I’m confused about the RTX Ada 5000-6000 line-up. It seems they compare poorly to much cheaper GPUs such as an RTX 4090 (not taking into account VRAM) and I figure if a company wanted to scale out nodes they’d go for A100/H100 instead. What makes them attractive for companies?
We use Nvidia enterprise cards in our workstations.
The cost vs. a 4090 is irrelevant. We only care that our software is supported and that the amount of VRAM is sufficient.
I cannot emphasize enough how unimportant the ~$4,000 cost difference is when the workstation it is in costs double that, the engineer using it costs more than that each week, and the software running on it is tens of thousands of dollars per seat.
We have a three-year desktop refresh cycle so over three years once you total up all of the expenses for everything-- from the per-square-foot rent for the office to the software license costs to the salary of the engineer the price difference is not worth even thinking about.
edit: conversely, if there is anything that causes an issue, like the consumer version of the card not being supported by application we use, even if it is only unsupported because the vendor didn't bother testing it, and that leads to a back-and-forth with support while diagnosing a problem, that $4,000 becomes very relevant because you end up wishing you spent double that to avoid the distraction.
They don't perform poorly, necessarily, just differently. It has to do with the workload and feature set.
I run an A6000 for a live video mixer. It barely touches the RAM and does nothing with the RTX cores but absolutely chews up the CUDA cores.
I also run a quadro sync card that allows the GPU to be genlocked (vsync for the video world) with the rest of the studio.
A feature that I don't use but that's also offered- you can connect multiple GPUs across multiple systems and have them all be frame synced to each other this creating a single output without tearing.
