Intel Xeon 6 Boosts DGX B300: Up to 64 Cores, PCT, SST-TF and +30% Memory Speed

Intel has officially unveiled the new Xeon 6 server processors, which will be used in the NVIDIA DGX B300 platform, focused on training neural networks and working with AI. The Granite Rapids line includes Xeon 6776P - 64-core 128-thread chip, specially designed for paired work with GPU, ensuring the highest possible output from video cards in systems with AI loads.

One of the main advantages of the new chips is two technologies: PCT (Priority Core Turbo) и SST-TF (Speed ​​Select Technology – Technology Frequency). The first allows increasing the frequency of priority cores when performing critical tasks, reducing delays between the CPU and GPU. The second dynamically adapts the core frequency to the current load, improving the balance between performance and power consumption.

Xeon 6 processors also received 30% faster memorythan the previous generation. This is achieved thanks to MRDIMM support and the CXL bus, which is especially important in AI tasks - where memory bandwidth is critical for the speed of processing large models. The new architecture also provides 20% more PCIe lanes, which increases the connectivity of accelerators and speeds up data exchange between components.

DGX B300 based on Intel Xeon 6776P and GPU NVIDIA Blackwell is one of the most powerful AI systems of the new generation. Its architecture minimizes latency between the CPU and GPU, eliminating traditional bottlenecks that limited scalability in model training. Intel stresses that new Xeon 6 are designed for the future of large-scale AI systems, helping companies accelerate the adoption of AI in areas ranging from healthcare to telecommunications.