Sony PS6, Microsoft neXt Series - 10th gen console speculation [2020]

[rntongo]

I'm not saying that 32GB of VRAM are impossible, but ignoring costs, it could end up like the 12GB's of the Xbox One X, which is, mostly unused. Nvidia's low-mid end GPU's feet's are firmly planted on low VRAM amounts, and that's a big chunk of the market. Then you got current gen consoles and Switch 2.

The forward looking bet is probably following in Nvidia's footsteps, and that is, AI hardware to upscale textures and/or reduce VRAM consumption.

My prediction is 32GB of unified RAM for the PS6, not 32GB VRAM since thats not been a thing on consoles for a while(Xbox 360 standardized unified memory so its not going to change). And no it wont be similar to the Xbox One X's 12GB bump because that was a midgen refresh. There wasnt a major architectural shift especially a developer ecosystem driven rethinking in how memory was to be utilized. Devs were still targeting the base 8GB slower memory when designing games. Top it all off it still had a spinning HDD which meant the memory hierarchy was bottlenecked by disk I/O. So even with a 384 bit bus, you'd only get about 70% of the 326GB/s bandwidth utilized due to mismatch between disk throughput and actual software design considerations related to the base hw.

Even more important, if you read my prior posts, the major thing devs clamored for(during the 8th gen) was for an introduction of SSDs like SATA, not more memory or memory bandwidth. Memory bandwidth wasnt the biggest request because the main bottleneck was the disk IO and then the CPU. Which is why Sony and MS went above and beyond to provide NVMe SSDs with highly customized hw compression blocks this gen. Again if you read my prior posts, Sony (and to some extent MS) 100% determines console specs by taking developer input and making long term bets on a 7-8 year console cycle, they dont base console specs on PC consumer averages.

This gen, whats different is devs have signaled a need for more physical memory, even MS's impressive virtual memory optimizations with Sampler Feedback Streaming wasn't enough especially with the Series S. With the PS6, Cerny already announced in the PS5 pro technical seminar that the focus of the PS6 will be not just a larger GPU( rasterization block) but also larger die area dedicated to hw acceleration for RayTracing and AI Upscaling.

Now add a more powerful CPU with at least 16 CPU cores to issue instructions/fully utilize the aforementioned 3 hw accelerators for graphics and you have a serious requirement for memory and memory bandwidth. You need to feed the CPU, and these 3 accelerators with enough memory bandwidth. As I discussed earlier with the example of the base Xbox One, cutting corners with ESRAM, couldnt beat Sony simply going wider+ faster memory. There is just no way around this. For a console supposed to last 7-8 years the last thing you want is to cut corners on memory.

Remember also the PS6 pro will likely have to use the same amount of unified memory. Pro models have always used the same amount of memory as the base console because of fundamental Software design constraints of the base model, its unlikely devs will spend the extra time to fully utilize extra physical memory on a pro model(but they can use the extra memory bandwidth on a pro model). So I just dont see anything less than 32GB of memory for the PS6 which will also be 2GB of memory for the PS6 pro. You cant handicap the PS6 pro like that.

 

[function]

Been having a few thoughts about a next gen lineup.

I know I'm not the only person to think about the same chip being used in both a handheld and a separate tv based console. But today I've been thinking more about the suitability of a 'handheld' chip in a conventional console, and how the console might differ.

At the simplest level you would use the console in the handhelds' docked performance level. But why stop there? With sufficient power and cooling a (for instance) 15W mobile chip could run at 60W or more, and roughly double performance over the 15W mode. More aggressive memory clocks could help boost bandwidth, or perhaps it might be possible to make a chip that could support both LPDDR5 and DDR/GDDR (I think AMD did something like this with an early APU, but the GDDR support was never used on it).

MS's strategy of Series S and Series X didn't work out how they wanted, but it wasn't fundamentally a bad idea they just executed poorly. Sharing a chip between a handheld and a Series S like console would be a good way to manage development costs, benefit from economy of scale, and potentially give two binning options for the chip.

Of course, MS might not make another console, and Sony don't seem keen on a Series S like cheaper home system, so maybe there'll be no-one to try this next gen.