UnitedGerbilNation's drive to the Top

[Gerbil Jedidiah]

It's Ubuntu actually. Both of my folding rigs are everyday PCs. I don't have any folding nodes per say. Currently getting 21k ppd from them and that will go up once i install the next two cards. I just hope heat doesn't become unbearable.

The 2 cores with the SMP client were at almost 100%. The 2 cores with gpu client were in the 20s or so. My experience is that the 240 shader count cards need more CPU time than the GTS 250 cards( or those with less than 128 shaders).

[Ragnar Dan]

Hm. Didn't you have a thread somewhere about your troubles with running Unix or some variant thereof? I could swear that was you.

As for the CPU usage of the GPU client, I suppose if it finishes more quickly then the client has to write the old data out to the temporary files and send the next bunch of data to be calculated with more offen, which may mean more CPU time. Still, it seems odd that there's that much difference.

[Flying Fox]

Hm. Didn't you have a thread somewhere about your troubles with running Unix or some variant thereof? I could swear that was you.

Jed just seems to equate Linux to Unix. I just didn't bother to correct him.

[Gerbil Jedidiah]

Hm. Didn't you have a thread somewhere about your troubles with running Unix or some variant thereof? I could swear that was you.

I had one a over a year ago where I got the virtual environments set up. Then later had one where I networked the environments to Vista64 so I could run FAHMon. I'm not sure I recall any recent UNIX escapades.

As for the CPU usage of the GPU client, I suppose if it finishes more quickly then the client has to write the old data out to the temporary files and send the next bunch of data to be calculated with more offen, which may mean more CPU time. Still, it seems odd that there's that much difference.

Yeah I agree. The first time I noticed it was when I upgraded my 8800GTSG92 to the 9800GX2. If I didn't run the SMP clients the GX2 would do much better. Then I went back to the 8800GTS for a bit and it ran fine with the SMP clients in virtual. Then I got the GTX280 and again PPD would go down when the SMP Clients were hitting all 4 cores.

I agree with FF that if I were to tweak it a bit more I might be able to get the overall PPD up with the SMP client hitting two cores. But I plan on adding another 240 shader GPU in a week or so, so it doesn't seem worth it to play with the SMP client when more than likely that 240 shader card is going to want it's own core to get max PPD. Still, the one machine will be getting somewhere in the 22k to 23k ppd range, so I'm OK with that. I may even try to get those 185 beta drivers from Guru3D working to see what they can do with all the GPU goodness. (Probably make my computer explode or melt I bet )

[Gerbil Jedidiah]

Hm. Didn't you have a thread somewhere about your troubles with running Unix or some variant thereof? I could swear that was you.

Jed just seems to equate Linux to Unix. I just didn't bother to correct him.

Yes I realized when he asked me what linux I was using that I had goofed

When I audit them all I just make sure

1) ROOT is locked and you have to SU to use it
2) Someone reviews the SU log
3) Take a look to make sure no one else actually has a ROOT level ID (User authority of 0)
4) I take a look at etcPassword to see who is on the box and what authorizations they have

Hmm that's about it... Depending on the flavor of UNIX there are different "etc" type files to look at, but with the reduction in testing that we have to perform the above is about all that is necessary now.

So, if you guys are system admins for companies that have their financial apps audited, now you know what we look for

I guess that was a long winded way of me saying that I don't get to see too many differences in the flavors of UNIX based on the limited amount of work I do with it. Therefore I goofed

[jeffry55]

I, for one, am anxious to see your numbers once you get the two other GPUs folding.

[Gerbil Jedidiah]

Me too Jeffry, hoping to add about 12k to 14k to the mix, which puts me in th 35k PPD range you'd predicted.

We may hit 100k today

Unrelated story, I have the stomach flu this week. Today I 've eaten half bagel, 10 saltine crackers, and 1.5 servings of chicken soup... Fever hitting 102 and my neck is so sore it's not funny. The FLU SUCKS!!!

/rant

All five of these videocards are in our home office where we work, so I may have to do something to quieten them down. Can you undervolt a card, lower FRQ, and raise the shaders to make the card.. Perhaps I will break down and add a folding node to get the heat out of the office.

Also a little concerned about the life expectancy of the GTX280. I've read other people's experiences where they get multilpe RMAs. I'm hoping they send me a GTX285 that might run cooler and last longer.

[Gerbil Jedidiah]

I'm just hanging out for a few more minutes to see if we hit 100k today. I'm a little loopy from all the flu meds I'm taking, just fyi

[jeffry55]

Me too Jeffry, hoping to add about 12k to 14k to the mix, which puts me in th 35k PPD range you'd predicted.

We may hit 100k today

Unrelated story, I have the stomach flu this week. Today I 've eaten half bagel, 10 saltine crackers, and 1.5 servings of chicken soup... Fever hitting 102 and my neck is so sore it's not funny. The FLU SUCKS!!!

/rant

All five of these videocards are in our home office where we work, so I may have to do something to quieten them down. Can you undervolt a card, lower FRQ, and raise the shaders to make the card.. Perhaps I will break down and add a folding node to get the heat out of the office.

Also a little concerned about the life expectancy of the GTX280. I've read other people's experiences where they get multilpe RMAs. I'm hoping they send me a GTX285 that might run cooler and last longer.

I concur with you Jed. The flu SUCKS the big root!! I hope you are feeling better soon.

[Gerbil Jedidiah]

And the wait was worth it! Perhaps this year we should stay away from TFrankenbot and let the other folders make a go for us as TRFrankenbot.

[jeffry55]

That's a nice spike! If we can maintain this level as our minimum performance level, we should be on top of the TR Folding heap by Halloween ........................................................ if Seggy-poo does not break out of his stupor.

Fold my Gerbil brothers!

[Gerbil Jedidiah]

That's a nice spike! If we can maintain this level as our minimum performance level, we should be on top of the TR Folding heap by Halloween ........................................................ if Seggy-poo does not break out of his stupor.

Fold my Gerbil brothers!

Don't get your hopes up for today. I'm crunching on some real dogs at the moment. PPD is down 3k on main rig. Haven't checked wifey's.

But, on the bright side, I still have 2 GPUs to add to the fold

[jeffry55]

That's a nice spike! If we can maintain this level as our minimum performance level, we should be on top of the TR Folding heap by Halloween ........................................................ if Seggy-poo does not break out of his stupor.

Fold my Gerbil brothers!

Don't get your hopes up for today. I'm crunching on some real dogs at the moment. PPD is down 3k on main rig. Haven't checked wifey's.

But, on the bright side, I still have 2 GPUs to add to the fold

Excellent! *taps ends of fingers together*

[jeffry55]

I'm lovin' this new mix of WU's! With the 5903, 4777 & 4776 WU's, my farm PPD is up 5k!!

[jeffry55]

Aaaaaand my GTX 260 (a wimpy 192 shader version ) is rockin' a 5903 WU for 6800 PPD!!!

[Gerbil Jedidiah]

Aaaaaand my GTX 260 (a wimpy 192 shader version ) is rockin' a 5903 WU for 6800 PPD!!!

One of my GTX295 GPUs does about that much. I guess your higher clock speeds makes up for the lower shader count.

[Flying Fox]

And I thought my 1400+1000ppd from my 2 SMP clients are good already. Come on Stanford, give me some A2's on my E2160 box and get it up from 1400 to somewhere higher!

[ReAp3r-G]

My PPD is gonna suffer a bit, both personal and the UGN contribution...at least for awhile

Does the SMP always have like 500k steps? If it does, it's conflicting way too much with the GPU client, it should be slightly faster than this...I remember running the 6.22 with the GPU client and it was giving me 4000PPD on alternate days...

Any non-destructive suggestions? Right now sytem stability is A-OK...

[Flying Fox]

Does the SMP always have like 500k steps? If it does, it's conflicting way too much with the GPU client, it should be slightly faster than this...I remember running the 6.22 with the GPU client and it was giving me 4000PPD on alternate days...

I really don't care about how many steps. Sounds like you don't have a monitoring tool like FahMon. Install that and check what WU the SMP client is working on? Also you can monitor all your clients at once so you can tell if the GPU client production is dropping off too much.

[ReAp3r-G]

I really don't care about how many steps. Sounds like you don't have a monitoring tool like FahMon. Install that and check what WU the SMP client is working on? Also you can monitor all your clients at once so you can tell if the GPU client production is dropping off too much.

I do have it installed

It's currently working on GROMACS Core 2653...I guess from the total it is dropping somewhat, which is sad...time for an upgrade maybe?

[ReAp3r-G]

I guess referring to your previous comment, I shall just run the SMP client as a non-SMP contributor it's faster that way and has minimal impact on the GPU client...will have to wait once this thing is done which is gonna take forever

[Flying Fox]

Project 2653 takes about 20minutes per step on my E2160@3GHz, for ~1400ppd. What is your current rate of production according to FahMon?

Say if your GPU client does 4000ppd for a given WU type, what is it doing now according to FahMon? Those 2 numbers will determine if the "drop offs" are worth it.

Example:
GPU client only = 4000ppd
If GPU and SMP clients = 3700ppd+1200ppd = 4900ppd => better
If GPU and SMP clients = 3000+1000ppd = 4000ppd => break even
If GPU and SMP clients = 2500+900ppd = 3300ppd => worse, might as well stay with GPU client only, or GPU+single core client.

That is why you need to know your numbers to make a correct decision.

[ReAp3r-G]

Somehow both my clients are only giving 1200+ PPD...

I should probably say that the GPU client usually uses half my CPU cycles...kinda odd really, not sure what the usage for the NVIDIA GPU client is

4000PPD with the GPU client alone isn't at all close to my actual production , it's only an HD4670 after all...I was expecting somewhere around 2000 with both my clients up and running but that isn't the case with this new SMP and I've clocked my CPU to 3.0GHz too

Well I've calculated at least 50 minutes average if I left the PC alone to do its thing for one step in the SMP client to finish...

[jeffry55]

Hey Jed! How's that Striker II Motherboard working for you? Is it a fast mamajama (all short 'a' sounds)? We have inquiring minds and we want to know?

Are you fully recovered from the flu? Are you trim and lean like Lance Armstrong? I mean, without the broken collarbone.

[Flying Fox]

Somehow both my clients are only giving 1200+ PPD...

I should probably say that the GPU client usually uses half my CPU cycles...kinda odd really, not sure what the usage for the NVIDIA GPU client is

4000PPD with the GPU client alone isn't at all close to my actual production , it's only an HD4670 after all...I was expecting somewhere around 2000 with both my clients up and running but that isn't the case with this new SMP and I've clocked my CPU to 3.0GHz too

Well I've calculated at least 50 minutes average if I left the PC alone to do its thing for one step in the SMP client to finish...

50 minutes for a 2653 WU (1760 points) work out to be 506ppd, yikes. If the GPU client itself is now doing 1200ppd instead of 4000ppd, then it is definitely not worth it to run both the GPU and SMP clients.

[ReAp3r-G]

Thing is Fox, my GPU client could NEVER do 4000PPD...EVER! It's just a puny HD4670 and I think my PPD production was less hit when the SMP client was ran without the -smp switch, altho the points gained by that is absolutely minimal in comparison...

The client running at 50% cycle usage (GPU) is sorta odd don't ya think? Shouldn't it run lower? I've got no choice but to sit this out for now...gotta wait for the SMP client to finish folding before I take the switch off...

The previous version was actually better than the current one, or it could just be because of the specific WU...

[jeffry55]

If GPU and SMP clients = 2500+900ppd = 3300ppd

-1 for math error

[jeffry55]

Thing is Fox, my GPU client could NEVER do 4000PPD...EVER! It's just a puny HD4670 and I think my PPD production was less hit when the SMP client was ran without the -smp switch, altho the points gained by that is absolutely minimal in comparison...

The client running at 50% cycle usage (GPU) is sorta odd don't ya think? Shouldn't it run lower? I've got no choice but to sit this out for now...gotta wait for the SMP client to finish folding before I take the switch off...

The previous version was actually better than the current one, or it could just be because of the specific WU...

I thought the most recent ATI GPU drivers cut down CPU usage to the 5-15% range? Which GPU driver are you using?

[Flying Fox]

Thing is Fox, my GPU client could NEVER do 4000PPD...EVER! It's just a puny HD4670 and I think my PPD production was less hit when the SMP client was ran without the -smp switch, altho the points gained by that is absolutely minimal in comparison...

The client running at 50% cycle usage (GPU) is sorta odd don't ya think? Shouldn't it run lower? I've got no choice but to sit this out for now...gotta wait for the SMP client to finish folding before I take the switch off...

The previous version was actually better than the current one, or it could just be because of the specific WU...

Then why are you throwing out the 4000ppd number? Keep FahMon up, have it watch your 2 clients, and tell me what they are doing now? Do you even know what that lone GPU client does in ppd when you run it by itself?

The GPU client takes up 1 full core so you see it as 50% overall (Task Manager oddness). I remember this topic being discussed suggesting that something on the CPU side needs to keep polling the data from the GPU or something. That's why it takes up a full core.

[ReAp3r-G]

I've actually mentioned it many times before Fox, that 4000PPD isn't even close to being the PPD that my system is capable of...and I thought you were using that as just an example

So anyway, my driver is the very recent one...9.3 to be precise, and no it never decreased CPU usage at all since the last few CCC and ATI driver updates for some odd reason...

Just to clarify and hopefully no more confusion after this, my GPU client COMBINED with the PS3 is ~2000PPD right now even with 3 clients running total projected PPD is still ~2000PPD...