Harry6363 / pyrit

Automatically exported from code.google.com/p/pyrit
0 stars 0 forks source link

Nvidia gtx 780ti benchmark less than 7970? #447

Open GoogleCodeExporter opened 8 years ago

GoogleCodeExporter commented 8 years ago
What steps will reproduce the problem?
1.pyrit benchmark
2.
3.

What is the expected output? What do you see instead?

Nvidia 780ti with 2880 cores benchmark of 60.000 PMKs only??  
if ati 7970 go more 100.000 PMKs and card is less than Nvidia 780ti!
what could be wrong?

What version of the product are you using? On what operating system?
kalinux

Please provide any additional information below.

Dual 7970's running on an i7 2600k with 16GB RAM. Beats the hell out of the 
75,000 PMKS.

Pyrit 0.4.1-dev (svn r308) (C) 2008-2011 Lukas Lueg http://pyrit.googlecode.com
This code is distributed under the GNU General Public License v3+

Running benchmark (222730.57 PMKs/s)... /

Computed 222730.57 PMKs/s total.
#1: 'CAL++ Device #1 'AMD GPU DEVICE'': 113695.7 PMKs/s (RTT 1.1)
#2: 'CAL++ Device #2 'AMD GPU DEVICE'': 114080.1 PMKs/s (RTT 1.1)

http://www.overclock.net/t/1256450/post-your-pyrit-benchmarks-here

Original issue reported on code.google.com by deltomaf...@gmail.com on 23 Mar 2014 at 12:37

GoogleCodeExporter commented 8 years ago
780ti should do about 100000 PMK/s
7970 should do about  115000 PMK/s

Original comment by parasid...@gmail.com on 1 May 2014 at 9:48

GoogleCodeExporter commented 8 years ago
what might be causing this loss?
will be a slot for PCIe 2.0x16? 

Original comment by deltomaf...@gmail.com on 24 May 2014 at 10:13

GoogleCodeExporter commented 8 years ago
which 'loss' are you referring to ?, I benched a 780 the other day @ about 
65,000 PMK's for its 2300 cuda cores, my 760 only pushes about 25,000 for 1100 
odd cores, do you mean why are nVidia cards far below ATi ? well its in the 
code .... and no one has updated it for years, the cuda code was 'not perfect' 
for the earlier generations of cards, and your basically just throwing more 
power at it ... why is ATi faster ? another user submitted a patch that made 
the ATi module more effecient, so it does more in less time, if the same were 
done for the cuda core, you would probably see 50% increase .... but dont hold 
your breath, no one is doing it, ive been trying, but I cant make it work ... 
and to learn, well, if you dont know C / python, its a pretty steep curve .... 
im trying, but I cant make it work, and I acknowledge i'm out of my depth .... 
nvidia's tools can update the cuda kernel code, and it goes from being 180k to 
3.4k, at 1/60th the size, it has to take less time executing, if it can ever be 
make to work ....

Original comment by shaneper...@gmail.com on 21 Aug 2014 at 8:00

GoogleCodeExporter commented 8 years ago
loss referring to is Nvidia 780ti with 2880 cores benchmark of 60.000 PMKs only.
by the amount cores should be greater.

Original comment by deltomaf...@gmail.com on 13 Dec 2014 at 11:19