What happened
TechRadar reports that Intel unveiled its GPU D-series, formerly codenamed Crescent Island, at the recent Open Compute Project Global Summit. The accelerator uses LPDDR5X memory and offers up to 480GB of capacity, alongside specialized AI computing hardware. According to TechRadar's report, Intel says one GPU can support language models with up to 300 billion parameters, while an eight-GPU system can handle models exceeding one trillion parameters. Intel lists 1.3 petaflops of FP8 performance, 1.5TB/s of memory bandwidth and a 350W power envelope. TechRadar says Intel has not disclosed a public retail price or specific general-availability date. The report's specifications and performance claims come from Intel and have not been independently benchmarked here.
TechRadar reports that Intel introduced the GPU D-series at the recent Open Compute Project Global Summit. The accelerator, previously known by the Crescent Island codename, is designed for AI rather than conventional gaming use and can contain up to 480GB of LPDDR5X memory. TechRadar says this gives it a higher stated memory capacity than Nvidia's 288GB Blackwell Ultra B300, though no independent comparison is provided.
The report says Intel rates the D-series at 1.3 petaflops for FP8 computation and 1.5TB/s of memory bandwidth, with a 350W power envelope, PCIe form factor and air cooling. Intel says the product can support models with up to 300 billion parameters on one card and models exceeding one trillion parameters in an eight-GPU system. TechRadar reports that Intel plans to pair the accelerator with Xeon processors for coordination, data movement and CPU-handled tasks.
TechRadar does not report a public retail price or a specific general-availability date. The article also does not provide independent benchmarks, customer results or confirmation that enterprises can order the accelerator.
Source details: techradar.com ↗
Why it matters
The product targets a practical bottleneck in enterprise AI: keeping large model weights and associated data available during . More memory could help organizations serve long prompts, concurrent requests and large models without splitting them across as many accelerators. However, capacity is not a complete measure of usefulness; real-world performance will also depend on bandwidth, software, compute resources, workload characteristics and cost. If Intel's stated specifications translate into competitive deployed performance, the D-series could give enterprises another option beyond Nvidia for inference infrastructure. Its stated PCIe form factor, air cooling and planned Xeon integration may reduce deployment changes for some data centers, but TechRadar provides no independent testing, customer deployments or evidence of actual price-performance. The product's practical impact is therefore not yet established. The report confirms a concrete accelerator announcement, while availability, purchasing terms and comparative performance remain unknown.
Large memory capacity can be useful when an system must keep model weights and related data accessible while handling long prompts or multiple requests. That makes the D-series relevant to enterprise AI services and some robotics workloads described in the report.
The stated deployment characteristics could matter to data-center operators: Intel says the accelerator uses a standard PCIe form factor and air cooling, while its planned Xeon integration is intended to support multi-stage AI systems involving , reasoning, orchestration and external tools. These are Intel's stated goals, not independently verified outcomes.
The main limitation is that memory capacity does not establish serving speed, latency, throughput or total cost. TechRadar reports Intel's specifications but does not establish how the product performs against competing accelerators in practical workloads.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
The key unknowns are Intel's general-availability date, public pricing, supported software stack and independent performance results. Those details will determine whether the D-series is a product enterprises can deploy now or primarily a announced platform for future systems. Watch for evidence about performance on long-context , simultaneous requests and model-serving workloads, as well as the cost and operational tradeoffs of using one high-capacity GPU versus multiple lower-capacity accelerators. Intel's integration with Xeon processors and compatibility with existing data-center infrastructure also warrants validation in customer deployments.
Intel's general-availability timing and public pricing are not known. Those details are necessary to assess whether the D-series is accessible to enterprises and how its economics compare with alternatives.
Independent tests should examine latency, throughput, power use, software support and performance across long-context and concurrent-request workloads. No such results are included in the source.
Future customer deployments may clarify whether the claimed PCIe, air-cooling and Xeon-integration benefits simplify adoption in existing data centers.