AMD Radeon AI PRO R9700
A 32 GB professional workstation GPU for local AI inference and development, used singly or in pairs.
Price: On request
Availability: Confirmed at quotation
Cooling: Active, workstation
Manufacturer: AMD
Specifications
Manufacturer figures, checked against the sources below.
| Specification | Manufacturer figure |
|---|---|
| Architecture | AMD RDNA 4 |
| GPU memory | 32 GB GDDR6 (ECC supported on Linux only) |
| Memory interface | 256-bit |
| Memory bandwidth | 640 GB/s (peak) |
| Compute units | 64 (4,096 stream processors) |
| AI accelerators | 128 |
| Host interface | PCIe 5.0 x16 |
| Board power | 300 W total board power (750 W minimum PSU recommended) |
| Power connector | 12V-2x6 |
| Cooling | Active (cooler design varies by board partner) |
| Form factor | Double slot, desktop |
| Display outputs | DisplayPort, up to 4 (layout varies by board partner) |
| Peak AI performance (manufacturer) | 1,531 TOPS (INT4 matrix with structured sparsity) |
| Intended use (AMD) | Workstations; not designed or validated for datacenter use |
Sources Last verified 27 Oct 2026
Is it right for your workload?
A good fit for
- Local models that fit in 32 GB, or 64 GB with two cards.
- Developers and small teams starting with private AI on a budget.
- Workloads whose software supports ROCm.
- Local LLM
- AI agents
- AI video & 3D
Consider something else if
- Tools that only support CUDA: check your software first.
- Models that need more than 64 GB without heavy quantization.
- Rack or data-centre deployment: AMD states it is not designed or validated for datacenter use.
Cooling, power and installation
- Actively cooled card for workstations and towers. Board partners' cooler designs and dimensions differ; we confirm the exact card's fit.
- ECC memory is supported on Linux only; we enable it where your results depend on it.
- ROCm support is confirmed for your exact frameworks and versions before quoting.
- Two-card builds need slot spacing and power planned together.
What is included
- Supplied as part of a configured system, or as a component with compatibility checked against your platform.
- ROCm and driver stack installed and tested when we build the system.
Last reviewed 27 Oct 2026. Warranty terms are confirmed per quotation.
Reference builds with this GPU
Questions about this GPU
Will my AI software run on it?
PyTorch and several popular runtimes support ROCm on supported AMD GPUs. We check your exact tools and versions before recommending it.
Can I use two together?
Yes, with runtimes that split a model across GPUs. Two cards give 64 GB of combined memory.
When can I get one?
Availability changes; we confirm it in the quotation rather than show a stock level we have not verified.
Plan a system around it
We check compatibility, power and cooling for your workload before anything is quoted.