Intel’s D270 data-center GPU offers up to 480 GB of LPDDR5X memory per card. Intel targets production AI inference, with enough memory to keep larger models, longer contexts and more concurrent sessions on the accelerator.
The card uses LPDDR5X rather than high-bandwidth memory, or HBM. Intel pitches that choice as a way to run more inference with fewer GPUs.
A memory-heavy card for existing servers
Intel lists 32 Xe3P cores, 256 XMX matrix engines and a 350 W power rating. The D270 uses PCIe 5.0 and passive air cooling. Intel says it reallocates chip area from graphics and ray-tracing pipelines to inference compute, and names vLLM and PyTorch as serving software.
Long prompts and concurrent sessions consume memory alongside the model. Intel’s page specifically cites growth in the KV cache during inference as a reason for the capacity.
The Nvidia comparison needs workload results
Intel’s disclaimer says performance projections against RTX Pro 6000 use public information and internal analysis as of February 2026. It qualifies the memory comparison as being at comparable bandwidth. The page gives no numerical bandwidth specification or reproducible benchmark results.
