Tenstorrent Tensix: small-SRAM, modest-bandwidth, open-source architecture

Tenstorrent Wormhole (n150) has 108 MB SRAM, 12 GB GDDR6 at 288 GB/s, 74 TFLOPS FP16. Blackhole (p150) has 180 MB SRAM, 32 GB GDDR6 at 512 GB/s, 664 TFLOPS BLOCKFP8. The architecture uses Tensix cores with dedicated RISC-V control cores and a network-on-chip. SRAM acts as a L2-like feed tier. The design bets on very low power (160-300 W) and open-source software rather than winning the bandwidth race. For inference, it is decisively bandwidth-bound at batch 1.

Tenstorrent's key differentiator is the fully open-source software stack. TT-Forge is an MLIR-based compiler that works with PyTorch, JAX, and ONNX. TT-NN provides pre-optimized operations. TT-Metalium is a low-level C++ SDK. The hardware is modest — 288-512 GB/s bandwidth is far below GPU HBM — but the power efficiency (160-300 W) and open-source approach are the value propositions. For inference, the bandwidth bottleneck is the primary limitation. The block-float formats (BLOCKFP2/4/8) have native hardware paths. HARDWARE-PARAMETER ROWS ADDED 2026-10-05 (schema authority pass, agent w5-schema). `affected_by_hardware` names PRODUCTS (accelerator record ids); these rows name the hardware QUANTITIES that move this class's cost, which had nowhere to live until flop.hardware_parameters was added in this commit. Every effect below is lifted verbatim from this record's own existing prose and no figure here is new - `magnitude` reproduces numbers the record already quoted. `basis` is 'unverified' where the effect was asserted with nothing concrete behind it, rather than being dressed up. The rows were pulled out of free-text notes; they are not a change to what this record claims.