Big Island, China's first "cloud-trained GPU," has been under development for nearly four years and is finally ready for mass production. The manufacturer, Shanghai Tianshu Intelligent Semiconductor Co.

According to Tianshu Zhixin (as the company is commonly known in Asia), this sleek-looking 7nm AI and HPC-focused monster could deliver "twice the performance of mainstream manufacturers' products."

Not only will Big Island's performance be comparable to commercial chips from AMD and Nvidia, Tianzhu Zhixin claims it will offer a more impressive price/performance ratio and lower power consumption. This could be a major boost for sectors such as education, medicine, and security, which utilize such powerful graphics solutions for critical tasks.

Big Island supports many floating point formats. These include, but are not limited to, FP32, FP16, BF16, INT32, INT16, and INT8. as noted by Tom's Hardware, Big Island products can achieve up to 147TFLOPS on the FP16 test platform There is potential for this to happen. However, the current generation Nvidia A100 and AMD Instinct MI100 are likely to be competitive at 77.97TFLOPS and 184.6TFLOPS, respectively. However, both Nvidia and AMD have tricks up their sleeves to further power their respective cards.

Still, 147TFLOPS is not a bad number. No price has been announced yet, so we can't comment on price vs. performance.

After years of development, the Big Island design was actually finalized sometime in 2020, but then stalled for some time. This may have been due in part to the fact that the technology industry ran out of parts during the pandemic process. The company has since managed to get back into decent shape and production is expected to begin as soon as possible.

It is not yet clear if Big Island will be released outside of China. Given the company's focus on domestic manufacturing and reduced reliance on foreign technology, it is likely that the chip will remain in the country.