The NVIDIA A800 Lets You Put a Supercomputer in Your Workstation

If you like teraflops, then NVIDIA has the card for you. The A800 has it in spades, but if you’re looking to run AI applications in your workstation or even data center, then these cards offer a mind-boggling amount of performance that can literally sit on your desk and make miracles happen.



How Fast?


The A800’s numbers are wild, and we’ve seen some pretty wild numbers over the years. In terms of raw compute, you’re looking at 9.7 TFLOPs, which doesn’t sound like much, but that’s before you realize these are FP64 numbers!

In terms of tensor core performance, NVIDIA claims 1247 AI TOPS, which is more than thirty times what you need to run something like CoPilot+, for a sense of scale.

All of this parallel processing power really comes into its own when you consider the 1.5 TB/s (yes terabytes) of memory bandwidth.

That’s just one card, but if you use NVLink, you can hook up two A800s using a 400GB/s bidirectional connection.

What’s It Good For?

NVIDIA is all-in on AI, and the Tensor core performance of these cards make them perfect both for inference training and generative AI workloads, but despite all the AI hype scientific and engineering simulations will also benefit mightily, and “regular” data science and analytics are also on the table.

Don’t forget that with supporting software, NVLink can pool memory, so having multiple cards with 40GB of VRAM each opens interesting possibilities.

Sharing Is Caring

These are enormous GPUs, and let’s not kid ourselves about the price, but thanks to multi-instance GPU technology you can share a single A800 with up to seven users. So you can offer multiple users a small piece of the action, or pool it all for time-critical jobs.

Power Efficiency Matters More Than You Think

You may have read that AI applications are real power hogs, and that’s true. It’s not just an environmental issue either. Getting AI power consumption down is crucial to making the technology profitable.

Which is why we were impressed to see that the A800 has a peak power draw of 240W! Keep in mind that’s the most the card will draw, so most real-world applications will come in under this number, depending on the job in question.

This low power consumption figure also means having two cards in one desktop system doesn’t mean blowing your budget on an extreme power supply or cooling solution. Which is why we’re confident we can put the A800 to good use in cost-effective build here at Titan Computers.