Provide chain reliability is one other underappreciated benefit. Hyperscalers desire a credible second supply, however provided that Intel can provide steady, predictable roadmaps throughout a number of product generations.
Nevertheless, the corporate runs into a serious constraint on the software program layer.
“The decisive bottleneck is software program,” Rawat stated. “CUDA capabilities as an business working normal, embedded throughout fashions, pipelines, and DevOps. Intel’s problem is to show that migration prices are low, and that ongoing optimization doesn’t turn out to be a hidden engineering tax.”
For enterprise consumers, that software program hole interprets instantly into switching danger.
Tighter integration of Intel CPUs, GPUs, and networking may enhance system-level effectivity for enterprises and cloud suppliers, however the dominance of the CUDA ecosystem stays the first barrier to switching, stated Charlie Dai, VP and principal analyst at Forrester.
“Even with sturdy {hardware} integration, consumers will hesitate with out seamless compatibility with mainstream ML/DL frameworks and tooling,” Dai added.
