id Software CTO talks about a Ryzen optimised game engine

[preter_s]

T....surely AMD could have kept certain architectural features in-line with Intel so that every engine and piece of code didn't need specific Ryzen optimisations? ...

It's not really a case of making Ryzen specific optimisations; the article doesn't even mention a single Ryzen-sepcific optimisation. It just says the next engine will be more parallel, and Ryzen's higher thread count will get the best out of it.


The ideal situation is for developers to make no vendor-specific optimisations. The issue for AMD for the last decade or more is that most developers will have been working on Intel systems, so they're going to have automatically optimised for Intel (even if they didn't do it consciously) just to get the game running as fast as possible on their dev systems; and AMD processors were so far behind they probably didn't care whether they were getting every last drop of performance.

Ryzen does share a lot more architectural features with Intel than the construction and cat cores did. I wouldn't be surprised if some of its performance jump over FX is because it's actually being used more efficiently by the code.

Actually, its all still x86 ain't it.

On one hand, it is true, that its a generic issue about more core/thread counts where higher thread counts will always work tremendously well with software built to harness each core/thread. Already the scientific and R&D communities are flocking to AMD Ryzen for what they can do for computational modelling which require massive parrallel processing.

However, there are specific Ryzen optimizations that are separate/different from Intel which can eke out even greater performance.

[preter_s]

Reading a lot of the comments here, you'd get the feel that somehow folks are saying that the AMD Ryzens are unable to run 1080p legacy single threaded games. Which is so bollocks. It runs fine!

Further they outperform anything else in the market when it comes to other productivity/workstation tasks. Even gaming and simultaneously game streaming.

All at half the price. Which will perform even faster over the next 3 years.

Amazing how ... some just don't(won't want to?) get it...

[scaryjim]

.. What I am wondering though is how they can make an AM4 low cost option. ..

Maybe they reckon that the cost/benefit ratio of running a lot of quad core chips and binning is good enough to throw into the low end? After all, when Llano first came out there was talk of a "little Llano" die having reduced cores and shader count, but it never materialised; they just served the lower end of the market by disabling full dies.

I'm expecting Zen APUs to cut out some of the IO anyway, to keep it closer to Bristol Ridge in feature parity. 16 PCIe lanes would give 4 for the chipset, 8 for dGPU and 4 for storage (and I reckon that's half of a full Ryzen compliment, based on the 128 they reckon they can run off a 4-die naples). I also can't help but wonder if they've got a higher density layout library up their sleeve for the APUs, which would lead to reductions in die area and power consumption, but probably at the cost of peak clock speeds...

I reckon the biggest problem is the cost of designing a smaller die. Sure, each sale will give you a higher profit margin for that part, but will you sell enough, even with the higher margin, to recoup the cost of designing and masking a smaller chip. Intel almost certainly will - they must sell low end dual cores by the bucket load. AMD would need a pretty dramatic uptick* in OEM design wins and market share to sell low end APUs in any real volume...


*of course, ironically having a smaller, cheaper, lower power processor might be one of the factors that would help them get those design wins. Who'd be a semiconductor designer, eh?

[DanceswithUnix]

Maybe they reckon that the cost/benefit ratio of running a lot of quad core chips and binning is good enough to throw into the low end? After all, when Llano first came out there was talk of a "little Llano" die having reduced cores and shader count, but it never materialised; they just served the lower end of the market by disabling full dies.

They could offer the lower end market the cat cores though, so cutting down the big APUs wasn't so important. Now the cat cores are basically dead, it makes me wonder what AMD will do next.

Possibly they could offer a slower DDR4 interface, that could cut down the pin buffer sizes and so consume less area. It does seem like a pad limited design though atm, hence the rectangular die shape.