APU Phoenix eliminates mobile graphics low-end

After the current AMD management put together an offer of large processors followed by a revival in the graphics segment, it seems to focus more on the APU. In the current situation, it also makes good sense. Large processors will receive small graphics that will allow the high-performance notebook segment to be covered by large chipsets, so there will be no need for the APU to increase the number of cores as fast as before. This gives more room for integrated graphics, which could put the APU segment back to its roots, a graphics where new games can be run with reasonable detail and resolution.

AMD once managed to implement this plan only marginally. APU Llano was delayed due to incompatibility of the K10.5 core design with the 32nm process, and the next generation – due to the absence of a significantly better process – did not bring such a performance shift that the position vis-à-vis stand-alone graphics could be maintained. With the advent of processor cores Zen Although the graphic part also experienced a revival, due to the need to increase the area for processor cores, it stood on the second track and remained with the improved Take it.

Rembrandt (Ryzen 9 6900HS) with 2 × higher graphics performance compared to Cezanne (Ryzen 9 5980HS)
source: ComputerBase

The graphics core of this year’s APU brought a real performance revolution Rembrandtwhich, thanks to a combination of RDNA 2 graphics architecture deployment, a 50% increase in functional units and a 6nm process, has resulted in an intergenerational doubling of performance, enabling it to retire the mobile GeForce GTX 1650M, a widespread notebook graphics.

APUyearproc.CPUGPUflat
Llano201132nm4/4 × K10.5400 SP VLIW-5226 mm²
Trinity
Richland
2012
2013
32nm4/4× Piledriver384 SP VLIW-4246 mm²
Kaveri201428nm4/4× Steamroller512 SP GCN2245 mm²
Carrizo
Bristol Ridge
2015
2016
28nm4/4× Excavator512 SP GCN3245 mm²
Raven Ridge
Picasso
2017
2019
14nm
12nm
4/8× Zen(+)704 SP Vegas210 mm²
Renoir
Lucienne
2020
2021
7nm8/16× Zen 2512 SP Vega+156 mm²
Cezanne
barceló
2021
2022
7nm8/16× Zen 3512 SP Vega+180 mm²
Rembrandt20226nm8/16× Zen 3+768 SP RDNA 2208 mm²
Phoenix20235nm8/16× Zen 4RDNA 2+?

If the APU Cezanne liquidated the basic offer in the form of mobile GeForce MX and Rembrandt eliminated the need for cards like the GeForce GTX 1650M, then currently upcoming Phoenix according to new information, it will eliminate mobile cards such as the Radeon RX 6500M or GeForce RTX 3050. In order to achieve something like this, it would need Rembrandtu further doubling of graphics performance. If we trust RedGamingTech’s resources, AMD is targeting Phoenix even higher. The APU should also retire mobile cards such as the GeForce RTX 2060 and put some pressure on the GeForce RTX 3060 / Radeon RX 6600M. This does not necessarily mean that it will outperform them, but that their performance will simply not be so high that it pays to spend extra money and reduce battery life with separate graphics.

Something like this would require more than an intergenerational doubling of power, rather a 2.5x increase (or more). How could AMD achieve such a thing? Primarily Phoenix will be 5nm. The transition to a new generation gives room both for extra transistors and reduces their energy requirements. As a result, the number of stream processors shifts from 768 to 1536, doubling. To this more significant shift, which was expected only from Zen 5 APU Strix Point therefore, it will happen a year earlier Zen 4 APU Phoenix.

However, doubling the number of function blocks alone was not enough to achieve such a performance goal, so there will be two more changes. The RDNA 2 architecture will be enriched with elements from RDNA 3. This hybrid, which I call “RDNA 2+” in the table above, should perhaps take over the computational blocks from RDNA 3 (and probably leave the back-end from RDNA 2?). Apart from the half-generation shift in the architecture, there will be a significant shift in the clock frequencies. While GPU Rembrandt runs at a maximum of 2.2 GHz, in the case of Phoenix it should be up to around 3 GHz. A simple multiplication of the number of stream processors by frequencies indicates a 2.7 × intergenerational increase in computing and texturing performance.

What about data throughput? The part will solve faster DDR5 and LPDDR5 (LPDDR5X support is still unclear) and the other part will go to the cache account. On the one hand, it should not be a “classic” Infinity Cache, on the other hand, it should Rembrandt larger graphics cache. Personally, I would explain that the Infinity Cache, which is currently understood as a large graphical L3 cache, will not be present, but will instead be enlarged L2 cache. There is a certain sense in this – than adding another small L3 cache to the L2 cache, it may be more advantageous (in terms of performance per unit area) to increase the L2 cache. Recall that Cezanne had 1MB, Rembrandt 2MB, so in the case Phoenix 4-8MB is possible.

For a better idea of ​​the performance shift: A game that, in a certain setting, ran smoothly on the integrated GPU Vega v 720p (Renoir, Cezanne) will run at 1080p on the current APU at the same FPS Rembrandt and in even higher FPS at 1440p per APU Phoenix. What runs at 40 FPS in 1080p on an APU Rembrandtcould run on APU Phoenix at ~ 60 FPS at 1440p. We could continue with FSR, RSR or FSR 2.0.


The processor part, however interesting, works as usual in this contrast: 8 cores and 16 Zen 4 threads, the basic clock so far looks like about 4 GHz, single-core boost around 5 GHz. Although desktop Zen 4 can handle even more (5.2-5.3 GHz as a conservative estimate for the top model), AMD will have no reason to squeeze every MHz of single-core boost at the APU, as the APU will not drive separate powerful high-end graphics – hence the chiplet Zen 4. Processor part of APU Phoenix will probably be equipped with a 16MB L3 cache.

Phoenix AMD will introduce it to laptop manufacturers at the end of this year, and will probably introduce it to end customers in January at CES 2023.


Source: Diit.cz by diit.cz.

*The article has been translated based on the content of Diit.cz by diit.cz. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!