By Hamza in Quantum Weekly — Jun 1, 2026

Quantum #114

Issue #114 of the weekly HPC newsletter by HMx Labs. More silicon news this week with a benchmarks for Nividia’s Vera CPU, ByteDance planning their own CPU and yet another AI inference option.

Nothing for decades and suddenly tens of them come along at once. I think this must be the third week in a row that the interesting HPC/ supercomputing news is dominated by developments in silicon.

AWS has had Graviton for about 8 years, Microsoft Azure and Google Cloud both also decided to offer their own ARM based CPUs and even Meta has developed its own designs. Now ByteDance becomes the latest big tech company to plan to tape out its own custom chips. Though they are currently hedging their bets between it being an ARM or RISC-V architecture.

Meanwhile Phoronix has managed to get their hands on Nvidia’s Vera CPU. The selection of benchmarks results they’ve chosen is, shall we say, interesting. Not sure we’re likely to get our hands on any Vera kit any time soon to benchmark ourselves but we do currently have our mitts on some Intel CPUs and are considering if we include some of the Phoronix benchmarks in our analysis going forward. Specifically, the HPC related benchmarks and probably quanlib too. Maybe we should include SPEC too? All available for cloud or physical selection decisions via FLOPx.

A new startup, FuriosaAI, seems to have an alternative approach to accelerating inference, choosing to focus on high bandwidth data movement and tensor operations rather than a multitude of small cores. Sounds promising but modern GPUs sound awfully similar to that these days, no?

To address my concern from the previous two weeks about the difficultly in targeting multiple accelerator architectures it seems there are some people out there trying to solve this. I mean I kind hoped and assumed this would be case but hadn’t seen anything till last week in the form of ZML for one. I guess there may also be others, so if you know of any please do point me to them. I sincerely hope someone cracks this, not so much from a technical perspective, but more from a general adoption point of view.

TSMC seems to think efficiency might be the most critical metric its customers are now asking for. I’m actually surprised it took this long, given that the most bottlenecked resource seems to be the energy to power enough of these things.

Lastly, we’re still hiring. We have open roles for both a HPC software engineer and marketing. Oh, and expect this to be the last issue of this newsletter branded as Quantum. A new look (but with the same content) coming next week!