AMD Ryzen AI Max+ 395 Analysis - Strix Halo to rival Apple M4 Pro/Max with 16 Zen 5 cores and iGPU on par with RTX 4070 Laptop

AMD expands its portfolio of APUs with more powerful Ryzen AI Max APUs. It is still a Strix Point CPU architecture (4 nm TSMC) and RDNA 3.5 GPU architecture. AMD calls the new chips Halo series, their flagship models. Hence the popular name Strix Halo, which is often used for Ryzen AI Max APUs. Below the Halo series we have the familiar Ryzen AI 9 APUs in the Premium series (like Ryzen AI 9 365) and Ryzen AI 7/5 APUs in the Advanced series (like Ryzen AI 7 Pro 360). All Ryzen AI APUs are equipped with the same XDNA 2 NPU with a performance of 50 TOPS, so all models get the Copilot+ certification.

While all the Strix Point APUs so far used a combination of Zen 5 cores and more compact Zen 5c cores with less cache, the Strix Halo APUs only use full-fledged Zen 5 cores with the same maximum clock of 5.1 GHz as before. The top model Ryzen AI Max+ 395 gets 16 cores with 32 threads and 80 MB cache; the Ryzen AI Max 390 gets 12 cores/24 threads with 64 MB cache and the Ryzen AI Max 385 gets 8 cores/16 threads with 40 MB cache. The RAM (LPDDR5x-8000) is also part of the APU and you can get 32, 64 or 128 GB.

The Strix Halo APUs are also equipped with powerful integrated GPUs based on the RDNA 3.5 architecture. The top model Ryzen AI Max+ 395 gets the Radeon 8060S with 40 compute units, while the other two models get the Radeon 8050S with 32 compute units. Depending on the RAM configuration, you can allocate dedicated VRAM for the GPU (up to 24 GB for 32 GB SKUs, 48 GB for 64 GB SKUs and 96 GB for 128 GB SKUs).

AMD specifies the TDP range between 45-125 Watts, so the new Strix Halo APUs are primarily designed for powerful multimedia laptops with corresponding cooling solutions. According to the spec sheet of the HP ZBook Ultra 14, there will be a fourth Strix Halo chip called Ryzen AI Max 380 with 6 cores as well as an even slower Radeon 8040S iGPU.

To ensure a meaningful comparison between different processors and graphics cards, we evaluate power consumption alongside synthetic benchmark performance, allowing us to determine overall efficiency. Our power consumption measurements are conducted using an external display, eliminating any influence from varying internal display power usage. However, it is important to note that we measure total system power consumption, rather than just comparing raw TDP values.

We start our CPU performance section with single-core scenarios. The new APU uses the familiar Zen 5 cores with the same maximum clock of 5.1 GHz, so it is not surprising that we get comparable result to other Zen 5 APUs like the Ryzen AI 9 HX 370 in the Cinebench tests. The results are on par with the old Dragon Range HX series (like the Ryzen 9 7945HX3D), but also slower than Intel's latest Arrow Lake and Lunar Lake chips (at least the fast Core Ultra 9 variants). The fast Snapdragon X Elite SoCs with dual-core turbo (X1E-80-100 and faster) are also ahead, but the ARM models X1E-78-100 and below are slower. Apple is in a completely different league with their M4 SoCs and the latest desktop CPUs from AMD as well as Intel are also faster than the Ryzen AI Max 395.

The new Strix Halo APU performs better in the latest Geekbench 6 test and it can beat all the current mobile processors from AMD, Intel, and Qualcomm. The old Raptor Lake Core i9-14900HX is comparable; the current desktop CPUs and especially Apple's M4 chips are once again ahead.

For the single-core efficiency comparison we use Cinebench 2024 as well the older Cinebench R23 where we just have more comparison data (especially with HX CPUs). The new Ryzen AI Max+ 395 is clearly worse than the other current mobile CPUs in Cinebench 2024 (both Ryzen AI as well as Intel Arrow Lake-H/Lunar Lake-V). The ARM competition from Apple and Qualcomm is also much more efficient in single-core scenarios. Only the desktop chips are even worse in terms of efficiency.

The older Cinebench R23 test also shows an advantage for the current mobile processors, but the older HX series chips (which are much closer to the Ryzen AI Max in terms of overall performance) are clearly beaten by the Strix Halo APU. This is also the case for the current desktop CPUs from AMD as well as Intel.

As expected, the new Ryzen AI Max+ 395 performs much better in multi-core scenarios, but please keep in mind that we tested the new chip in a 13-inch convertible with limited power limits (86/70 Watts). The performance would be noticeably better with the full 125 Watts. The current mobile chips from AMD, Intel and Qualcomm do not stand a chance, only the HX series CPUs are on a comparable level (but usually with much higher consumption values, especially for the Core i9-14900HX rivals). Apple has a small advantage with the M4 Pro, and the majority of current desktop chips is also faster.

The new Ryzen AI Max+ 395 once again performs much better in Geekbench and is one of the fastest models. Apple's M4 Pro and M4 Max SoCs stay in front, but then we have the Ryzen AI Max+ 395 surrounded by current desktop chips like the Intel Core Ultra 9 285K or AMD's Ryzen 9950X.

We once again start with the Cinebench 2024 efficiency, where the Ryzen AI Max performs well and is only beaten by the new Arrow Lake-H chip Core Ultra 9 285H among the x86 competition. All the other models are either on a comparable level or worse, but the ARM competition is once again more efficient.

The older R23 Multi test is more interesting for the comparison with the x86 rivals since the ARM CPUs emulate this benchmark, which affects the performance. In this scenario, the new Ryzen AI Max APU is one of the most efficient chips and beats all of the HX competitors as well as desktop models.

The performance you get depends heavily on the laptop model itself and its power limits, so we also performed the Cinebench 2024 Multi benchmark at preset power limits. We used tools like Throttle Stop or Universal x86 Tuning Utility, but that means we are limited to AMD and Intel processors.

We also included the results of the two current chips Ryzen 9 AI HX 370 as well as Core Ultra 9 285H in our comparison to get a better picture, especially in the range between 45-55 Watts. This comparison clearly shows that the new Ryzen AI Max+ 395 is already very capable at low power limits and offers a noticeable performance advantage over the other two mobile chips. The older Ryzen 9 7945HX3D starts to be competitive at 80 Watts, while the Core i9-14900HX falls behind. We know from our reviews that the 14900HX simply needs quite a lot of power (much more than 100 Watts) to offer good performance numbers.

In addition to the powerful CPU, you also a powerful integrated graphics card called Radeon 8060S. AMD claims it is the fastest x86-iGPU and we can confirm that right away, since neither the Radeon 890M, nor the Intel Arc Graphics 140V/140T stand a chance against the new Radeon 8060S.

We start with the different 3DMark benchmarks and they clearly show that Radeon 8060S is on par with the GeForce RTX 4070 Laptop with TGP value between 65-75 Watts. This is the case for 3DMark 11 as well as 3DMark Fire Strike and Time Spy, but the Radeon 8060S loses some ground compared to the RTX 4070 Laptop in the Speed Way as well as Steel Nomad tests.

We also tested the new Radeon 8060S against the ARM competition from Apple as well as Qualcomm. The fastest version of the Qualcomm Adreno X1-85 (4.6 TFLOPS) does not stand a chance against the new Strix Halo iGPU. The situation changes compared to Apple’s GPUs, and depending on the benchmark the 8060S is sitting right between the two M4 Pro GPUs or slightly ahead, respectively. The M4 Max GPU on the other hand is out of reach.

We have some gaming benchmarks that we perform in every laptop review and the Radeon 8060S is on par with the Radeon RX 7600M XT and slightly ahead of the fastest version of the GeForce RTX 4050 Laptop (90W TGP) in the Lenovo IdeaPad Pro 5 16. The two comparable RTX 4070 Laptop devices Asus ROG Flow Z13 as well as Asus ProArt PX13 have a slight advantage, but this is mainly caused by some of the results in older benchmarks like GTA V. This is also the case for fast versions of the RTX 4060 Laptop. The two newer titles Cyberpunk 2077 and Baldur's Gate 3 show better results for the Radeon 8060S. All in all, you can play almost every game in 1080p and maximum details, while some titles are a bit too demanding for the QHD resolution (more on that in the following paragraphs).

We also had the old Flow Z13 with the RTX 4070 Laptop on hand, so we used it for a direct head-to-head comparison in modern titles (Turbo mode on both devices). The situation looks better for the new Radeon 8060S in 1080p and maximum details. Titles like Cyberpunk 2077 or Call of Duty Black Ops 6 show an advantage of 13-17 % for the AMD GPU, while many titles show comparable results. There are also some negative examples like GTA V or F1 24 (-12 up to -21 %). All in all, the two devices are pretty much on par.

The situation is very similar in the QHD resolution, even though the GeForce Flow Z13 has a slightly bigger lead (+5 %). Many titles are generally playable in QHD and maximum details, only very demanding games like Black Myth: Wukong, Star Wars Outlaws, Indiana Jones or Monster Hunter Wilds drop below the 30 fps mark.