AMD Gonzalo APU spotted – Zen 2 plus Navi 10 Lite?

[HEXUS]

Expected in next gen consoles, this 7nm octa-core APU has a base / boost of 1.6 / 3.2GHz.

Read more.

[Bagpuss]

Well You can't put a CPU in console that has boost clocks.. That would play havoc with people having differing experiences in game performance,depending on temperatures and the silicon lottery.

[DanceswithUnix]

Well You can't put a CPU in console that has boost clocks.. That would play havoc with people having differing experiences in game performance,depending on temperatures and the silicon lottery.

I'm sure your're right, though a BIOS tweak could knock out any boosting to fix that.
Still, that base clock smells of laptop part to me.

[gagaga]

Well You can't put a CPU in console that has boost clocks.. That would play havoc with people having differing experiences in game performance,depending on temperatures and the silicon lottery.

Don't think it will be an issue - VRR will be standard on next gen so the console will just adjust frame rate to meet the performance. They'll be developing software that in the future will run on the Xbox 2X / PS5pro so will anticipare varying screen resolution/performance.

The CPU will also adjust CPU vs GPU balancing for the TDP according to a known strategy so the dev can balance non-time critical CPU tasks away from peak GPU load.

[edmundhonda]

Yeah, base/boost that far apart reads like a laptop part choked down to a lowish TDP to me as well.

The console version will presumably be related to this but semi-custom to whatever design they want, same as the Jaguar APUs all were.

[Shirley Dulcey]

AMD uses its die designs for multiple purposes. With the low base clock this is dialed in as a laptop CPU. But with a higher fixed clock (or a well tuned boost strategy that only down-clocks the CPU while you're not playing a game, keeping down power consumption while you are navigating the menus or watching a movie) it could also serve as a console part.

So far as I know, AMD has only released three Ryzen die designs to date: the first generation, the second generation, and the APU. Every Ryzen, Threadripper, and EPYC variant is binned from one of those designs - in multiples for Threadripper and EPYC. Similarly, I expect all the third generation parts to use a small number of chiplet die designs that are combined in various ways.

Besides being a clever way of using the strengths of different fab nodes, I wonder if the decision to separate out the I/O die was a way of fulfilling contractual obligations to buy some parts from GlobalFoundries? (GloFo is making the I/O die while TSMC makes the CCX and APU parts of the third generation Ryzen.)

[DanceswithUnix]

Besides being a clever way of using the strengths of different fab nodes, I wonder if the decision to separate out the I/O die was a way of fulfilling contractual obligations to buy some parts from GlobalFoundries? (GloFo is making the I/O die while TSMC makes the CCX and APU parts of the third generation Ryzen.)

Beyond contractual obligations, it sounds like TSMC 7nm is going to be in demand for some time which makes it a possible production bottleneck. Moving part of the design away from 7nm could allow them to make more CPUs.

There is of course nothing to stop them making a new 7nm I/O die in the future and re-packaging. They could even do that as a stepping stone to AM5 if AM4 is supposed to be EOL in a year or so. It also allows an easier split of teams for development, and will have reduced risk on any parts of the design that were carried over from the current Ryzen design and would have needed to be ported to 7nm in a traditional monolithic die.

I've seen this sort of thing often enough in the past though, engineering get pushed to make things more flexible allowing more products with more options. The next step is that the bean counters realise that most of the sales are a couple of parts that are now more expensive than they could be if they were a monolithic die, so I only expect this to last a few years