COMPUTEX 2024 : AMD releases the Radeon PRO W7900 Dual Slot

AMD has just announced the Radeon PRO W7900 Dual Slot Workstation Graphics Card at COMPUTEX 2024.

The Radeon PRO W7900 Dual Slot combines AMD's ultra high end graphics processor based on the RDNA3 architecture with a 2 slot form factor. This new workstation desktop card features 48GB of fast GDDR6 ECC memory with a bandwidth of 864GB/s. A total of 192 AI Accelerator units and 96 Compute Units delivers 123 TFLOPS of peak half precision compute performance, making this new card ideal for generative AI and scalable multi GPU workstations.

AMD has qualified the use of up to 4 of these cards in a single workstation chassis. This allows the creation of a high density desktop system for compute intense workloads.

The W7900 Dual Slot was designed to work with leading content creation applications which can leverage the computational performance of the GPU, such as Adobe Creative Cloud, AutoDesk, Blender, Da Vinci Resolve, Maxon Cinema 4D, RedShift and many more.

Compared to other ultra high end workstation cards in the market, the Radeon PRO W7900 Dual Slot provides comparable performance but at a significantly lower cost. This results in a up to 52% better performance per dollar, when compared toe the NVIDIA RTX 6000 Ada Lovelace card on the SPECviewperf 4K Geomean benchmark.

When looking at application specific benchmarks, the AMD Radeon PRO W7900 was shown to perform even better, reaching about 2x the performance per dollar of other cards.

AMD has also introduced the AMD ROCm 6.1 for Radeon GPU. ROCm is an open compute software stack. With ROCm 6.1, developers can now use TensorFlow with their AI workloads. Developers will also get beta level support for the usage of their GPUs on the Windows Subsystem Linux, allowing them to run ROCm on their Windows systems. All these new features are made optimized for the latest qualified Radeon products, including the Radeon PRO W7900 Dual Slot.

With the new AMD ROCm, researchers and developers can make use of the powerful Radeon PRO W7900 Dual Slot to tap into power of Large Language Models (LLMs). LLMs can help companies in better way of doing business by making processes smarter, being more consistent, and generally improving productivity.

Using LLM on local workstations can also allow developers to improve their work while keeping their source codes private and secure. The solution can also run with multiple GPUs in data parallelism mode to serve multiple users across various departments or work groups. The Radeon PRO W7900 has been shown to be about 38% better in performance per dollar with inference workloads on LLaMA 2 or 3 models, when compared to the NVIDIA RTX Ada 6000.

The AMD Radeon W7900 Dual Slot will be available from the 19th June 2024 onwards. It is a workstation graphics card for enterprises and developers for Multi-GPU setups and catered for Generative AI workloads.