AMD Claims Strix Halo's iGPU Memory Bandwidth Is Same As Mobile RTX 4070

hero amd ryzen ai max series render
In a recent interview, with Notebookcheck, AMD's Ben Conrad made a bold claim: Strix Halo's integrated GPU offers memory bandwidth equivalent to NVIDIA's GeForce RTX 4070 mobile GPU. Well, maybe it's not that bold, really; both data points are known, so it's easy to compare. Let's do that.

The Ryzen AI MAX, codenamed "Strix Halo," is a powerhouse processor. It pairs up to sixteen desktop-class Zen 5 CPU cores with a staggering 40 RDNA 3.5 compute units. That's more than some discrete graphics cards, like the 32-CU Radeon RX 7600. To back up this level of performance, Strix Halo uses a 256-bit LPDDR5X memory interface, delivering up to 256 GB/sec of memory bandwidth, or up to 275 GB/sec with the fastest-supported 8533 MT/s LPDDR5X RAM.

legion rtx4070 laptop
This Lenovo Legion Slim 5 with a mobile RTX 4070 is quite a bit thicker than a Strix Halo system.

For comparison, the mobile GeForce RTX 4070 has a 128-bit memory bus—half the width, but with double-speed 16 GT/sec GDDR6 memory. So saying, it achieves a similar level of memory throughput, just as AMD said. Of course, it's worth noting that the mobile GeForce RTX 4070 is more like a desktop GeForce RTX 4060 Ti, given that both are based on the AD106 GPU.

z2 mini lifestyle
HP's Z2 Mini workstation featuring Strix Halo seems very promising for the size.

This level of bandwidth is unprecedented for an integrated GPU, and it sets Strix Halo apart as a serious contender for high-performance laptops and mini-PCs. The platform also integrates 32 MB of Infinity Cache to further boost GPU performance, particularly in gaming and compute-heavy tasks. While discrete GPUs remain dominant in raw power, Strix Halo's iGPU is shaping up to be one of the most capable solutions ever seen in a single package.

However, Strix Halo isn't without its trade-offs. Unlike AMD's previous mobile platforms, the Ryzen AI MAX series doesn't support discrete GPUs at all. Furthermore, external connectivity is limited to just twelve PCI Express Gen 4 lanes, which underscores that this platform is designed to rely entirely on its integrated GPU.

strix halo copilot
AMD claims that Strix Halo can be 87% more efficient than a GeForce RTX 4090 at AI inference.

Strix Halo's versatility makes it an attractive choice for more than just gaming, though. With the ability to allocate up to 96 GB of system memory to the GPU, it's highly suited for AI and machine learning workloads. This positions AMD's new processors as a competitor not only to Apple's high-end Mx-series chips but also in fact to NVIDIA's GPUs, which can be constrained by memory capacity in AI workloads.

How all of this translates to real-world performance remains to be seen, but we won't have to wait long: our full review of a Strix Halo-powered tablet is coming next month. Meanwhile, Ben Conrad had many other interesting things to say about both the Ryzen AI MAX 300 series processors as well as the larger Ryzen AI 300 family and the Ryzen AI 200 chips, as well. Head over to Notebookcheck if you'd like to read the full interview for those details.