AMD Radeon PRO W7900 Dual Slot Unleashes Multi-GPU Power for AI

During the Computex 2024 keynote, AMD announced the launch of the Radeon PRO W7900 Dual Slot, designed to meet the computational demands of multi-GPU workstations and generative AI. This card promises to deliver superior graphical performance at a more cost-effective rate.

The Radeon PRO W7900 Dual Slot features a dual-slot design with a length of 280mm, integrating 192 AI acceleration units capable of achieving up to 123 TFLOPS in FP16 (half-precision floating-point) computation.

Additionally, the Radeon PRO W7900 Dual Slot is equipped with 48GB of GDDR6 ECC memory, providing a memory bandwidth of 864GB/s. It supports output via DisplayPort 2.1, offering a maximum data transfer rate of 774 GBit/s, while the overall power consumption is 295W.

AMD highlights that the Radeon PRO W7900 Dual Slot GPU offers a performance-per-dollar ratio 52% higher than the NVIDIA RTX 6000 Ada. It can seamlessly integrate frameworks and AI models such as ONNX, PyTorch, and TensorFlow through the AMD ROCm 6.1 software, which also supports various databases, compilers, and execution tools.

The AMD ROCm 6.1 software is natively compatible with Ubuntu 22.04.32 Linux and the Windows Subsystem for Linux (WSL2). It supports usage with the Radeon RX 7900 XTX, Radeon RX 7900 XT, Radeon RX 7900 GRE, Radeon PRO W7900 Dual Slot, Radeon PRO W7900, and Radeon PRO W7800.

In terms of AI inference performance, the Radeon PRO W7900 Dual Slot leverages Meta’s Llama 3 70B-Q4 or the ROCm 6.0 paired with vLLM (35GB), maintaining a performance-per-dollar advantage over the NVIDIA RTX 6000 Ada. Its substantial memory capacity facilitates the easy deployment of large natural language models, making it well-suited to current generative AI computational needs.

The Radeon PRO W7900 Dual Slot is scheduled for official release on June 19, with a suggested retail price of $3499.