I installed CUDA 6.5 on Nvidia GeForce 210 and O.S. windows 7 ultimate. When I run my Maple code, Maple shows CUBLAS error. Can anyone help me?
I think you might be in the wrong board in this forum. This covers mostly programming topics.
The Geforce 210 is a Tesla microarchitecture device (compute capability 1.2) with just 1 GB of RAM. The later CUDA development kits (beyond CUDA 6.5) have actually deprecated this old architecture. Furthermore it lacks features that some software might rely on (such as warp shuffles, double precision support). So even software written with older CUDA development kits may refuse to run on Tesla class devices if it needs to make use of said features.
here’s a reference for what I’ve just stated:
If possible, upgrade to something like a current generation nVidia GTI 1030, 1050, 1050Ti card or better. This is, if you aren’t on a laptop where graphics cannot be upgraded.
For actually decent compute performance in single precision floating point arithmetics you might want a GTX 1070 card at least.
hmm, according to this support page, Maple might support old hardware with compute capability 1.2 hardware or better.
nVidia drivers release R396 or newer have dropped CUDA driver support for Fermi cards.
More details about the driver changes here:
Driver support for the Tesla microarchitecture of graphics cards (compute capability 1.x) would have been ended earlier, but I can not pinpoint the release number now. EDIT: Version: 342.01 WHQL (late 2016) seems to be one of the last drivers to support this card.
NOTE: do not confuse Tesla series of GPUs (expensive cards for professional compute applications) with the Tesla microarchitecture (earliest generation of CUDA capable cards).
Thanks. But I saw in one site that CUDA was worked on Older systems.
You might want to contact that site to learn how they did it. Presumably by running with old CUDA versions and old NVIDIA driver packages from about five years ago.
As cbuchner1 says, it would be very much advantageous to upgrade to an affordable modern consumer card based on the Pascal architecture.
Yes, your solution is to run ancient driver that still contained CUDA6.5 runtime and support for your old card.
You must act as if it is whatever year that card was released. This may include not installing service packs if it broke compatibility with an ancient WDDM driver… etc
Driver version was given above, install that driver, and try again.