Major NVidia cards comparing PPD/W and PPD/$

[Michael_McCord,_M.D.]

Dear Fellow Folders:

As encouraged by Bruce of Pande Group--thank you BTW, here is my study of some high-end Nvidia cards and their PPD/W data I compiled. These cards were all in Quad-core CPU mobos with 2Gb RAM each, using Windows OS, 32-bit, some XP, some Vista. The no-load column is a measurement of the OS load only. All power measurements were made using a Kill-a-Watt meter on the PSU outlet. The load column is a measurement of the load of the GPU2 client and OS only. The PPD column is as measured by FAHmon. The current pricing is that available today on pricewatch.com.

Graphics Card-- No load (W)--GPU load (W)-- PPD-- PPD/W-- Price-- PPD/dollar

1. GTX 280 #1 158-------- 237-------- 6583------- 27.78---------- $445-------- 14.8

2. GTX 280 #2 135-------- 225-------- 6689------- 29.73---------- $445-------- 15.0

3. 8800GTS-G92 #1 101-------- 175-------- 5317------- 30.38---------- $182--------- 29.2

4. 8800GTS-G92 #2 112-------- 170-------- 4996------- 29.39---------- $182-------- 27.5

5. 8800GTS-G92 #3 103-------- 172-------- 5184------- 30.14---------- $182-------- 28.5

6. 8800GTS-SSCG80 149-------- 222-------- 4275------- 19.26---------- $169-------- 25.3

7. 8800GTX-G80 198-------- 264-------- 4557------- 17.26--------- $375-------- 12.2

There you have it. Of these cards tested I believe the G92 8800 GTS is the way to go at present.I welcome questions, comments and more data from others. Also the G80 cards suck about 50 more watts continually than the G92 which adds up over time.

For some reason-I can't get a nicely spaced table of data here. Looks great until I post it haha.

.

[legoman666]

viewtopic.php?f=43&t=3193&st=0&sk=t&sd=a&start=570#p44411

ftp://legoman666.dyndns.org:47024/scores_nvidia.zip

http://www.atomicmpc.com.au/forums.asp?s=2&c=7&t=9354



results from 413 data points.

[Foxery]

There is a thread for exactly this purpose here:
http://foldingforum.org/viewtopic.php?f=43&t=3193
It includes an enormous spreadsheet with all of this information neatly formatted.

[Michael_McCord,_M.D.]

Wow-thanks to legoman for that data! Foxery-I don't see the PPD/W data I was interested in on that big spreadsheet. With power costs on the rise....and I think that will continue...power costs are important to most volunteer folders, which was my main drive to do this little study. I also wonder what will happen if the WUs get bigger with more atoms to move around will a more powerful GPU/more/faster GPU memory be important?

[legoman666]

Wow-thanks to legoman for that data! Foxery-I don't see the PPD/W data I was interested in on that big spreadsheet. With power costs on the rise....and I think that will continue...power costs are important to most volunteer folders, which was my main drive to do this little study. I also wonder what will happen if the WUs get bigger with more atoms to move around will a more powerful GPU/more/faster GPU memory be important?

ftp://legoman666.dyndns.org:47024/scores_nvidia.zip As I posted before. It's on the 3rd worksheet.

[Carel123]

Good work, but i cant imagine that a 8800GT saves 28 w with onlly reducing 256MB? Memory hardly uses energy. The clockspeeds must be almost the same, cauz of results.

I think its also handy to place preconditions for the CPU-power to reach those points.

[toTOW]

For this kind of measure (PPD/$) you have to include the cost of electricity during the board lifetime (let's say 2 or 3 years 24H/24) to best reflect real costs, and also take into account the purchase of the entire machine.

You'll see how all PPD/$ values will collapse

[bollix47]

@legoman666

The Watts figures you are using in the above table appear to be the 'at idle' usage. Obviously, there are zero PPD being produced when the GPU is idle. Thus, the PPD/W is kinda meaningless as it stands now. For a true indication the calculation would have to use the wattage figures 'at load'.

IMHO

[legoman666]

@legoman666

The Watts figures you are using in the above table appear to be the 'at idle' usage. Obviously, there are zero PPD being produced when the GPU is idle. Thus, the PPD/W is kinda meaningless as it stands now. For a true indication the calculation would have to use the wattage figures 'at load'.

IMHO

Negative, the values I took from the chart are max 3d draw (which are actually higher than fah draw). Read the legend carefully.

Good work, but i cant imagine that a 8800GT saves 28 w with onlly reducing 256MB? Memory hardly uses energy. The clockspeeds must be almost the same, cauz of results.

I think its also handy to place preconditions for the CPU-power to reach those points.

I thought the same thing about the power usage numbers for the 8800GT 256. I'll scrounge around to see if I can come up with a better value.

For this kind of measure (PPD/$) you have to include the cost of electricity during the board lifetime (let's say 2 or 3 years 24H/24) to best reflect real costs, and also take into account the purchase of the entire machine.

You'll see how all PPD/$ values will collapse

Indeed. That's why I included PPD/W.

[bollix47]

Negative, the values I took from the chart are max 3d draw (which are actually higher than fah draw). Read the legend carefully.

Okay, thanks for clearing that up. It appears that your figures are for the GPU only and Michael's are for the entire system. That's where my confusion came from.

[MtM]

Legoman, do you think you could assemble a same chart for the Ati cards? I know you have an ati ppd spreadsheet but not sure if holds the wattage.

And the same goes for the smp client, I'm looking into some projects on my own but either I will have to rely on your collected data, look around myself ( which I don't think so, to time consuming ) or make a data miner tool for people to run so they don't have to go through allot of hassle to get the info we need.

[Foxery]

An ATI chart would look pretty poor right now, as the current FAHCore doesn't take advantage of the 4000 series' extra shaders. They look like horrible deals until the next FAHCore arrives. So far, the 3850 & 3870 are probably clear winners for all metrics.

[MtM]

Doesn't influence the reason why I requested the info one tiny bit

I still would like to know ppd/watt for any given card with current wu's. And ppd for any cpu running the smp client, with hw nfo ( cpu type,clocks,ram ) and power usage info. If I can find the same info for Ati and Cpu's running smp I can create a simple calculator to predict ppd, and running costs for any combination of hw.

[Xilikon]

Nice idea of having a PPD/W chart, this will help people get the best bang for the buck while staying very efficient in terms of power use.

[grumpydaddy]

@legoman666

The Watts figures you are using in the above table appear to be the 'at idle' usage. Obviously, there are zero PPD being produced when the GPU is idle. Thus, the PPD/W is kinda meaningless as it stands now. For a true indication the calculation would have to use the wattage figures 'at load'.

IMHO

Don't you think that the info about Idle power usage is relevant ? When machines are already running for some other purpose this figure can be deducted from the total and therefore the additional cost of folding is what counts.

My current setup x38 q6600 g0 @3.2 2gig ram 4,4,4,10
xp sp2 forceware 177.79 6.20 core not changed (1.09?)

EVGA 9800gtx ssc (770,2101,1150)core shader mem

all projects so far (5006,5007,5008,5215,5216,5229,5232,5504,5505,5506)
6286ppd (last 3 frames) Benchmarks shows different

EVGA 9800gx2 ssc (757,1949,1050)both GPUs

all projects so far (5006,5007,5008,5215,5216,5229,5232,5504,5505,5506)
5924ppd (last 3 frames) Benchmarks shows different

Rig idle=285watts
Full folding=495watts folding load =210w
just gtx = 370watts =85w
just 1*gx2=350watts =65w
full gx2= 405watts =130w
These figures don't quite add up but then the quality of the watt meter may be to blame