Researchers Reverse-Engineer Apple M4 Max GPU Tensor Operations, Find FP8 Matmul Emulated Rather Than Accelerated
A preprint introduces Rigel, an empirical benchmarking framework that reverse-engineers Apple's undocumented Metal 4.1 tensor compute path on the M4 Max GPU, uncovering eleven hardware behaviors the official specification conceals or misrepresents. The key finding is that the Metal 4.1 FP8 (E4M3) matrix multiplication operation is software-emulated rather than hardware-accelerated, achieving only 0.94x the throughput of FP16 despite using half the data. This matters because developers targeting Apple Silicon for AI inference workloads may be misled into expecting performance gains from FP8 precision that do not materialize on current M4-generation hardware.
Researchers present Rigel, a checksum-gated, provenance-tracked microbenchmark harness designed to empirically characterize Apple's Metal Performance Primitives (MPP) matmul2d operation on the M4 Max GPU — a component whose interface is publicly documented but whose hardware behavior is deliberately withheld by Apple. The central finding is that FP8 (E4M3) matrix multiplication in Metal 4.1 is emulated in software rather than executed on dedicated matrix hardware, yielding only 0.94x the throughput of FP16 despite halving operand data volume, making it a memory-footprint optimization rather than a speed advantage. Using three independent signals — throughput ceilings, comparison against simdgroup_matrix, and per-rail power attribution — the study concludes that matmul2d runs entirely on GPU shader cores with no dedicated matrix datapath and no evidence of routing to the Apple Neural Engine. The researchers also determine that the accumulator operates at FP32 or wider precision, and they reconstruct the opaque 8×8 cooperative_tensor fragment layout that Apple does not document. Acting on these findings, a hand-fused GEMM + bias + GELU kernel outperforms the decomposed Metal API path by 6.5–12.9% in cache-resident workloads. The work is limited to a single M4 Max device and predates Apple's neural accelerator generations, so results may not generalize to future or higher-tier Apple Silicon. All code and per-cell benchmark data are released under an MIT license.
What's missing
The study is conducted on a single M4 Max device, leaving open whether findings apply to M4 Pro, M4 Ultra, or future Apple Silicon variants. It is also unclear whether Apple intends FP8 acceleration to arrive in a future Metal or hardware revision, or whether the emulation behavior is a deliberate, permanent design choice. The paper has not yet undergone peer review.
What different sources said
- arXiv cs.CLCenter
Rigel: Reverse-Engineering the Metal 4.1 Tensor Compute Path on the Apple M4 Max GPU
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.