Hawk Point with Zen 4 and Ryzen AI

Although CES 2024 is happening next month in Las Vegas, AMD has lifted the lid on their latest Zen 4 mobile processors, the Ryzen Mobile 8040 series. Codenamed Hawk Point, the Ryzen Mobile 8040HS series is set to be the successor to this year’s ‘Phoenix’ mobile chips. Two new families of Ryzen Mobile chips have been announced, including the mid-ranged 8040HS APUs and the lower-powered 8040U series with a lower TDP designed for thin and light notebooks.

Ranging from 8C/16T down to an entry-level 4C/8T SKU, AMD’s Ryzen Mobile 8040 series has nine SKUs on offer (at the time of writing) and is a direct refresh of the previous Phoenix 7040 series rather than a complete overhaul. This means AMD’s Zen 4 core manufactured on TSMC’s 4 nm process looks set to retake center stage in their notebook portfolio for at least the first half of next year. Another interesting element of AMD’s Ryzen Mobile 8040 series is that they incorporate the Ryzen AI NPU, developed by Xilinx, into seven of the nine announced SKUs. AMD acquired Xilinix last year and looks set to use Xilinx IP again to bolster AMD’s charge towards AI generation.

During last year’s CES 2023 trade show in Las Vegas, AMD announced their first Zen 4-based mobile SKUs, namely the Ryzen Mobile 7040 series. While users did have to wait a considerable portion of the year before these SKUs hit retail via notebook partners, they delivered solid portable performance and efficiency, something which AMD’s Zen 4 microarchitecture has delivered in desktop, mobile, server and, more recently, in the workstation space. The Ryzen 8040HS and 8040U series look to further push the adoption of AI features with more on-chip Ryzen AI-enabled NPUs than the previous generation.

Just quickly recapping AMD’s product number naming scheme for 2023 to 2025, it was announced at the beginning of the year, and as such, it is now in full swing. Keeping in line with the previous Ryzen Mobile 7000 series announcement in January, we can see AMD has kept within its naming structure.

With three classes of new Ryzen Mobile 8000 series SKUs, we have two primary product types: the HS (between 35-54W TDP) and the U (15-30W) series. The main difference between the two families is the TDP, with the HS series acting as the mid-range chips with sufficient power to deliver processing power in various workloads. In contrast, the U series is designed more for ultra-thin and lower-powered notebooks at just 30 W at the top end of the power spectrum. Dissecting the model scheme further and using the Ryzen 9 8945HS as an example, the 8 signifies a 2024 model, with the 9 indicating a Ryzen 9 part. The 4 shows the chip uses Zen 4-based silicon (TSMC 4nm), and the 5 means it’s a higher-performance chip representing AMD’s positioning in their higher segmentation.

AMD Ryzen 8040HS Mobile: Phoenix Die Refreshed, Now Hawk Point For 2024

AMD’s drive looks to push on-chip AI capabilities in mobile right from the outset of 2024, but Hawk Point isn’t new silicon, nor is it a new architectural design. Instead, they have opted to directly refresh their existing ‘Phoenix’ APUs and repackage them. Using the same TSMC-built 4 nm Zen 4 cores, which is AMD’s current flagship mobile CPU architecture, at least for the foreseeable future, the Ryzen 8040HS series has five new additions to an ever-growing mobile product lineup. This includes the top SKU, the Ryzen 9 8945HS with 8 x Zen 4 cores, the slightly slower Ryzen 7 8845HS with 8 Zen 4 cores, and the Ryzen 5 8645HS with six Zen 4 cores. This also includes two lower-powered 28 W additions, the 8C/16T Ryzen 7 8840HS and the 6C/12T Ryzen 5 8640HS.

Looking at the specifications, starting with the Ryzen 9 8945HS, it has a base frequency of 4 GHz, with turbo frequencies of up to 5.2 GHz. Compared to its predecessor, the Ryzen 9 7940HS, no difference exists in any of the specifications, not at least compared to what AMD has provided us. Everything down to the core/turbo frequencies, core count, and the 16 MB of L3 cache remains identical. AMD uses the same Radeon 780M we’ve previously seen in the 2023 Phoenix lineup, which both share the same 12 x RDNA 3 CUs and 2.8 GHz graphics frequencies.

The same can be said about the AMD Ryzen 7 8845HS, with 8C/16T of 4 nm Zen 4 cores and a base frequency of 3.8 GHz, topping out at 5.1 GHz at turbo frequencies. This aligns with the previous Ryzen 7 7840HS, which, again, has the exact specifications right down to the ground. Looking at the last of the three Ryzen Mobile 8045HS series SKUs, the Ryzen 5 8645HS has 6C/12T, 16 MB of integrated L3 cache, with a 4.3 GHz base frequency, can hit up to 5.0 GHz at turbo clock speeds, and has slightly lower spec integrated graphics than the other two chips, with just 8 x RDNA 3 CUs clocked to 2.6 GHz. These are the same specifications as the previous Ryzen 5 7640HS, which not only denotes that these three chips are an identical refresh of the Phoenix variants, but they are essentially the same barring changes to the Ryzen AI NPU.

There are also two new low-powered additions to the HS series, the Ryzen 7 8840HS is an 8C/16T part with a base frequency of 3.3 GHz and a turbo frequency of up to 5.1 GHz. It has 16 MB of L3 cache, with 12 x RDNA 3 CUs clocked to 2.7 GHz, all within a TDP of 28 W. The second of these is the Ryzen 5 8640HS, which includes 6C/12T of Zen 4, a 3.5 GHz base frequency, can boost up to 4.9 GHz, and has slightly lesser integrated graphics with 8 x RDNA 3 CUs at a frequency of 2.6 GHz. The Ryzen 5 8640HS also has 16 MB of L3 cache, with a TDP of 28 W, which is more akin to the U-series chips than it is to the HS series.

All five of AMD’s Ryzen Mobile 8040HS series SKUs include AMD’s XDNA-based NPU designed for AI inferencing, which is more widely known as ‘Ryzen AI.’ It’s worth noting that technically, all of AMD’s Ryzen 7040 series (Phoenix) CPUs had the NPU there physically, but AMD didn’t enable it in most of them. The NPU itself isn’t new hardware, it uses Xilinx IP, which AMD now owns due to their acquisition of Xilinix last year. AMD looks to be pushing AI capabilities directly into the silicon as vendors, manufacturers, software developers, and the world adopt the valuable benefits and abilities AI can bring in the future.

While the core fundamental reasoning for these chips signifies a refresh, there’s more to this than meets the eye. The first comes through the AI performance, with AMD promising up to 1.4x the performance using the Llama 2 and Vision models compared to the previous Ryzen 7040 series chips. Pivotal to the performance increases in Llama 2; this model uses PyTorch in Eager mode, ONNX Runtime, and the ONNX performance tool quantized to INT8. AMD has boosted the frequency by 60% on the NPU within the 8040 silicon compared to the 7040 series, as AMD is claiming up to 16 TOPS of performance from the NPU on the Ryzen 8040 series compared to the 10 TOPS on the 7040 series.

This shows that AMD used a more conservative clock speed on the NPU for the Ryzen 7000 series mobile chips, and this could translate directly into new frequency tables for 8040. Upping the NPU frequency while keeping the same core frequencies on the Zen 4 cores could lead to efficiency issues, although AMD hasn’t disclosed this information.

The other difference comes under the hood of the architectural nature of the Zen 4 cores, including things such as firmware updates, bug fixes, and minor optimizations. While AMD hasn’t explicitly stated what bugs have been fixed and what things have been optimized, we can see that the XDNA-based Ryzen AI NPU has undoubtedly been upgraded. In the above slide from the Ryzen 8040 series slide deck, there are a lot of integrated experiences to be gained from on-chip AI, including many Adobe implementations within Photoshop, Premiere Pro, Lightroom, and After Effects.

There are also plenty of features within Black Magic DaVinci Resolve, including auto-balancing, DNN-based deinterlacing, and AI-driven voice isolation features. We expect to see many more as software developers adopt AI inferencing in their software, which should hopefully be driven by silicon-level AI implementations and engines within hardware to help things move forward. AMD seemingly aligns with what Microsoft is optimizing for in regards to AI, especially with its optimizations for software, including ONNX Runtime and PyTorch.

AMD Ryzen 8040U: 5 New 28 W SKUs, All On Zen 4 (4 nm) for 2024

Also announced by AMD isn’t so much a refresh of the previous Ryzen Mobile 7035 (Zen 3) series based on their Rembrandt Refresh, which was based on TSMC’s 6 nm node. This family consisted of two distinct series in HS and U flavors. For 2024, AMD has decided to go the entire Zen 4 route for both of these families for this generation, which means all of today’s newly announced chips are based on Zen 4 and built using TSMC’s 4 nm node, which effectively makes the entire series part of the new ‘Hawk Point’ family.

The last of AMD’s Ryzen Mobile 8040 series SKUs to be announced is the lower-powered 28 W U-series, with four new U-series chips ranging from 4C/8T to an 8C/16T model. Opening with the top SKU in the U-series, the Ryzen 7 8840U has 8C/16T of Zen 4 (4 nm), with a 3.3 GHz base clock and turbo frequencies up to 5.1 GHz. It shares the same 12 x RDNA 3 CUs and 2.7 GHz graphics frequency as the Ryzen 7 8845HS but with a lower 28 W TDP. The Ryzen 7 8840U also includes AMD’s Ryzen AI XDNA block within the silicon, giving it on-chip AI inferencing capabilities.

The Ryzen 5 8640U and Ryzen 5 8540U both have 6C/12T. However, there are obvious differences between the two. The Ryzen 5 8640U is a faster chip with a 3.5 GHz base clock and has 8 x RDNA 3 CUs at 2.6 GHz. In contrast, the Ryzen 5 8540U has a slightly lower base frequency of 3.2 GHz, with just 4 x RDNA 3 CUs clocked at 2.8 GHz. Both chips have boost frequencies of up to 4.9 GHz, as well as both have the same 16 MB of L3 cache. What primarily sets both chips apart is that the Ryzen 5 8640U includes Ryzen AI brought to the table by AMD’s acquisition of Xilinx, which offers integrated AI capabilities. The Ryzen 5 8540U does not include this, although both chips have a TDP of just 28 W.

Lastly, the AMD Ryzen 3 8440U is the entry-level and, ultimately, is the lowest spec chip in all of today’s announcements. It has 4C/8T, with just 8 MB of L3 cache, which is half the amount of the other chips announced today. The Ryzen 3 8440U has a base frequency of 3.0 GHz, with turbo clock speeds of up to 4.7 GHz, with 4 x RDNA 3 CUs clocked at 2.5 GHz. It is also one of two chips announced today that doesn’t include the XDNA Ryzen AI block integrated within the silicon, as, like its predecessor, it’s based on the NPU-less Phoenix 2 silicon.

AMD hasn’t provided a specific launch date for the Ryzen 8040 mobile series of processors, although AMD is currently shipping to OEMs, and it’s expected that they will be launched sometime in Q1 2024. We expect AMD to make a broader announcement, possibly with more details attached, either just before or during CES 2024, which is set to take place in Las Vegas from the 9th to the 12th of January 2024.