark will probably win).
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
I have everything computer related all on one 20A circuit and it has never blown even when I had 4 more machines than I do now. Unfortunately, I do go through surge protectors with regularity but I keep spares.

Recently my game playing has significantly been impinging upon my folding productivity (Skyrim). Over this week-end, leaving now and be back late Sunday, I'm traveling to Seattle so unless something blows a gasket expect an additional 40K PPD for the next 3 days. Unfortunately my 580's need monitoring else they like to randomly down-clock on p8900's till they are rebooted.

I consistently get 39K PPD on 580's@800-825 running p8900's. Certainly better than 27K. Other's get 100K - 111K PPD on 7970's depending on OC'ed. CougTek, why do your machines run so much slower than others?

DDrueding, when is your dual-Titan machine going to start to fold? The 2x Titans should be able to add 250K PPD.
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
I consistently get 39K PPD on 580's@800-825 running p8900's. Certainly better than 27K. Other's get 100K - 111K PPD on 7970's depending on OC'ed. CougTek, why do your machines run so much slower than others?
I did not enabled beta units on my GTX 580 because I had an issue with one several months ago. I've just re-enabled them. I've also set the control panel to max 3D performances for the FAHCore, so that it doesn't throtle while the FAH client is working.

Regarding my HD 7970, which runs at 950MHz (slight overclocked), I simply installed the latest driver and started the FAH client, with beta units enabled. That's it. If there's a way to get 30% more points out of it, it's not configured by default.

And despite my 5 systems operating 24x7, I've never blown an electrical circuit. It's a 120V/15A one and they all are on the same. 4 of the 5 systems have graphic cards capable of using ~200W each. Two are Nehalem Core i7. One is a tri-core Athlon II. The other two systems are a Sandy Bridge and an Ivy Bridge. I sometimes start another computer or two on the same circuit and nothing wrong happened.
 

ddrueding

Fixture
Joined
Feb 4, 2002
Messages
19,511
Location
Horsens, Denmark
The Titans are folding now, but without an overclock and without beta units. That system is pulling ~900W and the other (dual 690s) is around 850W. Add that to the silent machine (250?) and some auxiliary equipment I had going, and it was likely 2500W+. 20A@120VAC=2400W
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
Ok, so with a Project 7625 unit, my GTX 580 achieve 31.7K ppd and with a project 8900 unit, they reach 35K ppd. Still not 39K, but it's closer. They do 27K ppd with a project 8074 unit.

And the post #105 of this thread should have been made in the generic Folding@Home thread. Sorry for that.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
What has been your failure mode?

They just stop working and anything plugged in gets no power. I get a replacement (I stock extras now, just in case) and everything runs again. Most of the time there is no power outages or known potential causes. I suppose that surges may be the cause but I'm not seeing any effect (other than surge protectors failing). I do not monitor power at any level that would be helpful in recording a surge. The protectors tend to be brand name and high-end units...
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
The Titans are folding now, but without an overclock and without beta units. That system is pulling ~900W and the other (dual 690s) is around 850W. Add that to the silent machine (250?) and some auxiliary equipment I had going, and it was likely 2500W+. 20A@120VAC=2400W

beta WU's or client-type advanced. You really want those core_17 WU's for high PPD.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
I did not enabled beta units on my GTX 580 because I had an issue with one several months ago. I've just re-enabled them. I've also set the control panel to max 3D performances for the FAHCore, so that it doesn't throtle while the FAH client is working.

Regarding my HD 7970, which runs at 950MHz (slight overclocked), I simply installed the latest driver and started the FAH client, with beta units enabled. That's it. If there's a way to get 30% more points out of it, it's not configured by default.

Be very conservative with OC'ing with CORE_17. It is very sensitive and will bring your PPD down a lot as it rejects p8900's (giving only pro-rated base points) or down-clocks your cards (killing of TPF's and QRB) and at the worst Core_17 claims it progresses in the client but the log stops recording the WU and it never finishes causing no points to be credited and a reboot will restart at the last know checkpoint (typically very early). These are all common failure modes of OC'ing with Core_17.
 

CougTek

Serial computer killer
Joined
Jan 21, 2002
Messages
8,724
Location
Québec, Québec
I think you mentioned that the Linux NVidia GPU client was pretty stable. I currently have two systems running the Windows GPU client, both on NVidia GTX580. If I convert both of those to Linux, should I expect similar or slightly better production? One is an AMD Athlon II X3 435 and the other one is a Core i7 960. I know the Linux GPU client is still in beta.

I'd like to move my Windows licenses to other systems.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
There should be no performance difference between Linux and Windows for the Nvidia GPU client. For the CPU client, Linux runs slightly faster.
 

timwhit

Hairy Aussie
Joined
Jan 23, 2002
Messages
5,278
Location
Chicago, IL
I'm attempting to run the F@H Client with GPU folding, but can't seem to get GPU folding to work.

My GPU is a GTX 650 Ti, which I believe is supported.

Log file is attached.

View attachment fah.txt

Any help is appreciated.
 

P5-133XL

Xmas '97
Joined
Jan 15, 2002
Messages
3,173
Location
Salem, Or
9 seconds of log isn't much to go on but I have questions:

Are you running a 64 or 32 bit Linux kernel? X64 Kernal is required for GPU folding in Linux and you also have to have the x64 bit Linux folding client.
Next, what happens if you try to add a GPU slot (There isn't one defined in your config) using the Advanced control (FAHControl)?
 
Top