AMD plans to launch Radeon RPO W7900 designed specifically for AI

Last April, AMD launched two workstation graphics cards based on the RDNA 3 architecture, the Radeon Pro W7000 series, specifically the Radeon PRO W7900 and Radeon PRO W7800, designed with three-slot and dual-slot thickness, respectively. Both feature the Navi 31, employing a small chip design with a 5nm process GCD (Graphics Compute Die) and six 6nm process MCDs (Memory Cache Dies).

Recently, a netizen revealed that AMD plans to introduce a new Radeon PRO W7900 workstation graphics card aimed at artificial intelligence (AI) applications, reduced to a dual-slot thickness. This release is tentatively scheduled for early June and may debut at Computex 2024. The new product retains the original specifications while adopting a turbo cooling design, optimized to reduce its size, allowing users to fit more Radeon PRO W7900 cards in the same space, thus enhancing cluster efficiency.

The original Radeon PRO W7900 boasts 96 CU compute units, a peak single-precision compute performance of 61 TFLOPS, a 384-bit memory interface, 48GB of GDDR6 memory, a memory bandwidth of up to 864GB/s, and 96MB of Infinity Cache, with a total power consumption of 295W, utilizing dual 8-pin power connectors. Additionally, the new multimedia engine supports H.264, HEVC, VP9, and AV1 video codec formats.

Notably, the Radeon PRO W7900 features DisplayPort 2.1 interfaces supporting UHBR20 mode, with a rate of 80Gbps, enabling higher resolutions and refresh rates. This contrasts with the consumer-grade Radeon RX 7000 series graphics cards, which provide DisplayPort 2.1 interfaces limited to UHBR13.5 link rates, achieving 54Gbps.