Parts/Accelerators/Meta MTIA v2
Accelerators · MTIA · 2024
Meta MTIA v2
Meta's second-generation training and inference accelerator: 177 TFLOPS BF16 dense, 128 GB LPDDR5, 90 W.
Why this part
Inference-oriented recommendation part with 256 MB on-chip SRAM. Meta's 2026 MTIA 300 to 500 specs are published only as images.
Systems using this part
No systems in the current dataset record this part. It is carried in the taxonomy so the graph is complete, and so edges can attach to it as coverage grows.