128 core CPU comes out: 12-channel DDR5-PCIe5.0, super-top Intel xstrong

Tachyum, a Slovakian start-up founded in 2016, today launched a bombshell, announcing a new 128 core Prodigy processor, which it says “can surpass Intel’s fastest and power consumption by only one-tenth of a percent.” Let’s start with Tachyum’s background: the company was founded in Slovakia and received $17 million from the Slovak government, but several of its founders are American and have made a big deal of it.

In particular, CEO Radoslav Danilak, with 25 years of experience in the semiconductor industry, was founded by SandForce, the once-red-and-purple SSD master giant, and was acquired by Seagate, then by Seagate, and later by Skyera, where he continued to study SSD mastertechnology and was acquired by Western Digital in 2014.

Tachyum Prodigy is billed as the world’s first “universal processor” because it integrates a common purpose processor, high-performance computing, AI artificial intelligence, DML deep mechanical learning, interpretable artificial intelligence (Explainable AI), bio-artificial intelligence (Bio AI) and more, based on a parallel multiprocessor environment that simplifies programming models and environments.

128 core CPU comes out: 12-channel DDR5-PCIe5.0, super-top Intel xstrong

The latest top-of-the-line model is the Prodigy T6128, which integrates 128 physical cores, a sequential execution architecture with 4 instructions per clock cycle, supporting 64-bit addressing, 512-bit vector operation, AI/ML vector and matrix multiplication acceleration, virtualization, advanced RAS, and up to 4GHz.

Cache, each core 32KB first-level instruction cache, 32KB first-level data cache, all support ECC, the last level of the cache share 64MB, support decTED ECC.

In terms of memory, 12 channels are supported by DDR4, DDR5, and the highest frequency DDR5-4800, but there can only be one memory bar per channel, a maximum capacity of 512GB per channel, a total of 6TB, and support for advanced error correction and RAS.

For the IO side, up to 36 PCIe 5.0 controllers are integrated, up to 48 channels, and two sets of 400G (400,000 M) Ethernet controllers are integrated.

What’s even more striking is that all these powerful specifications, with tSMC’s 7nm process added, have a package area of just 85 x 85 square millimeters, slightly larger than Intel LGA2066’s Core, but smaller than the AMD SP3 thread tearer.

Tachyum did not disclose The specific architecture of Prodigy, did not know whether based on RISC-V, MIPS, ARM or self-study, but said that both single-threaded and multi-threaded applications have surpassed Intel’ strength, but it is smaller than ARM.

According to reports, The Prodigy T6128 processor is suitable for large-scale supercomputers, big data, large AI applications, can provide 262TFlops AI training and reasoning performance, 16TFlops HPC high-performance computing performance.

In terms of development environments, Tachyum also offers a range of tools, including FPGA simulators, software simulators, binary translators, C/C?/Fortran compilers, debuggers and profiles, and TensorFlow compilers, all under the Linux operating system.

128 core CPU comes out: 12-channel DDR5-PCIe5.0, super-top Intel xstrong

128 core CPU comes out: 12-channel DDR5-PCIe5.0, super-top Intel xstrong

If the customer does not need the 128 core so tall specifications, Tachyum also offers 64/32/24/16 core and other different configurations.

The 64 Core has two models, one tadle, supporting eight-channel DDR4/DDR5 memory, 72 PCIe 5.0 channels, two sets of 400G Ethernet, two sets of HBM3 (optional), 32MB fully consistent level / Three-stage cache, operating frequency 4GHz, core voltage 0.8V, thermal design power consumption of 180W, core area of 290 square mm, package area of 66 x 66 square mm, can replace single/dual-way XE7, Xeon E5.

The other is TH24, which is dedicated to AI/HPC, four-channel DDR5 and/or 32GB HBM3, which can also be used as a cache or stand-alone memory and requires high precision water cooling.

The 32 core models are T432, four-channel DDR4, 32 PCIe 4.0, and two sets of 100G Ethernet. The core 16 is T216, dual-channel DDR4, 32 PCIe 4.0, and two sets of 50G Ethernet. Both are small packages, low cost and low price, suitable to replace The E5, E3, And EGD Series.

Of course, most of these products are still on paper, currently only streaming success of 64 core T864, is expected to be put into mass production this year.