AMD Strix Halo: Ryzen AI Max with up to 16 cores, 40 CUs and 256 GB/s 130 comments
Image: AMD
AMD’s Strix Halo “monster APU” is official: four different variants are launched under the name Ryzen AI Max(+) 300. The performance available on this base enables a whole new class of compact solutions with 16 Zen processor cores 5 and 40 CU of the RDNA 3.5 Graphics Solution.
Table of Contents Ryzen AI Max with up to 16 cores, 40 CUs and 256 GB/s Four configurations across seven models at launch AMD Ryzen AI Max (Pro) 300 in detailDual CCD meets a large GPU with I/O capability Up to 256 GB/s Memory bandwidth In AI benchmark, faster than an RTX 4090 Four models: Key technical data First laptop models from partners
Four configurations in seven models to launch
AMD Strix Halo is commercially referred to as AMD Ryzen AI Max 300 or AMD Ryzen AI Max Pro 300 for the industry. Behind this is the three-digit numerical sequence that indicates the classification in this product line – and at the top is a “+” for the high-end model.
AMD Ryzen AI Max (Pro) 300 (Strix Halo) at a Glance Model Architecture Cores/
Takt of discussions
(Base/Turbo) Graph L2+L3-Cache
(MB) TDP
(Watts)NPU
(TOPS) AMD Ryzen AI Max 300 (Strix Halo) Ryzen AI Max+ (Pro) 395 Z5 16/32 3.0-5.1 GHz Radeon 8060S
RDNA 3.5, 40 CU, 2.9 GHz 80 45-120 50 Ryzen AI Max (Pro) 390 Z5 12/24 3.2-5.0 GHz Radeon 8050S
RDNA 3.5, 32 CU, 2.8 GHz 76 45–120 50 Ryzen AI Max (Pro) 385 Z5 8/16 4.6-5.0 GHz Radeon 8050S
RDNA 3.5, 32 CU, 2.8 GHz 40 45–120 50 Ryzen AI Max Pro 380 Z5 6/12 3.6-4.9 GHz Radeon 8040S
RDNA 3.5, 16 CU, 2.8 GHz 22 45–120 50
AMD Ryzen AI Max (Pro) 300 in detail
The first image presented by AMD reveals not only the basic technical details, but also the structure of the large chip.
Dual CCD meets large GPU with I/O function
Two CCDs, each with eight cores, are placed next to each other at the top, with the rest enormous below: the large graphics unit as well as the functionality that an I/O chip offers on the classic Ryzen and the NPU are located there.
AMD Ryzen AI Max aka Strix Halo (Image: AMD)
Up to 256 GB/s storage bandwidth
So that the large graphics unit of the chip is not starved, the memory bandwidth should reach 256 GB per second thanks to a revised memory interface. Previous APUs had and still have exactly this point as a weak point.
Strix Point regularly offers 120 GB per second of bandwidth via LPDDR5-7500, so Strix Halo would be more than twice as fast in this case. With LPDDR5-8000 and practically doubling the interface, the exact value given by AMD could be obtained mathematically.
AMD Ryzen AI Max aka Strix Halo (Image: AMD)
Faster than an RTX 4090 in the AI benchmark
What AMD presents are benchmarks that make the new solution shine in the AI benchmarks, even against a GeForce RTX 4090 with 24 GB of memory. But AMD also sees itself ahead compared to Intel Lunar Lake aka Core Ultra 200V and Apple’s M4.
In this case, however, it depends again on the context, as the AI benchmark against the RTX 4090 shows. The configuration in this test used a configuration with 128 GB of RAM (up to 96 GB can be addressed by the iGPU), and the large memory is what allows the Strix Halo to get ahead in this case. All other tests use 32 GB of RAM of unknown speed. The TDP comparison is based on 55 watts for the Strix Halo and 450 watts for the RTX 4090.
AMD Ryzen AI Max aka Strix Halo in manufacturer benchmarks (Image: AMD)
AMD is not yet giving any further details on the chip announcement – as is the case with many topics at CES. This should be followed by a comprehensive technical update when the first laptops are launched in the first and second quarters. For now, only the basic specifications of the four models are available.
Four models: Key technical data
The new graphics unit will be known as the Radeon 8060S (40CU), Radeon 8050S (32CU) and Radeon 8040S (16CU) in APUs. It is intended to be the bridge to the new Radeon 9000 series, which will debut on desktop with the Radeon RX 9070 XT and RX 9070 in the first quarter.
AMD Ryzen AI Max aka Strix Halo at a glance (Image: AMD)
The table below shows the technical specifications of AMD’s mobile APUs for 2025 in comparison. For ranking:
Halo series: AMD Ryzen AI Max 300 (Strix Halo)
maximum. 16 Zen 5 cores, 40 CU RDNA 3.5, 50 TOPPremium series: AMD Ryzen AI 9 300 (Strix Point)
maximum. 4/8 Zen 5/Zen 5c cores, 16 CU RDNA 3.5, 55 TOPSAdvanced-Series: AMD Ryzen AI 7/5 (Krackan Point)
maximum. 4/4 Zen 5/Zen 5c cores, 8 CU RDNA 3.5, 50 series TOPSEinstiegs: AMD Ryzen 200 (aka Hawk Point)
maximum. 8 Zen 4 cores, 12 CU RDNA 3, 16 TOPS Ryzen AI Max 300 (Strix Halo), Ryzen AI 9 300 (Strix Point), Ryzen AI 7/5 300 (Krackan Point) and Ryzen 200 (Hawk Point) in the model comparison architecture kernels/
Takt of discussions
(Base/Turbo) Graph L2+L3-Cache
(MB) TDP
(Watts)NPU
(TOPS) AMD Ryzen AI Max 300 (Strix Halo) Ryzen AI Max+ (Pro) 395 Z5 16/32 3.0-5.1 GHz Radeon 8060S
RDNA 3.5, 40 CU, 2.9 GHz 80 45-120 50 Ryzen AI Max (Pro) 390 Z5 12/24 3.2-5.0 GHz Radeon 8050S
RDNA 3.5, 32 CU, 2.8 GHz 76 45–120 50 Ryzen AI Max (Pro) 385 Z5 8/16 4.6-5.0 GHz Radeon 8050S
RDNA 3.5, 32 CU, 2.8 GHz 36 45–120 50 Ryzen AI Max Pro 380 Z5 6/12 3.6-4.9 GHz Radeon 8040S
RDNA 3.5, 16 CU, 2.8 GHz 22 45-120 50 AMD Ryzen AI 9 300 (Strix Point) Ryzen AI 9 HX 375 4×Z5 + 8×Z5c 12/24 2.0-5.1 GHz Radeon 890M
RDNA 3.5, 16 CU, 2.9 GHz 36 15-54 55 Ryzen AI 9 HX 370 4×Z5 + 8×Z5c 12/24 2.0-5.1 GHz Radeon 890M
RDNA 3.5, 16 CU, 2.9 GHz 36 15-54 50 Ryzen AI 9 365 4×Z5 + 6×Z5c 10/20 2.0-5.0 GHz Radeon 880M
RDNA 3.5, 12 CU, 2.9 GHz 34 15-54 50 AMD Ryzen AI 7 & 5 300 (Krackan Point) Ryzen AI 7 (Pro) 350 4×Z5 + 4×Z5c 8/16 2.0-5.0 Radeon 860M GHz
RDNA 3.5, 8 CU, 3.0 GHz 24 15-54 50 Ryzen AI 5 (Pro) 340 3×Z5 + 3×Z5c 6/12 2.0-4.8 GHz Radeon 840M
RDNA 3.5, 4 CU, 2.9 GHz 22 15-54 50 AMD Ryzen 200 / 8000 Mobile (Hawk Point) Ryzen 9 270 Zen 4 8/16 4.0-5.2 GHz Radeon 780M
RDNA 3, 12 CU, 2.8 GHz 24 35-54 16 Ryzen 7 260 Zen 4 8/16 3.8-5.1 GHz Radeon 780M
RDNA 3, 12 CU, 2.7 GHz 24 35-54 16 Ryzen 7 (Pro) 250 Zen 4 8/16 3.3-5.1 GHz Radeon 780M
RDNA 3, 12 CU, 2.7 GHz 24 15-54 16 Ryzen 5 240 Zen 4 6/12 4.3-5.0 GHz Radeon 760M
RDNA 3, 8 CU, 2.6 GHz 22 35-54 16 Ryzen 5 230 (Pro) Zen4 6/12 3.5-4.9 GHz Radeon 760M
RDNA 3, 8 CU, 2.6 GHz 22 15-30 16 Ryzen 5 220 (Pro) 2×Zen 4 + 4×Zen 4c 6/12 3.2-4.9 GHz Radeon 740M
RDNA 3, 4 CU, 2.8 GHz 22 15-30 – Ryzen 3 210 1×Zen 4 + 3×Zen 4c 4/8 3.0-4.7 GHz Radeon 740M
RDNA 3, 4 CU, 2.5 GHz 12 15-30 –
First models of partner notebooks
Not only laptops, but also mini PCs are equipped with the Strix Halo. AMD is already teasing the first, it comes from HP. Here too it becomes clear: the focus of AMD or its partners with Strix Halo is not on gaming at the moment.
AMD Ryzen AI Max aka Strix Halo from early partners (Image: AMD) Updated January 7, 2025 at 2:21 p.m.
Launch partner Asus has clarified the issue of memory and high bandwidth: it is a four-channel interface. With LPDDR5-8000 as the preferred memory type, this gives exactly the advertised bandwidth of 256 GB per second.
Techtip received information about this item from AMD under NDA. The only requirement was the earliest possible publication date.
Topics: AMD Press Conference AMD CES 2025 Processors AMD Strix Point APU CES 2025 Ryzen
Marc deciphers processors by testing their performance for gaming, content creation, and artificial intelligence.
