Gpu2 Client Is Here!!!! This Is Not A Joke!!

[Marvin_The_Martian]

Marvin!!!! The thing that worries me is ... I then set the card at 900/900. Same time needed roughly... 91s?

Clock the cpu higher, this client needs it more then the last one even

It's also mentioned in the faq

45s * 100 = 4500

86400 / 4500 = 19.2 wus/day = (19.2*97) 1862PPD

Still not so good but allot better

So.. 2900xt's suck?? Cpu's 400mhz diffrence can't account for that. The SS from riptide is abit small can't read it clearly but dual core with no other cpu processes running I assume ( I do see vmware's I guess for dimes? )

Edit: Dragonorta nailed it before me

[Mechromancer]

Marvin!!!! The thing that worries me is ... I then set the card at 900/900. Same time needed roughly... 91s?

This client is HEAVILY CPU dependent. Overclock it as high as you can ! Also use FahMon to calculate accurate PPD. If I'm doing 1350PPD with an Opteron185+2900XT combo, I'm sure you Intel guys will do much more.

[Marvin_The_Martian]

Remember it just got released. Sometimes they put out really tiny units to begin with to be sure things are going smoothly before they put out the real deal.

All GROGPU2 projects listed are 97points, let's hope there are projects which just run faster.

[[XC] riptide]

Ah.... so... I'll postulate for now that this is nearly as much a new CPU client as it is a GFX client?? Cos the GFX is not under any pressure at all!

Also I should point out I'm on Cat 8.1

Win XP 32bit.

[[XC] riptide]

Looks like you need to stop :banana::banana::banana::banana::banana: footing around with that CPU overclock.

Ha! My SS is on standby for E8500. Then we'll see who's :banana::banana::banana::banana::banana:ing around. nOOb.

[[XC] riptide]

Also i noticed... that the client in taskmanager is NOT taking over 50% ie one core, when nothing else is grabbing it?? When I load up something else, the CPU is NOT 100% used. Something like 2-5% is not been used. This reminds me of the last client when combined with another project.

[MikeB12]

just adding my data to the thread.
2900pro 802/800 GPU client
Q6600 3420mhz (2 smp clients on affinity changer Vista 64)
good news and bad news...

good news is it only takes 4-5% cpu away from my dual smps.
good news is CC reports 50-51c full load gpu temp folding.

bad news is a measly 343ppd.
bad news is doesn't look like this thing is even pushing the 2900pro, temps of 50-51c sound like idle temps.

anyway, I'm gonna run it for a day and see what it does to the ppd on the X4 labeled pc in fahmon below ((the pc that has the 2900pro)... the smp ppd is hurtin right now because it's my main pc, and I've been using it this afternoon, which eats cpu away from smp. but for 343ppd, it doesn't seem worth it. my 1950gt in my old sempron pc was turning 550ppd on the old gpu client. it may be the whole affinity changer and smp combo, but the low gpu temps really make me think this thing is not using my gpu's potential. it's like it's putting along on battery power or something. plus if it's taking 300ppd or more away from my dual smps, then it's a no brainer. I won't be on it much longer.

[MikeB12]

oh, btw, for those looking for the fahlog.txt file to point fahmon to, it's in your profile/app data. that threw me for a minute, since all the other clients put it in the working app directory...

[wickedld9]

just adding my data to the thread.
2900pro 802/800 GPU client
Q6600 3420mhz (2 smp clients on affinity changer Vista 64)
good news and bad news...

good news is it only takes 4-5% cpu away from my dual smps.
good news is CC reports 50-51c full load gpu temp folding.

bad news is a measly 343ppd.
bad news is doesn't look like this thing is even pushing the 2900pro, temps of 50-51c sound like idle temps.

anyway, I'm gonna run it for a day and see what it does to the ppd on the X4 labeled pc in fahmon below ((the pc that has the 2900pro)... the smp ppd is hurtin right now because it's my main pc, and I've been using it this afternoon, which eats cpu away from smp. but for 343ppd, it doesn't seem worth it. my 1950gt in my old sempron pc was turning 550ppd on the old gpu client. it may be the whole affinity changer and smp combo, but the low gpu temps really make me think this thing is not using my gpu's potential. it's like it's putting along on battery power or something. plus if it's taking 300ppd or more away from my dual smps, then it's a no brainer. I won't be on it much longer.

Why don't you try dedicating a core to it to get accurate results?

[Mechromancer]

Why don't you try dedicating a core to it to get accurate results?

+1

Even one of your full cores@3.4Ghz may not be enough to keep your GPU fed full time. This client is a monster on the CPU and requires as much speed as you can throw at it.

[MikeB12]

I'm more of a dual smp fan. If I dedicate a core, that will severely handicap smp, and plus I'm not sure how affinity changer would deal with that.. plus fahmon is already showing about a smp 400ppd drop.. if I dedicate a core that will rise to about 1000ppd smp drop.. not worth it to me.

the same deal with my old 1950gt and the old gpu client, the ppd losses and gains cancel each other out.

I'll let someone with a slower dual core test it out as a dedicated gpu folder, my Q6600 is too efficient as a dual smp machine. my x2 4600 would be a better candidate, but it's not stealing my 2900pro from my main pc. it's on linux using onboard video as a dedicated smp and turns about 1100ppd at 2.6ghz. it's in my fahmon ss above.

but for a test and to get some accurate data, I'll stop my smp's and just let the gpu fold on the 2900pro at 827/874 and post some benches for thread data.

[MikeB12]

okey doke. killed the smp's and just ran the gpu for a 2900pro bench.

Q6600 3420mhz / 2900pro 827/874 catalyst 8.3
1250ppd and uses 100% one core (25% cpu on a quad). temps ares till low for the gpu, so I guess it really doesn't push it too much.

[beard]

They got this out incredibly fast. Next big thing is to utilize both gpus on 3870x2. So, it's like crossfire computing or making a tool that has control over the PLX chip to disable cf (to run 2 clients)?


...hmm, waiting for those "all in 1" times. Actually feels kind of bad because this is far from energy efficient way to make things better. I'm talking about crazy cpu usage for kind of nothing. Well... i guess this has been said before (haven't read these folding forums): a greater cause justifies my electricity bill ...though my wife may disagree...

[xPliziT]

I have a 2900XT with 1GB DDR4 clocked @ 800/1000
and a AMD 64 @ 3500MHz.

[23:26:10] Completed 60%
[23:26:59] Completed 61%
[23:27:48] Completed 62%
[23:28:37] Completed 63%
[23:29:27] Completed 64%


I get around 49-50 seconds per 1%

Dunno how good that is, just for information.

[[XC] riptide]

Gius^^^^^^^^^ What Cat version you all using???

[MikeB12]

fixed cat 8.3

[xPliziT]

Cat 8.3 is a must i heard. I am using it.

[[XC] riptide]

Yep. Thought that was the issue. I was on Cat 8.1 LOL

[MikeB12]

I just read this on fah forum... for those interested in running smp with gpu2..

Here's the best config I found for GPU2 + SMP on a quad core machine :

- I run one SMP client, and I answer "idle" when it asks me if I want idle or low priority (advanced options).
- I run one GPU client, and I select "Slightly higher" for Core priority parameter.

With this configuration, Fahcore_11 use one entire core, and the 4 other Fahcore_a1 use the 3 remainings cores.

[Stijndp]

I'd try 2x SMP in idle. 1x GPU2 on slightly higher. Or even 2x SMP plus 2x GPU2 if you really clocked it good or it's a 100% cruncher.

[MikeB12]

Q6600 3420mhz / 2900pro
TOTAL PPD COMPARISON for folding with just 2 smp's or 2 smps+gpu2.
notes*
-running 2smps with gpu2 on a higher priority core. (higher priority core selection only assigns 9-17% cpu, instead of a true dedicated core which is 25% on a qaud, so I see slightly less gpu ppd than when I run the gpu alone in above post for 1250ppd. the gpu is turning 1000-1200ppd on the slightly higher priority setting depending on where the cpu util fluctuates on the core assigned to gpu2-it seems to jump around between 9-17%) conclusion: the 10% loss in cpu core really only accounts for less than 100ppd loss on the gpu2 ppd. another note is my 2900pro oc's didn't make a difference in frame times, stock clocks produced the same frame timess as oc clocks on the gpu.

but for a quick summary of the below data:
2 smps with affinity changer by them selves = 4253ppd

2 smps (AC) and GPU2 set with higher priotity core = 4343ppd

in conclusion for me, it doesn't make any noticeable difference in maximizing ppd by running all 3 clients. they tend to cancel each other out concerning total machine ppd production. anyway, that's what I thought would happen, just needed to see it in action and make sure I wasn't dreaming...
*also note this is just a window of one test; I'm sure it goes up and down as wu's always do. in fact as I was typing this one of my smp clients took a big frametime hit and went down to 738ppd, probably something to do with AC; looks like one smp started hogging cpu time, and one was getting neglected.. it'll pick back up once AC catches on to it and moves priority to the other client.


Anyawy, here's the DATA I snapped from the test:


2 SMP's (AC) without GPU:

Project : 2653
Core : SMP Gromacs
Frames : 100
Credit : 1760

-- X4-A-Vista-Q6600-3.4 ghz --

Min. Time / Frame : 11mn 41s - 2169.24 ppd
Avg. Time / Frame : 11mn 50s - 2141.75 ppd
Cur. Time / Frame : 12mn 03s - 2103.24 ppd
R3F. Time / Frame : 11mn 56s - 2123.80 ppd
Eff. Time / Frame : 12mn 10s - 2083.07 ppd


-- X4-B-Vista-Q6600-3.4ghz --

Min. Time / Frame : 11mn 52s - 2135.73 ppd
Avg. Time / Frame : 12mn 00s - 2112.00 ppd
Cur. Time / Frame : 12mn 31s - 2024.82 ppd
R3F. Time / Frame : 12mn 23s - 2046.62 ppd
Eff. Time / Frame : 13mn 00s - 1949.54 ppd




2 smp's with GPU2 using higher priority on 1 core:

Project : 2799
Core : Unknown
Frames : 100
Credit : 97

-- X4-GPU 2900pro --

Min. Time / Frame : 1mn 06s - 1269.82 ppd
Avg. Time / Frame : 1mn 13s - 1148.05 ppd
Cur. Time / Frame : 1mn 19s - 1060.86 ppd
R3F. Time / Frame : 1mn 18s - 1074.46 ppd
Eff. Time / Frame : 17mn 23s - 80.35 ppd



Project : 2653
Core : SMP Gromacs
Frames : 100
Credit : 1760

-- X4-A-Vista-Q6600-3.4 ghz --

Min. Time / Frame : 20mn 10s - 1256.73 ppd
Avg. Time / Frame : 20mn 10s - 1256.73 ppd
Cur. Time / Frame : 20mn 10s - 1256.73 ppd
R3F. Time / Frame : 20mn 10s - 1256.73 ppd
Eff. Time / Frame : 12mn 27s - 2035.66 ppd


-- X4-B-Vista-Q6600-3.4ghz --

Min. Time / Frame : 13mn 01s - 1947.04 ppd
Avg. Time / Frame : 13mn 04s - 1939.59 ppd
Cur. Time / Frame : 13mn 07s - 1932.20 ppd
R3F. Time / Frame : 13mn 07s - 1932.20 ppd
Eff. Time / Frame : 12mn 54s - 1964.65 ppd

[7im]

Eh, don't get too hung up on the PPD from this very early look at the new GPU2 beta client. And as you've seen, these are very small test WUs. Larger WUs will perform better, IMO. Also watch for client tweaks in peformance as new beta revs are released. IMO, the PPD will go up.

[[XC] riptide]

Well as it stands this is virtually a new CPU client. I'm getting about 60sec dead per % at 3.6 E6850. Changing the Card clocks does nothing from 743-900

[MikeB12]

that makes sense, 7im... the 97 point unknown wu's definitely act different than the 330 point wu's that the 1950's fold...

the 97 point beta's surprised me when load temps didn't look like the card was loading up, and oc clocks didn't affect frame times from stock clocks.

I'm sure the real production wu's will be different.. or at least hope they will once this client gets transferred to the main client dl page..

[[XC] riptide]

I agree Mike. I reckon the best is yet to come. I hope to see my card warp with pressure.