Intel Core Ultra X9 Panther Lake Outguns AMD’s Strix Halo In New Leak

It's really important to keep things in context when you're looking at hardware benchmarks. Desktop CPUs often operate with essentially unlimited power and thermal budgets, so it's simply not fair to compare them against laptop chips that must constantly keep power draw in check to deal with thermal constraints and to avoid draining batteries too quickly. With that in mind, let's compare this fresh Panther Lake leak against AMD's current flagship mobile part, the Ryzen AI Max+ 395.


Above are the scores for a leaked Intel Core Ultra X9 388H Geekbench result. For comparison's sake, the "official" scores from the Geekbench browser for the Ryzen AI Max+ 395 are 2792 on a single thread, and 17680 for all threads. Now, that result represents an average score for the processor based on user-submitted benchmarks; our own testing found that the same chip in the ASUS ROG Flow z13 delivers scores of 2990 and 21269 for single- and multi-thread.

Even going with our considerably above-average scores, the Panther Lake part is still ahead in single-threaded performance. That's impressive considering that AMD's Strix Halo is based on the same Zen 5 CPU chiplets that are used in desktop Ryzen 9000 processors as well as the company's EPYC server CPUs. Of course, we have to point out that if we compare against HP's Z2 Mini G1a workstation sporting the same Strix Halo part, it scores 3134 on a single thread and 23849 on all threads. But that's exactly why you can't compare desktops to laptops.

If we instead compare against AMD's actual fastest mobile-only part, the Ryzen AI 9 HX 370, the deltas are large; that chip scores 2591 on a single thread and 13699 across all 24 threads, both of which get destroyed by the Core Ultra X9's scores above. In fact, even comparing against Intel's own Core Ultra 9 285H, it's pretty brutal: that chip scores 2603 on a single thread and 14783 across all threads.


A lot more goes into system performance than just the CPU, and while Geekbench results are typically sorted by CPU scores -- as that's the single biggest factor -- other performance metrics such as overall system latency, memory latency and throughput, and even the speed of the connection between the GPU and the host system can have an impact—sometimes quite an outsized impact. As an example, the 256-bit memory bus of Strix Halo allows those parts to essentially match the multi-core performance of the Ryzen 9 9950X despite a clock rate disadvantage that can exceed 1 GHz in Geekbench results. They perform better on different parts of the multi-core test, but the final scores are pretty similar.

While we know that both Cougar Cove and Darkmont architectures will feature architectural improvements over Lunar Lake and Arrow Lake, these benchmark results give us hope that Intel has resolved its platform latency issues that affected Arrow Lake and gave it relatively poor gaming performance in some scenarios. After all, the clock rates visible in the full result are running right around 5.05 GHz—exactly the same as most Lunar Lake results. We'll know for sure once we get our hands on the parts, which will likely happen not long after their CES launch. Stay tuned.