Graphcore is announcing the launch of the C600 PCIe card – a new way to access the power of the IPU for artificial intelligence inference applications.
Based on Graphcore’s industry-leading MK2 IPU, C600 adds FP8 for low and mixed-precision AI, delivering up to 560 TFLOPS (FP8) / 280 TFLOPS (FP16).
C600 is fully supported, from launch, by our Poplar software stack.
Versatile form-factor
We are making the IPU available on a PCIe card in response to customer demand in markets where datacentre configurations, including rack size and power delivery, vary widely.
This highly versatile form-factor enables Graphcore customers to tailor their system setup, including host server / chassis, to their exact requirements.
C600 is available for pre-order now in China and Singapore.
Additional technical information
C600 is a PCIe Gen 4, dual-slot card with a TDP of 185W.
The IPU features 1,472 processing cores, capable of running 8,832 independent program threads in parallel. Each IPU has 900MB of on-chip SRAM memory, located adjacent to the processing cores.
Users can directly connect up to eight cards in a single chassis, bridged with high bandwidth IPU-Links.
