China’s First 7 Nanometer GPU “Big Island” Nears Mass Production to Challenge AMD and NVIDIA

The FPS Review may receive a commission if you purchase something after clicking a link in this article.

Tianshu Zhixin has revealed that Big Island, China’s first 7-nanometer GPU, is about to enter mass production and commercial delivery. The company claims that the GPU is capable of offering nearly twice the performance of mainstream manufacturers’ products at a lower power consumption, which seems impressive since it only uses half the chip area of similar products. Big Island features a 32 GB HBM2 memory configuration and a peak FP32 rate of 37 TFLOPS.

GPU NameAMD Instinct MI100NVIDIA A100Big Island
Process NodeTSMC 7nmTSMC 7nmTSMC 7nm
ArchitectureCDNA 1AmpereUnknown
Transistors50 Billion54 Billion24 Billion
Memory32 GB HBM240 GB HBM232 GB HBM2
Memory Bandwidth1.2 TB/s1.6 TB/s1.2 TB/s
FP32 Compute23.1 TFLOPs
46.1 TFLOPs (Matrix)
19.5 TFLOPs
156 TFLOPs (Tensor)
312 TFLOPs (sparsity)
BFLOAT16 Compute92.3 TFLOPs312 TFLOPs
624 TFLOPs (sparsity)
147 TFLOPs

As the first flagship product launched by Tianshu Zhixin, BI is the first fully self-developed GPGPU cloud high-end training chip based on the universal GPU architecture. It adopts the industry-leading 7nm manufacturing process, 2.5D CoWoS package, and accommodates 24 billion. Transistor, supports FP32, FP/BF16, INT32/16/8 and other multi-precision data mixed training, integrated 32GB HBM2 memory, storage bandwidth up to 1.2TB, single core can perform 147 trillion FP16 calculations per second (147TFLOPS@FP16)

Sources: Tianshu Zhixin, Wccftech

Tsing Mui
News poster at The FPS Review.

Recent News