AMD announced its Ryzen 8040 series of mobile processors, codenamed “Hawk Point,” here at its Advancing AI event in San Jose, California, claiming the new processors deliver up to a 1.4X increase in some generative AI workloads. AMD also teased its future “Strix Point” processors that will arrive in 2024 with more than three times the AI performance of their predecessors, which comes courtesy of a new and improved XDNA 2 NPU engine. AMD also revealed its new Ryzen AI Software, a single-click application that allows enthusiasts and developers to deploy pre-trained AI models that work with the XDNA AI engine.
The new Hawk Point lineup is the first to come with Ryzen 8000 branding, but the chips feature the same 5nm Zen 4 cores, RDNA 3 graphics, and XDNA AI engine as the existing Ryzen 7040 “Phoenix” processors. As we typically see with refresh generations, AMD has introduced these models to give OEMs a new lineup of revamped processors for updated laptop designs. However, aside from a few models that have reduced specs, many of the new Ryzen 8040 processors appear to be nearly carbon copies of their predecessors — core counts and frequencies remain the same on most of the lineup.
That makes for what appears on the surface to be an underwhelming launch — even considering the typical lackluster improvements we expect from a refresh generation. Still, AMD has improved the performance of its in-built AI engine, and it also has plenty of other exciting developments in flight. AMD says the Ryzen 8040 series is already shipping to its partners, with devices expected to be on the market in the first quarter of next year. Let’s dive in.
AMD teases Strix Point
AMD’s previous-gen Ryzen 7040 series was the first to feature an NPU (Neural Processing Unit) for AI workloads. The NPU is a low-power dedicated accelerator that resides on-die with the CPU cores, consisting of the XDNA AI engine. This engine is designed to execute lower-intensity AI inference workloads, like photo, audio, and video processing, at lower power than the CPU or GPU cores using INT8 instructions. It can deliver faster response times than online services, thus boosting performance and saving battery power.
AMD teased Strix Point, its next-generation processor lineup that will arrive later in 2024. The company says these processors will feature a next-gen XDNA 2 engine that will deliver up to 3X the performance of its first-gen XDNA 1 NPU. AMD also provided performance claims for XDNA 1, saying the NPU alone delivers 10 TOPS (teraops INT8) of performance in the Phoenix 7040 series, and that increases to 16 TOPS in the Hawk Point 8040 series. AMD didn’t divulge the test methodology it used for these metrics in its deck, but we do know that the total TOPS metric in the slide represents the CPU, GPU, and NPU all working in concert.
Ryzen Hawk Point 8040 mobile processors
The 8040 series continues to focus heavily on integrated NPUs, with all but two of the lowest-end models equipped with the engine. AMD points to improved NPU performance as one of the key selling points for the 8040 series, saying the XDNA engine delivered 10 TOPS of performance in the 7040 models but improved to 16 TOPS in the new 8040 series. AMD cites increased NPU frequencies and efficiency as a source of the improved performance. You’ll have to take AMD at its word, though, as it doesn’t publish the NPU clock speeds for either the inaugural 7040 lineup or the new 8040 series, so we can’t make direct comparisons.
AMD has reshuffled its product branding to adhere to its new naming convention, with three of the “HS” models now having a ‘5’ in the product name to identify them as upper-tier models within the HS product stack. Yes, it’s more than a bit confusing.
The eight-core 16-thread Ryzen 9 8945HS has the same 45W TDP as its direct predecessor, the 7940HS, and it also has the same peak 5.2 GHz boost and 4.0 GHz base frequency. All the other specs appear identical as well, which also carries over to the 8845HS and 8645HS — we see no significant changes other than the new product naming to merit the step up to Ryzen 8000 branding. We’ve asked AMD for more details about what exactly precipitated the move to denote these as Ryzen 8000 models, and the company indicated that improved AI performance is the primary motivation; this refresh is designed primarily to boost performance in AI workloads.
We spotted a few down-clocked models in the lineup as well. These two Ryzen 8x40HS models are 28W processors designed for ultrathin laptops, but their direct predecessors used to fit into the 45W swim lane. As a result, we see the ‘0’ in the product name to denote these are now lower-tier models within the HS lineup, and we also see the reduced specs that reflect AMD’s tweaks to enable the lower power budget.
The Ryzen 7 7840HS has the same 5.1 GHz peak boost clock as its predecessor, the 7840HS, but AMD reduced the base clock by 500 MHz to accommodate the new 28W TDP. Meanwhile, the Ryzen 5 8640HS gets a substantial 800 MHz reduction in its base clock compared to the previous-gen Ryzen 5 7640HS, and AMD also shaved 100 MHz off its boost frequency.
The U-series Ryzen 5 and 7 models also slot into the 28W category. These four processors all feature the same CPU and GPU clock speeds as their predecessors, except for a 300 MHz clock rate increase for the Radeon 760M graphics engine in the Ryzen 5 8540U. Aside from AMD’s unspecified changes to NPU frequencies, these also appear to largely be the same processors as their predecessors, but with new branding.
Notably, the 'new' Ryzen 5 8540U and the Ryzen 3 8440U are amazingly fast refreshes of the previously-announced Ryzen 5 7545U and the Ryzen 3 7400U, both of which employ AMD's Zen 4c cores. The 8540U comes with 2 standard Zen 4 cores and four Zen 4c cores, whereas the Ryzen 3 8440U comes with 1 standard Zen 4 core and three Zen 4c cores.
Ryzen 8000 Series performance
As always, we should take vendor performance claims with a grain of salt. We’ve included the test setup notes at the end of the above album.
AMD provided the above benchmarks, highlighting a 1.4X increase in the Llama 2 and Vision Maker generative AI benchmarks over the Ryzen 7040 series. Again, AMD cites improved NPU frequencies and efficiency as the source of these improvements, but it hasn’t provided the clock rates for the NPU.
AMD didn’t provide gen-on-gen comparisons for gaming and productivity workloads, which isn’t surprising given that the Ryzen 8000 models are either the same as the prior-gen or have reduced specs. Here, we can see AMD pitting what it lists as the Ryzen 9 8940H (this SKU doesn’t exist, but we think it's supposed to be the 8945HS) and its Radeon 780M integrated graphics against the Intel Core i9-13900H with integrated Xe graphics.
AMD ran nine different gaming benchmarks, including titles like Borderlands 3, Far Cry 6, and Hitman 3, among others, to derive one metric to quantify gaming performance. Overall, AMD says its processor is 1.8X faster than the Core i9-13900H in 1080p gaming with low fidelity settings. AMD also claims a 1.1X improvement over the 13900H in the Cinebench R23 and Geekbench 6 multi-threaded benchmarks. Finally, AMD says its silicon is up to 1.4X faster than the 13900H in a spate of content creation workloads, including Blender, POV-Ray, and PCMark 10.
AMD Ryzen AI Software
The age of AI is upon us, and AMD has already shipped over a million processors with an in-built XDNA NPU to enable the software ecosystem. However, deploying AI models for local use, which confers performance, security, cost, and efficiency benefits, can be a daunting task.
AMD’s Ryzen AI Software suite is designed to allow both enthusiasts and developers to deploy pre-trained AI models on its silicon with a one-click approach that greatly simplifies the process. Users can simply select machine learning models trained on frameworks like PyTorch or TensorFlow and use AMD’s Vitis AI quantizer to quantize the model into an ONNX format. The software then partitions and compiles the model, which then runs on the Ryzen NPU.
The Ryzen AI Software is available now for free, and AMD also has a pre-optimized model zoo on Hugging Face available for users. The software only works on Windows for now, but we’re told that a Linux version will be made available in the coming quarters.
AMD also recently announced its Pervasive AI contest that provides developers with hardware and then issues cash prizes of up to $10,000 for developers that develop winning AI applications for robotics, generative AI, and PC AI platforms.
AMD outlined its approach to broadening the ecosystem of applications that leverage AI. As we’ve covered before, AMD’s general strategy has been to enable the hardware first and get it into the market, to the tune of over 1 million XDNA-enabled products so far, thus providing developers with both the motivation and the hardware to develop new applications.
The next step is perhaps the most critical — enabling the development community to leverage the in-built AI acceleration. AMD already has a host of top-tier partners working on leveraging the NPU, like Adobe, Microsoft (there are rumors that Copilot will leverage the NPU in the future), and OBS studio, among many others. Those efforts will eventually result in more applications that fully leverage the NPU, joining over 100 AI-driven applications already present for PC users.
