← All hardware
Open RISC-V AI server: 23 PFLOPS from 32 Blackhole chips
Pros
- Open, Ethernet-native scale-out
- No proprietary interconnect lock-in
- Strong price/performance
- Adopted by Cirrascale, Equinix, ai&
Cons
- Smaller software ecosystem
- p150 card tensor-core firmware downgrade noted
- Newer, less battle-tested
- Self-integration effort
✓ Where it shines / best for
- Open-hardware AI labs
- Ethernet-fabric datacenters
- Inference cost optimization
- RISC-V enthusiasts
✕ Not the best fit for
- CUDA-dependent teams
- Turnkey simplicity seekers
- Tiny deployments
- Edge
Features
- ✓ 32 Blackhole RISC-V chips
- ✓ 23 PFLOPS Block FP8
- ✓ 6.2GB on-chip SRAM @ 2.9 PB/s
- ✓ 1TB DRAM @ 16 TB/s
- ✓ Up to 56x 800G Ethernet ports
- ✓ Open-source software stack
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| Galaxy Blackhole server | ~$110,000 | one-time | Supercluster of 4 from ~$440,000 |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| use | Data center inference/training |
| power | Server rack scale |
| memory | 1TB DRAM + 6.2GB SRAM |
| performance | 23 PFLOPS Block FP8 |
| architecture | Blackhole (RISC-V Tensix) |
Sponsored
A full review is being generated for this product and will appear here shortly.