Easily gone higher huh? So how high do you expect PC RDNA 2 chips to clock?Because they had a set power draw in mind. Cerny said they could have easily gone even higher implying that they just didn't want to slap a big ol' PC PSU in there.
Easily gone higher huh? So how high do you expect PC RDNA 2 chips to clock?Because they had a set power draw in mind. Cerny said they could have easily gone even higher implying that they just didn't want to slap a big ol' PC PSU in there.
. We dont know how low it will clock but Cerny did state that they couldnt achieve a locked 2 ghz, which is just over 10% less. By far the most reasonable deduction to be made from the information we have is that under heavy load the GPU will likely need to drop below 2 ghz.
That's what Cerny said and I'm sure they did the tests, otherwise he wouldn't have claimed that.Easily gone higher huh? so how high do you expect PC RDNA 2 chips to clock?
He did say that with a prefix "with old design". But you can always decide to ignore what he said and fixate on something else to suit your opinion.
Yes on the OLD design, so how the hell can someone concludes that means ps5He did say that with a prefix "with old design". You can always decide to ignore what he said and fixate on something else to suit your opinion.
Not much, maybe not even close. At that frequency you're already past the point of diminishing returns. Silicon have a power/performance exponential curve that after a certain point the power needed to increase performance will skyrocket. People like to say that the PS5 GPU is "optimized" or "efficient" when in reality at that frequency it's neither of them. N7 process doesn't fix this, it actually makes it worst.Easily gone higher huh? So how high do you expect PC RDNA 2 chips to clock?
This thread is 26 pages long because people are twisting words and concluding whatever the hell they want. NXGamer couldn't have simplified it any further and yet here we are.Yes on the OLD design, so how the hell can someone concludes that means ps5
People rather read it the opposite way.
"Holy shit ! Cerny is so good that he can reduce power consumption by 10% just with a few percent downclock".
It's able to more closely track a power budget than a system with a fixed clock, so in that sense it's an efficient use of die space and the available parts budget. Yes, you could run the whole system at half the speed for a small fraction of the power. Of course, then you'd have a system running at half the speed and you're still paying just as much for it. So just about everyone runs silicon as fast as they can get away with reliably unless cost is no object or power efficiency is incredibly important, as when considering battery life. It doesn't matter if it's an Intel CPU, AMD APU, or Nvidia GPU, power scaling around the clock speeds devices ship at scale at roughly the cube of the clock. This chip isn't an exception, it's just following the same curve as everyone else.
I wonder what the system's total power draw is. In my mind it would have been way more efficient to clock the chip a couple percent lower and just boost to that frequency whenever the amount of processing load demanded it.
Doesn't it seem incredibly wasteful though to spend 10% of the system's total power budget, on a platform that is bound to be constrained by power limits, just to add a couple of percentage points to to the GPU's already stratospheric clocks?
If there's one thing people say about Cerny, it's that he's all about the marketing
If there's one thing people say about Cerny, it's that he's all about the marketing
I mean look at him - Mark Cerny is the slick face of Playstation marketing. He's not at all like a mild-mannered professor of EE/Game design/SWEIf there's one thing people say about Cerny, it's that he's all about the marketing
I mean look at him - Mark Cerny is the slick face of Playstation marketing. He's not at all like a mild-mannered professor of EE/Game design/SWE
It doesn't mean Cerny is a marketing person nor that he has no tech knowledge. The person clearly knows his job and he's really talented at engineering and such. But it's not incompatible with a few marketing stuff here and there. And yes, a big 10Tflops in a slide is a better sell than 9. The same way "Supercharged PC architecture" is a good sounding word to sell the tech they're making.
Doesn't it seem incredibly wasteful though to spend 10% of the system's total power budget, on a platform that is bound to be constrained by power limits, just to add a couple of percentage points to the GPU's already stratospheric clocks?
Github leak really did numbers on people. They still believe it was 9.2tf and was pushed to 10tf at the last moment while also replaced RDNA1 with 2, putting hardware RT in and replacing the whole cooling system.That's not how it works. Why do people at this point still not understand how it works?
I don't really understand this "Sony is desperate to get over 10 TFlops, so they OC'd their chip like crazy." If Sony was as concerned about getting double digit TFs for marketing reasons, then going wide and slow like MS was an option they could have easily taken at the start. You don't design a 36 CU chip and expect to get super high double digit TFlops. You'd just be barely touching double digit TFlops using such a config, and that's only after using super high clocks.
But it's only 10% of the system's total power budget when the power budget has been reached. There will be many instances where it looks like this:
- Processor load 100%, power consumption 50%, clocks 100%
or
- Processor load 100%, power consumption 75%, clocks 100%
At the end of the day they have a power budget that they are sticking to. I think they have made their choices based on affordability while also squeezing as much out of their hardware as possible and I think they have done this in an elegant and efficiënt way.
Cheaper chip, cheaper cooler, cheaper power supply and only 15% less perfomance.
That's what Cerny said and I'm sure they did the tests, otherwise he wouldn't have claimed that.
Whether you believe him or not is your choice but it won't change the reality no matter what it ends up being.
However, based on what we know about RDNA2 so far seems to point in the direction that Cerny wasn't lying.
/edit: This guy has a take on clock speeds towards the end of his video:
"Gotta push the Tflops over 10 for the marketing slides."
So you think the console is 9tf?
And Cerny is marketing a 9tf console to developers who will work on it as 10tf? Lol
Genius lol
Higher clocks increase temperature & power draw exponentially. It's very much possible the XSX runs cooler and is more power efficient despite being a bit faster. That's the big downside of going for higher clocks over going wider with CUs. The upside is that it's cheaper when it comes to the chip.Cheaper chip, cheaper cooler, cheaper power supply and only 15% less perfomance.
Not really, it means their current strategy (fixed power budget) allowed for higher clocks most of the time than a fixed clock speed strategy would have, so why not take advantage of it?That's not what I'm saying. I'm not saying it's a 9tf machine in practice but a 10tf one. But according to Cerny, a few % of clock reduction means around 10% of power consumption drop, which indicates the GPU was clocked beyond its optimal clockspeed ratio compared to power consumption, to reach a certain figure.
Not really, it means their current strategy (fixed power budget) allowed for higher clocks most of the time than a fixed clock speed strategy would have, so why not take advantage of it?
That's why there is actually a tangible benefit to doing variable clocks, you can get a higher clock speed at least sometimes than if you had to lock the clocks to what would be stable in a worst case scenario.
1. What is this ' optimal clock speed ' figure you have in your mind?That's not what I'm saying. I'm not saying it's a 9tf machine in practice but a 10tf one. But according to Cerny, a few % of clock reduction means around 10% of power consumption drop, which indicates the GPU was clocked beyond its optimal clockspeed ratio compared to power consumption, to reach a certain figure.
I'm looking at it from a different perspective. They have a specialized cooling system, and they know what that system can handle. They decided to allow the system to always use the amount of power/heat their cooling system can handle at a specified noise level.I mean, didn't Cerny said it'll be at that clock most of the time ? A console has a limited power budget usually. When you're willing to go for a few percent power "most of the time" for 10% more power consumption, that screams "attempt to get as much power no matter what". This reminds me of AMD releasing GPUs like the RX 590 which was the same as RX 580 but pushing clocks higher and TDP far higher for the sake of numbers.
I'm looking at it from a different perspective. They have a specialized cooling system, and they know what that system can handle. They decided to allow the system to always use the amount of power/heat their cooling system can handle at a specified noise level.
It may only be a few percent gain, but it's "free" in this case so why not use it? The other alternative was to have it locked to a lower clock speed so that even in the worst case the system would never go above what its cooling system can handle, but they'd lose performance the vast majority of the time versus their current solution.
1. What is this ' optimal clock speed ' figure you have in your mind?
2. You fail to take in the fact that the design he went with allowed this amount of continuous boost hence your thinking of ' optimal clock speed ' it could go even go higher according to Cerny
3. Sony doesn't seem worried about the power draw, why should we?
Edit: Think of the process first. not the end figure, you're clearly thinking 10tf then process after.
I'm looking at it from a different perspective. They have a specialized cooling system, and they know what that system can handle. They decided to allow the system to always use the amount of power/heat their cooling system can handle at a specified noise level.
It may only be a few percent gain, but it's "free" in this case so why not use it? The other alternative was to have it locked to a lower clock speed so that even in the worst case the system would never go above what its cooling system can handle, but they'd lose performance the vast majority of the time versus their current solution.
That's why I said he has a take. Not saying it's a fact.You shouldn't take anything MLID says as a fact. This guy is wrong more often than not, by a lot.
2. You fail to take in the fact that the design he went with allowed this amount of continuous boost hence your thinking of ' optimal clock speed ' it could go even go higher according to Cerny
1. Basically, most GPUs have a sweet spot between performance and power consumption. After a certain threshold, any gain in clockspeed and performance will result a far higher power consumption.
As an exemple, you have a GPU which's perfect sweetspot would be 1600mhz for 150W. At 1700mhz, it might be at 180W. You have a 6% gain in performance/clockspeed but a 20% increase in power consumption. Because you're higher than the sweetspot in performances.
2. "The design he went with". Yeah, you can clock GPUs higher. RX 5700XT for instance is usually clocked at 1800mhz. But you can push it to 2200mhz. So yes, you can always push high and higher. But as Cerny's own admission: "A few percent drop of clockspeed leads to a 10% drop in power consumption". This doesn't mean that it's a well optimised hardware, but that they pushed even further for a few percentage of performances in spite of a lot higher power consumption. They're targeting over that optimal sweetspot.
As for why not use it, I'm not against that. But it's not efficient at all. It's not a "smart design" as some people seems to imply. It's fine to go over that optimal clockspeed, let's not just pretend that it's something well done or optimized.
As for why not use it, I'm not against that. But it's not efficient at all. It's not a "smart design" as some people seems to imply. It's fine to go over that optimal clockspeed, let's not just pretend that it's something well done or optimized.
Parts of the chip could go faster, but that's not really relevant. Every chips clocks are limited by their weakest component.
Agreed, it's pretty obvious Sony didn't just make all these decisions last second.I don't really understand this "Sony is desperate to get over 10 TFlops, so they OC'd their chip like crazy." If Sony was as concerned about getting double digit TFs for marketing reasons, then going wide and slow like MS was an option they could have easily taken at the start. You don't design a 36 CU chip and expect to get super high double digit TFlops. You'd just be barely touching double digit TFlops using such a config, and that's only after using super high clocks.
Thanks, I stand corrected regarding quad data rate. Overall, I think my larger point stands: that your way of describing things is overprecise for most discussion contexts. A lot of people are confused by simple terms like "bandwidth", so when you start pulling out the niceties of "QDR" and "MT/s" and suggest complex formulas they lose the plot.
Either way, the system can only move 256 bits at a time. If the CPU needs some of that bandwidth, then it can't be used by the GPU. Hence contention, and lowered GPU bandwidth. On PS4, it seems that somehow this process of CPU access lowered GPU bandwidth by more than just what the CPU was taking. This is out of my depth, so I'm not sure either why that is, or whether it might have been fixed for PS5.Wouldn't the content on the XSX affect it more then the PS5 because of the split bandwidth, contention will cause lower bandwidth on both but on the PS5 at least all the memory runs at the same speed (as far as we know for now), meaning that any access should be at the full speed.
Which is ultimately why it can't reach clockspeeds as high.Agreed. We just don't know what "a little bit" is on the PS5.
We know what the "little bit" is on the Xbox: zero. It won't drop the clocks at full load.
What you and many others miss is the scenarios Mr. Cerny referred to as being the usual absolute maximum power draw. We're used to thinking of power draw increasing as game complexity and visual ambition rise. This is true. But past a certain point, that very complexity starts to make games inefficient. Parallelization means work dependent on other unfinished work stalls, which is why you see framerates or IQ drop in demanding sections. Counter-intuitively, the very highest power usage is when complex games run simpler loads. The engine has been optimized to shove as much work through as possible, so when the work is simple--like a menu screen--there are no stalls. Every single transistor ends up in use, and power peaks.What evidence would that be? The first "leak" was an engineering sample and is meaningless as they were never going to run at 1 ghz. If they are having issues even locking at 2 ghz what do you expect the actual sustained clocks to be? Lets assume by a couple Cerny did mean 2. 2% of 2.23 ghz is 45 mhz. That brings the clock down to 2.185 ghz which is well above the 2ghz clock they cant sustain. How do you envision this scenario playing out
1. Basically, most GPUs have a sweet spot between performance and power consumption. After a certain threshold, any gain in clockspeed and performance will result a far higher power consumption.
As an exemple, you have a GPU which's perfect sweetspot would be 1600mhz for 150W. At 1700mhz, it might be at 180W. You have a 6% gain in performance/clockspeed but a 20% increase in power consumption. Because you're higher than the sweetspot in performances.
2. "The design he went with". Yeah, you can clock GPUs higher. RX 5700XT for instance is usually clocked at 1800mhz. But you can push it to 2200mhz. So yes, you can always push high and higher. But as Cerny's own admission: "A few percent drop of clockspeed leads to a 10% drop in power consumption". This doesn't mean that it's a well optimised hardware, but that they pushed even further for a few percentage of performances in spite of a lot higher power consumption. They're targeting over that optimal sweetspot.
As for why not use it, I'm not against that. But it's not efficient at all. It's not a "smart design" as some people seems to imply. It's fine to go over that optimal clockspeed, let's not just pretend that it's something well done or optimized.
My personal objection to calling Sony's approach efficient is that Cerny himself said that a big amount of power is needed to push those clocks. "A couple percent of downclocking saves 10% power" means that you have to spend 10% of your power budget to get those clocks 2% higher, right? I wouldn't characterize this as efficient design.
I disagree. The way it was implemented, it has no major drawbacks. The tricky bit was making the performance predictable and consistent in all environments and in all PS5's. That's what was well done and clever.
Well considering it should run at those clock speeds most of the time vs the traditional boost and the added space on the main chip this leaves for the i/o customizations they added seems to be smart design to me.
I disagreed with the 'optimal clock point' as the design allows them the flexibility
If your argument is strictly about power to clock relation then I'll agree to disagree cause it doesn't seem like they're worried about power draw so I'm not sure why we should?
Consoles will not always work at maximum capacity anyways
I don't really understand this "Sony is desperate to get over 10 TFlops, so they OC'd their chip like crazy." If Sony was as concerned about getting double digit TFs for marketing reasons, then going wide and slow like MS was an option they could have easily taken at the start. You don't design a 36 CU chip and expect to get super high double digit TFlops. You'd just be barely touching double digit TFlops using such a config, and that's only after using super high clocks.
All this complexity could be eliminated by just putting in a cooling system capable of handling 100% saturation at 2.2GHz. But this must be huge and prohibitively expensive, because neither platform attempted it. Series X has a novel wind-tunnel design, a huge fan, and a heatsink that takes up a lot of the interior volume, and can't reach even 2GHz fixed. (Neither could Sony, by their own account.)
What traditionnal boost ?
Where's the smart design in pushing clocks higher than the efficiency point ?
I mean by itself, I have nothing against that. I just dont see where's the smart design here.
I mean, yes it does. It's inefficient. They're pushing over the efficiency of their GPU design.
What is "well done and clever" here ?
MS still can if they put an actual LED fan in their systemStupid Sony/MS. They could've hit 3GHz easy if they just stuck some RGB and icicle decals on the fans.
I was reading this thread late last night and woke up with a thought: could it all have started with Cerny (or someone) thinking about how weird it is for a main menu screen to make the console go loud? With their deterministic approach I understood that CPU and GPU will vary frequency based on their loads (meaning, based on the instructions they will be running at a given point?), so when on a main menu or map screen (of game paused) there is no AI running, no graphics to be pushed, etc, so it will automatically reduce the frequency in those situations, right?