TACHYUM
Tachyum has revealed new details about its Prodigy processors, which the company calls the world's first general-purpose processors.
In particular, the performance (FP64) of the 128-core Prodigy T16128-AIX is three times that of the latest monstrous Nvidia H100 accelerator! If Nvidia's solution is talking about 30 TFLOPS after the recent spec update, the Prodigy T16128-AIX offers 90 TFLOPS! Single precision performance (FP32) is also three times better: 180 TFLOPS vs. 60 TFLOPS. Also, in some other modes (INT8 and FP8) the processor is six times faster than the new Nvidia.
It can also be noted that a standard server rack with liquid-cooled Prodigy T16128-AIX processors offers a performance (FP64) of 12.9 PFLOPS versus 960 TFLOPS for the same rack with Nvidia H100.
Compared to conventional processors, the Prodigy T16128-AIX is four times faster than the Intel Xeon Platinum 8380 in Specrate 2017 INT benchmarks and 40 times faster in double-precision floating-point calculations. Also, the novelty is better than Google's fourth-generation specialized TPU.
Remember that Prodigy processors are universal, that is, they are able to conditionally execute any code using a dynamic binary translator, so they can perform typical tasks for both CPUs and GPUs or TPUs.
Tachyum expects to receive the first samples from its processors before the end of this year, and mass production will begin in the second half of next year.
mundophone
No comments:
Post a Comment