A microprocessor developer with a proprietary architecture claims its Prodigy universal processors can outperform Intel and Nvidia chips in HPC and AI workloads. In addition, they can run code designed for other architectures using a dynamic binary translator without any performance degradation. Emulating x86, Arm, and PowerPC using general-purpose CPU hardware is something chipmakers have been doing for years, but with substantial performance degradation that was prohibitively expensive compared to running on high-performance hardware. native hardware. In fact, in the mainstream space, only Apple managed to emulate PowerPC using Intel's x86 processors in the second half of the 2000s, but Apple's Rosetta dynamic binary translator has been so successful largely because Intel's processors in at the time they were considerably more advanced than those based on the PowerPC architecture.
A processor for every workload
But Tachyum says its software emulation technology is so efficient that its Prodigy universal processor can run Arm and RISC-V code better than modern processors based on these architectures. Plus, it can run x86 code well enough to run legacy apps, Tachyum says. Tachyum Prodigy are homogeneous processors with up to 128 cores based on a proprietary architecture that can run different types of workloads (AI, HPC, data center, etc.) smoothly, reducing software development complexity and hardware architecture. According to the developer, when the Prodigy runs native code, it can outperform the fastest Intel Xeon processors with 10 times less power and leave Nvidia's A100 GPUs behind in HPC tasks, training, and AI inference. . Tachyum claims that 125 Prodigy HPC racks can deliver 32 EXAFLOPS tensors of performance, but does not disclose the number of Prodigy processors required and their expected power consumption. One of the interesting implications of universal processors like Tachuym's Prodigy that can be used for different workloads is that they can be used more widely by hyperscalers like Amazon Web Services or Google than systems based on traditional processors and hardware accelerators. . like GPUs. As a result, they can earn more money and lower their maintenance costs. Tachuym's Prodigy universal processors are not quite right. The company uses site-programmable gate arrays to emulate the chips and currently does not have a fully functional FPGA prototype. Currently, Tachyum lists a range of processor models on its website, although it doesn't appear that any of the chips can be purchased. Tachyum expects to register its Prodigy later this year (i.e. ship photomasks to the factory) and then begin volume production of processors sometime in 2021.