British chip designer Graphcore unveils new AI processor more complex than Nvidia’s

British chip designer Graphcore has unveiled a new processor. The Colossus MK2 is intended for performing AI workloads. It has 1472 cores and 8832 parallel threads. The company will start with large-scale deliveries in the fourth quarter of 2020.

Graphcore new AI processor

The Colossus MK2, also referred to as GC200 by the company, is a so-called ipu that is used for performing AI calculations. The chip features 59.4 billion transistors and is made on a 7nm process from TSMC. The GC200 thus has 1472 cores and 8832 parallel threads. The ipu also has 900MB ram in the processor itself. The whole is housed in a which of 823mm².

 Graphcore GC200 chip (left) and the M2000 system

Graphcore also supports additional exchange memory for memory. With this, every ipu can be provided with a maximum of 448GB memory with a maximum theoretical bandwidth of 180TB / s. Graphcore further claims that the GC200 performs about eight times better than its predecessor: the Colossus MK1. The latter was announced in 2018 and was made on a 16nm process.

Graphcore reports that the new ipu comes in an M2000 system. This M2000 contains four Colossus MK2 chips and has a 1U housing. That system must be good for a petaflops of computing power. Users can link up to 64,000 of these ipu’s for a total computing power of up to sixteen exaflops. The company also supplies a rack-scale ‘IPU-POD 64‘, which consists of 16 M2000 machines for a total of 64 Colossus MK2 chips. These Pod systems are placed in a standard 19″ server rack.

Graphcore M2000 system - frontview

Each M2000 system contains a GC4000 gateway chip for communication between the various IPUs. The company calls this ‘ipu-fabric’, which functions as an interconnect between different systems, with a bandwidth of 2.8Tbit / s per M2000 system. According to Graphcore, this can be scaled up linearly to “several petabits per second” using many ipus.

The system uses standard copper or optical osfp connectors for ipu fabric. M2000 systems can be individually connected to each other. Tunneling over Ethernet is used for larger configurations. It is also possible to use standard qsfp interconnects and 100Gbit / s Ethernet switches.

Companies can pre -order the M2000 systems and IPU-POD 64, Graphcore reports. Large-scale deliveries will start in the fourth quarter of 2020. The company does not disclose prices.