- It is claimed that the “sound wave” chip of AMD is an arm based on the arm
- It is believed that it has “RDNA 3.5+” CUS that are “improved” for automatic learning
- MLID sources claim that the chip will have 16 MB of infinity cache
New information about the rumored APU of “Sound Wave” AMD has been discovered, claiming that it will be more for automatic learning, with CUS “improved 3.5”.
The news comes from the reliable law of the Moore hardware filtration is dead in a new YouTube video, with a long list of updated specifications cited since “the end of March 2025”. It is a substantial update on the previous rumor that we cover in March 2024.
Previously, it was believed that AMD's sound wave processor would be a Zen 6 mobile chipset. However, it is now claimed that the processor will be an APU based on the arm, similar to the Qualcomm Snapdragon X line of Qualcomm and Qualcomm.
Specifically, MLID sources claim that AMD sound wave APU will be built in a 3NM TSMC node, aimed at 5-10 W factors, with two performance cores, four efficiency cores (six in total) and 4 MB of L3 cache. Crucially, it is alleged that there will be four CUS of RDNA 3.5 with “improved ML performance” to which the name of “RDNA 3.5+” has been given not officially, to denote a difference with the current RDNA 3.5 cus available in chips.
Their capacities for the computing of AI seem to be a main approach, as can be evidenced by the alleged “fourth generation AI engine” backed by 16 GB LPDDR5X-9600 RAM that is said to be for the “standard” systems. MLID does not know if the new RDNA 3.5 CUS will allow FSR 4. However, the games seems to be a secondary concern for this silicon.
An interesting development for the Sound Onda APU is that it will allegedly have 16 MB of access to memory at the last level (shopping center) (known by AMD users as infinity cache), which is unusual for a 5-10 W APU. It is curious because “Strix Point” does not have this, which, according to the reports, is “documented” by MLID.
Later, MLID points out how the AI engine, the CPU and the GPU will work in the same controller, so having the additional bandwidth of 16 MB of shared infinity cache between them could be beneficial, even in such a modest chip.
Ai computing without the energy raffle
While it is still early for the sound wave, all the rumors we have heard from the chipset sound as if it were mainly oriented around the workloads of AI without the (usually) rival hardware raffles. 16 MB of Infinity Cache does not sound too dramatic, however, that is before taking into account that the chip only has a rating between 5-10 W TDP.
In addition, RDNA 3.5+ CUS could be a great driving factor in AI workloads, depending on how optimized they are really for automatic learning. With a claimed 2026 launch date, it is unlikely that we receive an official AMD word in the short term, but now we know that it could be an optimized arms based instead instead of the chip that introduces in Zen 6 as we think above.
