AMD’s AI capabilities will continue to increase in 2024.
AMD unveiled the Ryzen 8040 series of laptop processors during the company’s AI-themed event, shifting the narrative from discussions centered on CPU speed, power, and battery life to one that emphasizes AI. In January, AMD introduced the Ryzen 7000 series (fave.co/3WnH8ev). Among this lineup, the Ryzen 7040 marked the debut of AMD’s XDNA architecture, which played a key role in powering Ryzen AI.
AMD’s latest Ryzen 8000 laptop CPUs are designed with an AI-oriented future in mind
When Intel revealed its Meteor Lake processor [fave.co/3PUqbHI] last summer, they started calling the Al accelerator an NPU, a name that has since become widely adopted. Over 50 laptop models currently come equipped with Ryzen Al, as mentioned by executives. The focus now shifts to AMD’s upcoming NPU, Hawk Point, featured in the Ryzen 8040. In AMD’s scenario, the XDNA NPU collaborates with the Zen CPU, while the graphics are driven by the Radeon RDNA architecture of the GPU.
All three components operate seamlessly together, contributing to the overall performance. Dr. Lisa Su, AMD’s CEO, expressed, “We consider AI to be the most revolutionary technology of the past decade,” as she initiated AMD’s presentation on advancing AI Currently, the battle is unfolding on multiple fronts. While Microsoft and Google advocate for AI computation in the cloud, major chip companies are arguing for local processing on PCs.
This involves identifying applications that can leverage the local AI processing capabilities and collaborating with software developers to create apps tailored for specific processors.
In essence, AMD and its competitors are obligated to furnish software tools that facilitate communication between applications and their respective chips. Naturally, AMD, Intel, and Qualcomm aspire for optimal performance of these applications on their own silicon. Consequently, these chip companies find themselves competing on two distinct fronts: not only must they develop the most potent AI silicon, but they must also guarantee that app developers can efficiently code for their chips.
For years, silicon manufacturers have attempted to persuade game developers to embrace similar practices. While AI might appear complex from an external perspective, familiar concepts can be discerned. For instance, quantization can be viewed as a type of data compression, enabling large language models designed for robust server processors to simplify their complexity and function on local processors such as the Ryzen 8000 series. Tools of this nature play a crucial role in ensuring the success of “local AI.”
Introducing the Eight Ryzen 8040 Mobile Processors.
AMD is currently unveiling the Ryzen 8040 series, and these chips are expected to make their way into laptops starting next year. The Ryzen 8040 series integrates AMD’s Zen 4 architecture, RDNA 3 GPUs, and the original XDNA architecture. Notably, these new chips incorporate AMD’s second NPU, Hawk Point, marking an improvement over the first NPU, Phoenix, with increased trillions of operations per second (TOPS) from 10 to 16 TOPS, contributing to a total of 39 TOPS when combined with CPU and GPU capabilities.
Donny Woligroski, senior mobile processor technical marketing manager, explained that the CPU utilizes AVX-512 VNNI instructions to execute lightweight AI functions efficiently. While AI can also run on the GPU, it tends to consume power at a higher and less efficient level—a perspective shared by Intel as well. Woligroski emphasized the importance of efficiency, stating,
“When it comes to efficiency, raw performance isn’t enough. You’ve got to be able to run this stuff in a laptop.” The Ryzen 8040 series comprises nine members, with the seven most powerful featuring the Hawk Point NPU.
The specifications range from 8 cores/16 threads with a top-end boost clock of 5.2GHz, to 4 cores/8 threads with a clock speed of 4.7GHz at the lower end. The Thermal Design Powers (TDPs) span from a minimum of 15W for certain processors, reaching up to 35W for high-end models and extending further to 54
There are nine new chips, and they consist of:
1. Ryzen 9 8945HS 8 cores/16 threads, 5.2GHz (boost); Radeon 780M graphics, 35-54W
2. Ryzen 7 8845HS 8 cores/16 threads, 5.1GHz (boost); Radeon 780M graphics, 35-54W
3. Ryzen 7 8840HS 8 cores/16 threads, 5.1GHz (boost) Radeon 780M graphics, 20-30W
4. Ryzen 7 8840U 8 cores/16 threads, 5.1GHz (boost) Radeon 780M graphics, 15-30W
5. Ryzen 5 8645HS : 6 cores/12 threads, 5.0GHz (boost) Radeon 760M graphics, 35-54W
6. Ryzen 5 8640HS 6 cores/12 threads, 4.9GHz (boost) Radeon 760M graphics, 20-30W
7. Ryzen 5 8640U : 6 cores/12 threads, 5.1GHz (boost) Radeon 760M graphics, 20-30W
8. Ryzen 5 8540U 6 cores/12 threads, 4.9GHz (boost) Radeon 740M graphics, 15-30W
9. Ryzen 3 8440U : 4 cores/8 threads, 4.7GHz (boost) Radeon 740M graphics, 15-30W
Using AMD’s model number “decoder ring” (fave.co/3pUA8Lu), which Intel recently criticized as “snake oil” (fave.co/47MmWcD), all the new processors are built on the Zen 4 architecture and are set to be available in laptops in 2024. AMD provides three integrated GPUs—the 780M (12 cores, up to 2.7GHz), the 760M (8 cores, up to 2.6GHz), and the 740M (4 cores, up to 2.5GHz)—based on the RDNA3 graphics architecture with DDR5/LPDDR5 support.
These integrated GPUs were featured in the Ryzen 7040 mobile chips (fave.co/3X3tCPc) that were introduced earlier this year, showcasing the Phoenix NPU. Notably, AMD has not yet revealed any HX parts specifically designed for premium gaming. Additionally, AMD is unveiling a third-gen NPU called Strix Point, expected to be shipped later in 2024, likely within the next-generation Ryzen processor.
Although AMD has not disclosed specific Strix Point specs, it has stated that it will provide more than three times the generative AI performance compared to the prior generation.
Performance of the Ryzen 8040 AMD included some standard benchmark assessments, comparing the 8940H to the Intel Core i9-13900H at 1080p on low settings, asserting that its chip outperforms Intel’s by 1.8 times. (AMD utilized nine games for a 2025 comparison without explicitly detailing the methodology.) AMD asserts a 1.4 times performance enhancement on the same chips, combining Cinebench R23 and Geekbench 6 results.
Furthermore, AMD stated a 1.4 times improvement in AI performance generation-over-generation on Facebook’s Llama 2 LLM and “vision models,” drawing comparisons between the 7940HS and 8840HS. Ryzen AI Software: A Fresh Tool for Enthusiasts of Artificial Intelligence Given the early stage of AI development, silicon manufacturers lack numerous benchmarks for traditional assessments.
AMD executives emphasized the significance of localized AI-driven experiences, citing examples like the neural filters in Photoshop, masking tools in Lightroom, and various features in BlackMagic’s DaVinci Resolve.
The underlying message underscores the importance of local AI processing as opposed to relying on cloud-based solutions. AMD is introducing the AMD Ryzen AI Software, a tool designed to enable the deployment of models developed in PyTorch or TensorFlow for workstations and servers on a local AI-enabled Ryzen chip. The challenge with running extensive language models (LLMs), such as local chatbots or voice-changing software, on laptops is that servers and workstations typically have more powerful processors and greater available memory.
Ryzen AI Software addresses this issue by transcoding the LLM into a simplified, less resource-intensive version, allowing it to run efficiently on the limited memory and processing power of a Ryzen laptop.
In simpler terms, much of what constitutes a chatbot or LLM involves the weights or parameters—the relationships between various concepts and words. LLMs like GPT-3 contain billions of parameters, and storing and executing these (inferencing) demands significant computing resources. Quantization serves as a form of compression, reducing the size of the weights without compromising the intelligence of the model.
AMD and Microsoft utilize ONNX (fave.co/47N19BE) for this purpose. ONNX is an open-source runtime featuring built-in optimizations and straightforward startup scripts. The role of Ryzen AI Software is to facilitate automatic quantization, preserving the model in an ONNX format ready for execution on a Ryzen chip.
According to executives, this Ryzen AI Software tool will be available today, offering both independent developers and enthusiasts a more accessible means to explore AI. Concurrently, AMD is launching the AMD Pervasive AI Contest, presenting awards for robotics AI, generative AI, and PC AI.
Prizes for PC AI, involving the creation of unique applications for speech or vision, range from $3,000 to $10,000. These initiatives contribute to advancing AMD’s position in the early stages of the AI race, particularly in the realm of client PCs. The coming year is expected to provide clarity on each chip company’s role as they navigate the initial challenges.