Changing GTX 760 to GTX 980

[MatWur-S530113]

Hello,

I will get a new GTX 980 card in a few days which will reset my older GTX 760.
For the 760 I'm using a compiled version of gpu-ecm by mklasson (many thanks for this btw ;) )
Do I have to change the gpu-ecm version for the new gc?
If I remember correct there were some issues about the cuda-version with the GTX 980.
And (if a changing of gpu-ecm is needed) which version of gpu-ecm I need for the new card?

tia

Matthias

[MatWur-S530113]

In the meantime my new gc arrived.
Isn't there a compiled version of gpu-ecm available compiled for
CC 5.2 and CudaRT 6.5?
Or simply a working version for a GTX 970, GTX 980 or GTX 980ti?

tia

Matthias

[wombatman]

Look here: http://www.mersenneforum.org/showthr...=16480&page=37

I posted both the exe and needed dlls there. I have a GTX 980 and Windows 7/10.

[MatWur-S530113]

oh, very nice, thank you very much ;)
I downloaded your last 2 posted zip's there and try them as soon as possible.
I will post some first results tomorrow or so, test mode... :D

Matthias

[MatWur-S530113]

first of all: wombatman's compiled ecm_gpu.exe runs correct and very good on my system, very appreciatet, thank you again ;)

for some speed tests I run curves on 2^1253-1, this number has a 268 digit composite cofactor.

old GTX 760:

B1=1e6 => 682 ms / curve througput
B1=260e6 => 185 s / curve througput

new GTX 980 with old prg. compiled for CC 3.0 (and CudaRT 5.0):

B1=1e6 => 285 ms / curve througput
B1=260e6 => 69.0 s / curve throuput

new GTX 980 with wombatman's prg compiled for CC 5.2 (and CudaRT 7.0):

B1=1e6 => 260 ms / curve throuput
B1=260e6 => 68.4 s / curve througput

thus the throuput is ~2.7 times of the old GTX 760
(the old card was given with 2300 GFlops (SP), the new card with 5000 GFlops (SP),
that's only a factor of ~2.2 ...)
the old card only used ~63% of the card's TDP (>170W) and became ~ 60°C
the new card uses ~94% of the card's TDP (>165W) and becomes ~ 66°C

both cards are from Inno with the triple-slot iChill x3 cooler, I like it :D (not the x4 version ;) )

and only for the record: now I not only have 2 times the stage1 capacity (compared to stage2 capacity),
no, now I have 5 times stage1/stage2 capacity... I need a xeon with 12 threads and 256 GB RAM.... next project ;)

greetings

Matthias

[wombatman]

Glad to hear it worked for you