AMD launches Ryzen 7040: 4nm APU with Zen 4 and RDNA 3 iGPU

Phoenix APU: monolithic 4nm Zen 4 with RDNA 3 graphics and new Ryzen AI accelerator

In addition to the AM5 Ryzen 7000 processors with 3D V-Cache and mainstream 65W TDP models, AMD also announced a wholly new offering at CES 2023: the first processors with Zen 4 architecture for laptops. They are based on the all-new Phoenix APU, which is the first 4nm Ryzen chip and the first processor with an integrated RDNA 3 GPU, which will also clock up to 3 GHz. So it’s going to be a very interesting laptop platform.

This mobile “APU” version of Zen 4 is officially designated as the Ryzen 7040 series. The mobile Ryzens will be a jungle of five different chips in the 7000 generation as a whole, so it will be challenging to navigate them, but the 4nm Phoenix APUs with Zen 4 can be distinguished by a numerical designation ending in 40, with the four indicating the Zen 4 core (the low-end Mendocino with Zen 2 Lite cores will have a designation ending with 20, refreshed 6nm Rembrandt models will end with 35 and refreshed 7nm Barcelo/Cezanne has a designation ending with 30).

Phoenix is not yet a chiplet design like the desktop Ryzen 7000s, but a monolithic 4nm chip, which should be beneficial for power draw and power efficiency. The silicon is manufactured on a TSMC process and contains more than 25 billion transistors in a 178 mm² area (not so small a die for this advanced and expensive process).

The main components of the Phoenix APU are eight Zen 4 cores in a single CCX block with 16MB L3 cache. It is half the size of the desktop chiplet version, which will reduce performance in games. The processors will also have slightly lower clock speeds due to the fact that there are only laptop processors in this series, for now.

Illustration showing the package and the Ryzen 7040 chip. The blocks shown in the chip probably do not correspond to reality (source: AMD)

The best integrated graphics in a PC to date

This is followed by integrated graphics, which should be the most powerful iGPU on the PC platform. Like the previous 6nm Rembrandt APU (Ryzen 6000), it has only 12 CUs (768 shaders), but it will reach very high clock speeds – the top model has a GPU boost clock of 3000 MHz in the specifications. In particular, it has the new RDNA 3 architecture from the Radeon RX 7000 discrete GPU generation, which has dual-issue execution units within a single CU (see the article on RDNA 3 architecture), which should advance performance even further.

Read more: RDNA 3 GPU architecture analysis: new CU and Media block, AI acceleration, DP 2.1 for next-gen 8K monitors

AV1 encoding and 8K monitor support

This graphics supports up to four monitors and HDMI 2.1 and DisplayPort 2.1 outputs, but with lower bandwidth than the Radeon RX 7900 (only UHBR10 mode with 38.7 Gbps bandwidth is supported). With DisplayPort , this APU supports display resolutions up to 7680 × 4320 pixels at 60 Hz, 3840 × 2160 at 240 Hz, and FullHD at up to 600 frames per second (these maximums are reached via DSC compression).

AMD Ryzen 7040 (source: AMD)

The GPU also has an integrated video encoder and decoder capable of decoding as well as encoding 8K video (7680×4320) at 60 Hz in both HEVC and AV1 formats. For VP9, 8K decoding is supported, but not encoding (this is a good compromise, VP9 encode capability doesn’t really matter).

The Phoenix and Ryzen 7040 are also supposed to have various generational efficiency and power draw improvements, making the laptops capable of playing video for more than 30 hours on battery. There will probably be a number of different tweaks and improvements across the main blocks and physical implementation of the APU, but they haven’t been detailed anywhere yet.

AMD Ryzen 7040 is said to enable more than 30 hours of video playback on a single laptop charge (source: AMD)

Ryzen AI

One of the main innovations (which you may not use much in practice however, at least not at first) of the Phoenix APU is a special dedicated unit for accelerating artificial intelligence (neural network) applications. This is labelled Ryzen AI and is based on the XDNA architecture that comes from Xilinx technology portfolio. In Phoenix, it has a performance of 12 TOPS, and its focus is apparently on efficiency and low power draw.

AMD Ryzen AI in Ryzen 7040 (source: AMD)

The performance itself is not that high considering that these TOPS should be achievable even on an integrated GPU, where it will however have higher power draw and spin up the laptop fans. The AI unit probably has the goal of giving the chip ability to run various AI algorithms with low power draw virtually all the time during normal usage. This assumes that such features will become a common thing in desktop applications, Windows File Explorer, web browsers and so on. Not so much of that around right now, but we’ll see in the future.

AMD Ryzen AI in Ryzen 7040 (source: AMD)

DDR5 5600 with ECC / LPDDR5X-7500, USB4

Phoenix has a new memory controller which seems to have better capabilities than the one in the Ryzen 7000 Raphael chipset – it supports DDR5-5600 instead of DDR5-5200, which puts it on the level of Intel Raptor Lake (this could also be just a bonus stemming from the fact that the processor is soldered on the board instead of in the socket, though). Of the mobile memory types, even LPDDR5X-7500 is supported, which would give decent bandwidth for an iGPU (120 GB/s). The memory controller is a classic two-channel (actually four-channel in DDR5 terms) 128-bit. However, ECC memory is supported if the board and firmware allow for them.

In contrast, PCI Express connectivity is less bleeding edge, Phoenix can still only do PCI Express 4.0 and provides 20 lanes. It is possible that in the AM5 desktop socket (although such a version is not announced yet) and perhaps in laptops, only eight of these lanes will be usable for GPUs.

Phoenix / Ryzen 7040 natively provides USB4 ports with 40 Gbps speeds as an equivalent of Intel’s Thunderbolt platform. Monitors can also be connected via USB4 (USB-C).

Just 35W–54W SKUs, for now

The company has only announced Phoenix APUs for the higher-performance laptop segment for now, not the mainstream 15W U-series. At CES 2023, three models were announced, branded as the HS series and using 35W–54W TDP – or perhaps better said, the TDP will be adjustable in this range. By the way, these processors will not support overclocking.

The most powerful model is the Ryzen 9 7940HS (recall that the 40 in the last two digits of the name is the hallmark by which you can identify all Phoenix APUs in listed specs of various laptops). It has eight cores and 16 threads and 16 MB of L3 cache. Note: the slides from the unveiling incorrectly list 32MB L3 cache and 40MB L2+L3 cache, but this is an error, unfortunately.

The clock speed of the Zen 4 cores is 4.0 GHz in the base (a lot for an 8-core, this shows the advantages of the 4nm process) and the maximum boost for single-threaded applications is 5.2 GHz. The graphics is labelled Radeon 780M and their 768 shaders have a listed clock speed of 3000 MHz – this is apparently the boost or maximum clock speed.

The second, cheaper option is the Ryzen 7 7840HS, where eight cores / 16 threads and 16MB L3 cache are retained, but the CPU clock speed is only 3.8–5.1 GHz. The graphics is also Radeon 780M and retain the full number of units (768 shaders). Just the clock speed is a bit lower, 2900 MHz. Overall, it seems you won’t lose much with this model. It’s possible that is this SKU that will account for most of the sales, while the higher Ryzen 9 model needs binned chips with the best parameters and it is a question whether it won’t be marred with poor availability.

AMD Ryzen 7040 processors for laptops: models and parameters. Note: the cache capacities are wrong, the figures should be 24 and 22 MB (source: AMD)

The third and for now last model Ryzen 5 7640HS is already partially cut down, and thus dies with some defect will find use through this SKU (by disabling the affected part of the chip – a classic practice in GPUs). Six Zen 4 cores (12 threads) at 4.3GHz in the base and 5.0GHz in the maximum single-core boost are present. The L3 cache remains at 16MB, but the Radeon 760M graphics have had their unit count reduced to 8 CUs (512 shaders) and will clock at 2800 MHz.

The TDP of all these processors is the same, the mentioned 35–54 W (or maybe 35W–45 W with the possibility of increasing it to 54 W). Also, all features and functions are apparently retained in all models. AMD states in the specs that this processor can be fitted to various laptop platforms – it is said to be compatible with the FP7, FP7r2 and the new FP8 BGA package infrastructure (assuming there was no mistake in the specs).

According to AMD, the first laptops with these processors will go on sale in March this year.

Sources: AMD (1, 2, 3, 4)

English translation and edit by Jozef Dudáš

Flattr this!

Leave a Reply

Your email address will not be published. Required fields are marked *