From Startup Fortune · · 1m 40s
Intel Names Nvidia Rival GPU D270 with 480GB Memory
Intel's new GPU D270 offers 480GB memory for AI inference, challenging Nvidia's dominance.
In brief
Intel's new GPU D270 offers 480GB memory for AI inference, challenging Nvidia's dominance. Intel's GPU D270 with 480GB memory targets AI inference workloads, competing with Nvidia's AI accelerators. Originally reported by Startup Fortune.
Intel Names GPU D270
Intel finally gave its inference chip a real name and memory number big enough to make hyperscalers look twice.
GPU D270 Memory Capacity
The GPU D270 can carry up to 480gb, triple what nvidia's most common inference cards typically ship.
OCP Global Summit Announcement
Intel used the ocp global summit on october 12 to drop the codename and introduce the real product.
Crescent Island to D270
For more than a year the chip was known only as crescent island, a placeholder that outlived its usefulness.
Memory vs. FLOPS Strategy
Intel went the other way: cheaper, more abundant lpddr5x, stacked high instead of chasing raw flops with gddr7 or hbm.
D270 Performance Metrics
The d270 delivers 1.3 peak petaflops of fp8 throughput across 32 xe3 cores and 256 xmx engines.
Power Efficiency
The d270 operates within a 350-watt envelope, a fraction of the cost and none of the current hbm supply crunch.
Market Shift to Inference
Inference now accounts for roughly two-thirds of ai compute demand, reshaping what chipmakers are racing to build.
Hyperscaler Custom Silicon
Google, microsoft, amazon, and meta have custom silicon tied to their own cloud stacks rather than merchant accelerators.
Intel's Merchant Silicon
Intel's d270 is merchant silicon, meaning any hyperscaler can purchase it like an nvidia or amd card.
The main idea
Intel's GPU D270 with 480GB memory targets AI inference workloads, competing with Nvidia's AI accelerators.
