← All hardware
256 GB of HBM3e for memory-hungry inference at lower cost than NVIDIA.
Pros
- More memory than H200/B200 for big models
- Competitive total cost of ownership
- Open-source ROCm software
Cons
- Superseded by MI355X (CDNA 4)
- ROCm ecosystem maturity gap vs CUDA
- No native FP4 (unlike MI355X / Blackwell)
✓ Where it shines / best for
- Running and serving large language models that benefit from very high memory capacity per GPU
- Memory-bound AI inference and training workloads
- Hyperscalers and neoclouds seeking an alternative to NVIDIA H200/H100
✕ Not the best fit for
- Consumers or small teams without data-center infrastructure
- Workloads locked into CUDA-only software with no ROCm path
- Edge or low-power on-device deployment
Features
- ✓ AI inference
- ✓ Data-center scale
- ✓ LLM
- ✓ HBM3E
- ✓ AI Training
- ✓ High Memory
- ✓ GPU Accelerator
- ✓ Large Model
- ✓ ROCm
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| List/MSRP | Not publicly listed | one-time | Enterprise data-center accelerator sold through OEM/ODM server partners and cloud providers, not retail. Street estimates run roughly $15,000-$20,000 per GPU; typically purchased in 8-GPU platforms. Contact AMD or server OEMs (Dell, Supermicro, etc.) for quotes. |
| Cloud access | Usage-based | hourly | Available via cloud/neocloud providers (e.g., Vultr, TensorWave, Oracle) on a per-GPU-hour basis; rates vary by provider. |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| clock | up to 2.10 GHz |
| power | ~1,000W peak board power |
| memory | 256 GB HBM3e, 6 TB/s |
| architecture | CDNA 3 (dual-chiplet) |
| stream_processors | 19,456 (304 CUs) |
Sponsored
A full review is being generated for this product and will appear here shortly.