Emulates everything, no more need for GPU, and supposedly has a first customer looking for an HPC setup with on towards a half million sockets.
Tachyum’s first chip Prodigy hasn’t even taped out - let alone gone into mass production - but one customer has, we’re told, committed to buying hundreds of thousands of the processors to power a massive 50 exaFLOPS supercomputer.
Usually when we see numbers like that, the obvious assumption is they’re talking about AI FLOPS using 8- or 16-bit floating-point math precision, not the 64-bit double-precision calculations typically used in high-performance computing. But Tachyum claimed the system will be capable of 25x the performance of “the world’s fastest conventional supercomputer built just this year.”
This appears to be a reference to the newly inaugurated Aurora supercomputer at Argonne National Labs, which boasts more than two exaFLOPS of peak FP64 performance.
If Tachyum’s claim wasn’t already wild enough, the processor designer claims the forthcoming system will be capable of eight zetaFLOPS of AI performance for large language models and will boast hundreds of petabytes of DDR5 memory, when it’s completed in 2025.
Very impressive chip. Is there any chance (or sense) in AMD buying them to strengthen their server and AI position?
I’ll be showing my ignorance here, I’m sure, but you can’t buy a start-up? If not, I’m sure Tachyum might like a big cash infusion from AMD, thereby giving AMD a large stake in its future.
If AMD is involved, that is great. I didn’t see that reported. Did I miss it?
When someone has a good idea for a company they put together a business plan and make presentations usually to venture capitalists for funds. They use those early funds to start the company and work out the technical details on their company.
The series B funding for Tachyum in 2021 suggests they are starting into the second stage. Company seems to be making progress. It probably means they met first milestones to qualify for more funds.
Venture capitalists often own a large share of the initial shares, but they are privately held and cannot be traded. Usually the venture capital firm assists in setting up the company, helping find officers and board members. When the company IPOs, the shares get market value and can be sold.
Individuals can participate in venture capital firms if they like. Usually it requires signing a statement that you have sufficient assets to tolerate the risk. IPOs are risky enough for most of us; venture investing can occasionally hit a home run but many new companies fail. Very risky.
It is not clear to me that this is the right fit for AMD right now. It would add yet another architecture for AMD to support with their limited software resources. I did notice an old AMD hand, Fred Weber, is involved in this project.
This is a dramatically different approach to microprocessors, graphics processors, everything. Intended to compete with everyone and everything. No real opportunity for synergy, I suspect. You would buy this to kill it or for the patents.
At the very least, there must be some good IP that would be worth assimilating and being able to use in possible future products. Oh well, it was just an idea, possibly a bad one.
There probably is some great IP here, but I doubt they are looking for a buyer. They will be worth A LOT more if they successfully deliver silicon and get a system up. And I’m sure the team is stunning.