← All hardware
Meta's RISC-V custom accelerators for ranking and GenAI
Pros
- Optimized for Meta's exact workloads
- Big efficiency gains
- Reduces NVIDIA spend
- RISC-V openness
Cons
- Not available to anyone outside Meta
- No public benchmarks/pricing
- Narrow workload focus
- Internal tooling only
✓ Where it shines / best for
- Meta's internal recommendation, ranking, and ads inference at massive scale
- Reducing Meta's dependence on third-party GPUs for target workloads
- Cost- and power-efficient internal inference (and emerging training)
✕ Not the best fit for
- External customers (not commercially available)
- General-purpose third-party AI training/inference
- Anyone needing to buy or rent the hardware
Features
- ✓ Data-center scale
- ✓ Inference
- ✓ Training
- ✓ Custom ASIC
- ✓ In House Silicon
- ✓ Pytorch
- ✓ Ranking
- ✓ Recommendation
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| Internal use only | Not for sale | n/a | Meta Training and Inference Accelerator is custom silicon used internally in Meta data centers; not sold or rented externally and has no public price. |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| use | Recommendation and generative AI |
| power | Data center scale |
| memory | On-package + HBM |
| performance | Up to 25x prior-gen (claimed) |
| architecture | MTIA RISC-V custom |
Sponsored
A full review is being generated for this product and will appear here shortly.