Jonathan Pearce, AheadComputing Co-Founder on AI Chips and CPU

Grace Gong (0:00)

Hi Jonathan, welcome to Venture with Grace.

Jonathan Pearce (0:02)

Thanks for having me, Grace.

Grace Gong (0:03)

I'm so excited. We got to chat at the AI Inference and Mystery Room.

Jonathan Pearce (0:07)

It was great to meet you.

Grace Gong (0:08)

So I want to start with, what are you guys building?

Jonathan Pearce (0:11)

AheadComputing is building really high-performance RISC-V cores. So we make CPUs. A lot of the co-founders and I came from Intel. We were making high-performance CPUs before, and now we're bringing that to the open specification that is RISC-V.

And that means we can give our CPU technology to a lot more people.

Grace Gong (0:32)

I want to start with CPU. Why CPU?

Jonathan Pearce (0:36)

So CPUs, there's a rule that Gene Omdahl coined maybe 60 years ago called Omdahl's Law, that you can't speed something up if you're only speeding up part of it. So if it takes this much time and you speed up the parallel portion this much, the sequential portion is going to be the upper limit of how fast it can be.

And when we started the company, we saw that people were investing billions of dollars in AI accelerators. They're really going after this parallel portion, trying to make that as fast as possible. But nobody was spending the time and the money to make CPUs a lot faster. So we started AheadComputing in 2024 with the idea that we know how to make really fast CPUs and we're going to do that in the RISC-V ecosystem.

Grace Gong (1:25)

Okay. So what is a RISC-V ecosystem?

Jonathan Pearce (1:28)

Oh yeah. So RISC-V is a architecture and a community. So RISC-V International is a organization that brings together a lot of different member companies like Qualcomm, TenStore, Microsoft, Intel, and a lot of different university researchers in the architecture space. So those people have gotten together and specified instruction set architecture, and it's the fifth version of RISC, which is Hennessy and Patterson, Berkeley, from ideas from the 80s, and this is the iteration that's really going to catch on and be commercially successful.

Grace Gong (2:16)

So I wonder, so like who would you identify as your core, like buyers in the RISC core?

Jonathan Pearce (2:22)

Yeah. So we're working with semiconductor teams that don't have a CPU core, and are probably licensing ARM cores right now, or unsatisfied with the power performance of ARM. When we have a design win, it's because we're offering higher performance than ARM and higher performance for a lot.

Grace Gong (2:46)

Okay. So you guys come from the Intel universe, like why don't we talk about, is there something that you guys have in the year past experience that kind of like transferred into like or like inspired into like starting the company?

Jonathan Pearce (3:00)

Yeah. So definitely there's a lot of frustration, the pace of innovation that Intel was going to allow.

We had a real difficulty there doing the kinds of designs we wanted to do. So that's why we went independent. And Intel has devoted many years, many billions of dollars to making extremely high-frequency CPUs.

Their CPUs are some of the highest frequency in the world. And those high-frequency designs require a lot of voltage. So it's not very power efficient. And at AheadComputing, we saw that there was a different way that did not rely on high voltage and instead relied on innovations in the instructions per clock cycle. So we could run at a modest frequency and still have incredibly high performance.

Grace Gong (3:56)

I guess what would you say is one of the whether it's mode or unfair advantage of this?

Jonathan Pearce (4:04)

So our secret sauce is definitely our performance per watt and the pace of innovation. So AheadComputing, all the founders believe very strongly that you have to have an innovation mindset when it comes to CPUs.

For too long, we see that academics have drifted away from doing research in CPUs. It's no longer the sexy thing for the past 12 years to publish papers and get a PhD in CPU microarchitecture or CPU architecture. But there's still a lot of really interesting work to do making CPUs faster and making general compute general purpose compute faster. So, AheadComputing, we're dedicated to our innovation mindset and teaching the team how to embrace that mindset so that we can accelerate the pace and progress of CPUs.

Grace Gong (4:54)

Okay. So, I have a question on some of the team. So, you mentioned some of the core founders are from Intel. But how do you think about what you can use agent for and what are something that you can't use agent for?