New GPU3 Client Available - 6.40r1 - Oct. 26th

T

Tobit

[H]ard|DCer of the Month - March 2010/May 2011
Joined
Dec 12, 2005
Messages
5,145
Sorry dudes, I am a couple days late in reporting this.

New GPU3 v6 clients (systray/console) are available at CONSOLE / SYSTRAY.

You will need to remove '_console' and '_systray' from the file names. You can then either update your shortcuts to use these new file names or simply rename the new executable to the existing name (having deleted, moved, or renamed the original first of course). Both methods work fine for me. I've been running this new client on my GPUs for almost two weeks now. Installing this mid-unit has also not caused any problems. Just stop the client first. :p

The major change to these clients is that the GPU species is now based on the compute capability of the board. In the earlier v6 client, the species was based on the installed version of the CUDA runtime libraries. The new GPU species values are

Compute capability -> GPU species

1.0 -> 10
1.1 -> 11
1.2 -> 12
1.3 -> 13
2.0 -> 20
2.1 -> 21
unknown -> 0

The motivation for these changes is that NVIDIA GPU3 cores (Core 15 at this time) cannot run on boards with compute capability 1.0. If a WU is run on these older boards, it will crash. The only cards with compute capability 1.0 are:

GeForce 8800 Ultra, 8800 GTX, GeForce 8800 GTS
Tesla S870, Tesla D870, Tesla C870
Quadro FX 4600, Quadro FX 5600
Quadro Plex 2100 Model S4, Quadro Plex 1000 Model IV

Source: http://en.wikipedia.org/wiki/CUDA
 
cool thanks, will check it out on the 1st of the month. shut off my gpu clients to save some power.
 
so, just to boil it down, the only purpose is to avoid distributing core_15 WUs to those old cards?
 
Tobit,

I am seeing an average of 2500 ppd on each of my GTX 460s on each of the prvious projects. Im guessing this core sees the other cuda cores avilable and is using them.

Has anyone else seen the same increases?
 
Only the client has changed, not the cores so I am not sure why you are seeing a PPD increase. The client itself is only an interface.

Edit: There are some fairly new WUs back on advmethods that run faster (10927-10978 and 11214-11265), maybe this is what you are seeing.
 
Last edited:
roguespearus said:
Tobit,

I am seeing an average of 2500 ppd on each of my GTX 460s on each of the prvious projects. Im guessing this core sees the other cuda cores avilable and is using them.

Has anyone else seen the same increases?
Click to expand...


you are seeing an increase of 2500 PPD or you were only getting an avg of 2500 PPD on each WU?
 
sirmonkey1985 said:
you are seeing an increase of 2500 PPD or you were only getting an avg of 2500 PPD on each WU?
Click to expand...
Yeah, I was kind of wondering the same myself.. he closed by asking if "anyone else was seeing the same increases" so I just assumed he is seeing a 2500 increase but I could be wrong as most assumptions usually are wrong. :confused: :rolleyes:
 
Tobit said:
Yeah, I was kind of wondering the same myself.. he closed by asking if "anyone else was seeing the same increases" so I just assumed he is seeing a 2500 increase but I could be wrong as most assumptions usually are wrong. :confused: :rolleyes:
Click to expand...


thinking the same but better off asking then not asking.
 
2500 ppd on each 460 gtx and the 1 GB version is running 1000 ppd faster than my 2 768 MB versions. Are some cuda cores disabled in the 768 MB 460?
 
roguespearus said:
2500 ppd on each 460 gtx and the 1 GB version is running 1000 ppd faster than my 2 768 MB versions. Are some cuda cores disabled in the 768 MB 460?
Click to expand...


ok what program are you using and how do you have it setup, because the gtx 460's stock should be doing 9k ppd and anywhere up to 12k PPD overclocked. so either your reading something incorrectly or you set something incorrectly in HFM if thats what you are using.

my guess is you set the Mhz settings in the client config on HFM so its showing PPD per mhz instead of the actual PPD. mhz should be set to 1.

and no the 768mb version doesnt have cuda cores disabled. just has less memory and memory bandwidth which doesnt matter to F@H.
 
You must log in or register to reply here.
Back
Top