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

[mrcorbo]

Cerny stated it during his reveal. Here's a Eurogamer article with a quote:

"Rather than look at the actual temperature of the silicon die, we look at the activities that the GPU and CPU are performing and set the frequencies on that basis - which makes everything deterministic and repeatable," Cerny explains in his presentation. "While we're at it, we also use AMD's SmartShift technology and send any unused power from the CPU to the GPU so it can squeeze out a few more pixels."​

If power delivery is constant and clocks are known variable and not tied to thermals, and unused power from the CPU has to be shifted to the GPU to achieve the highest clocks, then the highest clocks for the GPU and CPU can't be achieved at the same time. Otherwise clocks would be at the maximum and power wouldn't have to be shifted to the GPU to achieve those speeds.

I think you may be missing part of the formula there. CPU and GPU clocks could both be maxed at times when both aren't also at high load.

 

[Proelite]

Cerny stated it during his reveal. Here's a Eurogamer article with a quote:

"Rather than look at the actual temperature of the silicon die, we look at the activities that the GPU and CPU are performing and set the frequencies on that basis - which makes everything deterministic and repeatable," Cerny explains in his presentation. "While we're at it, we also use AMD's SmartShift technology and send any unused power from the CPU to the GPU so it can squeeze out a few more pixels."​

If power delivery is constant and clocks are known variable and not tied to thermals, and unused power from the CPU has to be shifted to the GPU to achieve the highest clocks, then the highest clocks for the GPU and CPU can't be achieved at the same time. Otherwise clocks would be at the maximum and power wouldn't have to be shifted to the GPU to achieve those speeds.

That's how I understand it based on AMD's explanation of SmartShift. You can see the power balance fluctuating to some extremes.

 

[psorcerer]

Sadly a lot of games still get bottle-necked (CPU-limited) by clock speed because of the main or render thread ... Thankfully Doom Eternal is here to save the day and it doesn't have a main thread or a render thread! WTF?! Apparently it's some kind of job graph system and all cores are used equally. Can't wait to see a talk on that. Also want to see some benchmarks between like 6 core high clock and 8 core low clock and see which one can push the most frames. Hopefully that becomes more common and core utilization is very high this gen.

https://imgur.com/a/y86TjsQ

 

[psorcerer]

We also have to remember that the PS4/XB1's CPU were quite anemic even by their launch date's standards, which is not the case this time.

Doom Eternal.
It wasn't anemic. It was just too hard to use properly for an average developer.
That problem will be a more exacerbated this generation: where to find developers that can use the hardware.

I wouldn't expect CPU to be a bottleneck this generation

You greatly underestimate bad SE practices.

 

[AbsoluteBeginner]

https://imgur.com/a/y86TjsQ

Everything we play today is designed with Jaguar cores in mind. They would go for 4C/8T Zen2 if they thought 8C was not required, which will be the case next gen. In fact, this was biggest bottleneck this gen and with SSD biggest request from developers.

 

[psorcerer]

Everything we play today is designed with Jaguar cores in mind.

Do you know what exact design decisions are related to the Jaguar?

 

[Scott_Arm]

https://imgur.com/a/y86TjsQ

It's beautiful.

 

[AbsoluteBeginner]

Do you know what exact design decisions are related to the Jaguar?

I know that Jaguars have to run it, if developers are to survive.

 

[psorcerer]

He's talking about view frustrum culling.

Nope. He's hinting that it's not even in RAM.

 

[psorcerer]

It's beautiful.

It should have been the f...ing norm this generation!

 

[BRiT]

I think you may be missing part of the formula there. CPU and GPU clocks could both be maxed at times when both aren't also at high load.

Which happens when during a game? Loading time? Movie playback?

 

[iroboto]

Which happens when during a game? Loading time? Movie playback?

likely maximum speeds unless it's doing some heavy lifting in the background.

He's right in that power != frequency.

But the type of load and the amount of load on a CPU/GPU will definitely increase power requirements.
All you need is 8 cores all going at once on the CPU doing different things, or you are pushing the limits of your GPU and making all sorts of async calls so that it's filling every gap with something.

Once that happens, the frequencies will need to drop, there's just not enough power available.

 

[psorcerer]

Which happens when during a game? Loading time? Movie playback?

Which happens at least 100 times in one frame. For example.

Once that happens, the frequencies will need to drop, there's just not enough power available.

Or power and temperature will rise.
We don't know yet which is better though.

 

[Rurouni]

Which happens when during a game? Loading time? Movie playback?

Obviously during every game except of God of War
For an extreme example, if I compare the temperature difference between running a power virus (Prime95, etc) vs offline rendering (Vray), if I look at the task manager, both are utilizing the CPU at 100%, but the temperature difference is substantial. So I believe during Prime95 it is using more amps (drawing more power) vs offline rendering.

 

[PSman1700]

Explains why the PS4 fan screams when there's a 60fps game going on. Doom 2016 stresses that thing more then any 30fps game.

 

[iroboto]

Or power and temperature will rise.
We don't know yet which is better though.

power is fixed. That's the whole point.
Temps will rise but you need power on the chip to keep it from crashing.
Lack of power under load = crash.

Try running a 2080 TI without a good PSU and see how stable your system is while playing games.

The PS5 is designed at a fixed power output, so there is no boost when there is power to share. It's always boosted. when you need more power to keep from crashing you have to draw from each other. If power is being drawn you need to reduce frequency. As the loads continue to increase and there is no more power to draw, it must also reduce frequency.

Theoretically you could be in a perfect storm scenario where you biased power towards the GPU, created the most insane pieces of graphics tech that saturated like 90% of the hardware at all times, and drain the CPU to very little power and still have to downclock itself just to keep from crashing.

 

[mrcorbo]

Which happens when during a game? Loading time? Movie playback?

Not being at high load doesn't mean being at low load. Where the bar is in between those two extremes where the power demands are too much to satisfy the demands of both devices along with the degree to which the clocks have to be throttled on one device vs the other to drop below that limit are the key questions.

I see 6 scenarios being possible:

• Both CPU and GPU are within their power budgets at max clocks.
• GPU is over it's power budget but CPU is underloaded and the GPU can borrow some of the power budget from the CPU to maintain max clock.
• GPU is over it's budget and CPU isn't sufficiently underloaded to the point where it has enough power to spare the GPU to maintain max clocks. GPU throttles clocks to the degree required to make up the difference.
• The reverse of the above two scenarios but with the CPU needing more power.
• Both exceeding their power budget at max clocks and both needing to throttle clocks.

 

[iroboto]

Not being at high load doesn't mean being at low load. Where the bar is in between those two extremes where the power demands are too much to satisfy the demands of both devices along with the degree to which the clocks have to be throttled on one device vs the other to drop below that limit are the key questions.

I see 6 scenarios being possible:

• Both CPU and GPU are within their power budgets at max clocks.
• GPU is over it's power budget but CPU is underloaded and the GPU can borrow some of the power budget from the CPU to maintain max clock.
• GPU is over it's budget and CPU isn't sufficiently underloaded to the point where it has enough power to spare the GPU to maintain max clocks. GPU throttles clocks to the degree required to make up the difference.
• The reverse of the above two scenarios but with the CPU needing more power.
• Both exceeding their power budget at max clocks and both needing to throttle clocks.

Nailed it.
I just don't like people using point (1) as the general basis of comparison. It's unlikely that all the titles will not push the hardware. And if it is like that, that's unfortunate.

 

[psorcerer]

power is fixed. That's the whole point.

Not on the competitor's console.

Temps will rise but you need power on the chip to keep it from crashing.

Yep. That's what I'm hinting at.

Theoretically you could be in a perfect storm scenario where you biased power towards the GPU, created the most insane pieces of graphics tech that saturated like 90% of the hardware at all times, and drain the CPU to very little power and still have to downclock itself just to keep from crashing.

Or exactly the opposite: the spikes in CPU and GPU power exactly cancel each other in the regular operation. And you're running full speed all the time.

 

[PSman1700]

power is fixed. That's the whole point.
Temps will rise but you need power on the chip to keep it from crashing.
Lack of power under load = crash.

Try running a 2080 TI without a good PSU and see how stable your system is while playing games.

That's true. For overclocking, try undervolt a CPU, when stressed things crash.