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

[BRiT]

I only recall seeing VRS in depth discussion about Gears Tactics, but that's a PC title. Here's the DevBlog about that particular implementation: https://devblogs.microsoft.com/directx/gears-tactics-vrs/

We probably will see VRS during the Microsoft July event.

 

[ultragpu]

So does that mean footage from Halo Infinite and Hellblade 2 trailers didn't have any either?

 

[iroboto]

So does that mean footage from Halo Infinite and Hellblade 2 trailers didn't have any either?

You won't see VRS kick in often for cinematics. Usually in gameplay. So it would be hard for us to look at cinematics and say yea VRS is on, it's a cost savings technique, not a graphical enhancement technique.

 

[BRiT]

I suspect Microsoft will highlight it when they reveal more information in July.

It's supposed to be nearly impossible to perceive, so you'd have to do some obsessive level of pixel inspection examining every aspect of every pixel of every frame to tell. It's more difficult to spot without any VRS ON / VRS OFF comparison shots.

They might even be using it dynamically, so it's only turned on if frame rates are barely exceeding a given target. Though it could be used to finely tune certain stress sections of the game to smooth out performance.

 

[chris1515]

You won't see VRS kick in often for cinematics. Usually in gameplay. So it would be hard for us to look at cinematics and say yea VRS is on, it's a cost savings technique, not a graphical enhancement technique.

VRS can be used behind depth of field I suppose it will be used in cinematic like this.

 

[Shifty Geezer]

You won't see VRS kick in often for cinematics. Usually in gameplay. So it would be hard for us to look at cinematics and say yea VRS is on, it's a cost savings technique, not a graphical enhancement technique.

You also won't see it in compressed video feeds as the small details VRS ignores are the same ones lossy encoding will reject.

And in fact, if VRS works really well, you won't notice it anyway! It should be very hard to pick out, maybe identifiable only as a difference in detail between a screenshot with and a screenshot without.

 

[Jay]

It's supposed to be nearly impossible to perceive, so you'd have to do some obsessive level of pixel inspection examining every aspect of every pixel of frame to tell. It's more difficult to spot without any VRS ON / VRS OFF comparison shots.

Depends on how aggressive they go.
https://www.eurogamer.net/articles/digitalfoundry-2020-gears-tactics-tech-analysis

Even in the default on mode, it degrades regions of shading more aggressively than Wolf's performance mode, with the impact to quality more pronounced in Tactics' own performance alternative.

Lot more about it there which is worth a read.

The goal is as you say to be nearly impossible to perceive, but as always depends on the implementation, this even effects the UI in tactics.
Considering this is a latest 1P title it should really be a poster child for its use. Although the game type may not help.
Needed a setting with it being less aggressive.

 

[BRiT]

The goal is as you say to be nearly impossible to perceive, but as always depends on the implementation, this even effects the UI in tactics.
Considering this is a latest 1P title it should really be a poster child for its use. Although the game type may not help.
Needed a setting with it being less aggressive.

IIRC, they're only using Tier 1 VRSS. It should be better with Tier 2.

 

[iroboto]

VRS can be used behind depth of field I suppose it will be used in cinematic like this.

Yea I think halo infinite may have been using this method.

 

[Jay]

IIRC, they're only using Tier 1 VRSS. It should be better with Tier 2.

Yea that didn't help. Still needed a less aggressive setting also.

Wonder what the breakdown of hardware is for tier 1 and 2.
Or do we have to now worry that tier 1 will be used a lot once next gen releases to capture wide range of hardware.

The thing is, I think they should've used 2 to be an example of what a good implementation can do.
If hardware doesn't support it, use DRS or lower different setting (it is pc after all)

 

[iroboto]

Yea that didn't help. Still needed a less aggressive setting also.

Wonder what the breakdown of hardware is for tier 1 and 2.
Or do we have to now worry that tier 1 will be used a lot once next gen releases to capture wide range of hardware.

The thing is, I think they should've used 2 to be an example of what a good implementation can do.
If hardware doesn't support it, use DRS or lower different setting (it is pc after all)

Everything should be tier 2 going forward. I believe the move for tier 1 Was to support intel. Otherwise all GPUs that support VRS will support up to 2.

and I agree, it will be really hard to spot tier 2. Good luck!

 

[Jay]

Everything should be tier 2 going forward. I believe the move for tier 1 Was to support intel. Otherwise all GPUs that support VRS will support up to 2.

and I agree, it will be really hard to spot tier 2. Good luck!

Yea, have to admit I do expect stuff to be tier 2.
Which is one of the reasons I find the use of tier 1 strange for tactics. If it had come out a year or so ago it would make sense.
It'll be like halo infinite using tier 1 for the same reason. It's same year of release even.

Anyway, was just something that I found to be crazy as a demonstration of the technology their promoting.

 

[iroboto]

Yea, have to admit I do expect stuff to be tier 2.
Which is one of the reasons I find the use of tier 1 strange for tactics. If it had come out a year or so ago it would make sense.
It'll be like halo infinite using tier 1 for the same reason. It's same year of release even.

Anyway, was just something that I found to be crazy as a demonstration of the technology their promoting.

Tactics is meant for a variety of devices. Laptops etc. It’s a strategy game. You probably want the widest spread.

Plus; learnings.

 

[Jay]

Tactics is meant for a variety of devices. Laptops etc. It’s a strategy game. You probably want the widest spread.

Plus; learnings.

Same for all pc games, laptops is just a thing now.

It's got so many settings that can be tweaked, I doubt would have problems getting it to run on the hardware that didn't support tier 2.

But VRS is one of their big messages, so you would want to show it in the best possible light.

Exactly, learning would be a really good reason for tier 2, especially if that's what we expect most to use.

Anyway, didn't want to drag this out as much. Was really about that vrs should be pretty invisible, but ms implementation for their most recent game wasn't.

 

[BRiT]

Well it is, as long as you dont have a means to compare with it on and off and if you dont pick to use the extreme performance variation.

 

[DuckThor Evil]

As I noted perhaps in another thread, just working out the wattage required, it's fairly significant.
To put it into perspective, the power required to get the same PS5 chip running to 1825 Mhz -- you need to double that amount to get it to 2230 Mhz.

I've seen you mention this a few times and I'm having difficulty understanding how you can state so? I saw the equations you put out in some thread, but you can hardly just toss out an equation that is able to point out real world power scaling characteristics of all chips, regardless of where they are at their power curve. You seem to think all chips behave the same way? That is not the case at all. Here for example is an AMD APU tested by Anandtech with different clock rates/voltages and the resulting power consumption. You can see the power scaling is very different to what your equation would state.

https://www.anandtech.com/show/12542/overclocking-the-amd-ryzen-apus-guide-results/7

 

[iroboto]

I've seen you mention this a few times and I'm having difficulty understanding how you can state so? I saw the equations you put out in some thread, but you can hardly just toss out an equation that is able to point out real world power scaling characteristics of all chips, regardless of where they are at their power curve. You seem to think all chips behave the same way? That is not the case at all. Here for example is an AMD APU tested by Anandtech with different clock rates/voltages and the resulting power consumption. You can see the power scaling is very different to what your equation would state.

https://www.anandtech.com/show/12542/overclocking-the-amd-ryzen-apus-guide-results/7

The formula includes a set amount of power draw as well, Po. If we are looking at total power, and include numbers we must calculate holistically.
Total power = Po + Pv
I don't know what Po is as it is the amount of minimal power to keep the chip running before you engage in actual activity. So basically, just keep it 'on'.
And Pv = CV^2fa

In this exercise, C, a are going to stay constant.
With frequency varying proportionally with V.
so Pv = f^3
no, not that far off actually.
3000 Mhz vs 4000 Mhz
that's 4/3
(4/3)^3
= 2.37 times more power.

So if I account for 30 cores, lets say the minimum power to keep each core active is 1W.
So it's really 30*(1W) + 39W @ 3000 Mhz. = 69W

Lets look at the variable now for 4000Mhz:
39W * 2.37 = 92.43W

Go back to our equation here
Po + Pv = Total power
30W + 92.43W = 122.43W

Not all that far off from 118W

That's not accounting for any other forms of power redistribution.

So I chose 30 just because. I really don't know what the 'Po' voltage is for each core. But it's likely a little higher than 1W. They say idle power is 18.9W (honestly I'm not sure where the power savings are, if they are shutting down idle etc) but it has 4 cores. So an idle wattage of nearly 4.725W per core.

A Ryzen 3600 is 51W idle with 8 cores. So 6.375W.

I'll try to plug in 4.725W per core
4*4.725W + 50.1*2.67 = 152.667W

Not all that far from the mark. I suspect Po is higher per core, but on idle they may be shutting down a lot of transistors.

 

[DuckThor Evil]

So if I account for 30 cores, lets say the minimum power to keep each core active is 1W.
So it's really 30*(1W) + 39W @ 3000 Mhz. = 69W

Lets look at the variable now for 4000Mhz:
39W * 2.37 = 92.43W

Go back to our equation here
Po + Pv = Total power
30W + 92.43W = 122.43W

Not all that far off from 118W

That's not accounting for any other forms of power redistribution.

Those are 4 core parts though... Then have the equation show results for 3.8Ghz vs 3Ghz. The 4Ghz is starting to be too much for this CPU, it's at the ceiling with high volts and can't go further, the efficiency loss only really starts there. Up till around 1.3v the power c0nsumption stays in check.

 

[iroboto]

Those are 4 core parts though... Then have the equation show results for 3.8Ghz vs 3Ghz. The 4Ghz is starting to be too much for this CPU, it's at the ceiling with high volts and can't go further, the efficiency loss only really starts there. Up till around 1.3v the power c0nsumption stays in check.

Yea I just used the 4 core wattage. i recalculated above.

Volts and Wattage aren't the same.

Okay lets do this right since I see a mistake above.
3800/3000 = 1.267
Assume Po = 4*4.725
PW then = 69 - 18.9 = 50.1
Then it becomes
4*4.725 + 50.1*1.267 = 81.9W

It's showing about 96W on Anandtech.
it's not all that far off, the issue is calculating Po correctly. I don't know what type of savings there are on idle.

I also can’t account for their dynamic voltage frequency scaling algorithms. Which will kick in at the higher end

 

[DuckThor Evil]

I think you forgot the "^3" on the last one. Not really wanting to hold on to this, but I've seen how different silicon behaves under different situations and they certainly do not scale identically. The results can be all over the place.