The Folding@Home GPU Statistics Database

[7im]

How much data are you looking for?

http://www.maximumpc.com/forums/viewtopic.php?t=83681
http://www.maximumpc.com/forums/viewtop ... 9#p1214269
https://spreadsheets.google.com/pub?key ... =100&h=750
http://www.evga.com/forums/tm.aspx?m=376316&mpage=1
https://spreadsheets.google.com/spreads ... yTGc&gid=4
http://www.bjorn3d.com/forum/showthread ... rence-list
https://sites.google.com/site/gpuppd/home
https://spreadsheets.google.com/spreads ... =en&gid=17

You could also harvest data from reviews like this...

http://hexus.net/tech/reviews/graphics/ ... 8/?page=10

[Dark Pulse]

When a FahCore is being developed/optimized/fixed, there can be significant TPF variations from one version to the next. (That's implied in the beta team rules of engagement.) Once a FahCore is released to the general public, variations can still happen but they're much less likely and/or are more likely to be quite small.

If you choose to collect beta information, you'll probably need to monitor it for changes and figure out a way to purge data associated with an earlier version of the FahCore, should that be necessary. I suppose collecting the FahCore version number (not the FahCore number) with the other data would help you manage that well enough to keep your results accurate.

(@kiore: you didn't report that information.)

Fair enough, but in my case, until the Kepler cores roll out more (I've still yet to get a single WU that needs it), I'm stuck folding on the Fermi stuff.


If it's kosher to disclose that, though, I'll certainly contribute results for my GTX 690 and, once it arrives, my GTX 680M on my laptop. While, of course, making sure the Core used w/ Version number is noted down and that the hardware is technically unsupported officially at the present time is noted as a caveat.

[Nathan_P]

Make that 4 to 0. I think this is a fine idea.

3) Some time back, VijayPande mentioned that they'd like to incorporate a bonus program for GPUs into the FAH projects. I have seen no recent statements about it and there's no prediction when, if ever, it will be rolled out but if a QRB system does come to pass, it may seriously impact the data you're able to gather. Rather than gathering data in PPD, perhaps there's another fundamental unit that can be gathered such as TPF and baseline Points Per WU that won't need to change if a bonus is later added.

Just a thought, unlike SMP where we have a bonus points calculator that we can look up expected PPD, there is not one for GPU projects, PPD would still be a good idea as that is the question that invariably gets asked.

I also secound the idea that info on all currently capable folding cards is included if possible, people still pick up older GPU's for cheap and knowing PPD makes cost/benefit calculations a lot easier

[artoar_11]

Just a thought, unlike SMP where we have a bonus points calculator that we can look up expected PPD, there is not one for GPU projects, PPD would still be a good idea as that is the question that invariably gets asked.

I also secound the idea that info on all currently capable folding cards is included if possible, people still pick up older GPU's for cheap and knowing PPD makes cost/benefit calculations a lot easier

Accidentally I saw the "Bonus Point Calculator" can calculate all projects. Click with the mouse on "Load all projects" (below) and will load all projects

http://www.linuxforge.net/bonuscalc2.php

[compdewd]

Thank you, 7im, for the sources! I have and will be entering data from them.

Dark Pulse, as for the your currently beta hardware, I don't think Bruce had a problem with having the data shared, so if you submit it I will enter it.

Also, all, note I have now made the database accessible: Folding@Home GPU Comparison Database

[Ripper36]

I would like to start gathering data on GPUs and their points per day (respective to each project) to make a comparison database.

I have an HFM.net database of approx 1000 units on GTX570s over last three months, all core 15 (2.22). Not always sure about clock speeds as I experiment a bit. Another determinant is PCIe (2.0 vs 1.1)

I also have a smaller number of units for GTX590, GTX550Ti and GTX220

Is there an easy way to export from HFM.net in a way that is useful to you? Or are single instance examples all you want?

Cheers

[compdewd]

I have an HFM.net database of approx 1000 units on GTX570s over last three months, all core 15 (2.22). Not always sure about clock speeds as I experiment a bit. Another determinant is PCIe (2.0 vs 1.1)

I also have a smaller number of units for GTX590, GTX550Ti and GTX220

Is there an easy way to export from HFM.net in a way that is useful to you? Or are single instance examples all you want?

Cheers

Don't worry about getting clock speeds. The database summary page doesn't take into account the clock speeds so it's not that big of a deal.

As for HFM.NET, I don't know of a way to export the list of work units completed, but if in HFM you go to Tools > Benchmarks Viewer and copy each of the boxes of text for each project into a text file and email it to me ([email protected]), I can sort it out and put in entries containing TPF, PPD, Project, and FAHCore. And, if you want, you can go to Help > View HFM.NET Data Files and email me the file "WuHistory.db3" and I'll see if I can find out how to decipher it's contents

[7im]

...
Don't worry about getting clock speeds. The database summary page doesn't take into account the clock speeds so it's not that big of a deal.

Clock speed makes a big difference in points. Are you averaging PPD or filtering non stock data, or showing the data as PPD per GHz?

[compdewd]

Not filtering non-stock data. I have two sheets of a spreadsheet. The first shows average TPF (if there is data for it) and average PPD for the sake of simplicity. The second sheet shows each individual entry of data which includes all that is stated in the original post (which includes clock speed). If someone wants to make a detailed comparison of GPUs, they'll have to look through the list of individual data entries which are sorted by GPU. I wish I could make people not have to do this, but I'm not a database expert by any means and there's only so much I can do with all the different kind of data that is characteristic of graphics cards.

The reason I told Ripper36 to disregard clock speeds is because I didn't want to inconvenience him more and/or possibly record invalid information if he happened to not recall his clock speed configurations correctly

[Dark Pulse]

Dark Pulse, as for the your currently beta hardware, I don't think Bruce had a problem with having the data shared, so if you submit it I will enter it.

I'll fire off a PM and ask him what the policy on it is, then. If it's okay for now, I'll certainly submit some numbers.

Even if it is allowed, though, obviously they will need to be re-tested once they're officially supported.

[kiore]

GPU: GTX 560 ti 384 shader version.
Overclock, yes 900MHz
Driver version: 301.42 .
Client Version: V 7. 1.52
Fah core 0x15 version 2.22
Project: 7623
TPF: 5.14
est PPD: 14272

2 identical cards doing this.
These units considerably bigger slower and hotter by about 10% than the 8008's. Fans set to full and foldon

edit to correct typo on core version.

[compdewd]

So question if I have one person say they have card A and it gets X amount of points on project B, and I have another person say their 100 cards A's get Y amount of points on project B, should the second person get 100 entries of their data entered? or should the second person's data be averaged and entered as one entry? The problems are that in the first method, the averaged data for card A, project B (and really all projects for card A) is skewed in favor of the second person's data, whereas in the second method, the data is skewed as a misrepresentation of the total amount of data received. So which method is better?

When presented with this problem on a smaller scale, I have gone with method 1 since it will give an accurate representation of all data received, however, on a scale as large as the above, the first person's data is overshadowed by the second person's and in the big picture might as well not even have been recorded. The situation close to the one as described above will be happening in the near future so I would like input on how everyone thinks it should be before I have to choose one

[7im]

What does it matter if if the 100 cards come from one user, or 10, or 100?

[compdewd]

You're right, 7im it was 3 AM for me. I meant to say what if the 100 entries were for the same card from the second person, should they be averaged and then put in as one entry? And I decided the answer to that is yes to make equal representation of each card. So, dumb questions on my part. Please forgive

[kiore]

GPU: GTX 560 ti 384 shader version.
Overclock, yes 900MHz
Driver version: 301.42 .
Client Version: V 7. 1.52
Fah core 0x15 2.22
Project: 8020
TPF: 4m 11s
est PPD: 19816

#2 card same specs same project..
TPF: 4m 12s
est PPD: 19738.