Skip to main content

Tenstorrent Grayskull / Wormhole

Product Overview

Tenstorrent, led by legendary architect Jim Keller, develops AI accelerators based on the RISC-V architecture. The Tensix core integrates a RISC-V CPU, matrix/vector/data transfer engines. Wormhole connects via standard Ethernet RDMA (no NVLink lock-in), and the software stack — TT-Metalium and TT-BUDA — is fully open-source.

Core Specifications

ItemGrayskull (e150/e300)Wormhole (n150/n300)
ArchitectureTensixTensix
ProcessTSMC 12nmTSMC 12nm
Tensix Cores120144
On-chip SRAM120 MB144 MB
FP32~50 TFLOPS~75 TFLOPS
INT8~300 TOPS~450 TOPS
InterconnectPCIe 4.08× 100GbE RDMA
TDP75W / 300W150W / 300W

Manufacturer Info

ItemDetail
ManufacturerTenstorrent Inc.
Websitehttps://tenstorrent.com
Product Pagehttps://tenstorrent.com/hardware/
TT-Metalium (driver)https://github.com/tenstorrent/tt-metal
TT-BUDAhttps://github.com/tenstorrent/tt-buda
Documentationhttps://docs.tenstorrent.com

Key Features

  • Tensix Architecture: RISC-V + Matrix + Vector + Data Transfer
  • No HBM: fully on-chip SRAM (suited for specific models)
  • Ethernet Interconnect: avoids NVLink vendor lock-in
  • Fully Open-Source Software Stack
  • Next-gen Blackhole: 2025 release, 2× performance

Use Cases

  • Open-source AI developers
  • Medium-scale AI training
  • Academic research
  • RISC-V ecosystem