What do you think about the merits of Intel's monolithic design approach vs AMD's chiplet based design philosophy?
That is a tough question. Intel and AMD as of late have been looking for "gaps" in each other's armor. With Zen 2 AMD was behind when it came to work rate but took the approach of more cores all the way to 64 in order to open up another "front" in the war, of course requiring the chiplet strategy.
With Zen 3 they are still using chiplets but starting to put them back together a bit with the shared cache among all 8 cores. How much does the slower memory transactions of the chiplet approach hurt AMD? I don't know. When you compare Zen 2 to Skylake on paper Skylake looks like the simpler design yet it still performs better overall. Is AMD making up for the chiplet approach with a more complex, wider design vs similarly performing Intel parts? Or does Intel have smarter, better written algorithms for things like prefetch and OoO scheduling? We've learned that those things are super important.
One way I think we can better understand design choices is by looking back. Where a design is today is probably where the company wanted it to be a generation or two ago but the process or design wasn't available at that point in time.
I don't think there is any doubt that in a perfect world monolithic designs are better. But in that perfect world yields are perfect, die space is abundant, profit doesn't matter, etc... Chiplets make our imperfect world a little more perfect.
I think there will come a time where AMD will not be able to compete core-for-core with a monolithic design using chiplets. They may have to surrender the crown for as many cores as Intel can squeeze onto one die, but they still may be the go to choice for higher core counts. Sunny Cove is 217 million transistors and looks to compete with Zen 3. Do we know how many transistors for a Zen 3 core? On paper it looks considerably more complex than Sunny Cove. AMD could be making up for those extra transistors with the denser 7nm TMSC process node.
This brings up another important question. Where is the bulk of the market going in terms of core count? I think most of us here would agree that 90% of users are fine with a quad core processor. But in the constant push for sales advertising will convince users that they need 8 cores, so that is where the market will go. But can they push users into 12, 16 or more cores in mass? If not then Intel can sit back with 8 or 10 core monolithic designs and swallow up a large percentage of sales.
Look at the TV industry as an example. The move from 480i (North America) to 1080p was huge in terms of a visual difference yet the adoption rate was pretty slow considering how easy it was to show consumers the huge difference in visual quality. Part of this was due to FullHD panels being expensive back then. Now the push is on for 4k and even though it's a much harder side-by-side sell they're still pushing it. Mainly because it's not really more expensive to manufacture a 4k panel (just check current pricing) and 4k could be that extra "push" that gets someone to upgrade. But the market is meeting much firmer resistance with 8k, and that may be how it might be for over 8 core processors, at least for a while. I think this is where Intel is currently. They're going to focus on the bulk of the market. If that is what most people want/need then they can produce 8 core monolithic parts cheaper than chiplet designs and they have the built in advantage of faster memory access within the chip.