AMD’s fierce ‘Strix Halo’ Ryzen AI Max+ debuts with radical new memory tech to power RDNA 3.5 graphics and Zen 5 CPU cores
AMD announced ‘Strix Halo’ Ryzen AI Max series laptop processors here at CES 2025 in Las Vegas, and by any definition, the new APUs look like absolute monsters for enthusiasts of the thin-and-light gaming and AI workstation laptop markets, showing what AMD bills as the fastest integrated graphics available in the Windows ecosystem courtesy of a new disruptive integrated memory architecture that unlocks new capabilities and a beastly 40-core RDNA 3.5 integrated GPU.
In fact, AMD says its AI Max chips deliver up to 1.4X faster gaming performance than Intel’s flagship ‘Lunar Lake’ Core Ultra 9 288V, and up to 84% faster performance to render than the Apple MacBook M4 Pro. AMD also says that the chip can provide an incredible 2.2X more performance in AI workloads than the smart desktop Nvidia RTX 4090 GPU, but at an 87% lower TDP.
If AMD continues the tradition, which we fully expect, we will see these awesome APU chips coming to desktop PCs in the future.
PICTURE 1 on 4
Ryzen AI Max Specifications
The 120W Zen 5-powered flagship Ryzen AI Max + 395 comes packing 16 CPU cores and 32 threads paired with 40 RDNA 3.5 (Radeon 8060S) integrated graphics cores (CUs), but perhaps more importantly, it supports up to 128GB of memory shared by the CPU, GPU, and XDNA 2 NPU AI machines. The memory can also be carved up into a separate pool dedicated to the GPU alone, thus delivering an astonishing 256 GB/s of memory throughput that unlocks incredible performance in AI workloads which is limited by memory capacity (details below). AMD says it provides groundbreaking capabilities for thin-and-light laptops and mini workstations, especially for AI workloads. The company also shares a lot of gaming and content creation benchmarks.
| Processor – 8060S iGPU | Cores/Threads | Raising the Clock | Cache (total) | GPU Cores | cTDP | Peak TOPS |
| Ryzen Al Max+ 395 / PRO | 16 / 32 | 5.1 GHz | 80 MB | 40 CU | 45-120W | 50 |
| Ryzen Al Max 390 / PRO | 12 / 24 | 5.0 GHz | 76 MB | 32 cu | 45-120W | 50 |
| Ryzen Al Max 385 / PRO | 8 / 16 | 5.0 GHz | 40 MB | 32 cu | 45-120W | 50 |
| Ryzen Al Max 380 PRO | 6 / 12 | 4.9 GHz | 22 MB | 16 cu | 45-120W | 50 |
As indicated by the ‘+,’ the Ryzen AI Max + 395 serves as the flagship of the new series, but it is joined by three other processors in different configurations from a 6-core 12-thread / 16 GPU core Ryzen AI Max Pro 380 to 12-core 24-thread / 32 GPU core Ryzen AI Max 390. Ryzen AI Max 390 and 385 are armed with Radeon 8050S graphics engine. AMD also has an unlisted 16-core 16-CU model for OEMs.
All AI Max chips have a 55W base TDP, but also a configurable TDP that goes from 45 to 120W to unleash more horsepower in designs that can handle thermal output. The top three models come in both standard and Pro series models, with the latter being available
including the RAS feature set for professional users. AMD plans to release these processors in Q1 and Q2 of this year.
As you can see above, the chip has a large central I/O die that houses the GPU and NPU, while two smaller dies contain the CPU cores. The chip connects to standard memory, but you can use the Radeon Adrenaline software to create a custom memory allocation dedicated to the GPU only.
For example, if you have 128GB of total system memory, up to 96GB can be allocated to the GPU alone, with the remaining 32GB dedicated to the CPU. However, the GPU can still read from the entire 128 GB memory, thus eliminating expensive memory copies through the integrated parallel memory architecture. However, it can only write to the directly allocated 96GB pool. The combination of a shared memory pool and a dedicated memory pool enables faster performance of a variety of workloads by leveraging memory as a slow VRAM of sorts, including disruptive performance in AI workloads that we will outline below. But first, playing and creating the content show.
Ryzen AI Max+ gaming and content creation benchmarks
PICTURE 1 on 3
The Strix Halo chips address the gaming enthusiasts and workstation markets, so gaming performance is key. AMD claims that Ryzen AI Max+ is 1.4X faster than Intel’s flagship Lunar Lake Core 9 288V, but the company hasn’t shared real-world gaming benchmarks. Instead, AMD shares synthetic 3DMark tests that often don’t correlate well with real-world gaming performance. We won’t know much about real-world gaming performance until Ryzen AI Max hits the market, and as always, you should approach vendor-provided benchmarks with appropriate skepticism (we include AMD test notes at the end of the article).
AMD also included several benchmarks in the rendering of its 16-core core against the 12-core Apple MacBook M4 Pro, claiming up to an 86% advantage in a v-ray workload. Of course, the 14-core M4 Pro, which is also included in the benchmarks, is more competitive, but AMD still holds a strong lead in Blender, Corona, and v-ray selection of benchmarks. However, the Ryzen AI Max+ was not as performant in the multi-threaded Cinebench 2024 test, beating the 12-core M4 Pro by a measly 2%, and trailing the 14-core M4 Pro by 3%.
AMD also took another swipe at Intel’s Lunar Lake Core Ultra 288V flagship, showing huge 3D rendering performance advantages of 2.6X faster on average, with advantages ranging from 340% to 402% in this selection of benchmark. The 288V is intended for low-power laptops, so comparisons against Intel’s newly launched Core Ultra 200H and 200HX series, the latter with dGPU, will be interesting.
AMD used a reference motherboard for its tests (a test board that usually has unlimited cooling capacity), so these results do not reflect the performance you would expect from a thermally constrained laptop. In contrast, Apple and Intel’s benchmarks were conducted using shipping laptops. AMD lists the tested TDP at the base of 55W, but the power consumption can be higher during heavy loads, especially with strong cooling of a reference motherboard. This can lead to a bad comparison – for example, the 288V has a 37W max TDP. All Ryzen AI Max configs were tested with only 32GB of RAM for these benchmarks.
Ryzen AI performance and system
PICTURE 1 on 2
AMD also shared some impressive results showing a Llama 70B Nemotron LLM AI model running both Ryzen AI Max + 395 with 128GB of total system RAM (32GB for the CPU, 96GB allocated of GPU) and a desktop Nvidia GeForce RTX 4090 with 24GB of VRAM (details of the setups in the slide below). AMD says the AI Max+ 395 delivers 2.2X the tokens/second performance of the desktop RTX 4090 card, but the company didn’t share time-to-first-token benchmarks.
Perhaps more importantly, AMD claims to do so with an 87% lower TDP than the 450W RTX 4090, with AI Max+ running at just 55W. That means systems built on this platform have exceptional power efficiency metrics in AI workloads. The Ryzen AI Max models all have an integrated XDNA 2 NPU AI engine capable of delivering up to 50 TOPS of performance, but AMD did not share benchmarks for the NPU.
Speaking of systems, AMD’s partners have systems built around AI Max chips coming to market this year. HP has the ZBook Ultra G1 compact workstation and the HP Z2 mini G1a workstation, but a release date for the systems has not been determined. In addition, Asus has its gaming tablet with a detachable keyboard, the ROG Flow Z13, which will hit the market in Q1 this year. That means a huge amount of gaming performance packed into a slim form factor. Here’s hoping we see it at retail, as AMD’s flagship gaming chips often suffer from poor usability.
https://cdn.mos.cms.futurecdn.net/Kb7ipLjAkMtvShGfbBHACf-1200-80.jpg
2025-01-06 19:27:00
