← All hardware
CDNA-Next flagship: 432GB HBM4 and 40 PFLOPS FP4
Pros
- Highest memory capacity in class
- More than 2x MI355X bandwidth
- Strong FP4 inference numbers
- Rack-scale ambitions vs NVIDIA
Cons
- H2 2026 availability
- Premium flagship pricing
- ROCm maturity gap remains
- Requires advanced cooling
✓ Where it shines / best for
- Hyperscalers planning 2026 frontier-model training buildouts
- Rack-scale AI deployments needing massive HBM4 capacity
- Operators evaluating non-NVIDIA rack-scale alternatives
✕ Not the best fit for
- Anyone needing hardware available today (pre-release)
- Small teams or on-device use
- CUDA-locked workloads without ROCm plans
Features
- ✓ AI inference
- ✓ Data-center scale
- ✓ LLM
- ✓ AI Training
- ✓ Rack Scale
- ✓ High Memory
- ✓ Next Gen
- ✓ Preview
- ✓ HBM4
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| List/MSRP | Not announced | one-time | Next-generation MI400 series (incl. MI455X) announced for 2026 as part of the 'Helios' rack-scale platform. No pricing disclosed; pre-release. Sold through OEM/cloud channels, not retail. |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| use | Data center training and inference |
| power | ~1,400W+ |
| memory | 432GB HBM4, 19.6 TB/s |
| performance | 40 PFLOPS FP4 / 20 PFLOPS FP8 |
| architecture | CDNA-Next (TSMC 2nm compute + 3nm I/O) |
Sponsored
A full review is being generated for this product and will appear here shortly.