By Hamza in Quantum Weekly — Sep 1, 2026

The SC #127

Issue #127 of The SC. Weekly supercomputing news. Updates from Hot Chips. AWS celebrates a birthday and Microsoft announces a death.

Hot Chips. The silicon conference as opposed to the calorie laden but rather tasty potato based food. That was the was the main source of supercomputing news last week. We naturally had the AI/ GPU related updates from the usual suspects but we also had insights into CPUs from Intel, Fujitsu and ARM.

Let’s talk Intel first, although it’s possibly the least exciting update its probably also the most relevant one for most of us. Xeon7, code named Diamon Rapids, will give us more cores than originally anticipate. A full 256 of them in fact. But that’s yer lot in terms of threads or logical CPUs too I’m afraid. For some incomprehensible reason, Intel, in their infinite wisdom decided that Hyperthreading (or SMT to give it its unbranded name) isn’t needed anymore. Oops. Slated for a 2027 release but if you want to wait for something with SMT you’ll need to hold out for Coral Rapids.

Sticking with CPUs and supercomputers for a bit longer, Fujitsu provided more details on its 144 core Monaka ARM v9 architecture chips. The previous generations powered Fugaku but with ARM’s increased popularity we might see these more widely used? They’re built on a 2nm process and have 2 SVE2 256bit execution units so could be pretty handy for your SIMD workloads. So long as you port them to ARM.

Talking of ARM, the company rather than the ISA this time, they also released more details on their confusingly named new CPU, AGI. As the name implies ARM seems to be chasing AI money here claiming that the CPU is targeting agentic workloads. Honestly, I still have no idea exactly what demands an agentic workload has on a CPU. Surely this is IO constrained anyway? What makes agentic special here?

Since we’re firmly into AI territory now anyway, Nvidia also provided more details on their NVL72 rack scale GPU solution. I had a funny thought here. As we see more rack scale deployment, are we about to see what happened in the consumer space with devices becoming smaller, more power efficient but also less repairable and upgradeable in the data centre space? DIMM failure? Time for a new rack. Hmm…

And of course, OpenAI gave us details of Jalapeno, their new inference engine. I’ll let you read the link below for the details on that one.

In more fun news, AWS EC2 was 20 years old last week. Happy Birthday! And Azure will give us 248 and 372 vCPU sized machines to celebrate 😁. But Microsoft giveth with one hand and taketh away with the other. HPC Pack, their faithful but aged scheduler is being retired.

Things from us at HMx have been pretty quiet still last week… I haven’t made any progress at all on my piece about compute markets I’m afraid so this week you’ll just have to have more fun guessing where HPC NErD is again.

Have a good week y’all.

All the News in Depth

Cloud and vendor releases for Supercomputing & HPC (and AI too if you change the filters)

Intel Diamond Rapids… no Hyperthreading but you do get 256 cores.

ARM shares details on AGI, its first foray into the CPU market

Fujitsu’s new ARM CPU

A closer look at Jalapeno, OpenAI’s new inference engine

More details on the NVL72

Happy Birthday EC2

HMx Labs Updates

If you didn’t guess (might have been a bit hard) last week HPC NErD was at Neuchatel at a temporary exhibit on the measurement of time. A new HPC Club T shirt if you can guess where he is this time.