China-based cloud and HPC system provider, Shanghai Tianshu Intellectual Semiconductor, has unveiled its first 7nm 'Big Island' GPU-based graphics card. Currently, in mass production, the GPU is set to offer nearly twice the performance mainstream offerings from competitors in half the die area and with much lower power consumption.
Chinese Big Island GPU Based on 7nm Process Node To Deliver Twice The Performance of Competitors In Half The Die Area, Graphics Card Pictured With Up To 37 TFLOPs Horsepower at 300W
During its board meeting, the Founder of the company, Zheng Jinshan, unveiled its Big Island GPU that features a monolithic design. The GPU is based on the TSMC 7nm FinFET process node & has the 2.5D CoWoS design considering it packs the GPU and DRAM on the same die. The GPU is made up of 24 Billion Transistors and features up to 37 TFLOPs of FP32, 147 TFLOPs of FP16/BF16, 371.0 TOPs INT32, 147 TOPs INT16, and 295 TOPs INT8 calculations per second. The exact core specs and clock speeds are unknown but they should be unveiled as the launch closes in.
For comparison, the AMD Instinct MI100 has a peak FP32 rate of 23.1 TFLOPs and a peak BFLOAT16 rate of 92.3 TFLOPs. The NVIDIA A100 GPU on the other hand has a peak FP32 rate of 19.5 TFLOPs with Tensor cores pushing it to 156 TFLOPs and 312 TFLOPs with sparsity. The peak BFLOAT16 rate of the NVIDIA A100 GPU is rated at 312 TFLOPs and 624 TFLOPs with sparsity.
China Big Island GPU Specs Comparison
| GPU Name | AMD Instinct MI100 | NVIDIA A100 | Big Island |
|---|---|---|---|
| Process Node | TSMC 7nm | TSMC 7nm | TSMC 7nm |
| Architecture | CDNA 1 | Ampere | Unknown |
| Transistors | 50 Billion | 54 Billion | 24 Billion |
| Cores | 7680 | 6912 | TBC |
| Memory | 32 GB HBM2 | 80 GB HBM2 | 32 GB HBM2 |
| Memory Bandwidth | 1.2 TB/s | 1.6 TB/s | 1.2 TB/s |
| FP32 Compute | 23.1 TFLOPs 46.1 TFLOPs (Matrix) | 19.5 TFLOPs 156 TFLOPs (Tensor) 312 TFLOPs (sparsity) | 37 TFLOPs |
| BFLOAT16 Compute | 92.3 TFLOPs | 312 TFLOPs 624 TFLOPs (sparsity) | 147 TFLOPs |
| TDP | 300W | 400W | 300W |
As for memory, the China Big Island GPU features 32 GB of HBM2 memory that operates with up to 1.2 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 themselves, the Big Islan GPU will come in both PCIe & the OAM form factors.
The China-based company unveiled pictures of both designs. The GPGPU based PCIe graphics card features a standard dual-slot design with a large aluminum heatsink underneath the silver-colored shroud. The card comes with a passively-cooled design considering that it will be integrated within servers with lots of active cooling. The card is powered by a single 8-pin connector and has a 300W TDP. The company also lists down 60A MOSFETs features on the card. The OAM offering ships with the standard mezzanine board form factor and the 2U rack shown has the OAM board featuring a massive aluminum heatsink with several copper heat pipes running through it.
The company hasn't provided any exact shipping dates or pricing for its Big Island GPU but it is likely that it would launch sometime by the end of this year or early next year given that it has already been moved to mass production. We expect more details from
