During the CES 2023 exhibition, AMD announced several lines of mobile processors at once. The AMD Ryzen 7040 series of chips attracts special attention, which seems to have absorbed all the most progressive improvements of the manufacturer at the moment.
The models of this line are based on 4 nm FinFET technological norms, use up to 8 computing cores with Zen 4 architecture, and also received a relatively powerful integrated RDNA3 graphics. In addition, it was AMD Ryzen 7040 that became the first processors of the manufacturer with a hardware unit for accelerating tasks computations with algorithms related to artificial intelligence. The AI engine is called Ryzen AI and is based on the scalable XDNA architecture.
For the unit, the development of the Xilinx company is used, the acquisition of which was completed at the beginning of last year and cost AMD 50 billion dollars. So the integrated Ryzen AI unit can be considered the first practical results of the strengthened team for consumer solutions.
Ryzen AI and XDNA architecture
XDNA is AMD’s new adaptive architecture for artificial intelligence, for the development of which developers already have a corresponding roadmap.
Neural networks function like neurons, where data flows from layer to layer. Suppose there are several layers of neurons, where each “point” performs some computational function – such as matrix multiplication or convolution – and then the data is passed to the next neuron for processing. There is quite a lot of sequential dependence from level to level. To follow this, the Ryzen AI engine architecture works as an adaptive data flow architecture where it can actually use a large array of computations and transfer data efficiently from array to array. This can be done without accessing external memory or even cache, which often consumes more power and results in longer latency.
Ryzen AI allows four AI threads to run simultaneously, thus realizing multi-threaded workloads. At the same time, all unit resources can be used to work on one task.
The presence of hardware Ryzen AI allows you to offload the CPU/GPU, as well as to use the AI load more efficiently, also from the point of view of overall power consumption and battery consumption. The work of a specialized unit on specific tasks requires significantly less energy than execution on general units.
The use of AI computing can well be considered the next “big thing” that involves significant technological changes. The areas of use here are very wide, starting with working with images during video broadcasts and games, ending with issues of security and predictive user design. The presence of specialized hardware units on user devices will only accelerate the spread of applications and solutions that benefit from the use of AI.
With the new Ryzen 7040 chips, AMD is once again opening a new niche, and will certainly try to build on its success. What can the main competitor offer? During the announcement of its Core 13th generation mobile chips Intel developers also touched on the topic of AI, mentioning the possibility of using Intel Movidius VPU (Versatile Processing Units). However, in this case we are not talking about an integrated solution, but a discrete module for accelerating AI computations.
In the context of VPU, the emphasis is primarily on auxiliary options for working with the camera during broadcasts – high-quality background blurring, eye tracking and automatic framing. It is not surprising that initially the developers focus on these applications, where the specialists of the company Movidius, which was merged by Intel, have a lot of expertise. However, VPUs will probably have a wider range of tasks in the future.
The capabilities of the Intel Movidius VPU are not known for certain. On the manufacturer’s website the announcement of the Intel Movidius 3700VC VPU module is mentioned, but without any technical details, except for the clarification about the possibility of using effects in Windows Studio.
However, it is obvious that if we are talking about an additional discrete module, mass distribution should not be expected. Even Intel’s slide states that Intel Movidius VPUs will only ship with select laptops based on 13th Gen Intel Core processors, primarily for developers and testers.
However, Intel is also going to actively develop the ecosystem of artificial intelligence, and the first processors with an integrated AI-accelerator VPU will be mobile chips Meteor Lake, which should appear this year. So while AMD has a certain advantage in time, it is interesting how developers will use this opportunity.
Ryzen 7040 model range
The series of Phoenix mobile chips will initially include three models – the 8-core Ryzen 9 7940HS (8/16; 4.0/5.2 GHz) and the Ryzen 7 7840HS ( 8/16; 3.8/5.1 GHz) and 6-core Ryzen 5 7640HS (6/12; 4.3/5.0 GHz). All chips have an L3 cache buffer with a capacity of 16 MB and an L2 cache of 1 MB per core. The specified TDP of all CPUs is configurable between 35 and 54 W.
Older 8-core models, at first glance, hardly differ from each other, except for CPU operating frequencies and 100 MHz faster iGPU operation in the flagship. However, unlike the top model, Ryzen 7 7840HS does not support AMD EXPO and Curve Optimizer technologies.
In turn, the Ryzen 5 7640HS, in addition to reducing the number of cores and operating frequencies, also lost the Precision Boost Overdrive (PBO) support announced for 8-cores. In addition, the 6-core chip is equipped with quantitatively simplified AMD Radeon 760M graphics with 8 RDNA3 executive modules.
AMD has already published detailed specifications of Ryzen 7040 chips. In addition to the mentioned features, a number of nuances deserve special attention.
Crystals for Ryzen 7040 chips are manufactured by TSMC using the most advanced 4-nanometer FinFET process. The total size of the monolithic silicon plate is 178 mm², and about 25 billion transistors can be accommodated on such a platform. This is about twice as much as the previous generation Ryzen 6000 processors.
Older 8-core Ryzen 7040 models will receive the new AMD Radeon 780M integrated graphics based on the RDNA3 architecture. The video core has up to 12 execution units with 768 CUDA cores, which operates at frequencies up to 3 GHz. Here we can single out support for SmartShift MAX and SmartAccess Memory. Support for DisplayPort 2.1 and HDMI 2.1 was announced, with the ability to output images in 8K at 60 Hz, 4K@240 Hz or 1080p@600 Hz mode.
For Ryzen 5 7640HS, the AMD Radeon 780M video core with 8 RDNA3 execution units containing a total of 512 CUDA cores with the maximum operating frequency of the graphics core at the level of 2800 MHz is provided. There are no functional limitations here, we are only talking about a relative decrease in productivity.
As for peripheral capabilities, Phoenix can offer two USB 4 ports (up to 40 Gb/s) and four USB 2.0 ports (up to 480 Mb/s). In addition, the Ryzen 7 7840HS and Ryzen 5 7640HS models can offer a pair of USB 3.2 Gen 2 (10 Gb/s), which are not mentioned in the description of the flagship. We assume that there is an error in the TX table.
The Ryzen 7040 processors did not receive support for the PCI-E 5.0 bus, instead offering 20 PCI-E 4.0 lanes for connecting discrete graphics and drives.
The memory controller will allow the use of DDR5-5600 or LPDDR5x-7500 in dual-channel mode. Interestingly, the processors support 4 modules (even two-rank ones), while the total RAM capacity can be 256 GB. Of course, it is not a fact that many manufacturers will implement similar configurationsand laptops, but we will probably see such interesting exceptions for specific tasks.
Judging by the equipment of the Ryzen 7040, it is likely that these chips will be used as the basis for future Ryzen 7000G desktop APUs.
AMD Ryzen 7040 performance
Data on the expected performance of Ryzen 7040 is still scarce. Only the indicators demonstrated in the presentation are actually available. When rendering in Cinebench multi-threaded mode, the 8-core Ryzen 9 7940HS (8/16; TDP 35-45 W) is 34% faster than the Core i7-1280P (6/12+8) with TDP 28-64 W, and is also significantly ahead of the Apple platform with the M1 Pro chip. In games, the advantage over the system with an Intel processor is claimed up to 21%, while the indicators of the solution from Apple are absent at this stage for obvious reasons.
Also according to the information in the presented presentation diagrams, the Ryzen 9 7940HS processor with the Ryzen AI AI engine in the MobileNet v2 test with neural network computations turned out to be 20% faster than the Apple M2 SoC, which also has a hardware Neural Engine AI accelerator. In turn, according to Apple, The Neural Engine version in the M2 chip is capable of processing 15.8 trillion operations per second, which is 40% more than the capabilities of the NE unit in the M1. So the advantage of the new Ryzen over SoC Apple of the previous generation in the relevant tasks should be even more noticeable.
During the presentation, the advantage of the Ryzen 9 7940HS over the mentioned opponents was also clearly demonstrated when rendering a scene in Blender. However, such comparisons require additional verification, because under similar conditions, Intel chips with a large number of additional E-Cores usually perform more powerfully.
Laptops on Ryzen 7040 battery life
During her report, AMD CEO Lisa Su did not spare favorable epithets for the laptops based on Ryzen 7040 chips battery life. The transition to a modern 4-nanometer technological process and the use of a new platform made it possible to significantly improve these indicators.
However, the announced possibility of battery operation for 30+ hours in video viewing mode looks like an exaggeration. We’d be happy to be wrong, but even with all the improvements, this figure is too optimistic. Given that we are not talking about record-breaking economical chips, but models with a TDP of 35–55 W. However, we will make sure from our own experience that future Ryzen 7040-based laptops can work for more than a day on a single battery charge. What if the integrated Ryzen AI can really do wonders?
Experiments will have to wait a bit. The first laptops on 4-nanometer AMD Phoenix chips with integrated Ryzen AI will begin to appear in March of this year.