15 replies to this topic

[maestro949]

Someone just reminded me about the electric bill impact. A good writeup. The PS3 suuuuucks compared to the video cards! (pun intended)

Seeking The Best Performance per Watt for Folding@Home

For the time-conscious, here are there conclusions :

# From the video cards we analyzed, GeForce 8800 GT is the one that provides the best cost/benefit and best performance/kWh ratios for running Folding@Home. Of course you will get a higher score with a GeForce GTX 260 or GeForce GTX 280, but they are more expensive and also will consume more. If you think only about the points/kWh ratio (i.e. efficiency), then GeForce GTX 260 is the best: it produces more points per kWh consumed than all other video cards.

# A “weaker” video card won’t necessarily consume less power than a “stronger” one. Just see how GeForce 8800 GTS produces a lower score and consumes more than a GeForce 8800 GT.

# ATI video cards should not be used for running Folding@Home: they have a far lower points/kWh ratio compared to nVidia cards. A GeForce 8800 GT provides almost double the efficiency of a Radeon HD 4870. If you are building dedicated systems for running Folding@Home, stick with nVidia: you will get a higher score and a lower electricity bill.

# Very low-end video cards like Radeon HD 3450 and GeForce 8500 GT are not efficient to run Folding@Home and should be avoided. From the mainstream market GeForce 9500 GT was the one with the best performance and efficiency index (points/kWh), being our recommendation on this segment.

# Running the SMP client together with the GPU client won’t necessarily increase the system performance. In our tests we saw two out of three systems where performance decreased. We found out that for each video card running the GPU client Folding@Home will completely use one CPU core (on quad-core CPUs it will use one core per GPU, so with a quad-core CPU and two video cards, two CPU cores will be constantly used; on dual-core CPUs it will use one of the two cores all the time independently of the number of GPUs you have installed). So when you run the SMP client at the same time, both clients will compete for CPU utilization, leading to a lower performance. On the other hand, consumption also decreases.

# If you want to build a system for only running the SMP client, it is more efficient if you run it with a motherboard with on-board video, because the base consumption will be lower as you won’t have a video card installed.

# The Playstation 3 achieved one of the lowest points/kWh ratio, meaning that you will feel an increase on your electricity bill without a meaningful increase in your Folding@Home score. We see lots of people praising the math performance of PS3, but this performance isn’t converted in a huge Folding@Home score because each PS3 work unit doesn’t give a lot of points.

# If you are building a system to run Folding@Home, we think that the points/kWh should be your metric for efficiency. Buying a digital watt meter will help you a lot finding what you can change on your setup to have a more efficient system. From our experience you should keep systems with a points/kWh index of at least 1,000. Systems with indexes below that should be reevaluated.

Edited by maestro949, 30 November 2008 - 02:58 PM.

[Krell]

BTW: In the last few days I have been getting a new type of work unit that give me much lower GPU PPD and higher Watts.

For instance, I am running a new (for me) 5756 WU right now on an overclocked 9600GSO with a FAHMON predicted PPD=2946 and my KILL A WATT reads 236 watts at the wall, compared to most previous WU with PPD~4500-5000 and wall powers at ~220 Watts.

[sentinel]

BTW: In the last few days I have been getting a new type of work unit that give me much lower GPU PPD and higher Watts.

For instance, I am running a new (for me) 5756 WU right now on an overclocked 9600GSO with a FAHMON predicted PPD=2946 and my KILL A WATT reads 236 watts at the wall, compared to most previous WU with PPD~4500-5000 and wall powers at ~220 Watts.

Nice post Maestro, cheers

Krell - I was going to start e new thread on this, my PPD has taken a kicking the last few days; down from 16-17k to c 14.5k, it's hit all 3 of my GPUs, I haven't isolated the work unit yet but even if I do, there's no way of avoiding it! I'm sure it will get flagged soon.

[Krell]

Sentinel - guess I am not the only one with lower PPDs!

Right now I seem to be getting a WU mix of about 50% new WUs.

There seems to be a lot of discussion of these larger work units and their effect on Nvidia GPUs at
the Folding Forum topic "New GPU projects 5749-5764"

Some reports suggest that GPUs like the 9600GSO with lower number of shaders are taking the
biggest PPD hits on the new WUs.

I have downloaded the latest Nvidia drivers and FAH client to see if that helps restore PPD.

One of my overclocked 9600GSO GPUs started rebooting with the new WU load,
so I had to throttle back on the overclock slightly.

Edited by Krell, 30 November 2008 - 04:13 AM.

[eternaltraveler]

so people actually buy dedicated computers specifically for folding@home.

that's ridiculous.

The whole point of these distrubuted computing projects is too make used of available unused computing capacity. Surely if one wants to make a difference in science or life extension there are better uses of resources than buying a few flops for folding@home.

[maestro949]

so people actually buy dedicated computers specifically for folding@home.

that's ridiculous.

Since when is human nature rational? Many of those people building dedicated folding rigs would never simply donate the money straight up to longevity science or the popular age-related killers. Some simply enjoy the more direct rewards of hands-on participation and involvement. For some, like this guy running 19 quad cores, it's a hobby. The effort needed to convince them to redirect their resources probably far exceeds their investment thus I say, more power to them. Literally

[Ghostrider]

so people actually buy dedicated computers specifically for folding@home.

that's ridiculous.

The whole point of these distrubuted computing projects is too make used of available unused computing capacity. Surely if one wants to make a difference in science or life extension there are better uses of resources than buying a few flops for folding@home.

Actually, the guy would probably be better off buying a few video cards and doing more work than 19 quad core CPUs.

BTW, distributed computing projects have had a small influence on my decisions to upgrade. I have thought about upgrading my motherboard to support a quad core CPU over my current dual core, mostly for DC (and to keep my condo warmer in the winter). Or adding extra RAM (which is dirt cheap right now) instead of simply not doing DC. I have thought about upgrading my video card to do GPU folding, but since I don't need to for the few games that I play, I haven't. I am waiting for a free after rebate video card...that and my motherboard only has AGP slot.

Edited by Ghostrider, 30 November 2008 - 11:08 AM.

[maestro949]

so people actually buy dedicated computers specifically for folding@home.

that's ridiculous.

The whole point of these distrubuted computing projects is too make used of available unused computing capacity. Surely if one wants to make a difference in science or life extension there are better uses of resources than buying a few flops for folding@home.

Actually, the guy would probably be better off buying a few video cards and doing more work than 19 quad core CPUs.

That's what this guy did with his 51 GPU Folding Farm! $80,000 worth of hardware that'll be obsolete in a few years. Ouch.

BTW, distributed computing projects have had a small influence on my decisions to upgrade. I have thought about upgrading my motherboard to support a quad core CPU over my current dual core, mostly for DC (and to keep my condo warmer in the winter). Or adding extra RAM (which is dirt cheap right now) instead of simply not doing DC. I have thought about upgrading my video card to do GPU folding, but since I don't need to for the few games that I play, I haven't. I am waiting for a free after rebate video card...that and my motherboard only has AGP slot.

Sticking with your current setup and just swapping in a new AGP video card would suggest going with an AG3850 according to this thread. The poster there suggests it would pull in 900 to 1900 points per day. Pretty damn good for AGP. It beats the PS3.

[Ghostrider]

so people actually buy dedicated computers specifically for folding@home.

that's ridiculous.

The whole point of these distrubuted computing projects is too make used of available unused computing capacity. Surely if one wants to make a difference in science or life extension there are better uses of resources than buying a few flops for folding@home.

Actually, the guy would probably be better off buying a few video cards and doing more work than 19 quad core CPUs.

That's what this guy did with his 51 GPU Folding Farm! $80,000 worth of hardware that'll be obsolete in a few years. Ouch.

BTW, distributed computing projects have had a small influence on my decisions to upgrade. I have thought about upgrading my motherboard to support a quad core CPU over my current dual core, mostly for DC (and to keep my condo warmer in the winter). Or adding extra RAM (which is dirt cheap right now) instead of simply not doing DC. I have thought about upgrading my video card to do GPU folding, but since I don't need to for the few games that I play, I haven't. I am waiting for a free after rebate video card...that and my motherboard only has AGP slot.

Sticking with your current setup and just swapping in a new AGP video card would suggest going with an AG3850 according to this thread. The poster there suggests it would pull in 900 to 1900 points per day. Pretty damn good for AGP. It beats the PS3.

I was thinking of going out to Frys when the offer a CPU + board combo for the same price as the CPU goes for on eBay, then keeping the board and selling the CPU. The rest of my hardware will carry over without a need for upgrade. It just takes a while, I would have to drive 24 miles out to that place, and hopefully if the board works spend a few hours rebuiding my system. Just takes time...

[poser]

so people actually buy dedicated computers specifically for folding@home.

that's ridiculous.

The whole point of these distrubuted computing projects is too make used of available unused computing capacity. Surely if one wants to make a difference in science or life extension there are better uses of resources than buying a few flops for folding@home.

You know, the only thing missing from your post is your pay pal address... ; )

I have given thought to donating money to science directly, but my conclusion was that since the people who get my money won't make any announcements (that's zero) on how efficiently the money is spent, I'm better off sticking to FAH and Rosetta that (especially the latter) continously commend contributors and tell in detail about the progress they are making. It's business and they have a better deal.

[poser]

How much do you guys read FAH forums? Haven't you seen this table that compares the folding output /W/$ of Nvidia cards?

http://foldingforum....php?f=52&t=4709

[eternaltraveler]

so people actually buy dedicated computers specifically for folding@home.

that's ridiculous.

The whole point of these distrubuted computing projects is too make used of available unused computing capacity. Surely if one wants to make a difference in science or life extension there are better uses of resources than buying a few flops for folding@home.

You know, the only thing missing from your post is your pay pal address

I prefer western union to my office in Thailand or just give your bank account number to my rep in Nigeria.

Edited by elrond, 04 December 2008 - 10:13 PM.
need work on my blackberry typing skills

[poser]

I prefer western union to my office in Thailand or just give your bank account number to my rep in Nigeria.

You could have been a scientist...

[kismet]

so people actually buy dedicated computers specifically for folding@home.

that's ridiculous.

The whole point of these distrubuted computing projects is too make used of available unused computing capacity. Surely if one wants to make a difference in science or life extension there are better uses of resources than buying a few flops for folding@home.

There's always something more efficient to be done with the money. If all folders sold their computers and used the money to buy or rent Bluegene/P clusters (I don't know if other systems à la roadrunner can beat bluegene/P in efficiency), we'd make much more of an impact when it comes to folding power (or any other number crunching) and a much greener impact at that (should be 100-1000 times more power efficient than the mix of general purpose desktop computers that is currently running FAH).

poser, if you donate to the mprize you know exactly how your money will be spent.

Edited by kismet, 05 December 2008 - 01:08 AM.

[poser]

Mprize-Yeah I guess it's transparent. But I'd rather give the money for a prize that rewards advances in animal-free medical study methods. They need it more.

Bluegene-nice to know. It's incredibly efficient for CPU folding as they can squeeze 13 teraflops from 40 kilowatts. But I wonder if GPU's are actually more efficient than that. I recall that some GPU (probably GTX280) would have 0.5 teraflops. 28 of them in 7 computers... might require a 1 kW PSU so 7 kilowatts...

Btw, the above figures were for one rack... How much do these require maintenance... workforce is far from free... so that's a disadvantage. Obviously not in the case of renting. And when they get old, I don't know how easy it is to sell them. I'll ask about this.

[poser]

Hmm, no edit button for my message..?!

Mprize-Yeah I guess it's transparent. But I'd rather give the money for a prize that rewards advances in animal-free medical study methods. They need it more.

Bluegene-nice to know. Pande has said that you can't compare Bluegene flops to FAH flops, for the benefit of the latter. But Bluegene uses 5.5MW for 1.8 Pflops (calculated from 40kW/13 teraflops for one rack). Erm... I checked that Bluegene/P costs $1.3M per rack. How on earth is that good PPD/W/$???

I recall that some GPU (probably GTX280) would have 0.5 teraflops continously. In a machine that has 4 of them and has a 1000W PSU, power required would be 3.6MW for 1.8 petaflops.

Alancabler said (In June) that if 310,000 computers used on average 100 watts for just folding, FAH would consume 30MW for 1.8 petaflops.

Edited by poser, 05 December 2008 - 01:46 PM.