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

Status
Not open for further replies.
But why have that if you can have a sustained level of performance at say 2% lower gpu clocks?



What about games that will tax both to the max? Late gen games could actually do that.

By now it's pretty clear that you view variable clocks as a compromise and nothing else. I suggest we stop beating that horse and move on.
 
I don't fully understand the coherency module and cache scrubbers, I assume it's another RDNA2 thing...
No, custom for Sony only, Cerny mentioned explicitly cache scrubbers were only in PS5. I am very intrigued with these as is what seems to make the gpu more efficient than stock RDNA2.
 
No, custom for Sony only, Cerny mentioned explicitly cache scrubbers were only in PS5. I am very intrigued with these as is what seems to make the gpu more efficient than stock RDNA2.
The PS4 had a customization where each L2 cache line had a volatile flag. Compute shaders could flag the lines they read or wrote for selective invalidation, rather than flushing the whole cache. This would on average reduce the time needed to flush the cache and the bandwidth cost of writeback to just compute-written lines.
It doesn't seem like AMD adopted that going forward, so it may have been considered too specific to a console or only modestly beneficial for the amount of complexity used.

Perhaps in the PS5 case the scrubbers are a modest benefit for a tightly integrated system, but it may not be generally applicable to discrete GPUs or PCs that may not have that level of integration at all for some time.
 
The 22 GB/s was given more as a peak value for some very lucky compression case, perhaps a lot of trivial or uniform values. That may point to a corner case in the existing decompression hardware that BCPack does not have. It's possible that if BCPack is more complex, they had other priorities to optimize for than a trivial decompression case that games wouldn't use.

Thanks, clears up alot :cool:

But you wont be able to run them all the time at max. I dunno, I would rather they hit 2150MHz and 3.5GHz locked, then variable. Few % up and down, when both will not work at max seems like meh solution...

Yes, why bother, that 2150mhz wont matter in real world performance?

By now it's pretty clear that you view variable clocks as a compromise and nothing else. I suggest we stop beating that horse and move on.

Then enlighten me, why it is better then having a 2% lower clock, and not having to have this boost clock? Is that 2% extra in clocks worth it? Did you see Alex from DFäs post on resetera and here about it?
To clarify, yes it IS a compromise, thats why variable clocks do exist. Its a compromise between TDP, power draw, and performance. I just find it strange that when this variable clock is basically never needed, and when its needed, only 2% in downclocks are being accounted for.
People then think there must be more to it. That leaves it open for discussions. To my question above, i have not seen any clear clarification to why it is better to have this variable clock boosts over a 2% downclock on the APU, as that 2% downclock would mean next to nothing in performance, yet gurantees stable clocks, a more stable system and less heat output?

There must be a reason why MS boasted about not having to have variable clock rates. They where awhere of the PS5's.
 
Then enlighten me, why it is better then having a 2% lower clock, and not having to have this boost clock? Is that 2% extra in clocks worth it? Did you see Alex from DFäs post on resetera and here about it?
To clarify, yes it IS a compromise, thats why variable clocks do exist. Its a compromise between TDP, power draw, and performance. I just find it strange that when this variable clock is basically never needed, and when its needed, only 2% in downclocks are being accounted for.
People then think there must be more to it. That leaves it open for discussions. To my question above, i have not seen any clear clarification to why it is better to have this variable clock boosts over a 2% downclock on the APU, as that 2% downclock would mean next to nothing in performance, yet gurantees stable clocks, a more stable system and less heat output?

There must be a reason why MS boasted about not having to have variable clock rates. They where awhere of the PS5's.

I actually don't see anybody claiming what you are claiming other people are claiming.

If you monitor CPU/GPUs to any extent in PC gaming, you'd see 2% clock (or even 5% clock) as a mere blip that gets lost in all the variables,
let alone a "conditional' downclock, and thus is not a topic that deserves too much attention (due to it being undetectable to the player) while you're making sound like it's a blasphemous thing to do.
You can disagree, but I've yet to see a 2% overclock do miracles in any game I played. I've had experiences where I changed the clock up to 10% and it I probably couldn't spot it in a double blind test.

We understand what the boost clock is.
It's a solution to push the envelope on the current hardware and it comes with compromises.
Sony chose this solution (with on hands data, of course) probably thinking the compromises are worth it, while you probably disagree.
Since we don't have further data or examples the discussion probably isn't going anywhere.
 
I actually don't see anybody claiming what you are claiming other people are claiming.

That's alot of claiming ;)

Sony chose this solution (with on hands data, of course) probably thinking the compromises are worth it, while you probably disagree.
Since we don't have further data or examples the discussion probably isn't going anywhere.

So, why didnt MS go for variable boost clocks? Why did they advertise it was a good thing not to have it?
Ye, actually we do, it is a speculation thread. Untill Sony provides us with real data of whats going on, people will speculate, discuss and wonder.
 

Although, he never said exactly 2% as in data. He said something like 'a couple of percent' downclock depending on the situation, to save 10% on power. That was an example. To save 15%, you would need to downclock more. Then we don't know how that is in relation to the CPU then. Like DF wonders, what if both are stressed, in like say, a 60FPS big open world game with alot of things going on.
 
That's alot of claiming ;)



So, why didnt MS go for variable boost clocks? Why did they advertise it was a good thing not to have it?
Ye, actually we do, it is a speculation thread. Untill Sony provides us with real data of whats going on, people will speculate, discuss and wonder.

Comes to their own HW design. You're actually questioning why any CPU/GPU varies their clocks on the fly/load, which I think they do by default.

Why isn't anybody questioning Sony going for lower clocks on the CPU but way higher clocks on the GPU compared to XBSX? That's a more interesting question IMO.
 
The PS4 had a customization where each L2 cache line had a volatile flag. Compute shaders could flag the lines they read or wrote for selective invalidation, rather than flushing the whole cache. This would on average reduce the time needed to flush the cache and the bandwidth cost of writeback to just compute-written lines.
It doesn't seem like AMD adopted that going forward, so it may have been considered too specific to a console or only modestly beneficial for the amount of complexity used.

Perhaps in the PS5 case the scrubbers are a modest benefit for a tightly integrated system, but it may not be generally applicable to discrete GPUs or PCs that may not have that level of integration at all for some time.

Depending on how the BC is setup on the PS5, could the scrubbers be a requirement to serve the same function for invalidation?
 
Why did they advertise it was a good thing not to have it?
Because they don't have (or need) it?

10% less heat is a big win. So there is no doubt it is worth it.
We probably can conclude Sony operates a bit far from the sweet spot. But if it works why not? It can increase performance and remains within some practical range.

Then enlighten me, why it is better then having a 2% lower clock, and not having to have this boost clock?
Because it can be more than just 2%, we get it.
What really matters is that Cerny guarantees the same performance for each box under regular room temperature conditions.
So it's not silicon lottery or any kind of problem - just some tweaking parameter more the devs will tune to their needs.
Ofc. this assumes we trust in Cernys words. If we don't, we would not talk about 2% but something else.
 
Comes to their own HW design. You're actually questioning why any CPU/GPU varies their clocks on the fly/load, which I think they do by default.

Ok, why is sony even mentioning it then?

Why isn't anybody questioning Sony going for lower clocks on the CPU but way higher clocks on the GPU compared to XBSX? That's a more interesting question IMO.

Aboslute limits regarding both CPU/GPU? If they could clock the CPU higher, they probably would have. They have shown us their ideas of clocking high now.

10% less heat is a big win. So there is no doubt it is worth it.

So its a big win, but loose only 2% that no one ever is going to notice. That is if the GPU doesnt allow for more savings.

It can increase performance and remains within some practical range.

How much is that 2% in upclock going to give in performance then?

Because it can be more than just 2%, we get it.

Aha, well then, then i understand it. Takes away all my confusions. Yes, it is worth it then.

What really matters is that Cerny guarantees the same performance for each box under regular room temperature conditions.

Thats.... a good thing :p We wont have to be lucky when buying on :)
 
Depending on how the BC is setup on the PS5, could the scrubbers be a requirement to serve the same function for invalidation?
Possibly. Cache invalidations are handled by some kind internal controller or internal microcode loop, which a scrubber would seem to behave like. As described, a scrubber is more concerned with an address range, but perhaps it can invalidate lines with less precision by tracking a limited set of ranges for compute shaders and falling back to a general invalidation if there's too much spread.
That might be close enough when emulating PS4 code if the original customization wasn't carried forward.
Brute force clock speed might also be enough for PS4 code, though.
 
https://www.resetera.com/threads/nx...eneration-is-born.176121/page-4#post-30070900

Devs will choose whether they want full Power to gpu or full Power to CPU where one or the other underclocks below the listed spec. So a game to game Basis. I imagine most cross gen games will choose to prefer higher clocked gpu Mode as they will be gpu bound even if the Zen cores are underclocked. Zen just runs around the Jag that most cross gen games are not going to worry about CPU time, especially 30 fps games.​

So unless the devs push both sides to the max at the same time it should be fine.
 
But why have that if you can have a sustained level of performance at say 2% lower gpu clocks?

What about games that will tax both to the max? Late gen games could actually do that.
Thermal Design Margins.

He explains they don't need a bunch of safety margins for unforseen code structure which is the case with a fix clocked. They need to add margins for the unknown.

Instead of a fix low clocking for the rare cases, and then add a margin on top, they can clock for the majority of typical cases because the worst that can happen is a small downclock for the rare cases, and more downclock possible for some catastrophic main menu trying to do 1000fps with AVX code, or anything they couldn't predict. Not just between games, but within the same game or even within the same frame. The cpu to gpu limits are more complicated I think.

There are massive advantages from this technique, and the reason it was never done in a console before is because it wasn't really possible to have perfectly deterministic outcome. AMD seem to have solved this recently.

Look at the ps4 games in the first year we had killzone touching 150W, I don't think we saw later games any higher than 170W, and still games above 150W have been very rare. With this as a reference, a 150W top wattage limit would have 2% frequency drop for a handful of later games. A 120W limit would have been more dramatic. A 170W limit would have practically never dropped in any game ever.

So the wattage limit and the cooling system they designed will have to support their claim of staying at or close to 2.23 the majority of the time. It's all about how efficent RDNA2 is.
 
Status
Not open for further replies.
Back
Top