Sunday 13 January 2019

A New Console Generation

So I've engaged in mild speculation on new console specs in the past. We're getting to the point where there should be some new consoles coming. It lines up pretty well with (large chip) mass production 7nm in the coming year or two so everyone will be ready to offer powerful SoCs with high power efficiency for console enclosures. A 2020 release is 7 years on from the current high fidelity consoles - very normal for the console upgrade cadence. Even the half-generation updates would be getting on by the end of 2020: four years after the PS4 Pro was released.

Phil Spencer was at CES talking up their continuing partnership with AMD, which certainly sounds like an ongoing use of their IP (makes a lot of sense - AMD have all the IP under one roof and a track record; Jaguar cores are going to get a substantial update with a Zen derivative; and the MCP design AMD are working with could be a good match for a console that will live through at least a couple of die shrinks while uncertain about how much of a monolithic SoC would benefit from that). Vega (and even lower end Polaris) are not leading the power efficiency war but at this point everyone is experienced in making GCN work well and there are still a few bad memories around nVidia. Everyone hopes that the next Vega is going to bring that power efficiency to where nVidia are at on 12nm (if they can't when a generation ahead at 7nm and not wasting any transistors on speculative features, something will be going very wrong). There is also the deep in development Navi so even without turning it around next month, the future may be bright. The only wildcard would be Intel trying to sell their 2020 GPU design to Sony or MS (or even offering an AMD GPU: did not see that coming a year ago) - very long odds in my book but worth mentioning as Intel make a big move to compete on high end GPUs to avoid seeing server farms filled with expensive nVidia chips (getting a foot in the door with Tensor etc specialised silicon to expand their use cases). [Those with a long memory will remember that nVidia were paid by Intel to relinquish any claim to be able to provide x86 chips and so can't offer a complete package that would enable a rolling window of compatibility.]

I'm going to say that odds are good for something like a CCX cluster (4 Zen cores) or two with a pretty large Navi GPU block. What I'm curious about: what if the next Xbox is two new Xboxes? Rather than releasing one console then doing a major redesign when the power requirements are lower, what if there is a constant pair of consoles, like we have now. The next Xbox as a high end system with at least 8 Zen cores and a big Navi block. $500 at the end of 2020 with enough fast RAM to avoid the issues that plagued the original XB1 [forced to waste die space on extra cache to make up for slow RAM] and hopefully at least a PCI-E SSD cache (128GB? 256GB? Big SSD premium SKU?) to avoid the loading times getting even worse (even if most downloads and game installs would live on a large platter drive for cost reasons, even with NAND getting cheaper - 500GB is floating around $100 right now so maybe prices can drop to make an SSD-only next gen happen with support for archiving to external storage via USB).

But the XB1X isn't that old and isn't a terrible target for an entry-level spec for games going forward from 2020. There would probably be a way to make a reasonably good approximation of that system using a smaller Zen cluster and much smaller Navi block (cheaper/less RAM and no SSD cache). Something that would run existing games just like an XB1X, with minimal shims to catch edge cases (in the low level API calls). Something that would effectively be sold as both the cheap entry to the next generation and the "Slim" edition of the XB1X. Something to ensure developers continue to make sure their new games work for existing XB1X customers (with whatever compromises were required for performance) while dropping support for new games on the anaemic XB1/XB1S system. The fact that the XB1S isn't identical to the original XB1 shows that the ecosystem is already built around supporting two low end specs that are only approximately the same.

When the next high end Xbox is ready for a slim-down, it might be time to look at another step forward (an "XB2X") and which devices get guaranteed access to 100% of new releases going forward (and which platforms are something a developer could optionally support in a "cross-buy" style system). Have two or three models always "current" that cycle out every three to five years until they can no longer continue to offer the right shims to ensure compatibility with their API choices. It would certainly be something not totally out of step with the current MS talk of slowly expanding their cloud gaming offerings (on TVs, tablets, and laptops) and making games work anywhere (Windows PCs and Xbox devices). It's the realisation of much of the "what if more like annual tablet/phone updates" talk from about five years ago.