Brother
Professional
- Messages
- 2,590
- Reaction score
- 483
- Points
- 83
256 cores are already available, and another 1600 are on the way — and all on one chip.
Scientists from the Institute of Computer Engineering of the Chinese Academy of Sciences recently presented the multi-chip complex Zhejiang Big Chip with 256 cores. The researchers plan to scale the design to 1600-core chips that use an entire crystal as a single computing device.
With each new generation of chips, it becomes more difficult to increase the density of transistors, so manufacturers are looking for other ways to improve processor performance. This includes architecture innovations, larger crystal sizes, multi-chip designs, and even full-chip chips. The latter has so far been achieved only by the American company Cerebras.
Chinese developers have already built a multi-chip design with 256 cores and are exploring the possibilities of creating a chip for the entire crystal. The Zhejiang Big Chip multi-chiplet design consists of 16 chiplets containing 16 RISC-V cores each, connected to each other in a symmetric multiprocessor configuration via an on-chip network, allowing chiplets to share memory.
Each such chiplet has several connection interfaces for communicating with neighboring chiplets via a 2.5 D interposer. The researchers claim that the design is potentially scalable to 100 chiplets or up to 1,600 cores.
Zhejiang chiplets are manufactured using a 22-nanometer manufacturing process, presumably by Semiconductor Manufacturing International Corp. (SMIC). How much power a 1600-core build will consume remains to be seen, but in theory it would significantly optimize power consumption and performance.
Multi-chiplet designs can be used to create processors for Exascale supercomputers, which AMD and Intel are already doing. Researchers suggest using a multi-level memory hierarchy for such assemblies, which may introduce difficulties in programming such devices.
The Big Chip design can also use technologies such as optoelectronic computing, near - Memory Computing, and a three-dimensional stack cache. However, the article does not provide specific details of the implementation of these technologies or solutions to the challenges that they may pose in the design and creation of such complex systems.
Despite all the difficulties on the way to implementing such projects, the desire to create more and more productive computers will eventually bear fruit in the form of new technological advances that will definitely change our world in the future.
Scientists from the Institute of Computer Engineering of the Chinese Academy of Sciences recently presented the multi-chip complex Zhejiang Big Chip with 256 cores. The researchers plan to scale the design to 1600-core chips that use an entire crystal as a single computing device.
With each new generation of chips, it becomes more difficult to increase the density of transistors, so manufacturers are looking for other ways to improve processor performance. This includes architecture innovations, larger crystal sizes, multi-chip designs, and even full-chip chips. The latter has so far been achieved only by the American company Cerebras.
Chinese developers have already built a multi-chip design with 256 cores and are exploring the possibilities of creating a chip for the entire crystal. The Zhejiang Big Chip multi-chiplet design consists of 16 chiplets containing 16 RISC-V cores each, connected to each other in a symmetric multiprocessor configuration via an on-chip network, allowing chiplets to share memory.
Each such chiplet has several connection interfaces for communicating with neighboring chiplets via a 2.5 D interposer. The researchers claim that the design is potentially scalable to 100 chiplets or up to 1,600 cores.
Zhejiang chiplets are manufactured using a 22-nanometer manufacturing process, presumably by Semiconductor Manufacturing International Corp. (SMIC). How much power a 1600-core build will consume remains to be seen, but in theory it would significantly optimize power consumption and performance.
Multi-chiplet designs can be used to create processors for Exascale supercomputers, which AMD and Intel are already doing. Researchers suggest using a multi-level memory hierarchy for such assemblies, which may introduce difficulties in programming such devices.
The Big Chip design can also use technologies such as optoelectronic computing, near - Memory Computing, and a three-dimensional stack cache. However, the article does not provide specific details of the implementation of these technologies or solutions to the challenges that they may pose in the design and creation of such complex systems.
Despite all the difficulties on the way to implementing such projects, the desire to create more and more productive computers will eventually bear fruit in the form of new technological advances that will definitely change our world in the future.
