3Dwarrior
SOCIAL:
Filed under: 7nm GPU, Big Island GPU, CHina, China GPU, cloud computing, Featured, Hardware, hpc, News, Report, Sticky

China Readies Its First 7nm ‘Big Island’ GPU For Cloud…

China-based cloud and HPC method company, Shanghai Tianshu Intellectual Semiconductor, has unveiled its initially 7nm ‘Big Island’ GPU-based mostly graphics card. At the moment, in mass generation, the GPU is set to offer practically 2 times the functionality mainstream choices from competition in fifty percent the die space and with much reduced power intake.

Chinese Significant Island GPU Primarily based on 7nm Approach Node To Supply Two times The General performance of Competitors In 50 % The Die Location, Graphics Card Pictured With Up To 37 TFLOPs Horsepower at 300W

During its board conference, the Founder of the organization, Zheng Jinshan, unveiled its Major Island GPU that features a monolithic design and style. The GPU is centered on the TSMC 7nm FinFET method node & has the two.5D CoWoS style and design thinking about it packs the GPU and DRAM on the same die. The GPU is made up of 24 Billion Transistors and characteristics up to 37 TFLOPs of FP32, 147 TFLOPs of FP16/BF16, 371. TOPs INT32, 147 TOPs INT16, and 295 TOPs INT8 calculations per 2nd. The actual core specs and clock speeds are not known but they should be unveiled as the start closes in.

For comparison, the AMD Instinct MI100 has a peak FP32 fee of 23.one TFLOPs and a peak BFLOAT16 charge of 92.three TFLOPs. The NVIDIA A100 GPU on the other hand has a peak FP32 price of 19.five TFLOPs with Tensor cores pushing it to 156 TFLOPs and 312 TFLOPs with sparsity. The peak BFLOAT16 level of the NVIDIA A100 GPU is rated at 312 TFLOPs and 624 TFLOPs with sparsity.

China Big Island GPU Specs Comparison

GPU Name AMD Intuition MI100 NVIDIA A100 Major Island
Course of action Node TSMC 7nm TSMC 7nm TSMC 7nm
Architecture CDNA one Ampere Mysterious
Transistors 50 Billion fifty four Billion 24 Billion
Cores 7680 6912 TBC
Memory 32 GB HBM2 40 GB HBM2 32 GB HBM2
Memory Bandwidth one.two TB/s 1.six TB/s one.two TB/s
FP32 Compute 23.one TFLOPs
46.1 TFLOPs (Matrix)
19.five TFLOPs
156 TFLOPs (Tensor)
312 TFLOPs (sparsity)
37 TFLOPs
BFLOAT16 Compute ninety two.three TFLOPs 312 TFLOPs
624 TFLOPs (sparsity)
147 TFLOPs
TDP 300W 400W 300W

As for memory, the China Huge Island GPU options 32 GB of HBM2 memory that operates with up to one.2 TB/s of bandwidth. The GPU also has a PCIe Gen 4 x16 interface which has a bandwidth of up to 64 GB/s. As for the products them selves, the Huge Islan GPU will appear in equally PCIe & the OAM sort elements.

The China-centered company unveiled shots of each types. The GPGPU primarily based PCIe graphics card capabilities a conventional twin-slot design and style with a big aluminum heatsink beneath the silver-coloured shroud. The card arrives with a passively-cooled layout considering that it will be built-in in just servers with lots of lively cooling. The card is driven by a one 8-pin connector and has a 300W TDP. The organization also lists down 60A MOSFETs characteristics on the card. The OAM providing ships with the typical mezzanine board sort variable and the 2U rack revealed has the OAM board that includes a substantial aluminum heatsink with numerous copper warmth pipes jogging by way of it.

The organization hasn’t furnished any actual delivery dates or pricing for its Large Island GPU but it is likely that it would launch someday by the close of this calendar year or early following year provided that it has previously been moved to mass manufacturing. We count on additional details from

The post China Readies Its To start with 7nm ‘Big Island’ GPU For Cloud Computing To Tackle AMD & NVIDIA – Graphics Card Pictured With 32 GB HBM2 Memory, Up To 37 TFLOPs at 300W by Hassan Mujtaba appeared first on Wccftech.