AMD has officially launched the Radeon AI Pro R9700, a professional-grade GPU designed to accelerate AI workloads and high-performance computing (HPC). Confirmed via a recent driver update, the R9700 targets AI developers, data scientists, and power users.
Built on the RDNA 4 architecture and powered by the Navi 48 chip, the Radeon AI Pro R9700 features:
- 64 Compute Units (4096 Stream Processors)
- 128 AI Accelerators
- 300W Total Board Power (TBP)
- 32GB of GDDR6 VRAM on a 256-bit memory interface
This doubles the memory of its predecessor, the Radeon AI Pro 9070 XT. Performance-wise, the R9700 delivers:
- 47.8 TFLOPs (FP32)
- 191.4 TFLOPs (FP16/BF16)
- 382.7 TFLOPs (FP8)
- 382.7 TOPS (INT8)
- Up to 1531 TOPS (INT4 with structured sparsity)
It also supports Wave Matrix Multiply Accumulate (WMMA) instructions, enhancing AI model throughput and efficiency.
Optimized for Large AI Workloads #
With 32GB of VRAM, the R9700 is optimized for handling complex AI models such as:
- DeepSeek R1 Distill Qwen 32B Q6
- Mistral Small 3.1 24B Instruct 2503 Q8
- Flux 1 Schnel
- SD 3.5 Medium
According to benchmarks, the R9700 completes DeepSeek R1 inference tasks twice as fast as the Radeon Pro W7800, and up to 5x faster than competitors like the RTX 5080 (16GB). AMD highlights that 32B 6-bit models benefit significantly from larger memory, offering higher accuracy than smaller 8B or 1B counterparts—especially in tasks like text-to-image generation.
Expandability and Compatibility #
Supporting PCIe 5.0, the R9700 can scale up to a 4-way multi-GPU setup, enabling up to 128GB of combined memory. This is ideal for running large-scale models like Mistral 123B and DeepSeek R1 70B, which require over 112GB of VRAM.
The card uses a dual-slot blower-style cooler and is expected to ship in July 2025 through partners such as ASUS, ASRock, Gigabyte, PowerColor, Sapphire, XFX, and Yeston.
ROCm Ecosystem and Product Integration #
The Radeon AI Pro R9700 plays a central role in AMD’s broader AI strategy, supported by the ROCm 7 software suite. AMD’s AI acceleration portfolio includes:
- Ryzen AI MAX APUs – For small to mid-sized large language models (LLMs)
- Radeon AI Pro GPUs – For edge inference and multi-GPU LLM workloads
- Instinct Accelerators – For large-scale inference and training in datacenters
CPU Companion: Ryzen 5 9600X3D and Ryzen 9000 PRO #
Alongside the R9700, AMD is introducing the Ryzen 5 9600X3D, a 6-core, 12-thread processor based on the Zen 5 architecture and enhanced with 3D V-Cache. It follows the legacy of the popular 5800X3D and is optimized for gaming and compute-intensive applications, featuring a large 64MB L3 cache to reduce latency.
The Ryzen 9000 PRO series targets business and enterprise deployments, offering advanced security and manageability for professional workstations and servers.
A Broader Vision for AI and Gaming #
AMD is also continuing its long-term investment in gaming, with a new chip roadmap tailored for Xbox consoles. The full specifications for the Ryzen 5 9600X3D and Ryzen 9000 PRO series will be revealed closer to their expected Q3 2025 launch.
With the Radeon AI Pro R9700, AMD delivers a compelling solution for professionals working with large AI models. Paired with the upcoming Ryzen CPUs, AMD’s hardware ecosystem is well-positioned to meet the rising demands of AI inference, gaming, and high-performance computing across both consumer and commercial segments.