AMD launches Radeon AI Pro R9700 to challenge Nvidia's AI market dominance

AMD unveiled the Radeon AI Pro R9700, a workstation graphics card based on the RDNA 4 architecture, targeting professional users and replacing the Radeon Pro W7800. The new card utilizes the Navi 48 silicon, a smaller but denser design compared to its predecessor, and boasts improved AI performance, achieving 96 TFLOPS FP16 and 1,531 TOPS of AI performance with 32GB of GDDR6 memory. Launching in July, the R9700 is positioned to compete with Nvidia's RTX 5080 in AI workloads, offering potentially up to 5x higher performance in certain large AI models, with pricing yet to be announced.

Analysis

AMD has unveiled the Radeon AI Pro R9700, a PCIe 5.0 graphics card targeting professional and workstation users, built on the RDNA 4 architecture and scheduled to replace the RDNA 3-based Radeon Pro W7800. The R9700 leverages the new Navi 48 silicon, a monolithic die produced using TSMC's N4P (4nm) process, featuring a 357 mm² die size and 53.9 billion transistors. This configuration results in an approximately 33% smaller die than the previous generation Navi 31 (529 mm², 57.7 billion transistors, MCM design) while retaining 93% of the transistors, leading to a transistor density increase of up to 38%, from 109.1 million/mm² to 151 million/mm². Although the Navi 48 incorporates 64 RDNA 4 Compute Units (4,096 Streaming Processors) and 128 second-generation AI accelerators—a 9% reduction in both compared to the Radeon Pro W7800—AMD emphasizes significant performance uplifts due to the advanced architecture. The R9700 achieves 96 TFLOPS of FP16 performance, a 6% improvement over the W7800, and 1,531 TOPS in AI tasks. AMD claims this new card offers twice the performance of the W7800 in DeepSeek R1 Distill Llama 8B and, notably, up to five times higher performance than Nvidia's GeForce RTX 5080 in specific large AI models. The card will feature 32GB of GDDR6 memory, support multi-GPU setups for up to 128GB total VRAM, and operate with a Total Board Power of 300W, which is 15% higher than the 32GB W7800. The Radeon AI Pro R9700 is set for a July 2025 launch, with pricing details yet to be announced; its predecessor, the W7800, launched at $2,499. AMD anticipates a robust supply chain supported by its partners.

Key Decisions for Investors

• Investors should monitor the Radeon AI Pro R9700's market reception upon its July 2025 launch, with particular attention to its official pricing and independent third-party performance benchmarks, especially in AI workloads against comparable Nvidia offerings.
• It is advisable to assess AMD's potential to gain market share in the lucrative professional GPU segment, contingent on the R9700 delivering on its claimed performance benefits, including the significant generational uplift and purported advantage over competitors like the RTX 5080 in AI applications.
• Consider the strategic implications of AMD's continued innovation in GPU architecture, such as the shift to a denser monolithic design with Navi 48, and its impact on AMD's competitive positioning and financial performance in the data center and AI markets.