The AMD Ryzen AI Max+ 395 is a powerful Strix Halo family processor that debuted in January 2025. The APU comes with 16 Zen 5 CPU cores running at up to 5.1 GHz, the 40 CU RDNA 3+ Radeon 8060S graphics adapter and the 50 TOPS XDNA 2 neural engine. Other noteworthy features include PCIe 4, USB 4 and LPDDR5x-8000 RAM support and a whole lot of L3 cache.
Architecture and Features
Unlike Strix Point, Strix Halo parts are powered by Zen 5 cores - no Zen 5c here. It's not clear if this is the desktop Zen 5 implementation with full AVX512 throughput or the mobile one. According to AMD, Zen 5 delivers a 16% IPC improvement over Zen 4 thanks to branch prediction improvements and other refinements.
Elsewhere, the AI Max+ chip supports RAM as fast as LPDDR5x-8000, and is natively compatible with USB 4 (and therefore Thunderbolt). It has PCIe 4.0 support for a throughput of 1.9 GB/s per lane, just like its 8000 series predecessors had. The integrated XDNA 2 NPU delivers up to 50 INT8 TOPS for accelerating various AI workloads.
Performance
From leaks, we know that the Ryzen delivers around 20,700 Geekbench 6.3 Multi points, meaning it trades blow with very powerful desktop CPUs like the Ryzen 9 9900X.
Graphics
The Radeon 8060S is the most powerful iGPU that AMD has on offer, as of Jan 2025. It features 40 RDNA 3+ architecture CUs (2560 unified shaders) that could see it outcompeting lower mid-range desktop graphics cards like the GeForce RTX 4050. This GPU will undoubtedly run any game at 1080p on Ultra, however, the ultimate question is whether or not the cooling solution of the laptop will be powerful enough to let the iGPU shine.
Naturally, the Radeon is capable of driving four SUHD 4320p60 monitors. It can also efficiently encode and decode the most popular video codecs including AVC, HEVC, VP9 and AV1. The latest addition to that list, the VVC codec, is not supported unlike it is with Intel Lunar Lake chips.
Power consumption
The AI Max+ 395 can eat up to 120 W depending on the system and its TDP power targets, with 45 W named as the minimum TDP.
The 4 nm TSMC process that the CPU cores are built with make for decent, as of Jan 2025, energy efficiency.
The Intel Core Ultra 9 285HX is a mobile high-end CPU for gaming notebooks based on the Arrow Lake architecture. It offers 24 cores consisting of 8 fast performance cores with up to 5.5 GHz and 16 smaller efficiency cores with up to 4.6 GHz clock speed. The CPU has access to 40 MB L2 cache and 36 MB L3 cache and is specified with a TDP of 55 watts.
The SoC integrates a small dedicated NPU called AI Boost with 13 TOPS (Int8) and optionally supports vPro Enterprise. The integrated memory controller supports up to 192 GB DDR5-6400 (dual channel, ECC optional). The integrated GPU (iGPU) is a small Intel Graphics graphics card with four Xe cores running at 300 - 2000 MHz.
Performance - High End
The performance of the Ultra 9 285HX should be slightly below the similar Desktop Core Ultra 9 285K (higher clocked with up to 5.7 GHz and more TDP with 125 watts). Thanks to the improved IPC of the cores (especially the efficiency cores), the CPU should outperform the old Raptor Lake Core i9-14900HX (8 + 16 cores, max 5.8 / 4.1 GHz). This makes the CPU ideal for very demanding tasks such as high-end gaming and content creation.
Production - TSMC and Foveros
The Arrow Lake-HX chips are based on the desktop Arrow Lake-S chips and are composed of several chiplets. The CPU part is manufactured at TSMC in the modern N3B (3nm) process, the GPU also comes from TSMC in the N5P process. The SoC and I/O tile are manufactured in the TSMC N6 process. The base tile is manufactured by Intel in the 22nm process and carries the individual chiplets thanks to Fovero's 3D packaging.
The Intel Core Ultra 5 235HX is a fast mid-range mobile CPU for gaming notebooks based on the Arrow Lake architecture. It offers 20 cores consisting of 6 of the 8 fast performance cores with up to 5.1 GHz and 8 of the 16 smaller efficiency cores with up to 4.5 GHz clock speed. The CPU can access 24 MB L2 cache and 26 MB L3 cache and is specified with a TDP of 55 watts.
The SoC integrates a small dedicated NPU called AI Boost with 13 TOPS (Int8) and optionally supports vPro Essentials. The integrated memory controller supports up to 192 GB DDR5-6400 (dual channel, ECC optional). The integrated GPU (iGPU) is a small Intel Graphics graphics card with three of the four Xe cores running at 300 - 1800 MHz.
Performance - High End
Thanks to the improved IPC of the cores (especially the efficiency cores), the CPU should be just before or at the level of the Raptor Lake Core i9-14500HX (6 + 8 cores, max 5 / 3.5 GHz). This makes the CPU very suitable for demanding tasks such as high-end gaming (2024 games) and content creation. The performance difference to the Core Ultra 5 245HX is only slight, as the CPUs only differ in the base clock of the performance cores (2.9 vs 3.1 GHz).
Production - TSMC and Foveros
The Arrow Lake-HX chips are based on the desktop Arrow Lake-S chips and are composed of several chiplets. The CPU part is manufactured at TSMC in the modern N3B (3nm) process, the GPU also comes from TSMC in the N5P process. The SoC and I/O tile are manufactured in the TSMC N6 process. The base tile is manufactured by Intel in the 22nm process and carries the individual chiplets thanks to Fovero's 3D packaging.
Average Benchmarks AMD Ryzen AI Max+ 395 → 100%n=2
Average Benchmarks Intel Core Ultra 9 285HX → 112%n=2
Average Benchmarks Intel Core Ultra 5 235HX → 63%n=2
- Range of benchmark values for this graphics card - Average benchmark values for this graphics card * Smaller numbers mean a higher performance 1 This benchmark is not used for the average calculation
v1.34
log 22. 23:29:05
#0 ran 0s before starting gpusingle class +0s ... 0s
#1 checking url part for id 18628 +0s ... 0s
#2 checking url part for id 18884 +0s ... 0s
#3 checking url part for id 18896 +0s ... 0s
#4 redirected to Ajax server, took 1761168545s time from redirect:0 +0s ... 0s
#5 did not recreate cache, as it is less than 5 days old! Created at Wed, 22 Oct 2025 05:18:45 +0200 +0s ... 0s
#6 composed specs +0.006s ... 0.006s
#7 did output specs +0s ... 0.006s
#8 getting avg benchmarks for device 18628 +0.004s ... 0.011s
#9 got single benchmarks 18628 +0.013s ... 0.023s
#10 getting avg benchmarks for device 18884 +0.004s ... 0.027s
#11 got single benchmarks 18884 +0.007s ... 0.033s
#12 getting avg benchmarks for device 18896 +0s ... 0.034s
#13 got single benchmarks 18896 +0.003s ... 0.037s
#14 got avg benchmarks for devices +0s ... 0.037s
#15 min, max, avg, median took s +0.046s ... 0.083s