Our GPU2 Results! / HOW TO INSTALL

[jcool]

Ya I tried clicking that expand to desktop thingy for every of the 8 screens showed, for some it worked for some it threw errors.

[Naja002]

Guys,

sorry, not much time atm. Thing is: I got a machine here that is running dual HD3870X2 in CF, and I need to really test it for stability for like a week or so. Now what would be better than the GPU client running on all 4 GPUs
Can anyone give me a quick rundown on how to set up the client? Running on Vista (not sure whether it's 32 or 64, gotta check later).
Thanks

Which OS, jcool? Before it was Vista, now its XP. It makes a difference....

[jcool]

Like I said in like the last 3 posts, I'm on XP. You gotta read my posts if you wanna help me, you know

[Marvin_The_Martian]

You got to read my posts to

Where are the log files??

[p2501]

Ugh, would anyone mind if we sort of revive this thread with the original purpose in mind, i.e. reporting PPD with the cards we're folding on. I'm asking this because with those new bigger GPUv2 WU I'm guessing that we have to rewrite the whole PPD/watt order of all cards.

Should we do this in a new thread or here?

[Naja002]

Sorry guys, that's it then, I don't have hours and days to try this, either it works or it doesn't. Clearly, it doesn't. Seriously they should make like a working installer for this stuff, as I did nothing wrong - double checked all the steps, it's a nogo

Personally, I don't know about Your question, p2501. I'll leave that up to ya'll. I've been feeling like jcool's situation needs its own thread for a number of reasons. And right now, I'm not quite sure where jcool's at with this:

Sorry guys, that's it then, I don't have hours and days to try this, either it works or it doesn't. Clearly, it doesn't. Seriously they should make like a working installer for this stuff, as I did nothing wrong - double checked all the steps, it's a nogo

So, where Ya at jcool? You wanna pursue this? Call it quits? Whatever ya wanna do...I've been busy with other stuff and weekends are my busy time, but we can work on getting things sorted out if You want......

[Marvin_The_Martian]

mhm ok no log filles still...

This forum requires that you wait 70 seconds between posts. Please try again in 22 seconds.

This limit is driving me nuts!!

[RaZz!]

why are ati cards so damn slow compared to nvidia?

my 4850 is only getting ~2100 ppd. even a gf8600 gets such ppd easily.

btw: i'm running the gpu and cpu client at the same time. when i open the f@h viewer it's always displaying the gpu-client?

[Marvin_The_Martian]

Because Ati's owner has less money spend on their R&D is my own take on it. They got the hw, gaming shows their competitive, and looking at raw numbers they should excell in massively parallel tasks like crunching. But like I said, their not backed up with the needed R&D.

But then again, think of it this way, that ppd you will not get running a cpu! Well maybe the one in your sig highly clocked with two linux clients on it, but normally no. So like you already said, it's relative performance about which you talk.

[RaZz!]

ah, nvm, after updating the stats in fahmon a few times the ppd of the 4850 are at 3177 now seems more reasonable. cpu is getting 466 ppd.

ppd total: 3644,42

seems like every project gets different ppd. ah, well i'm just messing around with this stuff. never used it before

[Marvin_The_Martian]

that quad core should be doing pretty close to 2K actually.. what os are you using.. darnit 2 at night, going for a smoke might respond when I get back might dive into bed

[RaZz!]

winxp pro sp3 (32bit)

dunno about the ppd. both clients are only using only 25% (the cpu client as well as the gpu client = 50% cpu usage overall), so it's not running on all cores i'd say. already changed the settings to 100% cpu usage. but it seems like i'm too tired now to really get how this whole stuff works
tomorrow, i'll give it another try

[Marvin_The_Martian]

RaZz I strongly encourage to read the stickies thread and the guides, and maybe even more about the types of clients lol

You're not optimally configured to be blunt! We can fix that though, but I love it even more when people try on their own to walk on the already paved roads layed out by the stickies, and just trying to follow them from a distance with a first aid kit to help get them going again when they hit a hole in the road. Which with f@h, is a certenty, not a question of if just when

You should see more cpu usage from the cpu client, it needs to load your entire cpu only leaving some for the gpu client to run with a slightly higher priority so it's not sufficated but the SMP client. What might be the case, you're running not smp but uniprocessor client, or you forgot to use the -smp flag.

[RaZz!]

lmao, you're right, i've downloaded the uniprocessor client - by mistake of course. lol, i told you, it was late already, and the beers i had didn't help either

thanks for your help though

[Marvin_The_Martian]

I'm not familiar with the ati client first hand and the newer intels either, curieus how many points you are able to squeeze out! Especially on the cpu, and maybe if it's even easier for you to run a single smp in linux vm taking two cores leaving you with a dual core which has to feed the gpu only. Maybe it will be more stable for the smp as well as I still think three cores are not that good to balance four threads but as said, no real first hand information.

[Paladin]

-Q6600 @ 3312 (8x414, 1.28v) Only 1 core for the GPU2 client
-8800GT 512MB (fan@100%, 65C temp ave)
-XP64 SP2
-177.41
-700core/950mem/1764shad
-5180ppd According to fahmon

[onex]

just went in..
E2160 1981Mhz
8600GT 1GB
XP Home Edition
186.18 driver (new drivers)+CUDA driver 2.2
540/924/333 - b. OC, up to 15+m 80% core usage,
585/1199/394 - a. OC 12-m @80% core usage.
going up to 10mpf at 100%.

quite dissapointing
or else, something has changed through the last 6-7 months in the frames/WU sizes..

note,
CPU is working 100% on WCG projects.

[OldChap]

Yep lots of different types of WU now...some take me 8 hours ish...some are deliberately switching off for a second or two as they run to reduce heat? etc.

[onex]

some are deliberately switching off for a second or two as they run to reduce heat?

it shouldn't be heat, it probably doesn't have any probe control, OS is using ACPI or other thermal monitoring tech to shutdown the CPU when overheated,
about GPU's..
stress tests programs doesn't work for such long periods normaly and still maintain standart temperatures when the processors are stable.
maybe it's related to the fact it's still a Beta or it doeses it for updating the server,
it should be probably more specificaly noted through the log..
or doesn't it?!

p.s - it's getting some 1100~PPD here while they got a 3600~ PPD mark on a stock 8800GT 3 months ago at the F@H forum:
http://foldingforum.org/viewtopic.ph...art=870#p88135
the differences between projects according to:http://foldingforum.org/viewtopic.php?f=52&t=3193,
can range up to 600PPD's,
though it doesn't say anything regarding differences in OC's etc,
maybe it has something to do with the CPU being overloaded..,
i'll try turning the WCG for a while OFF.

[OldChap]

Perhaps my explanation was poor, yes it is part of the beta trials and one of these has a range of wu's which "pulse" ....if you watch on a Kill-a-Watt the figures don't stable but are constantly variable. There have been a number of folders posting at fah forums complaining that folding is too stressful on their gpu's pushing the temps too close to the design limit...I'm thinking this is an experiment on pande's part to resolve this (there is another core out that allows the user to select the percentage of non processing to this end too)

[onex]

it is missunderstood what u say, as the 6.23 client does allow configuring the % of usage,
it ain't that accurate though but still, it gives the user the opportunity to be a head of the usage&heat..,
people complaining seems to be quite silly, aren't they?
if the core or shaders are being used, obviously the temperature will amplify and there is probably nothing to be done about it.. or is it being missunderstood too?!?
electricity flows, heat is generated and thats it ain't it?!?..
a GPU can operate at up to 125C threshold depending on the model, and it's actually a necessary part of the deal..?!?,
there shouldn't be any difference between using it 100% for gaming or for a folding thread..
corrently here, it's working OC'd at ~79-80 degrees and it doesn't seems to more or less bother it..,


results:

Pentium Dual Core E2160 & Geforce 8600GT G84 OC'd 580/1188/393 
Graphic card 186.18 drivers & CUDA's 2.2
F@H v6.23 Gromacs GPU client 
XP 32 Home

GPU@80% @~75 deg. & CPU@idle
12m15s~ Avg. per frame
873.09 - while in constant use

943.89 - ~idle usage
942.55

GPU@80% + WCG CPU@100%
 
897.91 - idle usage
899.12

GPU@100% @~78 deg. & WCG CPU@100%
9m52s~ per frame

1120.86 - idle
1117.09

GPU@100% &CPU@idle

1120.86
1124.66 
well, controversary results, differences between WCG on/off (80% ->100%)might be due to testing under usage (borwser etc.) & without noticing,
last 1120,1117,1120,1124 results were taken at +- idle.
seems there isn't any difference afterall..

all results are based upon
FahMon 2.3.99.1 "last frame only" PPD monitoring 

[OldChap]

I have a couple of gx2's that fold o'clocked to 1836 in the mid nineties if ambient is 25 and around here 25 is quite high so I've put a big cfm fan on the exhaust of these which helps alot.

I guess that folders who live in hotter areas could run into trouble with temps so they are looking at options for these people

My single cards (9800gtx, gtx260's) don't get anywhere near these temps (maybe low 70's on a 25deg day) I'm therefor sure that there is a moral to that story...If you live where you get temps consistantly over 30deg buy single cards!!

As for the running wcg and gpu folding together ...I just do it and am not really so anal about the points....if I were I'd probably move away from xp

[onex]

how can anyone use his/hers graphic card without getting it overheated? do you mean underclock it, or designing a different core that will use different aspects of the card?
leveling the usage % is still always possible at the GUI so there seems to be no sense for complains.. .
apart from that,
yeah, it's not realy about these 3-4 points,
if you have some multiple card system or dual processor MB's then these figures ain't exactly create any difference to you though as much as u get it's better (as much as u fold is better),
i just though about it yesterday,
if only a million people will join and fold for even just a day,
they can get up to few billion points!
few percents of the whole folding project through all of it's years!

it is also about the benchmarking abilities of the cards,
F@H is a great system testing & benchmarking tool , (and with a meaning, for a change)
note, that it may not be that important for you, but 5800~ with these 3 great cards seems just a bit low, just as a remark
hi di ho .

[BulldogPO]

Hoho, did find ole EVGA 8800 GTX ACS3 card that has extended warranty for years, not it happily folding and fahmon says that PPD is 5175.18
Running f@h only on GPU as CPU is crunching WCG.

System:
- i7 975XE@4.2GHz (21x200MHz 1.38V air)
- Gigabyte X58A-UD7
- 3x2GB
- EVGA 8800 GTX ACS3 (forcweare 197.25)
- Win 7 X64

[jcool]

how can anyone use his/hers graphic card without getting it overheated?

This is XS - we watercool
My 5870 stays at around 20C core during FAH, running at 1Ghz.
I am not a fan of folding on stock cooling - for high end cards at least. It's noisy, the card runs hot, and you run a risk of the VRM burning up & breaking down, especially with Volterra-based cards like the GTX 280 (not 285), 65nm GTX 260, and almost anything ATI.

About your low PPD issue: You have to set the FAH client's CPU usage to "slightly higher" if you run WCG on the CPU. That way, your PPD will be normal (for my HD5870 with a few tweaks, it takes 3-4% CPU off WCG - with NV cards it's even less, and no environmental table tuning needed)