Intel Names Crescent Island AI Accelerator the GPU D-Series
Intel's Crescent Island inference accelerator will ship as the GPU D-Series, offering 1.3 PFLOPS of FP8 compute and up to 480 GB of LPDDR5X memory.
AfroEuropa Newsroom
AfroEuropa desk
Intel has given a formal name to its Crescent Island AI inference accelerator, which will reach the market as the Intel GPU D-Series. The disclosure was made at OCP 2026, according to Tom's Hardware.
The company shared a theoretical peak performance figure for the first time: 1.3 PFLOPS of FP8 compute. Intel also confirmed memory bandwidth of 1.5 TB/s. The chip supports data formats ranging from FP4 to FP64.
According to TechPowerUp, the D-Series GPU carries 32 Xe3P cores, 256 XMX engines, 256 vector engines, and 32 MB of unified L2 cache. Each core includes a 1 MB register file and 512 KB of L1 cache. The XMX engines have been reworked from a four-deep systolic layout on the standard Xe3 to a 16-deep design on the Xe3P. A media engine with four encoders and four decoders is also included, which could support multimodal workloads involving video generation and understanding.
Keep reading
Nvidia Reportedly Redirects RTX 5090 Silicon to Professional and Data Center GPUs
Leakers say Nvidia has stopped feeding GB202 chips to the GeForce RTX 5090, reserving the silicon for pricier workstation and data center cards. Nvidia has not confirmed the reports.
Other highlights
Ripple-Backed Evernorth Debuts on Nasdaq as First Pure-Play XRP Treasury
Evernorth's XRPN shares opened higher on Nasdaq after its SPAC merger closed, though most SPAC investors exited and its XRP holdings have fallen in value since signing.
Follow the story
One newsletter, two continents
The Bridge brings you the tech, startups, and leaders moving between Africa and Europe in one sharp email each morning. No spam, unsubscribe anytime.
