On 03/26/2013 10:10 AM, Connie Sieh wrote:
> On Tue, 26 Mar 2013, Yasha Karant wrote:
>
>> Am I missing something here?  Does any other production vendor supply
>> GPU compute engine cards but Nvidia?       Are any GPU compute cards
>> fully
>> supported (including any additional interconnects beyond PCI) using
>> fully open source drivers and compilers/application support
>> generators/libraries?  To use the Nvidia GPU compute cards under CUDA,
>> it appears that the Nvidia proprietary driver is necessary.
>>
>> Yasha Karant
>>
>> On 03/25/2013 07:34 PM, Paul Robert Marino wrote:
>>> Um well
>>> Frankly the proprietary driver is never up to date with the kernel and
>>> it is well that's luck if it ever works with a new version of the kernel
>>> after you have reinstalled "recompiled the module with code you can't
>>> see against the new code"
>>>
>>>
>>> If you  have a problem with the proprietary driver take it up with
>>> ?Nvidia. In theory you pay them to make it work correct ?
>>> If you don't pay them for support then find a card that doesn't use
>>> proprietary code.
>>>
>>>
>>>
>>> -- Sent from my HP Pre3
>>>
>>> ------------------------------------------------------------------------
>>> On Mar 25, 2013 9:59 PM, Jeff Siddall <[log in to unmask]> wrote:
>>>
>>> On 03/25/2013 12:41 PM, Yasha Karant wrote:
>>> > We are forced to use the Nvidia proprietary driver for two reasons:
>>> >
>>> > 1. We use the switched stereoscopic 3D mode of "professional" Nvidia
>>> > video cards with the external Nvidia 3D switching emitter for the
>>> > stereoscopic 3D "shutter glass" mode of various applications that
>>> > display stereoscopic 3D images (both still and motion).
>>> >
>>> > 2. We need to load Nvidia CUDA in order to use the CUDA computational
>>> > functions of Nvidia GPU compute cards in our GPU based compute
>>> engines.
>>> > The Nvidia CUDA system appears to require the proprietary Nvidia
>>> driver.
>>>
>>> Yup, I run the proprietary driver for VDPAU support. If anyone knows
>>> how to get that from the open source driver I would like to know.
>>>
>>> Jeff
>>
>
>
> The issue with the video card driver is with the vendor of the card.
> Since you paid for the video card then you should contact the video card
> vendor and have them "fix" what needs fixing.  They got your money now
> they need to support their products.
>
> -Connie Sieh

An excellent suggestion.  The unfortunate reality is that for a least 
some of these products, the only well support environment is MS Windows, 
sometimes Mac OS X, and only lastly, open systems such as Linux or the 
BSD variants.  In the case of Nvidia, the CUDA compute engines as well 
as the professional switched stereoscopic 3D do have a strong advertised 
support for and deployment under enterprise Linux.  However, the reality 
is that the compatibility with existing Linux environments, e.g., SL, is 
not what one might desire for a "supported" product.

However, if enough of us who professionally use SL and other TUV EL 
variants sufficiently complain, it is possible that Nvidia will make all 
of this work -- provided the profit is there for the corporation. 
Unlike SL and related efforts, as far as I can tell, USA for profit 
corporations, such as Nvidia, only exist for one purpose -- that all 
overarching profit.  All else is just lipservice.  (This may not be the 
case in some EU nations in which for profit corporations are required to 
have real workers share in the shaping of policy and in real business 
management, unlike the Gompers model in which all the workers are 
concerned with are working conditions and financial compensation -- not 
the direction or societal value of the corporation.)

The above is not meant as a political statement or a statement of 
philosophy -- it is meant purely as a statement of factual reality -- a 
reality within which we must work if we want "them to 'fix' what needs 
fixing".

Yasha Karant