Intel has unveiled an AI inference accelerator with up to 480GB of memory, according to TechRadar. The D-series card is designed to run models of up to 300 billion parameters in a standard PCIe slot, with a 350W power envelope.
Intel AI Watch analysis
What happened
Intel presented its GPU D-series at the Open Compute Project Global Summit, TechRadar reports. The accelerator uses LPDDR5X memory, rather than the GDDR7 readers might expect from a conventional graphics card, and Intel is pitching it for AI inference in existing data centres.
The reported specifications are notable for their combination: high memory capacity, a standard PCIe form factor and a stated 350W power envelope. TechRadar says a single card can support language models with up to 300 billion parameters.
Why it matters
For running large models, memory capacity can determine whether a model fits on one accelerator or has to be split across several. Intel’s reported 480GB ceiling could make high-capacity inference possible on a single card, while the PCIe design is intended to suit existing data-centre deployments.
That is the promise, not yet a verdict on real-world performance. The supplied report gives capacity and power specifications, but no benchmark results, pricing or shipping timetable. Those details will decide whether this is a practical alternative or an impressive spec sheet with its sleeves rolled up.
Our read
This is a meaningful AI-computing announcement, not a gaming GPU story: the specific development is an accelerator designed for large-model inference. The most interesting test is whether Intel can pair its unusually large memory capacity with competitive performance and availability.
What to watch
- When Intel plans to ship the D-series and what it will cost.
- Independent inference benchmarks, including performance per watt.
- Which models can run on one card in practice, and with what limitations.
Discussion spark: Would you favour a single accelerator that can hold a very large model, even if its performance is still unproven, or wait for benchmarks before changing your infrastructure?
Sources and evidence
Independent WittyWires tracker for public updates about Intel AI. Not affiliated with or endorsed by Intel AI; this is not an official account.