Meta MTIA 200

Meta MTIA 200 is Meta’s in-house accelerator for ranking and recommendation inference.

Vendor: Meta
Architecture: MTIA, 8x8 PE grid, TSMC 5nm
Memory: 128 GB LPDDR5 (off-chip) + 256 MB SRAM
Memory bandwidth (TB/s): 0.205 TB/s
Power (W): 90 W
Form factor: Accelerator board (two per board, up to 72 per rack)
Status: Shipping

Meta detailed MTIA 200 in April 2024 as the next-generation MTIA, built on TSMC 5nm at 90 W. It uses 128 GB of off-chip LPDDR5 plus 256 MB of on-chip SRAM, and Meta’s racks hold up to 72 accelerators. Meta has since announced MTIA 300, 400, 450 and 500, deployed or scheduled for deployment in 2026 and 2027.

Meta lists INT8 GEMM at 354 TFLOPS dense (708 with sparsity) and FP16/BF16 at 177 dense (354 with sparsity). LPDDR5 bandwidth is 204.8 GB/s; on-chip SRAM runs at 2.7 TB/s.

Memory (GB): 128 GB
Precision note (sparse or dense): Meta lists INT8 GEMM 354 TFLOPS dense / 708 with sparsity and FP16/BF16 177 dense / 354 with sparsity. Bandwidth is the LPDDR5 figure (204.8 GB/s); on-chip SRAM runs at 2.7 TB/s.
Launch: Published April 10, 2024
Vendor source: ai.meta.com
Checked: October 2, 2026

Questions

What is Meta MTIA used for?

MTIA 200 is Meta’s in-house accelerator for ranking and recommendation inference.

How much memory does MTIA 200 have?

MTIA 200 uses 128 GB of off-chip LPDDR5 at 204.8 GB/s plus 256 MB of on-chip SRAM at 2.7 TB/s.

How much power does MTIA 200 use?

Meta lists MTIA 200 at 90 W, built on TSMC 5nm.

What comes after MTIA 200?

Meta has since announced MTIA 300, 400, 450 and 500, deployed or scheduled for deployment in 2026 and 2027.

Source

Every figure is taken from the vendor’s own datasheet, product page, technical documentation or announcement, linked on each chip with the date it was last checked. How we pick and check sources: editorial standards.