Tianshu Zhixin has developed the first Chinese GPGPU computing circuit
Source: IO Tech added 19th Jan 2021At this stage, the theoretical FP 16 performance is known from the performance of a computing circuit called Big Island, which falls between the computational circuits of AMD and NVIDIA.
Named Big Island by Tianshu Zhixin, the GPGPU is built from as many as 24 billion transistors and is manufactured using a 7-nanometer manufacturing process. According to Tom’s Hardware, the circuit manufacturer is not public information, but the information provided about the process is consistent with TSMC’s process. The so-called district. tape-out, ie the completion of the finished masks used in its manufacture, took place in May last year. According to a published rendering of the circuit, it is supported by four stacks of HBM memory.
So far, Big Island’s performance readings are still partially obscured. According to Tianshu Zhixin, the circuit would achieve TFLOPS performance with FP 16 accuracy 147 when its most direct competitors AMD’s Radeon Instinct MI 100 achieves 184, 6 TFLOPS and NVIDIA A 100 77, 97 TFLOPS with the same accuracy. In addition to FP 16 accuracy, Big Island also supports all other general calculation accuracies, including the relatively newcomer BF 16 or BFloat 16.
Source: Tom’s Hardware
brands: AMD Built Direct First It NVIDIA other Tom media: IO Tech keywords: Memory Radeon
Related posts
Notice: Undefined variable: all_related in /var/www/vhosts/rondea.com/httpdocs/wp-content/themes/rondea-2-0/single-article.php on line 88
Notice: Undefined variable: all_related in /var/www/vhosts/rondea.com/httpdocs/wp-content/themes/rondea-2-0/single-article.php on line 88
Related Products
Notice: Undefined variable: all_related in /var/www/vhosts/rondea.com/httpdocs/wp-content/themes/rondea-2-0/single-article.php on line 91
Warning: Invalid argument supplied for foreach() in /var/www/vhosts/rondea.com/httpdocs/wp-content/themes/rondea-2-0/single-article.php on line 91