Wednesday, August 10, 2022
HomeComputer HardwareChinese language Chip Firm Releases 77 Billion Transistor BR100 GPU To Rival...

Chinese language Chip Firm Releases 77 Billion Transistor BR100 GPU To Rival NVIDIA Hopper H100


birentech installed gpu

China’s very desperate to not solely rid itself of dependence on overseas silicon but in addition to show dominance of its home processors. That is probably the impetus behind a few of the extra bombastic claims we have seen from Chinese language chipmakers, just like the Fantasy One GPU. Innosilicon just lately debuted its Fantasy Two GPU that appears far more down-to-earth, but it surely’s hardly the one firm making such claims.

Have you ever ever heard of Biren Expertise? Established in 2019, the corporate was created to concentrate on design and creation of home compute accelerators for the Chinese language market. Although we are saying “GPU”, it is pretty unlikely that Biren’s new chip really has any graphics capabilities in any respect; it seems to be fully targeted on GPGPU-style compute.

biren tech mezzanine card

We had been initially skeptical upon studying of Biren and the BR100. In spite of everything, who would not be skeptical when listening to a couple of brand-new firm debuting a processor that may compete with NVIDIA’s A100 in AI workloads? Naturally, there are some caveats to this information, not least of which is that, so far as we are able to inform, Biren has not really publicly demonstrated the chip but. As such, that is principally a paper launch of the 77b-transistor BR100 processor.
Biren Expertise claims that the BR100 can attain 2048 TOPS—that is 2 exa-ops—in INT8 compute, with efficiency purportedly cutting down in linear vogue by knowledge dimension. That ends in a declare of 256 TFLOPS FP32 compute, one thing which makes even NVIDIA’s H100 (and its 60 TFLOPs spec) look puny.
br100 specifications

BR100 apparently achieves this efficiency via the usage of 64GB of HBM2E reminiscence, 300MB of on-die cache, and a chiplet-based design constructed utilizing 2.5 CoWoS fabrication on TSMC’s 7nm course of. The BR100 itself comes on a mezzanine card with a beefy heatsink, not not like NVIDIA’s quickest compute accelerators. Its Open Accelerator Module design makes use of Compute Specific Hyperlink (CXL) to connect with the host system at 2.3 TB/sec.
br104 specifications

Biren Expertise additionally debuted the BR104 processor, which we’re guessing might be a single chiplet from the dual-chiplet BR100 design. It is available in commonplace PCIe card type issue, and its creators declare it may well present precisely half the efficiency of the BR100 inside a 300W add-in card type issue. Biren says the chips are designed for medical imaging, molecular dynamics, and different bodily simulations.

birentech chiplet gpu

Biren didn’t come out of nowhere in a single day; the corporate has been accumulating funding since its founding in 2019 and introduced a profitable take a look at of the BR100 in April of this 12 months. Whereas the corporate does not appear to have executed a stay demo of BR100 but, it appears reliable sufficient to us. The query is whether or not the BR100 and its BiLiren structure can stay as much as its creators’ formidable claims.

We are going to most likely discover out extra about these processors when the corporate presents at Sizzling Chips 34, beginning August twenty first.
RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

- Advertisment -
Google search engine

Most Popular

Recent Comments