neoverse

Helium DPU, ET3000, ET3600, ET2500, ET1600

Educational guide for developers to understand Arm Neoverse processors and Armv9 architecture for server and cloud use cases

Neoverse-based Labs for 2026 IEEE Workshop

int8 vector search for Arm that reaches SDOT and SMMLA. 9.1x faster than the fastest working FAISS int8 mode at slightly better recall, with every number reproducible in CI on free Arm runners.

Arm64 LLM inference autotuner: measures quantization, threads and concurrency on your exact Arm CPU with Arm Performix evidence, then hands you the launch command.

Slash enterprise AI costs by routing simple LLM queries to right-sized local models running on Arm Neoverse CPUs.

One command to prove how much faster your LLM runs on Arm with KleidiAI. Arm AI Optimization Challenge 2026 — Track 2 (Cloud AI).

ARMONIC is an Autonomous agentic performance optimizer for Arm64 cloud AI. Profiles with Arm Performix (APX), feeds telemetry to an LLM, auto-generates validated patches, and commits optimizations to git — zero human code changes around 98% on moderate work load and upto 93% on heavy workloads speedup on AWS Graviton.

2.12x faster IQ4_XS prefill on Arm Neoverse: the missing smmla repack kernel for llama.cpp. IQ4_XS is the smallest standard 4-bit GGUF and had an accelerated matmul path on Intel AMX but none on Arm. Measured stock-vs-patched on free Neoverse N2 CI.

Prove whether an Arm binary can actually use the chip's matrix hardware. Static ISA verifier + CI gate for AArch64 AI runtimes (SME/SME2, i8mm, dotprod).

Arm Neoverse SVE dispatch audit: KleidiAI SVE kernels are dead code on 128-bit Neoverse (svcntb()==16 vs gate==32). Measured evidence + one-command reproduce.

First public reproducible KV-cache quantization tradeoff study on AWS Graviton4 CPU with PMU-level mechanism analysis. Arm Create: AI Optimization Challenge, Cloud AI track.