Intel´s Xeon 6 processors have emerged as a critical component in the new generation of NVIDIA GPU servers designed for Artificial Intelligence workloads, notably featured in the upcoming NVIDIA DGX B300. The DGX B300 incorporates the Intel Xeon 6776P CPU, a 64-core, 350W processor equipped with a substantial 336MB L3 cache. Historically, securing a place in NVIDIA´s reference server platforms signals wide market adoption, as server manufacturers typically mirror these reference system configurations for the broader NVIDIA HGX ecosystem.
Intel is now highlighting the capabilities of its so-called ´priority cores´ as a significant differentiator in Artificial Intelligence server infrastructure. According to Intel, the Xeon 6 lineup delivers up to 128 Performance-cores (P-cores) per CPU in its 6900P series, although the DGX B300 uses the 6700P variant. The processors are touted for both high core counts and strong single-threaded performance, purportedly enabling balanced workload distribution for data-intensive Artificial Intelligence applications. Memory performance is also a focus—Intel claims up to 30% faster memory speeds over AMD EPYC 9005 in dual-DIMM-per-channel (2DPC) configurations, although the article points out this comparison is nuanced; AMD provides more memory channels and higher overall capacity and bandwidth at lower per-DIMM speeds.
The industry´s preference for Intel Xeon in NVIDIA GPU servers is influenced not only by technical advantages but also by competitive alignment. NVIDIA, aiming to avoid market overlap with AMD in the graphics sphere, prefers to pair its GPUs with Intel CPUs, sidestepping AMD’s adjacent product lines. With AMD’s closest Artificial Intelligence-oriented CPUs largely absent from NVIDIA reference platforms, Intel consolidates its position. The article further notes that NVIDIA’s strategy includes standardizing not just the GPU baseboards (HGX 8-GPU), but also motherboard designs, raising the stakes for reference socket design wins. Intel´s presence in the DGX reference design thus ensures broad adoption across the Artificial Intelligence hardware ecosystem, even as the marketing around memory performance comparisons remains contentious among industry insiders.
