@ Tapani

The estimate is more around a 10% differential, dropping the FP32 performance from 10.3 Tflops, down to 9 Tflops.

@Chris12 Because you have to build tolerances in for all cooling solutions. Otherwise you end up with RROD size scandals. And no it wont be relative to heat, its relative to power. If the ambient temperature goes up the fans will go up until eventually (When super hot) it will turn off for safety reasons.

@Sophisticated_Chap can you substantiate that claim of 10%? I like how you casually reference 32bit Floating Point, it definitely made me think you were smart, and super "sophisticated".

And thats certainly reason why all cpus and gpus have variable freqency.

Its the opposite. What is variable is better suited for different scenarios.

Maybe that could be difference between (I hope) Sony classic chassis instead of MS fridgelike box.

As was said earlier. It doesnt mean, power of PS5 is randomly changing up and down. It means, devs can choose power profile and they get it. Also Mark Cernys "Race to zero" exactly means, PS5 tries to give only such power, what power app really needs.

Its different approach. I think there can be two reasons of this approach:
1)cooling and form factor
2)price (36CUs are cheaper then 56). XSX is premium model. PS5 is simply PS5.

If Sony can manage better price with better form factor and is only about 15% less powerful, that is huge success.

The question is why Mark choose the variable frequency instead of the fixed one...
I believe he has reason for that

Fixed would be using more power variable can clock down when target performance is met

Even when it can still meet the same frequencies if it were fixed or not? There's no inherently better solution for max performance. Max performance is just a spec. But, variable is technically superior because you're not wasting resources to meet a framerate, and improves longevity of hardware. As well as providing another measure for devs to notice that a scene still has more headroom

That's a baseless claim. Most PC GPUs use variable frequency. A 2080 ti will vary its frequency ~200 MHz without being overclocked.

@Sophisticated_Chap

lol....and where did that "estimate" come from?

Smart shift is just a way of trying to get more performance out of lesser chip sets. One does wonder why Mark Cerny will not say just how far it can drop. If it could maintain max clocks they would so I won’t be shocked to devs have to make compromises on the ps5 they won’t have to make on the series x. I don’t have much faith in most devs taking the time to get the most out of one machine when they got like 7 other platforms to target

@Donnie81

Performance is performance.

Why would we wonder how far it can drop? He said it stays max clock all but ~1-2% of the time. Dropping isn't an issue. It's a smart feature so the console doesn't run high clocks if it doesn't need to. This gives more power delivery to what needs it, like the GPU when it gets a sudden request for more power. It won't cause frame-pacing, stutter, or skipping in events where the power draw rises because it has more than enough access to power with Smart Shift.

It's extremely simple. If a game doesn't max out the console. The clocks lower automatically. If a game is demanding, then it will clock the CPU and/or GPU as high as they need to be, or max them out. And, when one or the other gets a sudden load, they can send resources such as power from that avenue to maintain consistent performance and have less hiccups. There is absolutely no negative to variable clocks. The max clock was stated as able to be constantly maintained because the cooling system is so efficient even in the most demanding situation.

Variable is only bad if that variance isn't always available. In the case of PS5, a dev can utilize that variance on demand.

If it's always available, the variance exists as a power saving function, and peak times generally are rarely used all the time anyways.

It's about performance per unit price

Yet just about every cpu/ gpu has some boost/variable aspect to them??

In this case, we're not talking about boosting clock speeds, we're talking about downclocking the GPU and CPU.