Cause for serious irritation - Specs lacking on NVIDIA site.

First of all I am uncertain where this post actually belong but in the end this is very much an issue for programming of CUDA.

I am looking to buy a few older GPUs and tried to discern which CM (Computing model) or std. is possible on the cards. Among other some GTX285 GeForce.

I have sent feedback to the website staff already but was informed not to obtain a response on my feedback.

Secondly, I have looked over two different sets of nvidia webpages with outlined specifications of this and other cards. The following holds true for all cards at all times with only a few recent exceptions to newer cards.

1: No spec of the X.Y level of CM.

2: No architecture specification.

These two apparently fundamental specs should logically be included in any and all specifications.

I can not say I am helped by the fact that the website states “CUDA is supported” gives any answer to the fact if it will work in my case.

Any help or a hint of possible contact to get some serious improvement of the website in motion would be highly appreciated.

And above all, is GTX285 and it’s cousins FERMI or something else?

The “X.Y” “CM” you are referring to is called compute capability in CUDA jargon.

Most GPU compute capabilities are documented here:

or here:

You will find the GTX 285 on the 2nd of those 2 pages. It is a cc 1.3 GPU. It is in a generation of GPUs that predates Fermi. Fermi GPUs have compute capability 2.x

Once you know the compute capability, the architecture name is straightforward:

cc 1.x - Tesla (not to be confused with the Tesla brand)
cc 2.x - Fermi
cc 3.x - Kepler
cc 5.x - Maxwell
cc 6.x - Pascal
cc 7.x - Volta

cc 1.x and 2.x GPUs are no longer supported by CUDA 9 or 9.1

cc 1.x GPUs had support dropped in CUDA 7, which means the last CUDA toolkit supporting that GPU is CUDA 6.5

You can get fairly detailed specifications for cc 1.x GPUs by retrieving the CUDA 6.5 toolkit, installing it, and referring to the programming guide contained therein. Here’s the compute capability section from the current programming guide:

(starts with cc 3.0 since CUDA 9.1 only supports cc 3.0 and higher)

General web site feedback can be delivered here:

Excellent information as per usual txbob :) /Thanks.

On the feedback, that is a check.
Submitted a suggestion to them to implement proper architecture ref. in specs.