AMD reports record revenues and says RDNA 4 will be 'on track' by early 2025. But why is it unable to get gaming GPUs right?

Why is it that AMD can't seem to get gaming GPUs right for PCs? This is the multi-billion dollar mystery that AMD's latest earnings report for professional bean counters revealed (via Seeking Alpha). Despite a promise from the chip maker that its next-gen RDNA 4 graphic was on track for early 2020.

Lisa Su, AMD's CEO who is celebrating her 10th year anniversary, revealed that AMD's revenues in the third quarter this year reached a record of $6.8 billion. This was achieved by a healthy increase in revenues for data centre CPUs, GPUs, and CPUs for PCs and laptops.

Su was forced to admit that gaming revenues had fallen. In AMD parlance "gaming" is a combination of graphics card for PCs and custom AMD APUs for consoles such as the Microsoft Xbox or Sony PlayStation.

AMD reported that revenue from gaming fell 69% on an annual basis to $462, or less than 10% of AMD's total revenues. This sounds like a catastrophe, and it is. This is largely due to the inevitable cyclical decline in console sales, as Sony and Microsoft's two biggest beasts continue to age.

Su cited the upcoming PS5 Pro for a reason to be optimistic in the near future. AMD does not break out PC graphics cards separately.

Su said that "revenue decreased year-over-year, as we prepare to transition to our Radeon GPUs next-gen based on RDNA 4 architecture. RDNA 4 not only delivers a significant increase in gaming performance but also significantly improved ray-tracing and new AI capabilities. We are on track to release the first RDNA4 GPUs in early 2020."

AMD has reported non-specific declines for PC gaming graphics since we can remember, but third-party analysts have continued to report declining market shares for the company against the only other significant competitor in the market, Nvidia.

AMD's inability of getting PC GPUs right, despite a company that appears to be performing so well on many fronts is a bit mysterious. One possible explanation could be that AMD made a large bet on a chiplet-based architecture that did not pay off.

AMD decided to divide GPUs into smaller pieces instead of using a large silicon die. AMD has seen great success with its CPUs. It makes CPUs more affordable and configurable. Chiplets working with GPUs presents a different challenge.

AMD's RDNA 2 graphics (the Radeon RX 6000 series) was a monolithic and competitive design across all market segments. RDNA 3, however, which was the first to use chiplet engineering in the RX 7700/7800/7900 model ranges has been much less successful.

AMD has also reportedly reversed its course for RDNA 4 and is now focusing on a simpler monolithic GPU design, which won't be aimed at the very top end. AMD originally planned to double down on RDNA 4 chiplets, but was forced to reconsider late in the development process. It didn't have the time to design a big monolithic GPU.

Most rumours suggest that AMD will try again with RDNA 5 chiplets. It remains to be seen. AMD's otherwise near-perfect copybook is marred by PC graphics.

You could also argue that AMD has been focusing its attention elsewhere, especially now that the company claims revenues from AI GPUs (for training and inferencing) have gone from almost zero to matching CPU revenue from servers, cloud and PCs in just one year.

The failure of PC graphics remains an anomaly, even for a company that is generally a roaring success. AMD has plenty of money to fix this problem. Here's hoping that it does, and it will.