AMD Pushes The Performance Envelope With Zen 5 At Computex 2024

At Computex 2024 there was a bevy of announcements from the world’s leading chip companies, including AMD. The company used its Computex keynote to talk up its AI capabilities, from the datacenter to the consumer. For AMD, it was especially important on the consumer side to show how the company’s processors are still competitive in this age of AI on the PC.

So far, AMD has struggled to claim ownership of the AI PC in ways that Qualcomm and Intel have, which I believe is a consequence of the company’s focus on AI in the datacenter. Intel’s AI PC story has predominantly been about scale, while Qualcomm’s has been about the NPU and overall efficiency, factors that have been supported by Microsoft’s push to launch Copilot+ first on Qualcomm’s Arm-based processors.

AMD is still considered one of Microsoft’s key chip partners for Copilot+ PCs, but until Computex AMD didn’t really have any processors to show that would even qualify for the Copilot+ specs that Microsoft has defined. While AMD did announce the Ryzen AI 300 series at Computex, it also formally announced the Zen 5 family of CPUs that power it. I believe that AMD’s focus right now is on performance leadership in all categories, which includes data center CPUs and GPUs and consumer CPUs and GPUs as well as AI.

AMD Ryzen AI 300 Aims For Maximum Performance

AMD calls the Ryzen AI 300 series its third generation of AI PC processors, following the Ryzen 7040 and 8040 processors it already has on the market. AMD says that the Ryzen AI 300 series is built for Copilot+ experiences thanks to its 50 TOPS NPU, which exceeds Microsoft’s minimum requirement of 40 TOPS for Copilot+ PCs. The new NPU is AMD’s second generation of its XDNA architecture, which it inherited from its acquisition of Xilinx. AMD is claiming up to 5x improved AI compute capacity and up to 2x better power efficiency than the first generation of XDNA. AMD also claims that this series has superior NPU performance compared to Qualcomm’s Snapdragon X series, but did not talk about efficiency comparisons.

There are two Ryzen AI 300 series parts, the Ryzen AI 9 HX 370 and Ryzen AI 9 365. These two very similar parts have minor specification differences but effectively the same TDPs in a range from 15 to 54 watts. The HX 370 is a 12-Core Zen 5 CPU clocked at 5.1 GHz while the 365 is a 10-Core Zen 5 CPU clocked at 5 GHz. There is also a 2MB difference in CPU cache and a slight difference in GPU performance. These are both powered by AMD’s RDNA 3.5, which seems like a new half-step in AMD’s GPU architecture that it hasn’t traditionally acknowledged before.

In the comparisons it made from stage at Computex, AMD went directly after Apple, Intel and Qualcomm processors, with a special focus on Intel’s Core Ultra 185H, Qualcomm’s Snapdragon X Elite and Apple’s M3. Against Qualcomm, AMD claimed an advantage of 5% in GeekBench single-thread benchmarks all the way up to 60% for graphics performance. Against Apple’s M3, it claimed 9% better productivity in Procyon and 98% better 3-D rendering in Blender. Against Intel, AMD claimed 4% better productivity in Procyon, 47% better Cinebench Multitasking and 73% better rendering on Blender. I believe that these improvements are powerful and that they help AMD continue to tell the performance story it seems to be pushing, but I am concerned that if it wants to compete with Snapdragon X Elite and Intel Lunar Lake, AMD will also need to tell an efficiency story with the Ryzen AI family, which is sorely missing so far.

AMD claims that it already has 100-plus PC models carrying the Ryzen AI 300 series that will be available from most of the major OEMs starting this month. Dell is surprisingly missing from this list. I saw some of these systems in person at Computex and was especially impressed with what AMD was able to achieve in partnership with ASUS on the Zenbook S 16, which is one of the most impressive notebooks in terms of thinness and weight that I have seen in years.

AMD Ryzen 9 9000 Series Aims For Desktop Superiority

With the new Ryzen 9 9000 Series, AMD has committed to continuing to use the AM5 socket platform. That doesn’t mean AMD won’t introduce a new X870 motherboard chipset for AM5, but it does mean, first, that there is a lot of flexibility for gamers to decide which motherboards they use and, second, that upgrade paths are not tied to the socket. Furthermore, AMD committed at Computex to releasing products with the AM5 socket until 2027; this might not mean that we’ll continue to see flagship processors with the follow-up to the 9000 series, but that AMD will release new processors for the AM5 to keep it fresh and accessible to more audiences. With that decision comes support for PCIe 5 and DDR5, as well as enhanced support for PCIe 5 storage on X870/E.

The Ryzen 9000 series also follows the example of the Ryzen AI 300 series by using Zen 5 CPU cores, which AMD says bring up to 2x better AI performance. For a broad set of workloads, AMD is also claiming a geomean IPC uplift of 16%, which is especially impressive when you consider that the company has brought double-digit IPC improvements to every generation. AMD’s new 9000 series ranges from six cores up to 16 Cores and offers four different nicely configured products, with the Ryzen 9700X looking like the best performance-per-watt story and the 9950X being the pinnacle for performance. That said, the company didn’t introduce any X3D SKUs with 3-D memory for enhanced caching for gaming, which means those products will probably come later in the year. AMD was able to keep TDPs roughly the same as the predecessor 7000 series while still increasing performance. Clock speeds are also roughly the same.

While AMD didn’t make many comparisons to its own Ryzen 7000 series, it did compare the Ryzen 9000 series against the Intel 14900K, where it claims better performance anywhere from 7% to 56% for productivity and content creation. It also claims better game performance in certain titles ranging from 4% in Borderlands 3 up to 23% in Horizon Zero Dawn.

One of the benefits that AMD is touting with the new X870 family of motherboards is USB4 support standard across all models. This is an upgrade over USB 3.2 Gen 2 (10 Gbps) that brings a 4x improvement in bandwidth. Additionally, AMD is no longer differentiating between the two tiers of motherboard chipsets when it comes to PCIe 5 SSD support, as it did between the X670 and X670E. I believe that this is a net positive for consumers and will result in a better experience for gamers with a limited budget.

The X870 series will also come with support for higher memory clock speeds using AMD’s EXPO profiles, which is another way for AMD to encourage high-end gamers to make the switch. That all said, the new Ryzen 9000 series will still likely be a drop-in upgrade for anyone using a high-end X670E motherboard already capable of supporting a 170-watt TDP CPU. The motherboard will likely require a BIOS update beforehand, but this is still one more way that AMD’s socket compatibility enables flexible upgrade paths for gamers.

In addition to the new Zen 5 CPUs, AMD is also introducing the Ryzen 9 5900XT and 5800XT for AM4 to satisfy markets that are still firmly using AM4. These tend to be more developing markets such as Latin America, where the cost profile of AM4 and DDR4 are much more attractive than forcing users to switch to AM5 and DDR5. These markets are also much less likely to benefit from the added bandwidth of PCIe Gen 5, which is mostly for high-end graphics cards and flagship NVMe SSDs. These processors will join the rest of the new Ryzen processors with July availability.

More (And Better?) To Come

As we watch AMD launch all these products, it is quite apparent that AMD has so far continued to focus on the performance aspect of its product line with XDNA2 NPUs and Zen 5 CPUs—at the expense of an efficiency story. I believe that AMD will continue to keep its performance leadership with the Ryzen 9000 series in desktops but will have a more complex story to tell with the Ryzen AI 300 series. AMD may be able to claim performance leadership with the Ryzen AI 300 series, but when these are deployed in laptops and it is comparing itself to Apple, Intel and Qualcomm, AMD will need to tell a much more robust power efficiency story.

Right now, AMD’s performance story is fairly credible, but one has to wonder what the cost of that performance is in terms of battery life. Without good answers to those questions, some (potential) AMD customers might remain unsatisfied.

That said, AMD did rack up a lot of gaming design advances at Computex; gaming is an area in which power efficiency is less of a concern, but that will not help AMD compete against Intel and the Arm ecosystem more broadly. We’ll have to wait until Ryzen AI 300-based systems can be benchmarked independently to give us a better idea of how AMD stacks up on the efficiency front.