MENU

Gyrfalcon launches second AI accelerator

Technology News |
By Peter Clarke


The 2803 demonstrates increased performance and efficiency compared with the 2801S. Whereas the 2801S has been pitched at applications at the edge the 2803 is intended to used on boards of multiple chips and support inference server operations in data centers although it can also address with Gyrfalcon calls the “advanced edge.”

The single chip Lightspeeur 2803 delivers  up to 16.8TOPS at 700mW, with latency as low as 2 milliseconds. This provides a TOPS/watt figure of 24. This compares with the 2801S which has a TOPS/W of 9.3 by offering up to 2.8TOPS while consuming 300mW. The Lightspeeur 2803 supports a PCIe interface, and includes ResNet, MobileNet, ShiftNet and VGG neural networks to support AI model sizes from 4.4 to 17.6MB for inference and training.

The GAIN series 2803 board includes a single 2803 but allows data center operators to add the chip into existing racks, alongside their current hardware, and use software to integrate the new AI accelerator board into their operations.

Gyrfalcon claims the GAIN Series 2803 consumes 10 times less energy than comparable chip.

Samples of the Lightspeeur 2803 chip and the GAIN Series 2803 board are due to be available in 4Q18.

Related links and articles:

www.gyrfalcontech.ai

News articles:

Flex Logix tips move into NN processing

Google’s second TPU processor comes out

Is that Graphcore’s Colossus IPU in package?


Share:

Linked Articles
eeNews Europe
10s