https://developer.nvidia.com/vulkan-driver
Watch this, if u dont believe they are good.
Watch that ->
http://setiathome.berkeley.edu/forum_thread.php?id=79105
My GTX 670 runs fine on it, with shorter running times.
Greetings
Copyright © 2024 Einstein@Home. All rights reserved.
Nvidia Maxwell and Kepler users , wanna test new "powerfull" off
)
Isn't Vulcan an API just like CUDA and OpenCL? If the application still use CUDA then Vulcan is not used?
Maybe Nvidia rewrote the CUDA part of the driver as well?
No, its like DirectX or
)
No, its like DirectX or OpenGL.
A programable interface to the graphics card.
Cant say if its using cuda50, an equivalent maybe, or the same, i cant figure it out.
The timings must be better set and cache enhancements, also it throws cpu work on the MCU of the graphics card. (so i understand that, i am NOT a programmer.)
At this point , NO workunit have been invalid crunshed.
I've got a GTX960 running
)
I've got a GTX960 running three BRP6-Beta-cuda55's at a time.
I swapped the driver last night and there's not many validated results yet, but this is what I see:
Driver 361.91:
Run time (sec) ... ~ 16100
CPU time (sec) ... ~ 1470 (by eye)
Driver 356.43:
Run time (sec) ... ~ 16100
CPU time (sec) ... ~ 1240 (1212-1269)
Total run time hasn't changed or difference is only minimal, but CPU time has systematically gone down over 200 secs. Not a single exception among the validated tasks so far.
I guess this driver might be able to free up some additional power for crunching CPU tasks, if a heavily pushed host is running both GPU and CPU tasks at the same time.
@Richie Low tech like mine
)
@Richie
Low tech like mine , profits more :D its a GTX 670 and must be not so optmized by nvidia.
Maximal: shortest
https://einsteinathome.org/task/547025709
Minimal: longest
https://einsteinathome.org/task/546724891
RE: Driver 361.91: Run
)
btw there must be something wrong with your host, can you set the url-link
to the tasks please.
With my GTX 670 i have only the quarter time to crunsh a tasks like u.
There must something very wrong on your host.
RE: RE: Driver
)
He is running three Parkes per 960. He will get faster times from changing P2 Memory speed to 3500mhz. My 960s run around the 14K second range after increasing the memory speed.
The two tasks you link have
)
The two tasks you link have minimal runtime difference, 1h4min vs 1h12min. I'd say that's within the variance in length for workunits, not a significant improvement in driver speed.
I could repeat a workunit, if
)
I could repeat a workunit, if u wish and see. ok, will crunsh some samples...
https://einstein.phys.uwm.edu
)
https://einsteinathome.org/task/548016769
I must admit , if the driver really work it must be a disease...
2nd:https://einstein.phys.uwm
)
2nd:
https://einsteinathome.org/task/548016842
3rd:
https://einsteinathome.org/task/548016859
Random is excluded. :D