Low end gpu folding

[Nicklebon]

Anyone folding on low end nvidia gpus? I've got a 9300ge running 177.35 drivers and it is turning in 460ppd. This seems low even for this card. Are my expectations that out of whack?

thanks

 

[APOLLO]

Along the same vein as the OP, anyone know what to expect from an 8600GT or 8500GT?

 

[drgnfang]

My 8600gt got ~1400 ppd before over heating (it's passive cooled). That's at stock. Seems average based on the various "show your points" threads. No idea on the 9300ge.

 

[metallicafan]

I have an 8400GS (16 shaders) running right now on the 177.35 drivers. Its doing about 575 PPD at the stock clocks of 460/920/800. If I bump the shaders to 1000Mhz it only raises my PPD to about 620 PPD. Just not enough shaders on this card to do any real damage. (Thats why its being replaced!) Every WU counts though.

 

[Vediovus]

my 8600gt with the shaders cranked up to 1400 is ~1700ppd

 

[alan2308]

I have an 8400GS (16 shaders) running right now on the 177.35 drivers. Its doing about 575 PPD at the stock clocks of 460/920/800. If I bump the shaders to 1000Mhz it only raises my PPD to about 620 PPD. Just not enough shaders on this card to do any real damage. (Thats why its being replaced!) Every WU counts though.

Don't forget to sell it and offset the cost of its replacement. Plenty of folders would love to grab an extra 620PPD.

 

[APOLLO]

I have an 8400GS (16 shaders) running right now on the 177.35 drivers. Its doing about 575 PPD at the stock clocks of 460/920/800. If I bump the shaders to 1000Mhz it only raises my PPD to about 620 PPD. Just not enough shaders on this card to do any real damage. (Thats why its being replaced!) Every WU counts though.

Don't forget to sell it and offset the cost of its replacement. Plenty of folders would love to grab an extra 620PPD.

Yeah, 620PPD may seem like a low production level these days, but it's more than I was getting with the GPU1 client and my AGP 1950!

 

[BillR]

I think perhaps too many of you are relying on F@H monitor for your ppd numbers. The problem with that is if you take a look at your log files and start doing the math you will see that within each work unit you have frames. If you calculate the folding times of a lot of frames a few times you will probably find they frames folding at very different speeds.

I’ve got frames that finish in 30 seconds and others that run as high as 2 min all within the same work unit. The monitor program simply doesn’t deal with these variations very well at all.

Might I suggest you pick a time, say 8PM and open the log file and note where you are. 8PM the next day open the log and see how many work units you completed then compare that to the chart on Stanford to total your points.

The other simpler way is look at hardfolding.com and look at your PPD stats. You can also click down to your personal graphs and look at individual days, weeks, months to see your averages. Those are the numbers that actually count in the long run.

Just a thought to get a reality check

And, don't forget each and every point counts

Luck

 

[HighYield]

I was folding on a 8500GT 512 meg DDR2 foxcon $25 special. It was getting me about 850 ppd (just a little over stock-didn't overclock for shit, although I tried )

This is more ppd than my wife's notebook (600ppd on SMP and it takes 3 days per WU!). When you lose a WU that took 3 days to run the GPU clients look better all the time. Also I bet that 8500GT uses less power than the notebook when folding.

 

[alan2308]

I think perhaps too many of you are relying on F@H monitor for your ppd numbers. The problem with that is if you take a look at your log files and start doing the math you will see that within each work unit you have frames. If you calculate the folding times of a lot of frames a few times you will probably find they frames folding at very different speeds.

I’ve got frames that finish in 30 seconds and others that run as high as 2 min all within the same work unit. The monitor program simply doesn’t deal with these variations very well at all.

Might I suggest you pick a time, say 8PM and open the log file and note where you are. 8PM the next day open the log and see how many work units you completed then compare that to the chart on Stanford to total your points.

The other simpler way is look at hardfolding.com and look at your PPD stats. You can also click down to your personal graphs and look at individual days, weeks, months to see your averages. Those are the numbers that actually count in the long run.

Just a thought to get a reality check

And, don't forget each and every point counts

Luck

Calculating things out myself is where I draw the line. I do enough of that in class.

Besides, as soon as you download a WU from a different project, everything is subject to change. Any way that you tally up your PPD is just an estimate that won't mean much as soon as tomorrow.

 

[drgnfang]

I’ve got frames that finish in 30 seconds and others that run as high as 2 min all within the same work unit. The monitor program simply doesn’t deal with these variations very well at all.

And, don't forget each and every point counts

Luck

Is this unique to the GPU client? My SMP WU's are pretty rock solid for time to complete each frame... This is across a couple hundred WU (I'm a convert from another team), and multiple boxen. Same is true for the standard client (though I don't run as many of those).

 

[BillR]

Is this unique to the GPU client? My SMP WU's are pretty rock solid for time to complete each frame... This is across a couple hundred WU (I'm a convert from another team), and multiple boxen. Same is true for the standard client (though I don't run as many of those).

At the moment being unique is what GPU folding does best. Everyone started out with much higher output and now it has dropped off a bit but there doesn’t seem to be any real pattern to the results.

As for regular and SMP units as a rule they run very close time wise. As a for instance I’ve been doing some testing with the SMP client running alongside with the GPU client on two different machines running at exactly the same speeds.

At 2:00 AM yesterday morning I started both SMP clients at the same time, each got a 2665 and both sent the finished unit in at 8:30 PM the same day however after checking the GPU log files there was no parity there at all.

I think Stanford is still working things out or this just might be a characteristic of this particular client. Only time will tell.

Luck

 

[chrisf6969]

got a crappy Dell OEM 8300GS (stock) in my work computer... it only cranks out about 330PPD...

ie: it takes about day & half for 1 WU

vs. my 8800GT at home that cranks out the same WU in about 2.3 hours... however, I haven't been able to upload them. I'm going to see if I can't find a proxy to use maybe if I do that it will work?

 

[jws2346]

@BiiR, I'm almost afraid to say it, but I'm also one of those people that depend on a "easy breezy, no intelligence required" ppd reading programs like FAHmon. What I was wondering, bein' not too swift in the math department and also bein' lazy, how far off are these programs for reading ppd

I'm one of those "close enough for government work" type people and if it's not ridiculously in error I think I can live with it, kind of using it as a reference point, only IMO it's more accurate than software temperature progams

Thank you in advance for any reply and thank you for all your hard work for the [H]orde. .

FOLD ON!

 

[Pocatello]

Is this unique to the GPU client? My SMP WU's are pretty rock solid for time to complete each frame... This is across a couple hundred WU (I'm a convert from another team), and multiple boxen. Same is true for the standard client (though I don't run as many of those).

Hey, welcome to the Team!

Nice to have ya!

170,000+ points for the [H]ardest Team in the World!

 

[lelliott731]

All I have to say is, don't complain about your "slow" cards, even the "slow" GPU2 cards are doing massive amounts of work that would take a whole lot longer on a CPU. And in this fight, every single WU, whether it take an hour, or multiple months, is worth the time.

Every spare CPU & GPU & PS3 cycle count, and are worth the time & effort to get hooked up!!!

 

[Xilikon]

I think perhaps too many of you are relying on F@H monitor for your ppd numbers. The problem with that is if you take a look at your log files and start doing the math you will see that within each work unit you have frames. If you calculate the folding times of a lot of frames a few times you will probably find they frames folding at very different speeds.

I’ve got frames that finish in 30 seconds and others that run as high as 2 min all within the same work unit. The monitor program simply doesn’t deal with these variations very well at all.

Might I suggest you pick a time, say 8PM and open the log file and note where you are. 8PM the next day open the log and see how many work units you completed then compare that to the chart on Stanford to total your points.

The other simpler way is look at hardfolding.com and look at your PPD stats. You can also click down to your personal graphs and look at individual days, weeks, months to see your averages. Those are the numbers that actually count in the long run.

Just a thought to get a reality check

And, don't forget each and every point counts

Luck

I agree, the fahmon ppd is a bit inflated vs the real ppd because it doesn't account for upload/download times and variations within a WU. I only use fahmon ppd to give a rough idea of the production and it remove 10-15% to give a more realistic estimate.

 

[BillR]

@BiiR, I'm almost afraid to say it, but I'm also one of those people that depend on a "easy breezy, no intelligence required" ppd reading programs like FAHmon. What I was wondering, bein' not too swift in the math department and also bein' lazy, how far off are these programs for reading ppd

I'm one of those "close enough for government" type people and if it's not ridiculously in error I think I can live with it, kind of using it as a reference point, only IMO it's more accurate than software temperature progams

Thank you in advance for any reply and thank you for all your hard work for the [H]orde. .

FOLD ON!

Please don’t misunderstand, I don’t sit here daily and work out all the PPD. I’m only interested in that sort of thing when something major happens such as the GPU client or if I see some strange anomaly in my daily stats.

I’m not saying don’t use a monitor program, but overall its best use is to make sure none of your boxen have suddenly stopped.

The other side of things are simpler, if you are ultra confident that you did a good build and setup then you won’t feel the need to monitor stuff.

In my case I check stats twice a day and I can tell pretty fast if something is pooched.

The newness if folding can wear off a bit but not the enthusiasm.

 

[relic]

Monitoring programs like fahmon are ideal for 'monitoring'.
Looking for changes or deviations in status.

The PPD is about 10% off but it's always about 10% off.
So production should always show about the same, just 10% higher than actual.

If something goes wrong, you will know it, and I leave it running on my second monitor so I know it right away.

 

[BillR]

Monitoring programs like fahmon are ideal for 'monitoring'.
Looking for changes or deviations in status.

The PPD is about 10% off but it's always about 10% off.
So production should always show about the same, just 10% higher than actual.

If something goes wrong, you will know it, and I leave it running on my second monitor so I know it right away.

Well, we all know about your occasional "deviations" but running a second monitor to check your clients? Where is your confidence man?

 

[relic]

Well, we all know about your occasional "deviations" but running a second monitor to check your clients? Where is your confidence man?

Well, to be honest...the second monitor monitors my servers and WAN and folding clients. First monitor and third I use to run design software, or firewall logging/monitoring software...or just F'ing around on the forums. That's my "work system".

I don't game much so my "gaming box" is really what the kids use to surf facebook and MySpace when they're home.

It's getting an 8800GT upgrade very soon. MySpace is much faster with an 8800GT vs the RADEON X300 currently installed...right?

 

[drgnfang]

Yes, that upgrade is essential to using MySpace to it's full potential.

Also, if the "hung" stuff bugs you just switch to F@HSpy (I actually use both... but I'm sick like that).