← All hardware
Reconfigurable Dataflow Unit with three-tier memory for trillion-param AI
Pros
- Excellent for large MoE inference
- Fast token generation
- Turnkey full-stack offering
- Intel-backed funding in 2026
Cons
- Proprietary dataflow stack
- Appliance/cloud model, not loose chips
- Niche vs GPU mainstream
- Smaller ecosystem
✓ Where it shines / best for
- Enterprise LLM inference
- MoE model serving
- Private AI appliances
- Fast inference seekers
✕ Not the best fit for
- CUDA-portable workflows
- DIY chip buyers
- Edge
- Tiny deployments
Features
- ✓ Reconfigurable Dataflow Unit (RDU)
- ✓ Three-tier memory (SRAM/HBM/DRAM)
- ✓ 640 BF16 TFLOPs
- ✓ 520MB on-chip SRAM
- ✓ Composition-of-Experts serving
- ✓ Full-stack appliance model
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| SambaNova Suite / DataScale | Enterprise (contact) | one-time | Appliance or cloud; pricing on request |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| use | Large-model inference and training |
| power | Data center scale |
| memory | 520MB SRAM + HBM + DRAM |
| performance | 640 BF16 TFLOPs |
| architecture | RDU SN40L (TSMC 5nm, 2.5D) |
Sponsored
A full review is being generated for this product and will appear here shortly.