Intel Xe2 Lunar Lake Graphics Compute / OpenCL Performance Looking Great

With the ProjectPhysX OpenCL Benchmark, right off the bat we see the huge uplift in FP64 performance with the Xe2 Lunar Lake graphics compared to prior Meteor Lake graphics, which were already much better than the older Intel integrated graphics.

The INT8 compute performance was also dramatically better with Lunar Lake atop this latest open-source Intel graphics stack.

The memory bandwidth with Lunar Lake is also significantly better thanks to its integrated memory approach on-package.

But moving past the synthetic OpenCL tests, with real GPU compute workloads like FluidX3D I was blown away by the difference in performance from prior generation Meteor Lake to now with Lunar Lake... FluidX3D as a reminder is an OpenCL-accelerated computational fluid dynamics (CFD) software package.

While the Core Ultra 7 256V was consuming slightly more power than the Core Ultra 7 155H, the Lunar Lake SoC easily won with delivering a huge performance-per-Watt lead for FluidX3D.