NVIDIA has designed a tiny accelerator for AI

A2 targets segments where the more muscular pieces in the product family are difficult to reach.

It’s been a year and a half since NVIDIA announced the A100 accelerator, which later became available with a PCI Express interface, but this spring, the more affordable A30 and A10 options also arrived.

It can be seen that the company covers the market relatively well, but in some areas even the A10 can be considered too fast, as the deduction phase of machine learning does not necessarily require powerful hardware, which is why A2, the smallest member of the entire product family, arrives. parameters are detailed in the table below:

NVIDIA’s New A-Series Accelerators (Ampere)
Type A2
GPU code name GA107
Alapórajel GPU 1440 MHz
GPU Boost Clock
1770 MHz
Memory size 16 GB
Memory bus 128 bit
Memory type GDDR6
Effective memory clock 12,5 GHz
Memory bandwidth 200 GB/s
TDP consumption 40-60 watt
PCI Express interface x8-os PCI Express 4.0
Computing power with simple accuracy
4,5 TFLOPS
Tensor performance with simple accuracy
9 TFLOPS
Tensor performance at halved accuracy
18 TFLOPS
8-bit fixed-point tensor performance
36 TOPS
4-bit fixed-point tensor performance
72 TOPS

A special feature of the novelty detailed in the table is that it is available in a half-height design, and although the default TDP consumption is 60 watts, this can be reduced to 40 watts individually, but this obviously results in a reduction in power. The development has one encoder and two decoder engines, one of which also supports AV1.

NVIDIA is primarily looking for a fresh A2 model for perimeter network servers.


Source: Hírek és cikkek – PROHARDVER! by prohardver.hu.

*The article has been translated based on the content of Hírek és cikkek – PROHARDVER! by prohardver.hu. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!