Chinese computing card in production – performance that competes with AMD and Nvidia

AMD and Nvidia are not alone in developing graphics and computing card circuits and in China are investing heavily in domestic solutions, with the goal of replacing all foreign hardware by 2022. Last year, for example, progress was reported on the processor front with Zhaoxin-signed models taking place in consumer-oriented computers and now more news follows, but with a data center in focus.

It is the company Shanghai Tianshu Intellectual Semiconductor that unveils a computing card with the code name “Big Island”, which now includes mass production on TSMC’s densely populated process 7 nanometers. Like Nvidia’s and AMD’s high – performance computing cards A100 and Instinct MI100, it is a monolithic circuit that is accompanied by large amounts of HBM2 memory, for a high-performance result on paper.

Big Island

A100

Instinct MI100

Circuit

Big Island

GA100

Arcturus

Transistors

24 billion

54 billion

50 billion

Manufacturing

TSMC 7 nm

TSMC 7 nm

TSMC 7 nm

Architecture

?

Ampere

CDNA 1

Kernels

?

6 192 st.

7 680 st.

Memory amount

32 GB HBM2

40/ 80 GB HBM2

32 GB HBM2

Memory bandwidth

1.2 TB / s

1.6 TB / s

1.2 TB / s

FP32

37 TFLOPS

19,5 TFLOPS
156 TFLOPS (Tensor)
312 TFLOPS (Sparsity)

23,1 TFLOPS
46,1 TFLOPS (Matrix)

TDP

300 W

400 W

300 W

Read This Now:   Gas supplies from Russia to France have been significantly reduced, the minister said

The graphics circuit has 24 billion transistors, which is more than halved compared to the Nvidia and AMD alternatives. Despite this, performance in the FP32 context with simple precision is claimed to reach 37 TFLOPS – a real step up from AMD’s 23.1 and Nvidia’s 19.5 TFLOPS. The latter, on the other hand, has more methods for FP32 calculations in the arsenal, where significantly higher capacity is listed.

The memory amount of the “Big Island” card is 32 GB HBM2, which is identical to the Instinct MI100 but lower than the 80 GB that the top model of the A100 is equipped with. In addition, few details about the new Chinese card are known, as neither architecture nor number of cores are listed. It is known, however, that the TDP value lands at 300 watts; once again a match with AMD and lower than Nvidia, which instead has a TDP value of 400 watts.

In connection with the unveiling of the card’s performance, the company shows images that show that the circuit takes place in both PCI Express-connected card format, as well as a form factor intended for mounting directly in servers. It remains to be seen when the strained manufacturing will allow the “Big Islands” to roll out into the server world on a broad front, and whether these will be competitors to the solutions from AMD, Intel and Nvidia outside China as well.

Source: Iluvitar (Chinese) via Wccftech

Read more about powerful hardware:


Notice: ob_end_flush(): failed to send buffer of zlib output compression (1) in /home/gamefeve/bitcoinminershashrate.com/wp-includes/functions.php on line 5420

Notice: ob_end_flush(): failed to send buffer of zlib output compression (1) in /home/gamefeve/bitcoinminershashrate.com/wp-includes/functions.php on line 5420