2x295 and 5902s?

[torin3]

I mentioned this in the 5902 thread, but I'm pointing this out in a different thread because nobody seemed to pick up on it.

I've got 2 GTX 295 cards in my main Vista 64 computer. So far, since I only have 1 monitor I can use at the moment, I can only run 3 GPU2 clients.

I turned on -advmethods in my clients and got 3 5902s in short order. However, 1 of the clients is producing almost twice as much PPD as the other 2. It completed the WU in 3hr 22min.

Project : 5902
Core    : Unknown
Frames  : 100
Credit  : 1680


-- GPU1 --

Min. Time / Frame : 1mn 58s  - 12301.02 ppd
Avg. Time / Frame : 1mn 59s  - 12197.65 ppd
Cur. Time / Frame : 1mn 59s  - 12197.65 ppd
R3F. Time / Frame : 1mn 59s  - 12197.65 ppd
Eff. Time / Frame : 2mn 02s  - 11897.70 ppd


-- GPU2 --

Min. Time / Frame : 3mn 22s  - 7185.74 ppd
Avg. Time / Frame : 3mn 25s  - 7080.59 ppd
Cur. Time / Frame : 3mn 31s  - 6879.24 ppd
R3F. Time / Frame : 3mn 31s  - 6879.24 ppd
Eff. Time / Frame : 3mn 27s  - 7012.17 ppd


-- GPU3 --

Min. Time / Frame : 3mn 28s  - 6978.46 ppd
Avg. Time / Frame : 3mn 33s  - 6814.65 ppd
Cur. Time / Frame : 3mn 34s  - 6782.80 ppd
R3F. Time / Frame : 3mn 34s  - 6782.80 ppd
Eff. Time / Frame : 3mn 32s  - 6846.79 ppd

It started with client GPU1, but after client GPU1 completed, the next WU (a 5902 as well) started going through at the normal high 6K PPD range.

However, client GPU3 then jumped up to almost 12K PPD. And then when client GPU3 finished its WU, client GPU2 went to nearly 12K PPD

It it somehow tapping the processing power of the idle core? And if so, how?

I'm pretty happy with the output, (Running about 25-26K PPD on all clients on this machine) and if I could just get 5902, I wouldn't need to try and figure out a way to get the 4th core running.

Anybody else experience the same?

 

[jonspd]

It it somehow tapping the processing power of the idle core? And if so, how?

that would be my only possible guess.


jonspd

 

[torin3]

that would be my only possible guess.


jonspd

I admit, it seems to be the case, but it is weird how it switches from core, to core, to core...

On the same card, I could see how it might be somehow not forced out of SLI mode, but switching between the cores suggests to me that it isn't the answer.

 

[ChasR]

Obviously folks on the FCF don't comprehend what you're talking about. I do, but was never able to get a 2nd GTX295 recognized on my GA-X48-DS4R even with 4 monitors (one HDMI on the 2nd card) connected.

 

[torin3]

Obviously folks on the FCF don't comprehend what you're talking about. I do, but was never able to get a 2nd GTX295 recognized on my GA-X48-DS4R even with 4 monitors (one HDMI on the 2nd card) connected.

I haven't checked there this morning yet, but from what you are saying, it looks like I'm going to have to add a long explanation.

Btw, I noticed I had a 5900 WU mixed in this morning, and one of the 5902s was still running at 12000 PPD.

I don't know if it has to ALL be core 14s for this to work, or if I only need 1 core 14 for it to work.

But yeah, I thought more people would be interested in this.

I'm tempted to turn off 1 of the clients and see if both of them go to 12K ppd.

 

[SniperXX]

I'm tempted to turn off 1 of the clients and see if both of them go to 12K ppd.

I'd be interested in the results even though I dont have a 295.

 

[torin3]

I'd be interested in the results even though I dont have a 295.

I'll have to wait until I get home tonight, but yeah, I'll do this for at least an hour (assuming I still have at least 2 5902 WUs running) and see what happens.

 

[SniperXX]

If the new core does indeed support dual gpu cards, that would be nice. One less instances of fah that need to be setup when going with a dual gpu card.

 

[ChasR]

The information published about these WUs and the core they run on said nothing about multi-gpu capability. There was mention of highly variable frame times.

 

[torin3]

Hmm..when I came home I had 2 5902s, and 1 384 point WU just starting. Shutting down the client that was working on the 384, and leaving the two 5902s running didn't get the second one up to the 12000 point range. However 1 of them was running at the 12K point range.

So, it doesn't scale, but it does seem to be doing SOMETHING.



It would be nice if somebody from Stanford would pick up on it though...

 

[torin3]

Apparently it isn't just the 5902s. 5900s also do it.