New cards with old drivers

Accelerated Computing CUDA CUDA Programming and Performance
1

Greetings,

We are considering using a GeoForce card for computing with CUDA on Linux in a new product, but we are concerned about stability. Once we release the product it will be on the market for about 5 years, and we do not want to upgrade the Linux distribution (we do the PC also, which will remain stable). Any modification of the linux distribution, or of the drivers that we ship with it, will be a major problem. I understand we cannot expect an assurance of supply of five years, but is it reasonable to expect that the new cards, as they reach the market, run on the old versions of the driver that will be on our product? I understand it wold not be optimal from the card point of view…

Any hint will be greatly appreciated, or any information on how it has been in the past.

Best regards,

jm

2

No, unfortunately typically new cards require new drivers as well. Especially over a timespan of 5 years, which will probably cover multiple binary incompatible architectures.