Nvidia Kepler

[bcp19]

:( :( :(

[firejuggler]

have you tried to update your cuda toolkit?

[Batalov]

CUDA driver version 4.20
name GeForce GTX 680
compute capability 3.0

recompile with sm30 !

[Dubslow]

How does 3.0 stack up in the whole 2.0/2.1/mfaktc schtick?

[Dubslow]

Wait a minute, nVidia says it's 2.1, not 3.0.

http://developer.nvidia.com/cuda-gpus

Edit: Maybe he just needs a driver update, to 295? What drivers do you have?

[flashjh]

Originally Posted by lalera

hi,
mfaktc v0.18 does not run with gtx680

mfaktc v0.18 (64bit built)

number of mutliprocessors 8 (unknown number of shader cores)

ERROR: cudaGetLastError() returned 8: invalid device function

This is likely due to the "eight next-generation Streaming Multiprocessors (SMX)":

 
The SM is the heart of NVIDIA’s unified GPU architecture. Most of the key hardware units for graphics processing reside in the SM. The SM’s CUDA cores perform pixel/vertex/geometry shading and physics/compute calculations. Texture units perform texture filtering and load/store units fetch and save data to memory. Special Function Units (SFUs) handle transcendental and graphics interpolation instructions. Finally, the PolyMorph Engine handles vertex fetch, tessellation, viewport transform, attribute setup, and stream output.
 
One of the keys to GeForce GTX 680’s extraordinary performance is the next generation SM design, called SMX. SMX contains several important architectural changes that combine to deliver unprecedented performance and power efficiency.

This *may* require some reprogramming in mfaktc and CUDALucas? Anyone else tried it yet?

Source: GeForce-GTX-680-Whitepaper-FINAL.pdf

[Karl M Johnson]

Ok, that's damn it number one.
I wonder why NV decided to call new-gen Kepler a sm_21 gpu.

[bcp19]

Originally Posted by Dubslow

Wait a minute, nVidia says it's 2.1, not 3.0.

http://developer.nvidia.com/cuda-gpus

Edit: Maybe he just needs a driver update, to 295? What drivers do you have?

Um, where do you see it as a 2.1? The 680 is not even on the list in that link.

[Dubslow]

Originally Posted by bcp19

Um, where do you see it as a 2.1? The 680 is not even on the list in that link.

CUDA-Enabled GeForce Products

GeForce 8, 9, 100, 200, 400-series, 500-series GPUs with a minimum of 256MB of local graphics memory.


GeForce Desktop Products
GPU	 Compute Capability
GeForce GTX 560 Ti	 2.1
GeForce GTX 550 Ti	 2.1
GeForce GTX 460	 2.1
GeForce GTS 450	 2.1
GeForce GTS 450*	 2.1
GeForce GTX 680	 2.1
GeForce GTX 590	 2.0
GeForce GTX 580	 2.0
GeForce GTX 570	 2.0
GeForce GTX 480	 2.0
GeForce GTX 470	 2.0
GeForce GTX 465	 2.0
GeForce GTX 295	 1.3
GeForce GTX 285	 1.3
GeForce GTX 285 for Mac	 1.3
GeForce GTX 280	 1.3
GeForce GTX 275	 1.3
GeForce GTX 260	 1.3
GeForce GT 520	 2.1
GeForce GT 440	 2.1
GeForce GT 440*	 2.1
GeForce GT 430	 2.1
GeForce GT 430*	 2.1
GeForce GT 420*	 1.0
GeForce GT 240	 1.2
GeForce GT 220*	 1.2
GeForce 210*	 1.2
GeForce GTS 250	 1.1
GeForce GTS 150	 1.1
GeForce GT 130*	 1.1
GeForce GT 120*	 1.1
GeForce G100*	 1.1
GeForce 9800 GX2	 1.1
GeForce 9800 GTX+	 1.1
GeForce 9800 GTX	 1.1
GeForce 9600 GSO	 1.1
GeForce 9500 GT	 1.1
GeForce 8800 GTS	 1.1
GeForce 8800 GT	 1.1
GeForce 8800 GS	 1.1
GeForce 8600 GTS	 1.1
GeForce 8600 GT	 1.1
GeForce 8500 GT	 1.1
GeForce 8400 GS	 1.1
GeForce 9400 mGPU	 1.1
GeForce 9300 mGPU	 1.1
GeForce 8300 mGPU	 1.1
GeForce 8200 mGPU	 1.1
GeForce 8100 mGPU	 1.1
GeForce 8800 Ultra	 1.0
GeForce 8800 GTX	 1.0
GeForce GT 340*	 1.0
GeForce GT 330*	 1.0
GeForce GT 320*	 1.0
GeForce 315*	 1.0
GeForce 310*	 1.0
GeForce 9800 GT	 1.0
GeForce 9600 GT	 1.0
GeForce 9400GT	 1.0
GeForce Notebook Products
GPU	 Compute Capability
GeForce GTX 675M	 2.1
GeForce GTX 670M	 2.1
GeForce GTX 660M	 2.1
GeForce GT 650M	 2.1
GeForce GT 640M	 2.1
GeForce GT 640M LE	 2.1
GeForce GT 635M	 2.1
GeForce GT 630M	 2.1
GeForce GT 620M	 2.1
GeForce 610M	 2.1
GeForce GTX 580M	 2.1
GeForce GTX 570M	 2.1
GeForce GTX 560M	 2.1
GeForce GT 555M	 2.1
GeForce GT 550M	 2.1
GeForce GT 540M	 2.1
GeForce GT 525M	 2.1
GeForce GT 520MX	 2.1
GeForce GT 520M	 2.1
GeForce GTX 485M	 2.1
GeForce GTX 470M	 2.1
GeForce GTX 460M	 2.1
GeForce GT 445M	 2.1
GeForce GT 435M	 2.1
GeForce GT 420M	 2.1
GeForce GT 415M	 2.1
GeForce GTX 480M	 2.0
GeForce GTS 360M	 1.2
GeForce GTS 350M	 1.2
GeForce GT 335M	 1.2
GeForce GT 330M	 1.2
GeForce GT 325M	 1.2
GeForce GT 240M	 1.2
GeForce G210M	 1.2
GeForce 310M	 1.2
GeForce 305M	 1.2
GeForce GTX 285M	 1.1
GeForce GTX 280M	 1.1
GeForce GTX 260M	 1.1
GeForce 9800M GTX	 1.1
GeForce 8800M GTX	 1.1
GeForce GTS 260M	 1.1
GeForce GTS 250M	 1.1
GeForce 9800M GT	 1.1
GeForce 9600M GT	 1.1
GeForce 8800M GTS	 1.1
GeForce 9800M GTS	 1.1
GeForce GT 230M	 1.1
GeForce 9700M GT	 1.1
GeForce 9650M GS	 1.1
GeForce 9700M GT	 1.1
GeForce 9650M GS	 1.1
GeForce 9600M GT	 1.1
GeForce 9600M GS	 1.1
GeForce 9500M GS	 1.1
GeForce 8700M GT	 1.1
GeForce 8600M GT	 1.1
GeForce 8600M GS	 1.1
GeForce 9500M G	 1.1
GeForce 9300M G	 1.1
GeForce 8400M GS	 1.1
GeForce G210M	 1.1
GeForce G110M	 1.1
GeForce 9300M GS	 1.1
GeForce 9200M GS	 1.1
GeForce 9100M G	 1.1
GeForce 8400M GT	 1.1
GeForce G105M	 1.1
Frequently Asked Questions

It is in the middle of the page, after all the "fancy" GPUs. (Alternately, just Ctrl+F.)

[bcp19]

Originally Posted by Dubslow

It is in the middle of the page, after all the "fancy" GPUs. (Alternately, just Ctrl+F.)

Ah, I missed it in the middle of all the 5's. I was wondering why they showed all the laptop ones but didn't seem to have the regular. I wonder why the early reports thought it would have a 512 bit bus when it has a 256. It appears identical to the 580 in memory bandwidth.

[LaurV]

Exchanging two 580's for... (I am thinking if should I put 4 keplers, or 3 keplers plus some cash change... :P)

Joking apart, I have read those papers (funny, I can't stop, they snagged-it the photos from winword documents and forgot to remove the red curved underlines used by winword to stress the words which are not in its dictionary, and all photos done this way look like shit because of those underlined words) and after reading those papers, I think that mfaktc should run about 2.7 to 3.9 times faster for the same power, and CudaLucas about 1.3 to 2.3 times faster for the same power. I say for the same power, because I believe they will release in the future some "GPU-Boost unlocked" versions (there is a GPU Boost inside which adjust the clock down when the power is over 170W, in such a way that the power should never go over 190W or so, as I understood, and you can do nothing against it, which is a big limitation for performance, which according with the larger number of cores, channels, clock, should be about 3.9 times of a gtx580, if allowed to "free heat"). Our mfaktc and CudaLucas would need to be re-thought, not only re-compiled. The reasons are missing the shader clock (for mfaktc??), different binding model and access logic (here they will have to realeas some new or amended cuFFT library, maybe).

I still don't plan to get one. But its release is for sure beneficial, I will get a third gtx580 with water cooling in two months, for the price I pay now for an air cooled one....