AMD’s Novel Ryzen AI Max Chip For Copilot+ PCs Was A CES Showstopper

It’s not uncommon to see major chip players introduce new and novel approaches to traditional computing paradigms at CES in Las Vegas in January. And if ever there was a limelight moment in that regard, AMD found it with the introduction of its new chip and platform architecture formerly known by the code name Strix Halo, and unveiled officially last week as Ryzen AI Max Series. This chip cojoins existing AMD CPU and GPU architectures, along with a high speed memory interface, in an unprecedented way, compared to traditional X86 laptop or desktop chip designs. And it’s safe to say this new Application Processing Unit, as AMD likes to call them, stole the show for many.

Underneath The Hood Of Ryzen AI Max, Aptly Code-Named Strix Halo

AMD aptly coined this part “Halo” with its internal code-naming because it truly does stand out versus traditional X86 PC chip architectures in the market currently. This power-optimized chip is comprised of up to 16 “desktop-class” CPU cores via two 8-core Core Complex Dies, up to a beefy 40 AMD Radeon Compute Unit integrated GPU, and a 50 Tops Neural Processing Unit. However, how these resources are all connected and how they’re strapped together with high speed memory is where the magic happens.

So, not only is Ryzen AI Max equipped with the largest integrated graphics engine of any X86 product on the market currently, but that engine and its companion 16-core AMD Zen 5 CPU complex is also infused with a new unified memory interface that affords some serious flexibility in system configuration. Details are still sparse on the high speed interconnect for the CPUs, with reports of a new low power, low latency “sea of wires” interconnect (I’ve reached out to AMD on this), but what the company was clear on is that it also developed a new memory interface that’s ready to feed the beast with a blistering 256GB/s of bandwidth.

MORE FOR YOU
TikTok Ban Live Updates: Trump Halts Ban For 75 Days—After CEO Attends Inauguration
Trump Inauguration Live Updates: Trump Pardons Almost All Jan. 6 Riot Defendants—And Targets Birthright Citizenship
Trump Renames Denali To Mount McKinley—Here’s What To Know

What’s more, with up to 128GB of LPDDR5x-8000 system memory on board, up to 96GB of that memory can be carved out for the GPU only, making this platform a serious workhorse for content creation, CAD, 3D rendering and of course AI workloads. In fact, AMD reports that it can handle Meta’s full 70 billion parameter Llama 3.1 Large Language Model in memory, and as a result, the system is actually over 2X faster at processing this model, versus a comparable desktop system with a discrete GeForce RTX 4090 card on board. Granted, AMD’s integrated graphics for this chip are no match for an RTX 4090 in compute performance, but when it comes to very large language model handling, Strix Halo’s massive GPU memory pool that can scale up to 96GB is obviously much more capacious for memory-hungry LLMs, versus the 24GB frame buffer of Nvidia’s previous gen gaming monster.

Ryzen AI Max Performance Claims Were Bold

With all this firepower in a single chip solution, AMD wasted no time pointing out Ryzen AI Max’s projected performance, with some specific benchmark numbers of its own. There were two high-level compares made here, versus Intel and Apple.

AMD touts a complete multi-threaded dusting of Intel’s highest-end Core Ultra 9 288V Lunar Lake laptop platform, when it comes to 3D rendering workloads like Blender, Cinebench, and VRay, claiming Strix Halo is some 2.6X faster. And when it comes to gaming graphics, the 40 CU Radeon engine on-board Ryzen AI Max+ 395 is claimed to be 1.4X faster on average, though I’ll point out that these are all synthetic benchmarks being listed here, versus actual game titles. It will be interesting to see how gaming performance pans out in triple-A titles as well.

The final compare was cross-platform, which is a little more difficult, versus Apple’s 12 and 14-core M4 Pro chips. Here AMD is showing close to parity performance in the Cinebench render but up to 86% faster in multithreaded rendering workloads like Vray for photorealistic ray-traced rendering.

Forthcoming AMD Ryzen AI Max Products And The Look-Ahead

Regardless, the proof of AMD’s claims will bare out once systems built in its Ryzen AI Max series ship in the Spring of this year. Specifically, I’ve hand hands on at least two of the systems in AMD’s initial unveil, though there should likely be more announced in the days ahead. AMD also announced a Ryzen AI Max Pro offering as well, signalizing the company’s support for advanced security and fleet management features that are prerequisites in the enterprise.

It’s important to note the relative positioning of this new AMD platform in the market, however. AMD obviously spent some serious engineering resources getting this product and platform out the door, and the way it’s architected is basically a no compromises approach — power-efficiency, ultra-performance, and a high level of integration. As such, the debut products above here from HP and ASUS are very much workstation class or ultra-premium consumer class machines, and they likely won’t be cheap. Specifically, the HP Z2 Mini G1a looks like a Mac Mini killer, with its pint-sized profile, while the ZBook Ultra G1a is a 14-inch clamshell laptop that beckons content creation types that need serious rendering and AI chops in their bag. And the ASUS ROG Flow Z13 is likely going to be a pretty wild hybrid detachable machine that can competently game as well.

AMD also had many other desktop and laptop offerings announced at CES this year, including exciting new processors for enthusiast gamers with new 3D V-Cache infused chips, along with its new Radeon RX 9070 series of graphics cards. Regardless, I really have to hand it to AMD for delivering something truly unique in the market for Windows Copilot+ PCs with Ryzen AI Max, and I look forward to taking one of these machines around our test track in the weeks ahead.