Analyse This: What would a mid-gen console refresh look like...?

There have been a lot of rumours and meanderings surrounding a potential mid-gen "pro" console for both Sony and Microsoft's current consoles. However, I've not really seen much analysis for what  form such a device would take or why such a device might exist. 

For this post, I'm going to delve back into my hardware speculation territory to see if we can't imagine some devices for both companies that might make some sense in the market.

Adapting to Reality...

I'm going to come right out and say the quiet part out loud: There isn't a really strong use-case for releasing a "pro" console for either Microsoft or Sony. The beauty of both the Series X and PS5 is that they are pretty powerful consoles and developers have yet to really master or stretch the hardware to its limit. Also, both consoles were in such short supply for the first two years of their lifecycle that pent-up demand is still being satisfied, with consoles selling-through pretty quickly to this day!

If anything, the fact that Sony increased the price of the PS5 in any market where there is little competition from the Xbox... with Microsoft following suit in two countries where they barely make a mark... puts paid to the notion!

Scratch that! During the writing of this blogpost, they've now done a Sony!

However, saying that, there are commercial realities aside from the nebulous "economic conditions" excuse. One of these reasons could be the shift away from the previously readily available and mass-produced 14 Gbps GDDR6 that is used in both Xbox and PS5 consoles: most new products are shipping with 16, 18, and soon to be 21 and 23 Gbps GDDR6 modules.

In all likelihood, sooner rather than later both companies will be purchasing the faster memory and downclocking it to match the listed specs of their consoles. For this very same reason AMD essentially updated their lineup with the 6X50 XT variants of cards, dropping the slower 16 Gbps spec on their memory in favour of the 18 Gbps spec.

The first improvement: increasing memory to the 18 Gbps spec.

This would bring a free performance boost to both consoles with the SX going from 560 GB/s to 720 GB/s maximum bandwidth and the PS5 increasing from 448 GB/s to 576 GB/s. Not too shabby!

Unfortunately, the compromises made in the Series S are quite large. Five modules of 2 GB each cannot be doubled or optimised better than they already are because the four memory controllers on the APU die are already over-provisioned. You can use 18 Gbps GDDR6, as with the other consoles but the increase is mediocre at best and does not address the capacity issue, at worst: 224 GB/s increases to 288 GB/s on the main portion and the single GDDR6 module increases from 56 GB/s to 72 GB/s.

This cannot save this heavily compromised design and so will likely never even be on the table for consideration... unless GDDR6W 4 GB modules enter the market soon... and cheaply: which is very unlikely!

The Series S has too many compromises in its design that are impossible to work around...

The other obvious choice, for Microsoft, is ditching the asymmetric design of the memory layout on the Series X. No more two pools of memory - just use 2 GB modules and be done with it. The cost difference isn't that large and, if we're honest, 1 GB modules will be being phased out of mass production over the coming couple of years with the upcoming newer JEDEC specs and GDDR6W on the horizon, anyway...

The second improvement: Series X moving to a unified memory design - 20 GB memory across a 320 bit bus.

This would remove any sub-optimal aspects to the Series X and give the best opportunities for developers to use the system well.

But what about the other parts of the consoles?

Zen 2 doesn't have the most optimal L3 cache layout... and the APUs reduced it by 3/4.

The Almost-Rans...

V-cache...

I'm sure people would love to see the application of 3D stacked L3 cache that we've observed in the desktop Zen 3 and Zen 4 lineups, especially given the paucity of the stuff on the console APUs*. Unfortunately, there are going to be at least three reasons why this will not be the case:

1. Zen 2 (as far as we know) never had any hardware designs with this overlaid structure in mind - i.e. the Zen 2 design did not include TSVs (Through Silicon Vias) that enable power delivery and communication to the silicon wafer stacked atop the main die. Zen 3 had these design elements from the start.
2. Stacked V-cache will cause cooling to become more of an issue. Loading up layers of silicon (especially active silicon) on top of each other reduces heat transfer and increases heat production on the chip. This could lead to lower clockspeeds or more expensive cooling solutions.
3. The cost of such an implementation would be quite large. The APUs in the consoles are significantly larger than the CCDs used in Zen 3. Not only is the packaging and layering itself an added expense, but if anything goes wrong with that manufacturing step, the loss of the much larger APU die is also a big cost to eat in terms of overall process efficiency...

*Zen 3 and Zen 4 both use 32 MB of L3 cache shared across 8 CPU cores. However, the consoles use the old Zen 2 design split the cache into two, sharing 16 MB for 4 CPU cores (x2 on the 8 core designs as in the image above). However, mobile parts shaved this L3 cache down to only 4 MB per 4 cores in order to save on precious space...

So, IMO, V-cache is off of the table when it comes to potential console updates. But what about using an updated Zen core design?

RDNA 3 has been a huge letdown - whether that's the chiplet-based N31 or the monolithic N33...

Zen 3 or 4... or RDNA 3...

Yeah, no... Seriously! 

The problem with this is that the consoles are not PCs. This idea of the "phone model" of console designs is just rubbish. Many commentators bring up the "end of console generations" time and time again. It's been literally eight years we've been talking about this and there is ZERO evidence that this is going to happen!

The reason is because console manufacturers have to write custom APIs and drivers to utilise the hardware in those devices. This isn't the situation where a phone can just use the same OS across different silicon? Oh, wait? That does happen, right? Well, no. Phones use layers of abstraction to "hide" the idiosyncracies of the hardware versions from the developers. Consoles do not do this - in fact, if anything, it's more akin to the opposite situation: developers of console games have to code to the idiosyncracies of the hardware to get the best out of it!

You see, the phone model relies on the fact that the applications are never really going to challenge the hardware in current or future iterations of the architecture. On the contrary, games* tend to be pretty demanding of the hardware they are run on! The end result is that the hardware manufacturers cannot (or do not want to) spend the money on providing very optimised abstraction layers for the game engines to hook into, and game developers would not want to support lots of different architectures within the same ecosystem**.

*I'm going to caveat this with a "this refers to graphically demanding and simulation-heavy titles... Like Cyberpunk and Starfield!"

**It's hard enough with various performance targets!

All of this adds up to the fact that neither Sony nor Microsoft will want to incorporate fully-fleshed Zen 3 or Zen 4 architectures into their current generation of consoles. Nor would they wish to do the same on the graphics front with RDNA 3 (relative flop as it is!)...

Optimisations...

There is, however, some light at the end of the tunnel: Process node optimisations.

Both TSMC 7nm and 6nm process nodes are said to be "compatible", ostensibly meaning that there is minimal work to get the same or similar designs working across them. We already have confirmation that Sony moved the PS5 APU onto the more optimised node which has allowed them to reduce the cooling systems in the console by a considerable amount, in order to save money (despite them increasing the price - as noted earlier!).

However, there is something that is possible: increased clock speeds.

It's no secret that both console vendors are limiting the frequencies of their APUs as much as possible in order to get the power and heat under control. However, the RX 7600 has shown us that given the same specs to the prior generation card (almost) can result in around a 7% increase in performance for 6% less power consumption. That's not nothing - especially for a theoretical Series X successor that can harness a properly implemented unified memory system.

If you kept the same power consumption, you could push that increase in performance up by, potentially, the same amount (even without increasing clock frequency*) and grant the CPU extra performance headroom, to boot!

*It seems to me, from my testing, that a lot of modern GPUs and CPUs are limited by power per unit time than they are frequency...

The Xbox One X had a pretty decent performance gain over the Xbox One S, but it didn't set the world on fire in terms of sales...

Conclusion...

These potential implementations would put us firmly in PS4 Pro and XBOX levels of performance increase over the base PS4 and XBOS console. The chief question there is whether Microsoft or Sony saw any real benefit from doing so last generation. Did the sales of the Pro and X offset the development costs - both on the hardware and software sides of things? My guess is, "no".

In that frame of reference, despite the fact that both console makers could easily do something to improve upon the base (main) console designs to provide equivalent mid-gen upgrades as they did last generation, the probability of them doing so seems slim, at best.

The phone model doesn't translate well to consoles. Sales do not correlate with sunk cost. The benefits of an increase in performance do not really work at this juncture when neither console is being pushed to its limits...

As a result, I would be shocked if any such mid-gen refresh materialised!