AMD introduced the new Instinct MI350P PCIe graphics accelerator designed specifically for artificial intelligence workloads. This model stands out as the company’s first PCIe-based Instinct design released after four years.

The new hardware allows corporate users to increase their artificial intelligence capabilities without changing their existing infrastructure. Standing out with its standard dual-slot and server-oriented structure, the card can be easily integrated into air-cooled systems.

Technical Specifications and Performance Capacity

Instinct MI350P, cDNA 4 architectureIt is built on TSMC’s 3nm production technology. This chip, which contains a total of 128 processing units and 8,192 stream processors, hosts 73 billion transistors.

Using HBM3E technology with a capacity of 144 GB on the memory side, the card offers bandwidth up to 4 TB/s. The device, which has a power consumption value of 600W, can be configured up to 450W if necessary.

The prominent performance data of the card includes 4.6 PFLOPs MXFP4 and MXFP6 computing power. Additionally, the 2.3 PFLOPs FP16 and 36 TFLOPs FP64 performance values ​​are optimized to support enterprise-level artificial intelligence operations.

Market Position and Competition

With this new product, AMD aims to compete with NVIDIA’s PCIe-based accelerators such as the H200 NVL. The card comes with an open source ecosystem and ROCm support, providing developers with a flexible working environment.

The product is designed to work in harmony with the power and cooling infrastructures of existing data centers. In this way, businesses can commission artificial intelligence projects locally without making costly infrastructure investments.

Instinct MI350P becomes an important option in the corporate artificial intelligence market with its high computing density and large memory capacity.

What do you think about this new PCIe accelerator from AMD?