Hacker News Reader

Comment by skohan | Hacker News Reader

It remains to be seen how the PS5 and the new XBox compare in terms of performance.

The specs of the new XBox are slightly higher on paper, but the PS5 is doing some really interesting things in terms of optimization. Basically the XBox is going for high constant clock speeds, and the PS5 is shifting priority between the CPU and GPU components of the SOC so each gets boosted performance when it's most relevant.

Both consoles are doing very interesting things with asset streaming and decompression of assets directly from the SSD to video memory which are going to open up new opportunities not only in visual quality, but in terms of how flexibly game worlds can be designed.

Probably we will have to wait and see how all of this plays out in terms of real-world performance.

View on HN

Highly dangerous of the PS people to build non-deterministic performance into their console. Developers who try to push every ounce of power out of it will hate it.

edit: Maybe deterministic isn't the correct word. What I mean is that you can design a physics system that you can ensure runs on the PS5 CPU. But then the graphics boys make an upgrade and suddenly some power is diverted to the graphics and the physics system is no longer working correctly. This is still deterministic. But a nightmare to work with when optimizing for hard-real time requirements. The only way this can be done sanely is if the developers can fix this power budget and thus restore "determinism".

It sounded like they managed to make it deterministic by keeping the power draw always the same and fixing the cooling solution to handle max load.

So developer can expect it to perform identical to whatever balance they've set between CPU/GPU. That's my understanding at least.

If the developer can control the balance then it's fine. If this balance cannot be controlled by the developers then it's a nightmare. It's not clear to me this is developer controlled.

Mark Cerny's tech talk emphasised that performance is determistic:

'So how does boost work in this case? Put simply, the PlayStation 5 is given a set power budget tied to the thermal limits of the cooling assembly. "It's a completely different paradigm," says Cerny. "Rather than running at constant frequency and letting the power vary based on the workload, we run at essentially constant power and let the frequency vary based on the workload."'

https://www.eurogamer.net/articles/digitalfoundry-2020-plays...

It's clear from the presentation that it's not possible to run the CPU at full power and the GPU at full power at once. This means the CPU will be stealing performance from the GPU or vice versa when the system is pushed to it's limits. This means the things you can graphically do are directly connected to how much the CPU is loaded. This will be very hard to optimize for. Since seemingly unconnected physical pieces of hardware are influencing eachother.

The only way I can see this going well if this balance is fixed by the developer. E.g. the developer specifies I want 60% of the power budget to go to the GPU and 40% to the CPU. If this is handled dynamically by the PS5... oh boy.

> the CPU will be stealing performance from the GPU or vice versa

Another way to think about it is that very few games are using 100% CPU capacity and 100% CPU capacity at the same time. This model gives the developer a combined compute budget, which can be allocated as required for the task.

> Another way to think about it is that very few games are using 100% CPU capacity and 100% CPU capacity at the same time.

Right, but to the GP's point, for those games which do aim to max out both at the same time (or at least get as close to that as possible), this could be a development pain point unless the developer has control over which gets prioritized. Once upon a time this sort of maximization was a common goal for console game development, though I don't know how true that holds nowadays.

So then you would tune your game to use 50% CPU and 50% GPU if the requirements are equal on both sides.

But I think this style of hardware also changes the way you make optimization decisions. I.e. in a world where you have a fixed CPU/GPU budget, you might try to think of ways too move work to the CPU if for instance your GPU is saturated. But if the relative capacity is variable, you can just do things in the most efficient way overall, and you just have to make sure you don't exceed your overall budget.

> Developers who try to push every ounce of power out of it will hate it.

How common is this though in an era where most releases are multi-platform?

The PS still sells itself heavily based on console exclusives (Uncharted, The Last of Us, Horizon Zero Dawn) that are ostensibly well-tuned to the hardware. Guess we'll have to wait until HZD's makes it to PCs to see how easily it's ported.

I think Death Stranding is on the docket first for a PC port; that uses the Decima engine as well.

I don't know to what degree it is deterministic or controlled by the developer. It might be interesting if, for instance, graphically heavy games can essentially trade CPU performance they're not using for additional GPU headroom in an intentional, explicit way.

does it have to be non-deterministic? couldn't they add an option to hint which should be prioritized? maybe

  hint_prioritize_cpu_begin()
  ...
  hint_prioritize_cpu_end()

or something?

Reply on news.ycombinator.com