AMD Sets Price and Launch Date for Radeon AI Pro R9700 Workstation GPU
AMD has confirmed the particulars of the price and launch of its most powerful RDNA 4 workstation graphics card, the Radeon AI Pro R9700. With its launch scheduled on October 27th, it is expected to retail for $1,299, confirming the earlier rumored price point.
Core Specifications and Memory-Focused
Based on the NAVI 48 die, it uses the same RDNA 4 architecture as the flagship gaming card, the Radeon RX 9070 XT. By doing so, it gains 4096 Stream Processors and 128 ROPs. Nevertheless, the defining factor for this professional card is its memory configuration.
- VRAM Capacity: 32 GB GDDR6
- Memory Bus: 256-bit
This doubles the memory capacity of its gaming counterpart, making the R9700 the cheapest 32GB workstation GPU available and positioning it to handle memory-intensive AI and creative workloads that can exceed the capacity of consumer cards like the 16GB GeForce RTX 5080.
AI Workstation and Multi-GPU Focus
The card is designed with the intent of supporting local AI at the workstation and multi-GPU setups. From AMD's perspective, the target profession with the R9700 is the one running AI models locally on workstation systems, ranging from medium to large, including large language models (LLMs) and diffusion models. The company claims that the GPU is two times faster than the previous-generation W7800 in specific benchmarks, such as DeepSeek R1.
The card has design and software support-oriented for this purpose:
- Software Support: Full support for AMD's ROCm open software platform enables multi-GPU scalability for high-throughput AI tasks.
- Physical Design: It features a compact, dual-slot design with a blower-style cooler, which is highly efficient for airflow in systems with multiple GPUs installed side-by-side.
Availability
The Radeon AI Pro R9700 will not be sold directly by AMD; it is not a reference-model GPU. The card will be available exclusively through board partners. Some partners may sell the card through system integrators only.
