Meet The Prodigy
Tachyum has disclosed the chip layout and the architecture for their Prodigy processor. Packed with 128 CPU cores, 16 DDR5-7200 memory controllers and 64 lanes of PCIe 5.0 expect top-notch compute power. The main units are connected via a 10 terabit/second mesh. The CPUs run at a peak performance of 5.7GHz which is on par with the latest offerings from AMD & Intel. The Prodigy is fabricated on TSMC’s industry-leading 5nm process node. This allows for a die size less than 500mm². The package itself is based on a monolithic design. The ISA (Instruction Set Architecture) is a custom version combining both RISC and CISC. This ISA effectively rules out all the inefficient instructions and all instructions are no more than 32/64-bits in width. The Prodigy CPU features an out-of-order architecture capable of decoding and retiring up to 8 instructions per clock. Similarly this CPU can issue 11 instructions per clock with an instruction queue that supports up to 48 instructions and a scheduler that supports 12 queues that are 15 entries deep. The CPU ships with 4 ALUs, one load unit, one store unit, one load/store unit, one mask unit, and two 1024-bit vector units. The core has an AI subsystem powered by a 4096-bit matrix unit. Performance wise, this all-in-one solution can offer 6x the power of NVIDIA’s cutting-edge H100 in AI FP8 Performance. Similarly, Tachyum took more shots by offering a whopping 9x lead in efficiency as compared to team green. It will be interesting to see how well these numbers hold up in a one-to-one comparison.
Release Date
All credits go to tachyum for these figures and numbers. The Prodigy CPU is planned to be released by next year, with mass production aimed for the second half of 2023.
