Intel has been preparing the launch of its next-generation Xe series of GPUs, codenamed DG2 (discrete graphics 2). It builds off the foundation laid by the Intel DG1, hopefully with significantly improved performance so it can compete with the best graphics cards from AMD and Nvidia. The supposed lineup for the Xe-HPG configurations will range from 128 to 512 Execution Units (EUs). Today, according to the leaked Geekbench entry, the DG2-128EU could potentially have a 2,200 MHz clock speed.
All of the results, including the reported frequency, should be taken with a huge grain of salt before we confirm it with the official specifications. Still, that's about a 40% boost from the DG1 in clock speeds, and the additional EUs, higher memory bandwidth, and architectural updates could let it reach even higher levels of performance.
The real question will be how the higher spec DG2 solutions stack up. Rumors indicate the DG2 family of Xe-HPG graphics cards may have as many as five SKUs, each configured with a different amount of EUs and memory. Today's leak is for the smallest of the bunch, a presumably budget-friendly solution.
Running on a test system that consisted out of an Intel Core i5-11400T Rocket Lake processor and 16GB of single-channel (?) memory, the system managed to achieve a score of 13710 points in OpenCL benchmarks. This is an entry-level performance segment usually targeting low-power configurations. All of this was achieved with the GPU allegedly running at 2,200 MHz clock speed, which is quite impressive for the new silicon. Of course, Geekbench's reporting could be wrong, so take that into account as well.
The test system isn't all that different from our DG1 test bed, so we ran that as a comparison point. It has a 65W i5-11400F, which has higher clocks and a higher TDP than the i5-11400T used with the DG2. We also ran with the original single 8GB DIMM to further eliminate advantages for our DG1 sample. The DG1 Geekbench 5 result proves yet again just how bad Geekbench 5 is as a GPU benchmark, scoring 19,176 — 40% higher than the supposed DG2 score of 13,710. The slower CPU might be partly to blame, but more likely early drivers and Geekbench 5 itself are the real culprits.
If we ignore the overall score and focus on the individual results comparison, the DG2 did score anywhere from 25% to nearly 175% higher than the DG1 card. The DG2 wasn't helped by essentially failing to run the Canny and Horizon Detection tests, where the DG1 scored 23X and 96X higher, obviously skewing the overall figure.
Anyway, since Geekbench 5 clearly doesn't provide an actually reliable estimation of performance, we'll have to wait for more leaks. Or just hold off any verdict until Intel's official announcement of the Xe-HPG DG2 family of cards and independent testing. That will be the best way to determine real-world gaming performance. Assuming it has better drivers and higher clocks, its gaming performance might actually be serviceable.
