Tachyum, which back in 2020 introduced the 128-core Prodigy universal processor with some outrageous performance, announced a workstation that will be affordable for many ordinary users.
The PC is called the Prodigy ATX Platform and will cost $5,000. This is a lot for a regular consumer PC, but for a workstation this is quite a normal price.
The Prodigy ATX Platform system is based, of course, it will all be on the same Prodigy processor, only cut down to 96 cores. The CPU will run at 5.7 GHz and offer an eight-channel DDR5 controller.
The system configuration will also include 1 TB of RAM and a 1200 W power supply. It is worth noting three PCIe 5.0 slots and three M.2 slots for SSD.
The company focuses its platform on working with AI. In particular, the manufacturer claims that a single Prodigy processor can run a ChatGPT4 model with 1.7 trillion parameters, while the same model requires 52 Nvidia H100 GPUs!
As for the 96-core CPU in the specified desktop system, the company calls it half-chip. Simply put, this is half of a large 192-core CPU. By the way, earlier during the announcement Tachyum talked about a 128-core processor, but now the site only has information about a 192-core processor.
All this looks very impressive and to some extent even fantastic, given the comparison with the Nvidia H100. Unfortunately, there are reasons to doubt that Tachyum will be able to offer the Prodigy ATX Platform for $5,000. For example, 1 TB of RDIMM memory retails for about $3,800. Even if the company buys modules cheaper, it is only RAM. And also the price of a complex motherboard, power supply and processor with cosmic performance.
It's also unclear whether the Tachyum processor is even real. Let's remember that the first announcement was back in 2020 with plans for release in the same year. Then plans changed several times, but the processor has not yet been released. The company is now talking about launching at the end of this year. Some oddities can also be attributed to a one and a half times increase in the number of CPU cores, but this requires additional effort to design a new chip.