EggXpert

Sidicas:

slo4sian:

Oh well now iegg pull me over, somebody help me to pay my light bill to run this project otherwise i will make very low score, XD

Where do you live? Some states, electricity is just really expensive.  AMD has some great 45 watt CPUs out that are really cheap these days. Take a look at the BE series.  Those chips make great Home Theater PCs and folding machines! Of course, those who want to maximize their points will probably hold out for the ATI 4870 card to come out and then they'll run simulations on that baby!

I'll probably be picking up some higher end gear in Fall sometime, right before it gets cold out.  

 

 

Im live in TX, my light bill rate at 14 cents/KW but im using X2 6400 now and everyone know its Electric hog, XD

Bah I cant believe iegg finished 4 WU and had 1244, while i got 7 but only 677 so far, XD

new beta version with simiulation btw, http://foldingforum.org/viewtopic.php?f=42&t=3188

slo4sian:

new beta version with simulation btw, http://foldingforum.org/viewtopic.php?f=42&t=3188

Installed this last night and it is working fantastic on my 8800gt. I'm running 2 x 5.04 clients along with this, cpu usage stays around 77-80% with an occasional spike up in the 90% range. I could probably run another 5.04 but I want to make sure my 8800gt is getting all of the proc power it needs.

Update: With the new client running on my 8800gt I am getting almost 500 points every 3 hours. I know I'm doing this to help an amazing cause but I almost feel like I'm cheating lol.

I wonder if the scoring for gpu work units will change soon, I can only imagine what a 4780 or a gtx280 will do

Ok folks I just dumped my F@H 5.04 4ea. and downloaded 1 of the new 6.12 beta's and man what a mind blower 10,000,000 steps and in the time it took me to order a new PSU my quad has done 1,904,503 steps. Now my question to you is how many of these can I run on a single quad computer and where do I go to adjust the core usage and memory usage?

Hey HH, question for you.  For your quad core, was it making 1.9 mil steps utilizing just 1 core with the GPU or was it all 4 cores?  If it was all 4 cores, do you think you would see lower numbers if you did it using each core separately?

No it was jumping around using all 4 cores but only at 25% and only 512 mbs of ram but I boosted it by going to normal usage instead of idle on priority settings. Thats why I posted the ? to find out if I could run 4 seperate installs 1 for each core.

HOGHAULER:

No it was jumping around using all 4 cores but only at 25% and only 512 mbs of ram but I boosted it by going to normal usage instead of idle on priority settings. Thats why I posted the ? to find out if I could run 4 seperate installs 1 for each core.

well the reason your performance jumped so much is because the 6.12 client is using your 2600XT to fold. so trying to run another instance is only going to help if you have another gpu. The cpu is not actually folding at all it is just feeding the 2600XT the data it needs.

What I do to max out my cpu usage is run the 6.12 client on my 8800gt and 2 instances of 5.04. I could probably run another instance of 5.04, 6.12 only uses about 25% cpu but every now and then jumps to 30%-40% and I want to make sure the gpu is being feed as much cpu power as it needs since it folds so much faster.

HOGHAULER:

No it was jumping around using all 4 cores but only at 25% and only 512 mbs of ram but I boosted it by going to normal usage instead of idle on priority settings. Thats why I posted the ? to find out if I could run 4 seperate installs 1 for each core.

Of course you can. The FAH Console client will use no more than a single core which means it will go between 0%-25% on a quad-core depending on how much that core is being used.  If you have a quad-core, with only 512 MB of RAM, you are better off running 4 copies of the console client.. Make sure you set your machine IDs to be unique and you install them in different directories!

Your best bet is to set it at idle and leave it at idle.. It will only use what is not being used on the core it is running on.

If you upgrade to 2 gigs of RAM, you could think about running the SMP client, which is the client specifically built for running all four quads at the same time on the same work unit.

 

Kardon:

Update: With the new client running on my 8800gt I am getting almost 500 points every 3 hours. I know I'm doing this to help an amazing cause but I almost feel like I'm cheating lol.

I wonder if the scoring for gpu work units will change soon, I can only imagine what a 4780 or a gtx280 will do

I think it will change. The reason nvidia cards are getting such higher scores than ATI is because nvidia is using CUDA. CUDA was built from the ground up for doing scientific calculations and getting every drop of performance out of the card.  Unfortunately, these CUDA drivers are often poorly suited to running the most recent graphics cards and the most recent games.  CUDA drivers are updated much less often than their regular gaming drivers.

AMD's counter to CUDA is CTM (Close-To-The-Metal) and a CTM supported F@H GPU client is in the works.

  Since all simulations are just a ridiculous amount of floating point math, and since graphics cards are so much better at it than CPUs, they will give a LOT of results in a short period of time.

AMD's 4850 does 1,000,000,000,000 (1 TFLOP) floating point operations per second and the 4870 does 1,200,000,000,000 (1.2 TFLOPs)..   Try doing that many additions/subtractions on a calculator by hand and see how long it takes...    Just to put the performance of your typical graphics card into perspective ..   It would probably take you a lifetime to render the first 30 frames of a Crysis game by hand with a pocket calculator.

The nvidia gpu client is currently running for me without installing any CUDA drivers. Maybe I should test out folding performance with CUDA drivers and see if I get a boost.
 

Listed in folding@home points per day.

PS3... PWNED????

A great quote for the F@H forums 

ToW:

Mike H said that they're currently optimizing the core to be able to use all the shaders ...

The WU we currently have are small (~550 atoms), and don't even utilize all the 320 shaders of [AMD HD]2900 or 38xx ...

We can't conclude on performance yet ... we don't know how NV boards will react to bigger WU (1000+ atoms), but we already know that such WU will make a better use of the huge amount of shaders on ATI boards.

Re: Sidicas's post about running multiple instances, here's a FAQ / instruction thread on ncix's forum.  It's great!

(edit - Mind your team numbers!)

can someone tell me whats going on here? I run the beta version that I posted up, me and GmS recieve large amount of WU, and sidicas seen to be in same situation so tell me how to fix this, please

slo4sian:

can someone tell me whats going on here? I run the beta version that I posted up, me and GmS recieve large amount of WU, and sidicas seen to be in same situation so tell me how to fix this, please

Depending on which projects you get, how much RAM you have, and other factors will determine the size of the workunit.. Larger workunits give more points and take much longer to do.

You can run the -configonly flag on your FAH client and enable "BigWU" which will allow you to pick up the bigger workunits if you have the RAM and bandwidth for it.. The uploads for these workunits are often upwards of about 25-35 megabytes.

Also, workunits for the SMP client can give about 2000-4000 points each, so it also depends in which clients you're running. The deadlines on the SMP client is very tight, usually about 2 days or so and it will normally require a quad-core system to complete the workunit on time and as such the SMP gives more points than the console client which has deadlines on each workunit about 2-3 months long.

 The fastest client, is the GPU client which runs on your graphics card.

Each client has their own set of workunits and projects that is most appropriate for the hardware (implicit linear vs. explicit parallel solving models depending on what the project is trying to accomplish).. Obviously, implicit linear runs great on a CPU while explicit parallel will run great on a GPU or Playstation Cell processor where it can run multiple parts of the workunit at the same time, but there's just much more work to do in an explicit project so it balances out overall.  Points per day is still the best indicator of how much science you're doing.   The more hardware you got, the more points you'll be able to earn a day, regardless of how many points your workunits give or how big your workunits are.

not to disagree with an esteemed uber but, i have been using th smp client or  about 6 months  on dual core machines and have had no trouble completing WU's on time and it list 4 days as the time limit. it takes my machine about 2 days to complete one, sometimes three if I have a lot of extra work. my score is not to bad considering the old computers I run on

Sidicas:

slo4sian:

can someone tell me whats going on here? I run the beta version that I posted up, me and GmS recieve large amount of WU, and sidicas seen to be in same situation so tell me how to fix this, please

Depending on which projects you get, how much RAM you have, and other factors will determine the size of the workunit.. Larger workunits give more points and take much longer to do.

You can run the -configonly flag on your FAH client and enable "BigWU" which will allow you to pick up the bigger workunits if you have the RAM and bandwidth for it.. The uploads for these workunits are often upwards of about 25-35 megabytes.

Also, workunits for the SMP client can give about 2000-4000 points each, so it also depends in which clients you're running. The deadlines on the SMP client is very tight, usually about 2 days or so and it will normally require a quad-core system to complete the workunit on time and as such the SMP gives more points than the console client which has deadlines on each workunit about 2-3 months long.

 The fastest client, is the GPU client which runs on your graphics card.

Each client has their own set of workunits and projects that is most appropriate for the hardware (implicit linear vs. explicit parallel solving models depending on what the project is trying to accomplish).. Obviously, implicit linear runs great on a CPU while explicit parallel will run great on a GPU or Playstation Cell processor where it can run multiple parts of the workunit at the same time, but there's just much more work to do in an explicit project so it balances out overall.  Points per day is still the best indicator of how much science you're doing.   The more hardware you got, the more points you'll be able to earn a day, regardless of how many points your workunits give or how big your workunits are.

 

 

Had u check ur score? sidicas, sound like u got same problem also, my computers are extremely fine

prophet42:

not to disagree with an esteemed uber but, i have been using th smp client or  about 6 months  on dual core machines and have had no trouble completing WU's on time and it list 4 days as the time limit. it takes my machine about 2 days to complete one, sometimes three if I have a lot of extra work. my score is not to bad considering the old computers I run on

Unfortunately, I don't think that will last. They are using a quad-core for the benchmark machine (used for determining deadlines) and as they optimize the work units for quad-cores, people with dual-cores have been falling off the edge of the map, unable to complete a workunit in time..   I'm surprised you hadn't heard about all those duallies complaining about not being able to complete some SMP projects anymore.. 

Originally, the SMP client workunits were

FahCore1: ~45%
FahCore2: ~20%
FahCore3: ~20%
FahCore4: ~15%

Which means that running on a single quad, some of the cores were idle a significant part of the time. For this reason, a dual-core would run the SMP workunit within the deadline that the benchmark machine set for a quad-core.

There is a new _a2 core out for the SMP client and it has the following distributions..
FahCore1: 798:18 / 2316:49 = 34.5% ~ 35%
FahCore2: 698:28 / 2316:49 = 30.1% ~ 30%
FahCore3: 496:06 / 2316:49 = 21.4% ~ 20%
FahCore4: 323:57 / 2316:49 = 14.0% ~ 15%

Which means less idle activity on each core. I do believe the new target deadline is 2-3 days (not sure on that, though, I might have it crossed with the next GPU client, but it was mentioned somewhere) .Some people with dual-cores were forced to give up the SMP client when these workunits were released as they could no longer complete the workunits within the deadline. In short, the quad-cores were returning workunits faster because they were using more of their cores, more often.

Which is closer to the ultimate goal of (25%, 25%, 25%, 25%) which would be firing on all 4 cores of a q6600 100% of the time and giving little legroom in the deadline, effectively cutting dual cores out of the SMP client altogether (it is designed for quads, but because it isn't optimized yet, duallies can complete workunits within the same time frame of a quad).

It's still a beta client..Not only are dual-core people falling off the map, but people with an overclocked quad-core who ran two SMP clients before (believe it or not, it met deadlines!) have also found themselves forced into 1 SMP client because of the efficiency boosts and resulting deadlines.

I could be wrong .. If one core of your duallie is faster than 2 cores of the stock quad combined, then you have no worries. Since they overclock so well, and the benchmark machine isn't, this is quite possible.

slo4sian:

Had u check ur score? sidicas, sound like u got same problem also, my computers are extremely fine

Yes, frequently   The low points are because they're Amber project workunits which take only a few hours to do each.. The other gromacs workunits that most console clients were doing before, took days to finish.  My overall PPD hasn't changed at all (still holding my position), I'm just completing much smaller workunits in a much smaller period of time..  It's what needs to be done so it's what the assignment server is giving out.  My guess is that they're leaving all the gromacs stuff to the high-performance clients like GPUs, SMP, and the PS3 which can get those units turned around very quickly (less than a week).   The Amber units are returned by regular CPUs very quickly (less than a day).. So it seems to me like they're just optimizing to get the results quicker. Msy be a temporary thing..