Attention GPU Folders: Make sure you are set up for QRB units

This may be a discussion best meant for a thread of its own, but what seems to be the best bang for the buck GPU now as in PPD/Watt/$? I'm thinking 660Ti but I don't know. I see 560Ti's are still scoring well, except that you have to have core 2.22 in order to do that... if you have the new 2.25 core, it comes without a few Fermi optimizations.

I have 4 SMP systems here that I could probably put to better use by adding a couple GPUs in each one. Right now they're running low end AMD cards and I would really like to switch them out with some high PPD, power efficient, NVidia GPUs to get in on the QRB goodness.

 

you can see what those that have reported are getting here. https://docs.google.com/spreadsheet/ccc?key=0Ap5lOkaO12jldG80aG03Z3BaNHI2czJ1YmZGaHVIeXc#gid=1

 

wait a sec. 74,000 ppd with a 460 and a c2q? Am I reading that right?

 

It looks like GTX 580 crushes the beta work units. I was hoping that because the GTX 680 has more than double the theoretical single precision performance that it would come out on top.

 

I clicked on the "Configure" button in the v7 client (7.2.9), but I only see three tabs named: Connection, Client, and Proxy.

Is there something else I need to do to make the slots tab appear?

-Edit, I found it, have to change it from Novice mode to Advanced or higher for that tab to show up. Depending on the size of the v7 client window, the dropdown box for selecting Novice, Advanced, and Expert may or may not be visible.

 

I get to be the wet blanket

This is beta and if you are not on the beta team you do not get support from PG.

(I understand the [H] is a helpful spot however)

That is all.

 

Wrong, Kepler is not optimized, you want fermi or bust. V7 reports 100-130k ppd on a c2d when only gpu folding on 8057 and gtx 670. 460 is 62-68k, 560 non ti 79k-ish, 560 ti 90-110k ppd.

The above depends on card speeds. I have a mix/mash of crap.

Make some PPD, gents.

 

Yup, and PG loses input/support from the non-beta folders running the beta flag.

 

I do have to say that we keep quite a bit of data here.

If PG or the beta team needs data, we have it. I'm sure EVGA and OCN does as well.

 

Well, it certainly looks like Fermi comes out on top still. Now, I'd have no problems pulling the trigger on a few more Fermi cards in the very near future but that whole core version thing has me leery.

It seems there are two choices... neither is really desirable.... please correct me if I'm wrong:

One can go with a Kepler (or Fermi) card and use the 2.25 core and have at least mediocre PPD... though Kepler seems to benefit the most from this core version. Things may change if a true Kepler optimized core comes out at some point in the future. But I personally am not holding my breath for that anytime soon.

Or one can go with the tried and true Fermi GPUs and get rather superior numbers (as far as GPUs go), but contend with having to manually downgrade your core to version 2.22 on occasion when you either see that your GPUs are using the improper version or that your PPD has taken a sudden nose dive (because the new core version gets downloaded automatically by FAHClient). This is not at all a desirable option for someone like me who has no access to their folding rigs to babysit them while at work (or just away from the house) for a good chunk of the day.

Is this really how it is? or am I missing something?

 

Nope you did not miss anything if you get assigned a 76xx WU it will force the 2.25 core and you will need to revert back to the 2.22 core. after the 76xx WU finishes. Unfortunately Stanford does not appear to be to concerned about the core crippling Fermi cards. They must figure the amount of Kepler cards it enables to fold off sets the loss.

 

If you watch the client it will update the core...I'm wondering what would happen if you made the file read only...

 

So, then the lesser of the two evils would seem to be "Kepler or bust" (and just hope that someone somewhere at Stanford has made progress on a Kepler optimized core... again, not holding my breath).

 

But what happens if the client downloads a 76xx unit and tries to run it on the older version 2.22 core which can't run 76xx units?

EDIT: Here's an excerpt from what was mentioned on the Folding Forum regarding the new 762x units (emphasis mine):

 

Your guess is as good as mine it may be the next generation of cards before Stanford gets it figured out, a good example of Stanford getting it figured out is ATI / AMD, who knows it is pretty much a shot in the dark.

 

Ok I am confused. Your post above says all that is needed is the -beta flag.

I thought I read somewhere NOT to use the -advancemethods flag with beta units. Is that true?

In any case I am turning my gpu's back on and try and get some of these points. Oh and using V6.34 on a GTX460 card.

 

-beta says bad argument .... did I miss something?

 

name = client-type
value = beta

 

oh, that is rich

 

I believe for 6.34 of the client, the argument is "-betateam" if I'm not mistaken.

 

deleted,

 

Hey Musky, I'm a Console Client guy and the "-beta" argument is not a recognized modifier for 6.34. You have to use "-betateam" and I'm speaking for Fermi cards here.

On my Kepler, using Console 6.34, I can (again) use the -betateam argument, it downloads a WU, appears to be running, but in fact never makes it past the first stage of running the WU. I'm still trying to figure out how to get a Kepler to Fold using the Console Client. From what I've read, I need to be using FAHCore15 (version 2.25) for a Kepler to Fold, so I'm still needing to play with that a bit on the Console 6.34.

Lots of discussion over in our EVGA forum on this whole subject, so feel welcome to stop by.

 

deleted

 

What about the -advancemethods flag with V6.34. Do we need it or not?

 

If it inst set in the guide, don't have it set. Some of these flags will conflict with each other.

 

I hope there is sarcasm here...AMD/ATI GPU folding has about as much point in the points contest as a one-legged man in an ass kicking contest.

 

Was it running a GPU as console or as a service that no longer worked in Vista and above? I can't remember. Might be the issue.

 

Hi Sticks,

I think the problem with using the 6.34 Console Client with Kepler is that it appears to assume that all GPUs hanging out there will run on the older FAHCores. I tried copying FAHCore_15.exe (ver 2.25, which is a required Core for Kepler) to my GPU Console Client folder. I was hoping to possibly receive a WU that would require the 2.25 Core and run.

The problem is...I'm being delivered WUs (like WU 10503) that require FAHCore11 (ver 1.31 from 2009) which get's automatically downloaded and executed. Kepler doesn't run under that Core, so the WU just hangs. Unless the 6.34 Client can look at my GPU, see that it is a Kepler and understand that I need the latest FAHCore_15 (ver 2.25) to run it, I've found no way to Fold on Kepler with the 6.34 Console Client.

It's V7 for now and probably all time with my Kepler.

 

Service, you cant run as service in Vista/7/8

You CAN run from command line.

What do you have against the V7 client? It's a better client, all-around.

 

It's not that I'm "against" V7 and I've been mainly trying to help with the conversation (here) associated with using the Console Client. I use what I find easy and effective to use and I also do use V7. I've found the Console Client easier to understand (for me) and it seems to behave better with other utilities like HFM across my network of PCs. It's really more personal preference based on use than disliking V7 as a tool. V7 has come a long way and as I said, I use it too.

 

Gotcha!

 

v6.34 is not the latest version of the client for GPU - you need 6.41r2, still don't know if that works but at least it rules out getting GPU2 work

 

You're correct Nathan, and I should have pointed out that I'm running 6.41r2. I'm so used to saying 6.34 when referring to the Console Client that it's an automatic thing I do. But yes, I've been running the latest client for awhile now and in my communications above, I was using the 6.41r2 version.

Thanks for bringing that up.

 

Yah me too. I meant to say 6.41... not sure about the r2 version will check that.

Meanwhile I have 3 GTX460's folding again but the points seen a little low. I used to get around 12-13K now getting max of 10K.
I noticed it is using ver 2.25 of the kernel and I think I saw someone here say something about not using it on 460's. Whats the info on that? Is that the reason my points are down?

Also ... these are all getting 8018's and 8043's at the moment. Hoping to get some of the big number ones to justify all the electricity I am burning.

Thanks,
Rick

 

Well I got my first 8057, on my laptop.

GPU 680m (essentially a lower-clocked GTX 670, GK104, 1344 shaders, 256-bit mem, 854/1800 core/mem)
TPF: 3m 45s
PPD: 60-61K

 

Wow.... Laptops just showed up.