Groq LPU vs Cerebras WSE-3
| Groq LPU | Cerebras WSE-3 | |
|---|---|---|
| Vendor | Groq | Cerebras |
| Architecture | SRAM | Wafer |
| BF16 TFLOPS | 188 | 125,000 ✓ |
| Memory (GB) | 0.23 | 44 ✓ |
| Bandwidth (GB/s) | 80,000 ✓ | 21,000 |
| TDP (W) | 300 ✓ | 15,000 |
| Cloud $/hr | — | — |
| TFLOPS / W | 0.6 | 8.3 ✓ |
FAQ
Groq LPU vs Cerebras WSE-3: which is faster?
Cerebras WSE-3 has higher BF16 throughput (188 vs 125,000 TFLOPS).
Which has more memory?
Cerebras WSE-3 carries more on-package memory (0.23 GB vs 44 GB).
Which is cheaper to rent?
Cloud rates aren't listed for both chips — see each chip's page.
Cost either chip in a fleet
Size a deployment or see the full priced rack BOM these accelerators sit in.
Specs + modeled rates maintained by Mintok's catalog watchers from public sources. Not investment advice. Methodology.