HPCwire catches up with SiPearl’s Craig Prunty at SC21. They discuss SiPearl’s role in developing foundational European exascale technology, recently announced partnerships with Intel and Graphcore and what’s next for the French company.

Transcript (lightly edited):

Tiffany Trader: Hello and welcome. I’m Tiffany Trader, managing editor of HPCwire in here with me is Craig Prunty from SiPearl. And we are at the SC21 show in St. Louis. Hey, Craig.

Craig Prunty: Hey, Tiffany.

Trader: How’s it going?

Prunty: Very well. It’s an interesting show. I think Addison [Snell] had it right on his podcast when he said the hallway conversations are why you come to see. And that’s certainly been true for us.

Trader: Great. Can you tell us about your role with SiPearl?

Prunty: Sure, I joined SiPearl over about a year and a half ago. Prior to that I was with Marvell Cavium, doing Arm based server processors. SiPearl is spun out of the European Processor Initiative, and we’re creating the European HPC processor for European exascale opportunities.

Trader: And I know that you have a strong relationship with EPI, the European Processor Initiative, as you just mentioned, do you have a broader broader market vision beyond that?

Prunty: Absolutely. So I mean, our goal is to first develop processors for HPC in Europe, and then beyond that scale, across the world. And then beyond that, we’ll look at other markets beyond HPC. In fact, HPC being written large, you look at today, it’s scientific computing. But with the birth of AI, it’s scaling beyond that to a lot of other markets.

Trader: What’s the status of the product development at SiPearl?

Prunty:  So we haven’t announced anything yet. We’re still in stealth mode. We are building Arm-based processors, multicore processors with HBM and SVE. So again, very much targeted at HPC. Looking for a very good performance per watt and power density. We don’t have a timeframe that’s a public timeframe yet, but we are targeting our processors to meet the European HPC exascale timeframe.

Trader: Right. I mean, you have shared some details of the roadmap. I don’t remember all the names, but there’s a Rhea processor. There’s some Arm in there. Some RISC-V in there, can you just share a few more details?

Prunty: Where is our first generation processor, again, based on the Arm V1 core, and then using SVE256 based acceleration. Arm is a very important partner for us. Again, if you’re building a server processor, you need something that has an ecosystem today. So we’re basing our server processors on the Arm V1 core.

Trader: Is it a second generation or third generation that you’re targeting for the exascale machines?

Prunty: No, it’s our first generation processor. Rhea is our first generation processor.

Trader: All right, and that’s the one that you you are also targeting to for the exascale.

Prunty: Correct.

Trader: And could you share a little bit about what the target is on those exascale systems? As far as generally speaking, is there one, is there two, what is the timeline?

Prunty: I don’t think anything has been announced in the EU. The tenders are ongoing to select the sites. There are certainly candidates that people have in mind for what potential sites could be. The RFP to select the site should come out shortly. I don’t have a specific time on that. And then once the sites are selected, then they’ll be looking at the the RFPs for the actual systems themselves.

Trader: And I believe either EPI or EuroHPC has said that there will be at least one exascale system.

Prunty: That’s been quoted in the press. I mean, I can’t speak for the EuroHPC…

Trader: I am probably one of the people quoting that in the press.

Prunty: Absolutely. That is the focus of the EuroHPC JU is to create technology in Europe, for European exascale.

Trader: And so you have some other partners that you’re collaborating with. Intel announced a few weeks ago that you’re working together with the Ponte Vecchio GPU, that’s forthcoming. How does that fit into your plans?

Prunty: So if you look at exascale systems, especially in Europe, most systems are modular. So they look at having a CPU module, and then an accelerated module and multiple accelerated modules, and GPU being the best candidate for acceleration today. So the announcement with Intel is around partnering up on their, our Rhea processor with their Ponte Vecchio GPU, and also porting OneAPI so that you have one software stack covering both the accelerator and the general purpose processor.

Trader: And conceivably, is the accelerator something that could work with either an Arm-based CPU or a RISC-V-based CPU?

Prunty: It could work with any processor. Yes, absolutely. For a server processor, you need an ecosystem in place. Today if you look at ecosystems, its x86, Arm and, and Power. So we’re an ARM based server processor. Arm is, let’s say more democratized in terms of the number of people who can potentially use it. That’s our focus. RISC-V is an interesting architecture but today doesn’t have an ecosystem for server processing. People are looking at it for acceleration, for embedded processing. We’ll see as and when the ecosystem develops, which we expected to do, we’ll certainly be looking at that as a technology as well. But today, we’re very much focused on Arm.

Trader: You have another partnership in the accelerator space as well. A few months ago, you announced a partnership with Graphcore?

Prunty: Exactly, at ISC. And again, it’s very important for us, we don’t make selections on acceleration, our end customers do. So partnering with multiple kinds of accelerators is very important for us going forward, especially for those accelerator modules. Right. Graphcore has a very strong AI story. And that’s becoming more and more important in all datacenter, but certainly in HPC datacenter.

Trader: Any other next steps you want to share? Things you are looking forward to and things you’re excited about?

Prunty: Well, I’d love to share… I’m looking forward to sharing with you more information on an on our Rhea product coming to market, again, we’re being quite quiet about that. And looking forward to other partnerships that we’ll be announcing going forward. But nothing I can explicitly call out today.

Trader: Well, we’ll check back with you in a few months or sooner. And we’re hear at SC, how’s your SC 21 experience going? Any highlights?

Prunty: Yeah, I had dinner with Jack Dongarra last night. That was really fun. I hadn’t met him face to face. So that was nice. And he’s the godfather of HPC. Frankly, it’s been the hallway conversations. I was working in Silicon Valley. I joined SiPearl just before Covid. This is my first trip out of Europe and back to the U.S. since then. You know, we were a little worried about the attendance at SC. But the number of hallway conversations I had with people I really want to talk to it’s been a fantastic show.

Trader: Fantastic, Craig. Well, thank you so much for joining us. And thanks all of you for watching. Take care.

To view more HPCwire exclusive SC21 video interviews, including our interview with Jack Dongarra, go here.