Intel has powered on the Ponte Vecchio A0 silicon and it appears to be working completely. That is nice information for Intel traders as a result of Ponte Vecchio was the corporate’s most formidable moonshot in fairly some time and contemplating it had no prior discrete graphics expertise it is vitally spectacular that Masooma Bhaiwala and her group, led by Raja Koduri appear to have gotten it proper from the get-go.
Intel additionally admitted a number of faults in its graphics facet throughout Intel Structure Day 2021 but additionally mentioned that that they had mounted each final a kind of with the Xe HPC structure. The graph beneath reveals a blue line (Intel graphics) and a inexperienced line (a competitor that’s seemingly NVIDIA). As you may see, earlier than 2021, the blue line was considerably beneath the inexperienced line however as of the Ponte Vecchio A0 silicon energy on, this has modified.
Intel revealed that constructing Ponte Vecchio had required a recreation and retooling of mainly *every part* concerned within the design course of.
The Ponte Vecchio package deal consists of 47 tiles which might be constructed on totally different foundries (some in Intel, some in TSMC) and require a ton of packaging applied sciences to return collectively as a cohesive entire.
The group at Folsom is measuring 45 TFLOPs of FP32 throughput, 5 TBps of reminiscence material bandwidth and a couple of TBps of connectivity bandwidth validating that each one components of the A0 silicon are working accurately and monitor to be a serious breakthrough in graphics.
No demo is ever full with out benchmarks and Intel even threw in some Resnet benchmarks that present that Intel has damaged by the world file rating with the A0 Ponte Vecchio silicon:
Intel’s Xe HPC will exist in 1 stack and a couple of stack designs and Xe Hyperlink shall be used to couple Ponte Vecchio’s in a number of subsystems.
In contrast to Xe HPG, an Xe HPC core within the Ponte Vecchio GPU has 8 vector engines (512 bit) and eight matrix engines (4096 bit).
It’s clear that Intel seems to have overcome the valley of stagnation and is making important progress in leaps and bounds relating to excessive efficiency graphics. The Aurora tremendous laptop will make the most of Ponte Vecchio GPUs and the corporate confirmed off a single Aurora Blade which can represent one computational unit of the supercomputing cluster.
The remainder of the Xe HPC and Ponte Vecchio presentation is given beneath: