//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
The leaders of VyperCore, a U.K. processor startup that received £4 million in seed funding this month, told EE Times they believe the firm has developed an innovation in memory allocation management that accelerates high-performance, general-purpose compute workloads by a factor of up to 10×—without modifying the original code.
In a record seed funding round for the U.K., investors believe that the startup, co-founded by Ed Nutting and Russell Haggar, has the potential to deliver significant cost and energy savings in the data center without rewriting existing code bases.
VyperCore’s key innovation is its memory-allocation–management technology, which moves away from the processor’s traditional view of its memory as being a single linear space. By defining an object-based view of the memory from within the core of the processor, substantial optimizations in execution of existing code can be achieved. A side effect of this is it also helps block memory-oriented cybersecurity attack vectors, such as memory leaks and buffer overflows.
Garbage collection deep in hardware
Ed Nutting, CTO and co-founder of VyperCore, said in an interview with EE Times, “The innovation is that we’ve taken what’s known as the ‘garbage collector’ algorithm and placed that into a hardware-state machine deep inside the processor, which then takes it out of software. Hence, it removes that element from the processor’s execution time.”
Garbage collection is essentially a feature within programming languages that frees up memory space allocated to objects no longer needed by a program. This helps overcome memory quota issues to free up a program’s memory.
VyperCore is putting that function deep in the hardware, which reduces thousands of processor cycles to just dozens of cycles. This is what gives rise to the claim of up to 10× performance improvement.
“There are some seven layers of software between the hardware and the top layer of software,” Nutting said. “This is incredibly slow and wastes a lot of processing time. In addition, we can run the existing code without changing the source code.”
He added: “Moore’s Law stalled for general-purpose compute several years ago, and processor architectures stopped evolving to meet the needs of modern programming languages, such as Python and C#. VyperCore releases processors to perform optimally for the current generation of languages without compromising their power, flexibility or compatibility.”
The processor is a result of Nutting’s research that started in 2017 with Professor David May and others at the University of Bristol. May is best known for his invention of the transputer, introduced in 1987 by Inmos, and co-founding XMOS in 2005. He also developed the parallel programming language occam. Nutting’s research involved proving an integrated hardware garbage collector that they worked on, implementing it in hardware and proving it was feasible in his undergraduate thesis. VyperCore was formed last year to commercialize that research.
Haggar recognized Nutting’s talent while the latter was doing his research. Haggar’s experience in the deep-tech—Haggar likes to call it “hard tech”—industry as an investor, advisor and mentor has certainly been a factor in helping establish VyperCore to commercialize the research and in raising the record seed funding for the U.K.
“VyperCore’s proprietary memory-management technology can be integrated into all leading processor architectures,” Haggar said. “Its ability to accelerate existing complex workloads in the data center delivers huge cost and energy savings in the data center, without any rewriting of existing code bases. We’re delighted to be working with the U.K.’s leading hard-tech investors to deliver breakthrough cost/performance benchmarks using VyperCore’s paradigm-setting processor technology.”
Not just RISC-V
VyperCore’s first-generation hardware will be based on the RISC-V architecture and will sample with partners in its early-access program in Q3 next year. The only reason the company is starting with RISC-V is because of the easy access.
“The longer-term goal beyond our initial product roadmap is that the technology has the potential to benefit every general-purpose application—whether it’s running on a server or in a data center or even at an endpoint as an embedded processor,” Haggar told EE Times. “So the innovation itself isn’t tied to RISC-V in any way at all. It applies to all the leading processor architectures, including ones we don’t even know about yet.”
Chasing general-purpose compute
VyperCore’s technology is aimed at accelerating general-purpose compute, irrespective of architecture, while using existing code, Haggar said.
“Our proposition to the customer is to give them the ability to run at 10× their current performance and do so with existing workloads, with existing source code and simply recompiling and getting all the benefits of acceleration and security without having to reopen any of their legacy code,” he said. “In our customer conversations so far, a lot of people are really interested in that proposition.
“So our go-to-market is to have our own silicon with our own processor or our own data card deploying into the data center exactly the same way that people deploy FPGA cards or your network accelerator cards,” Haggar added. “But the key goal here is we’re going after general-purpose compute, which is a much wider market segment than any of those specialty areas.
“Our goal, first of all, is to build out the processor as a technology, and then we go to market with trial customers to evaluate what sort of applications we get the best performance speed-ups on, bearing in mind we’re not looking to accelerate the entirety of the stack,” he said. “We’re looking just at applications and complex workloads in the data center where people will be looking to shift that workload off the main server CPU and onto the data accelerator card.”
The company has proof of concept already, and with the funding, it will now assemble a team of hardware and software engineers, both compiler engineers as well as runtime engineers.
Nutting emphasized that the software engineers they are hiring have deep software expertise. “We’re hiring engineers who can modify Python runtime rather than write applications,” he said.
Next year, the company hopes to have the technology in a server-class FPGA for customers to validate, and then ultimately in two to three years, they hope to sell their own data card with their own silicon for hardware acceleration—or, as Nutting put it, shifting workloads from the motherboard to the company’s card.
The investors in this £4 million round are Octopus Ventures, Foresight WAE Technology, Science Creates Ventures, British Growth Fund and Silicon Roundabout Ventures. The funding will be used to open design centers in Cambridge and Bristol, U.K., and to develop its first generation of accelerated compute silicon.
