Current Generation Hardware Speculation with a Technical Spin [post GDC 2020] [XBSX, PS5]

Status
Not open for further replies.
Sony's approach is too extreme and whatever exotic cooling solution they employ, it will impact yields and acoustic, but as a concept it's sound and something that I propose for some year now.
Of course on this forum, many explained to me how bad an idea it's, but this is the forum that considered idiotic middle generations, portable/home hybrid, games sold on solid memory, ssd in general, ssd as an expansion, smt, cloud gaming, and so it must, of course, be a good idea.
Whenever you limit your soc by maximum power or by thermal, you can switch from compute mode to graphic mode on the fly, for example, while loading a level you can prioritize the cpu, and then during normal play tiptoe between balanced mode and graphic mode.
These three preset represent guaranteed sustained clocks so any developer can target them, plus if there's some headroom due to refined process, better cooling, sheer lottery luck, the soc can temporarily turbo some MHz up when needed.
If vrs, vrr, and dynamic resolution are employed, the lucky one gets a less upscaled resolution or a slightly higher framerate.
Patent pending.
 
Then why even mention it? It will never downclock, but lets mention variable clocks.
Because this was an explanation on a new tecnology, and it was needed to explain how it worked, and how it handled overloads.

If they could have sustained those clocks, they would, but they can't, so downclocking will happen when both the CPU/GPU are pushed too hard. Developers probably get to choose which of either they can fully utilise.
You are the one saying they can’t sustain those clocks. Cerny did not say that. He mentioned they will be there most of the time.
Most is not because they cannot sustain them, but because workloads change every time, and since clocks are variable the GPU and CPU will adjust to workloads. This is a downclock, but one that does not affect performance in any way, since the extra processing power is not required.
A simple stupid example just to make my point... Xbox has locked clocks. On every scene it will have 12.1 Tflops, Imagine we are working at 4K and locked 60 fps, and the scene doesn’t require 12,1 Tflops, it would require only 8 for the designated target. Xbox could calculate over 60 fps on that scene because it has the extra power locked . Vsync will lock them out, but the power would be there.
PS5 will only calculate the required 60 fps, since it will downclock and produce only what is required. The system will see the target needed, and adjust the speed for that workload.
This is the standard case in which PS5 will downclock. But the effect on performance will be zero. If 10,28 Tflops is required, the GPU will give them.
The other case is on the rare occasions when the system goes over the power budget. Standard systems like Xbox will overheat, but PS5 will downgrade a couple of percentage points to keep power usage and prevent overheating. This is why Cerny talked about downgrades. To explain this!
 
No, we all have our interpretations, and you are victim of selective bias.

I just follow what DF has said about it, they have developer input. Please don't attack people personally, keep it technical.

You are the one saying they can’t sustain those clocks. Cerny did not say that. He mentioned they will be there most of the time.

He basically says they can't sustain it then. Its trival to MS's presantation where they claim they can sustain their clocks, and can gurantee consistent performance across the board. They mentioned this for a reason.

The other case is on the rare occasions when the system goes over the power budget. Standard systems like Xbox will overheat, but PS5 will downgrade a couple of percentage points to keep power usage and prevent overheating. This is why Cerny talked about downgrades. To explain this!

Sony couldn't even maintain 2Ghz/GPU, now they can attain 2.23ghz, thanks to variable clocks. Even the PS5 can overheat, every device can. The variable clocks are in place because 2300mhz is too high to sustain all the time in all workloads together with that Zen 2 8 core cpu at over 3ghz.

They went with a narrow and fast design, MS did not, they don't have to trick with variable clocks to maintain their performance targets.

And again, read Alex's post on it, it's right here on this forum. Theres no reason to doubt his thoughts on it. They have developer input and even talked with Cerny.
 
Nobody said it never downclocks. The quote to use is Cerny's. What's so difficult to understand?

This is hardware predicton. Maybe the average is 2% maybe it's more. I predict from this point in time all future comsoles will,use sony's method. It will become the most efficient use of silicon, power, and yield.

I predict it won't unless console memory clocks are also part of the variable clock paradigm. In the PC space, GPU overclocks are almost always combined with memory OC for perceptible performance gain, unless of course the memory bw had head room to begin with.
 
i predict it won't unless console memory are also part of the variable clock paradigm.
It's somewhat difficult to vary the clock of gddr6, because it's clocked as high as the parametric yield alllows, there's only a little power to save going lower compared to what it's connected to (uness it's using power management modes, which they already can) and not much room to go higher. There is no provisions for dynamic voltages either. The chips initialise the PHY with a complex equalization sequence, so varying the voltage would be incredibly difficult.

Power modes are like half clock, quarter clock, etc... And standby with refresh.

We do have gddr5/6 dies characterized for two different voltages yielding two gbps figures, but it would be a pain to even try to switch between the two voltage. Probably impossible.

Then we have boutique gpus pushing the memory beyond specs. This cannot be done on a console product scale, requiring identical units, and long term procurement, and multi sourcing.
 
Last edited:
I predict it won't unless console memory clocks are also part of the variable clock paradigm. In the PC space, GPU overclocks are almost always combined with memory OC for perceptible performance gain, unless of course the memory bw had head room to begin with.

I heard his talk and took it the exact opposite way, its not really a boost more a predictable brake that will be applied if needs be.

PS4 they made poor thermal estimates or devs got so good at beating the apu that cooling sometimes went into overdrive.

Avx2 means this overdrive could be a meltdown so the apu has a governor of sorts to drop the anchor temporarily if needs be.

If the extreme you have to account for is that extreme but also that rare why degredate your day to day running, why not flip the model and reign the apu in when you need to. They way they have done it, it's repeatable and predictable rather than a random per system dip.

So more don't leave performance on the table to make an overhead, just slow when you need to. The opposite of boost whenever you can.

Probably totally wrong however.
 
I predict it won't unless console memory clocks are also part of the variable clock paradigm.

True, MS would have used the same thing for their XSX otherwise. And not boast about not having to have it. Maybe future designs would be more probable, but not with performance impacts like the PS5 has in those cases (downclocks).
 
Sony couldn't even maintain 2Ghz/GPU, now they can attain 2.23ghz, thanks to variable clocks. Even the PS5 can overheat, every device can.

Man... this system was created to prevent overheat.
Heat is produced due to workloads and respective power usage. With locked clocks you have variable workloads and power usage and cant really have efficient control over any.
With PS5 you have locked power usage. With this you can control system temperature by analysing workloads.
This is way more efficient in thermal control, and it allows the system to be pushed to higher clocks. Because variations will not exist, and if they do the system can control them (the two percent downclock saving 10% power) and prevent overheat.
Decreasing clocks would be no feat at all... and you would not add it on a presentation about inovative tecnologies on your console.
 
Man... this system was created to prevent overheat.

Im sure the PS5 can overheat, not everyone lives in a colder climate. There must be some limit to when it shuts down, i hope.

With PS5 you have locked power usage. With this you can control system temperature by analysing workloads.
This is way more efficient in thermal control, and it allows the system to be pushed to higher clocks. Because variations will not exist, and if they do the system can control them (the two percent downclock saving 10% power) and prevent overheat.
Decreasing clocks would be no feat at all... and you would not add it on a presentation about inovative tecnologies on your console.

It is still a compromise between thermals, performance and power draw. They would most likely not have achieved such high clock speeds as they did with the GPU. The reason behind it could be to max the TF number on a narrow/fast chip, to be more competitive with the competition. However you put it, it makes not much sense to anyone that when they say 'It never happens, only in very rare cases, and when it does, its a downclock by 2%'. I wouldnt even bother those two procent which in practice means nothing, and have a stable system 100% of the time no matter what.
There's probably more to it, and seeing the monster GPU clock, it doesnt surprise me. Alex's post about it does make perfect sense.
 
I wouldnt even bother those two procent which in practice means nothing, and have a stable system 100% of the time no matter what.
And where is that stable system? Consoles use locked clocks since always... and they did overheat, lock, crash, and reboot.
Is that what you call a stable system?
 
Then why even mention it? It will never downclock, but lets mention variable clocks.
Because it will drop in frequency depending on whether or not power hungry instructions are used too much.
If they could have sustained those clocks, they would, but they can't, so downclocking will happen when both the CPU/GPU are pushed too hard. Developers probably get to choose which of either they can fully utilise.
Yes. Mark Cerny said that the frequency variation is deterministic. So developers can have the GPU running at 2.23GHz forever technically.
Exactly, because that 10.28TF for the GPU is not sustained, unlike for the XSX where they are (MS even advertised 'sustained' probably because they where up to date of Sony's variable clocks)
No, I am under the assumption that 10.3TF is sustained if the developer choses for it to be. The necessity of stating "up to" is because the devs can potentially change that on their own, but not beyond.
 
Actually, this is Microsoft's fault. Their marketing department's job is to make sure that their Google search stays at the top - this isn't free of course. In my previous company, we had to do a huge amount of work to make sure that our name would be the firsr result when searching for it (it wasn't always the case), or that when searching our industry you would get our company as the first result - or at least the first three. Forget about the second page of Google search results, that's where companies die.

Your naming scheme either makes this a easy job or an extremely complicated job.
Microsoft is pretty clueless in this department.

The only sane player in this regard remains Sony. PS1~5 easy to follow, easy to search.
Nintendo keeps their names distinct enough to not be an issue, but still...
 
Status
Not open for further replies.
Back
Top