Cuda compute capability identified incorrectly, reading 3.5 as 2.1 ?

Hi,

My Pc got GeForce GT 705 graphic card OEM, with Cuda compute capability 3.5 (and note that its not notebook graphic card which has GeForce 705M Cuda compute capability 2.1).

https://developer.nvidia.com/cuda-gpus (one can see that it is 3.5, * is OEM only)

I followed official instruction and install correctly.
A app which needs at least Cuda compute capability 3.0.

It needs CUDA 9.0, I installed CUDA 9.0 and cuDNN 7.0 as per apps instruction.

The error as follows:

[totalMemory: 2.00GiB freeMemory: 1.65GiB
2019-05-03 18:22:54.082584: I C:\tf_jenkins\workspace\rel-win\M\windows-gpu\PY\36\tensorflow\core\common_runtime\gpu\gpu_device.cc:1168] Ignoring visible gpu device (device: 0, name: GeForce GT 705, pci bus id: 0000:01:00.0, compute capability: 2.1) with Cuda compute capability 2.1. The minimum required Cuda capability is 3.0.]

My GPU is OEM version and officially Nvidia states it has Cuda compute capability 3.5. Why it is ignoring GPU and not using when it’s computing capacity.

I am not developer or programmer but I can handle some advice from expertes in changing script etc.

I found this

https://github.com/tpruvot/ccminer/wiki/Compatibility

it shows “nvcc flag” code.

What is the precedure to change the script or code in python or to make this program work with GPU properly use Cuda capability 3.5.

Please help me with this error.

I don’t know where you got the information that the GT 705 is CC 3.5.
Its chip is a GF119 (Fermi), so the CC 2.1 being reported is correct.

I got it from here,
https://developer.nvidia.com/cuda-gpus

look for CUDA-Enabled GeForce and TITAN Products section>>GeForce and TITAN Products
GeForce GT 705* 3.5

and its not GF119.

The Tensorflow reporting is correct. Your GPU is compute capability 2.1
This isn’t a software or script issue. That GPU won’t be usable with Tensoroflow

https://imgur.com/EKt0vJS

Here is the screenshot for the website. If you are saying its 2.1 then website is incorrectly informed.

GT 705, GT 705*, and GT 705(OEM) are not necessarily the same product.

If you have a GT 705(OEM) I would refer to the information here:

https://www.geforce.com/hardware/desktop-gpus/geforce-gt-705-oem/specifications
(you might also want to read the note at the top of the page)

which indicates 48 CUDA cores. It is not possible for a cc3.0 or higher device to have only 48 CUDA cores, because cc3.x devices have a minimum of 192 CUDA cores, and all other higher compute capability devices have a minimum of 64 or higher CUDA cores.

The GT 705(OEM) evidently belongs to the Fermi generation, with a compute capability of 2.1

There may exist other GT 705 variants that have a compute capability of 3.5, but that doesn’t appear to be applicable to your case.

Thank you that was very informative (CUDA cores). And thank you everyone for guiding me in right direction.I need to get new GPU, too bad nvidia control panal does not show all these detils. things would have been lot easier to get a software which can tell you what your GPU capable of, its computing version details.

Nvidia makes gpu and it has different variants of same model and does not say 2.0 version or anything so got confused. (*is acutally OEM). Therefore, tensorflow is correct.