A British startup promises to quintuple speed without changing a single line of code as it prepares to compete with Nvidia and AMD in the field of generative artificial intelligence.
Akurra from VyperCore is currently undergoing testing.
The startup VyperCore, located in Bristol, aims to revolutionize processor design. The British company is working on a 5nm chip and a card designed to accelerate server-level applications without the need to modify existing software code. By shifting the complexity of memory allocation management from software to hardware, the company claims it intends to eliminate up to 80% of the processor cycles that are typically required for these functions. This approach ensures memory safety at the gate level in the processor, enhances cache utilization, reduces latency in event processing, and decreases overall memory demand.
VyperCore's memory management system promises to double the performance in C and C++ and increase it by five times in Python without requiring changes to the original code. This is particularly useful for managing unoptimized code generated by artificial intelligence.
Russell Haggar, co-founder and CEO of VyperCore, commented on their mission: "We are a processor company, and we promise a fivefold increase in speed without changing a line of code and with hardware-level memory safety. This can be integrated into any CPU, from a toaster to a server."
Last year, VyperCore secured £4 million in funding and is currently seeking additional investments to accelerate the development of its products. The company is in the process of hiring hardware and software engineers, aiming to double its workforce in Bristol and Cambridge. Its first product, a single-core RISC-V processor called Akurra, is already operational on an FPGA. The startup plans to launch a single-core test chip next year, followed by a multicore commercial server chip and an accelerator card.
Haggar emphasizes that VyperCore's technology can be incorporated into various processors, although it will initially focus on accelerating data center applications. "Our goal is a 64-bit quad-core RISC-V processor, probably in 5nm, and server card hardware," he explained. This product is expected to be launched by the end of 2026.