We are during process of buying new work stations for our GIS specialists. Some of the GIS tools required CUDA Compute Capability on the specified level in order to experience better performance when dealing with large GIS data.
According to the GPU Compute Capability list (
CUDA GPUs | NVIDIA Developer) the NVIDIA RTX A2000 is listed only in “mobile” section and not in the “desktop” section. According to the NVIDIA RTX A2000 data sheet ( https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/rtx-a2000/nvidia-rtx-a2000-datasheet-1987439-r5.pdf) it supports CUDA API but the Compute Capability version is not provided.
Can anyone confirm that the NVIDIA RTX A2000 desktop has Compute Capability 8.6 and there is just error on the GPU Compute Capability list?
For what it’s worth, the information here states that the card is using the GA106 chip, which would make it Compute Capability 8.6:
NVIDIA GA106, 1200 MHz, 3328 Cores, 104 TMUs, 48 ROPs, 6144 MB GDDR6, 1500 MHz, 192 bit
Thanks! That’s some start point.
NVIDIA RTX A2000 Specs | TechPowerUp GPU Database NVIDIA RTX A2000 has GA106 and CUDA 8.6. Moreover according to NVIDIA GA106 GPU Specs | TechPowerUp GPU Database GA106 has CUDA 8.6.
Then according to
https://www.techpowerup.com/gpu-specs/docs/nvidia-ampere-architecture.pdf which looks like NVIDIA paper, GA100 has CUDA 8.0.
What is really frustrating I’m unable to find in NVIDIA RTX A2000 official specification that it is using GA106, moreover I’m unable to find GA106 whitepaper.
As far as It seems reasonable that NVIDIA RTX A2000 supports Compute Capability 8.6, without official specification it is difficult to justify the purchase.
You could contact nvidia support for clarification.
Or ask your workstation supplier to verify for you, I’m sure they’ll see this as a reasonable request to make the sale.
Unfortunately, getting response from support sometimes takes too much time. But of course we have send question for clarification.
According to NVIDIA RTX A2000 specification it is based on Ampere architecture. According to Ampere documentation, it is compatible with Compute Capability 8.0:
NVIDIA Ampere GPU Architecture Compatibility Guide :: CUDA Toolkit Documentation
I suppose this is enough official justification.
That’s an oversight. It is compute capability 8.6 in desktop form also.
I don’t personally have the authority to update the web page. If you need that sort of confirmation, you can
file a bug