Intel unveils D-series AI GPU with 480GB of LPDDR5X memory
The accelerator, formerly codenamed Crescent Island, is rated at 1.3 petaflops for FP8 and pairs with Xeon processors, but Intel has not set a retail price.
Intel introduced a data center GPU called the Intel GPU D-series at the Open Compute Project Global Summit, according to techradar.com. The accelerator, previously known by the codename Crescent Island, offers up to 480GB of LPDDR5X memory.
Intel chose LPDDR5X rather than GDDR7 for the card. The company says a single GPU can support large language models with up to 300 billion parameters, while an eight-GPU system can accommodate models exceeding one trillion parameters.
The D-series is rated at 1.3 petaflops for FP8 computation with 1.5TB/s of memory bandwidth. It carries a 350W power envelope and uses a standard PCIe form factor with air cooling.
Intel plans to pair the accelerator with its Xeon processors. The company has not announced a public retail price or a specific general-availability date for the GPU D-series.
Memory capacity compared with Nvidia
Nvidia's Blackwell Ultra B300 offered 288GB of HBM3e memory before the D-series took the highest-memory-capacity title for data center GPUs.
Quick answers
How much memory does the Intel GPU D-series have?
The D-series offers up to 480GB of LPDDR5X memory, which Intel chose over GDDR7.
When will the Intel GPU D-series be available and how much will it cost?
Intel has not announced a public retail price or a specific general-availability date.
What size AI models can the D-series run?
Intel says one card supports large language models with up to 300 billion parameters, and an eight-GPU system can handle models exceeding one trillion parameters.