At the end of last month a rumour surfaced about AMD's upcoming Ryzen 4000 series (Zen 3) desktop CPUs potentially being moved onto TSMC's N5P process. The story, based on information from inside industry sources in Taiwan, indicated that TSMC was ahead of schedule with its developments and refinements. With the new process ready to rock, it had approached AMD to make the necessary modifications to get Zen 3 on N5P, said the report.
AMD has just published (PDF link) a new set of roadmap and associated slides for investors that appear to put to flight those TSMC N5P Zen 3 rumours. If you grab the slides via the AMD investor site (or check my embedded screen shots) you can see that AMD is still asserting to investors that both its Zen 3 CPUs and RDNA 2 GPUs with be 7nm parts.
Above you can see the latest slide for investors makes it clear Zen 3 is 7nm, and we will have to wait for Zen 4 to see its first 5nm CPUs. Now that is done and dusted, it is worth a look at the latest GDNA 2 GPU slides. First of all, below, we get to see a slide showing off the performance per watt improvements between GCN, RDNA, and RDNA 2. Satisfyingly, AMD trumpets that both of these transitions will deliver a 50 per cent upgrade in perf/watt.
In addition to the perf/watt benefits, AMD highlights that thanks to 'microarchitecture innovation' users will benefit from improved IPC, logic enhancement, and increased GPU clock speeds. Taking this into account you should worry about being optimistic about the capabilities of RDNA 2.
Another slide worth highlighting today is the latest AMD GPU architecture roadmap. Above you can see this slide confirms the perf/watt improvements plus the implementations of advanced graphical technologies like ray tracing in hardware, variable rate shading (VRS) and more. RDNA 2 will be succeeded by RDNA 3 before 2022 is on your calendar. AMD teases that RDNA 3 will be on an 'Advanced Node' but no other details are given.
Around the same time AMD launches the first RDNA 3 GPUs it will be ready to provide data centres with an advanced node Compute GPU (with CDNA 2 architecture) utilising its 3rd gen Infinity Architecture, which extends to exascale applications.