AI COMPUTING · HARDWARE · SOFTWARE · SYSTEMS
Navigate the
AI accelerator landscape.
A curated, evidence-led map of the chips, software stacks and benchmarks shaping modern AI infrastructure — from silicon and IP to kernels, runtimes and serving systems.
Landscape snapshot · November 2026 · 48 company logos · Open full-size ↗
EXPLORE THE ECOSYSTEM
Five views. One connected landscape.
Move from a vendor or chip to its software support, then compare the benchmarks that matter.
Chip & IP directory
Explore GPUs, NPUs, custom AI silicon, edge accelerators and semiconductor IP vendors.
Browse the directory 02 / SOFTWARESoftware ecosystem
Map compilers, IRs, kernel DSLs, operator libraries, runtimes, communication and serving engines.
Explore software layers 03 / COMPATIBILITYCompatibility explorer
Search hardware-to-software integration records, version pins, support status and primary evidence.
Query compatibility 04 / WORKLOADSWorkload explorer
Connect model families and workload phases to candidate hardware/software paths, with evidence maturity made explicit.
Explore execution paths 05 / MODEL SUPPORTModel support explorer
Trace exact model checkpoints to accelerator products, serving stacks, precision modes and source-backed deployment recipes.
Explore model support 06 / BENCHMARKSBenchmark landscape
Understand model, serving, kernel, memory, interconnect, power and cost benchmarks.
Explore benchmarksEDITORIAL PRINCIPLES
Evidence before rankings.
Compatibility and performance are workload-, device- and version-specific. We distinguish upstream support, vendor plugins, experimental branches and vendor-reported results rather than collapsing them into a single score.
See how support is classifiedPROJECT & CONTRIBUTIONS
A living technical reference.
Primary sources are preferred. Version snapshots are dated, compatibility claims carry evidence links, and unresolved documentation conflicts remain visible for review.
View the project on GitHub