25% of 0 is measly ;)
For the quote, ctrl+f "frames". The exact wording is a few frames, I don't know where I read/heard 1-2.
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
PlayStation 5 uncovered: the Mark Cerny tech deep dive [Digital Foundry]
- Thread starter Wollan
- Start date
For the quote, ctrl+f "frames". The exact wording is a few frames, I don't know where I read/heard 1-2.
You are saying this, as well. But just to add a bit more, it has less to do with clock frequency and more to do with instructions used, type of workload, and GPU saturation. In certain workload conditions where both the CPU and GPU are being used heavily, they can both be running at their peak frequencies potentially.
I must've misread your post originally, I'll elaborate anyway.
When I say nothing's idle, I mean including cache misses and CU waiting for new data, so complex scenes don't count as transistor-heavy usage which is super counter-intuitive until you hear a GPU fan spin up like crazy while sitting on a menu screen.
Last edited:
You forgetting people buy off value, the consistent, high quality 1st party software line up and VR brings value to PS5.
Lets break this down by what has been given in the article:
race to idle - pointless increase in a GPU's frequency
"There's another phenomenon here, which is called 'race to idle'. Let's imagine we are running at 30Hz, and we're using 28 milliseconds out of our 33 millisecond budget, so the GPU is idle for five milliseconds. The power control logic will detect that low power is being consumed - after all, the GPU is not doing much for that five milliseconds - and conclude that the frequency should be increased. But that's a pointless bump in frequency,"
"The net result is that the GPU doesn't do any more work, instead it processes its assigned work more quickly and then is idle for longer, just waiting for v-sync or the like. We use 'race to idle' to describe this pointless increase in a GPU's frequency,"
"So, when I made the statement that the GPU will spend most of its time at or near its top frequency, that is with 'race to idle' taken out of the equation - we were looking at PlayStation 5 games in situations where the whole frame was being used productively.
Maybe you are you talking about this?
There isn't a lag where extra performance is available for several seconds or several minutes and then the system gets throttled; that isn't the world that developers want to live in - we make sure that the PS5 is very responsive to power consumed.
race to idle - pointless increase in a GPU's frequency
"There's another phenomenon here, which is called 'race to idle'. Let's imagine we are running at 30Hz, and we're using 28 milliseconds out of our 33 millisecond budget, so the GPU is idle for five milliseconds. The power control logic will detect that low power is being consumed - after all, the GPU is not doing much for that five milliseconds - and conclude that the frequency should be increased. But that's a pointless bump in frequency,"
"The net result is that the GPU doesn't do any more work, instead it processes its assigned work more quickly and then is idle for longer, just waiting for v-sync or the like. We use 'race to idle' to describe this pointless increase in a GPU's frequency,"
"So, when I made the statement that the GPU will spend most of its time at or near its top frequency, that is with 'race to idle' taken out of the equation - we were looking at PlayStation 5 games in situations where the whole frame was being used productively.
Maybe you are you talking about this?
There isn't a lag where extra performance is available for several seconds or several minutes and then the system gets throttled; that isn't the world that developers want to live in - we make sure that the PS5 is very responsive to power consumed.
Because it was a GDC talk.
Developers know what this all means.
You're correct, but that doesn't invalidate my statements. Clock Speed cannot compensate for lack of CUs. In the case of these machines, a 40% CU difference.
Idling isn't the same as reducing clocks. When a processor is idle it literally stops performing tasks until it is woken up by an interrupt indicating that there's more work to be done. On a performance focused device like the PS5 it will basically hit the ground running at full clocks. After all, clock speed doesn't determine power draw. It's switching transistors that draws power, and if you're not doing any work then the clock speed if basically irrelevant. Modern processors have a lot of different power states that allow various subsections to be shut down independently when not needed, further reducing power demand.
By being curious enough to look it up? Everyone should take the time to look up words and acronyms they encounter that they don't know. We used to have to refer to physical dictionaries and encyclopedias, but it's so easy to do a quick search that there's really no excuse. Google returns it as the first entry. Or if you're on macOS, just press Command+Space and type "define:FUD" and you'll see a definition. It's as easy as that.
They are actually.
But yes, the GDC talk was not really consumer-friendly. And well, that's because it was a GDC talk.
They'll properly unveil it with the box and all, and the tear down in the future.
Everything is getting delayed anyway. No rush.
Help me understand then, why would Cerny be concerned on eliminating race to idle if the idle state was the most efficient?
From what I can tell, he appears to be saying there are three cases for dropped clocks:
1.) multi-frame instances of high power draw GPU workloads
2.) Avoiding race to idle by utilizing the GPU for the entire
3.) Profile use in developer debug mode
where did Cerny confirm that the PS5 actually can do that all the time without any issues?
Why isn't it locked there then? like any other console? Why does it need to go down?
I mean Cerny himself even confirms that it gets throttled:
and
I mean, he even says "should" and "near" and not all of the time.
User Banned (3 days): antagonising other members, prior infractions for similar behavior
Which alt account are you?
The way I read it, he wasn't interested in eliminating race to idle. He just didn't want it to be part of the conversation when people talked about workloads; in other words, he wants people to understand that when he talks about the system usually running at max clocks, it's when the GPU and CPU are being used for the entire frame.
Getting work done is ultimately what presents an immersive experience to users. The most efficient thing you can do is nothing at all, which rocks are pretty good at. They're also not very entertaining.
Engineering is all a balancing act. The fact that most code doesn't demand full power draw means that you can usually run it a little faster. The fact that some code does demand that much power means you need to anticipate that situation and decide what to do with the situation. Consoles need to provide a consistent experience so historically they've gone with the simple answer: slow everything down so that you can cool even the most demanding situations. This means leaving potential on the table, so the PS5 goes for the approach that everything that can run at a higher clock speed does so, and only the most demanding workloads wind up running at the kinds of clocks you'd have wound up with if you went with a fixed clock speed. The average performance goes up as a result, but the designers of the console had to deal with the extra complexity to do so in a deterministic way that guarantees every PS5 handles the same workload at the same performance.
I think you're misinterpreting that second comment:
I think he is emphasizing that a fully rendered game can run at full speed, not just low powered situations.
It can run at both, not all the time.
The amount of time is dependent on how power intensive the instructions are.
It is probably possible to have a game always running at 100% for both, depending on which instructions are power intensive to cause a downclock.
Shoppers are not one dimensional beings, though.
Many people do not consider power to be the most important thing when buying a console. If they did then the X1X would have instantly outsold every console on the market, the Switch would have been DOA, etc, etc. I'm sure some people do care about power. I am also sure the market is large enough to cater to them.
What many people do tend to shop for is price vs performance. I have never bought a console just because it was the most powerful, especially if it really expensive compared to others. I have also never bought a video card because it was the most powerful. I tend to buy the one that offers the best performance at the right price, and PC enthusiasts do exhaustive price vs performance analysis all the time.
On the other hand some people tend to shop for the games and brands they are interested in. I buy Sony consoles day one because I have yet to really be burned by their games and consoles (PS3 had a bad time for sure but ended strong). I used to be a day 1 Nintendo purchaser, but got burnt a few times so I do a lot more waiting and seeing. I did, though, get a switch day 1 because of Zelda. I always end up with an Xbox, but usually when the available new games and price has gotten to the point where can justify the purchase.
I do agree that Lockhart existing will have some interesting effects. The PS5 will probably benefit from it, mostly because if the PS5 ends up being more expensive than the Lockhart and the XSX then I don't know what Sony is thinking.
Yes, ok that seems to be a reasonable interpretation, thanks.
This is exactly my fundamental understanding of the variable clocks and the design of the PS5.
I disagree with your interpretation of efficiency though. When a device is idle it still consumes the static DC power draw, yet it does not work and produces no output, therefore, efficiency is at it lowest possible value - 0. When the device is processing instructions it will have a variable efficiency based on the workload and frequency of operation. edit: just to clarify I am not interested in arguing this as it is an exercise in semantics. I understand the analogy you were trying to convey.
Last edited:
Where did you read "all the time"? Your post makes no sense.
You know, you could have watched the video. Or read the article. Or you could have watched the Road to PS5 video, in which this was also explained. Or read one of the dozens of articles that have been written about that video. Why ask a question that has been answered countless times?
I know the answer. It starts with an L, ends with a Y. The videos are rather interesting to watch about the technology and philosophy, especially if you're in a thread to talk the techno and philosophy.
Exactly. If power really mattered to the market, then many of the best selling consoles in history wouldn't have been the best seller. People want value and everyone has different ideas of value.
Some people might see it as "it's just $50 more dollars", others might see it as "$50 more bucks? Nah."
This power managment thing is still driving everyone nuts. It is fairly interesting, tho. In the end there will be no way to tell what effect it actually has. Besides that the thing can run high clocks perhaps with a quiet fan and hopefully a good price. In all likelyhood those would be the only tangible effecte
I'm not even sure why. Is everyone getting a PS5 dev kit or something? It's interesting to read from an academic sense, but as a consumer none of this should matter to you. Most of us aren't educated on how our phones, computers, or cars work. We just use them and don't think twice about it. I can't think of a time of when people cared so much about how the internals of a console worked until now. Will these new consoles completely blow away the old ones? Yes.
I tend to be believe that *everything* matters - but it's not ONE thing that determines success or failure. Power is important. So is form factor, price, storage space, content, back compat, services, etc. Equally or sometimes more important is where your friends play.
The person who talked about value above I wholeheartedly agree - customers (and I'm talking outside the core) tend to want a good value which is a combination of price, quality, and performance - on top of content and services that people enjoy. They don't generally want the top performer nor the cheapest thing either (although there are customers that care about these outer boundaries). There are other reasons to have "halo" product or low entry pricepoints so long as they are a good value.
I used to make a joke to my team that if you made a Lamborghini cost the same as a Camry, you still wouldn't outsell the Camry. For some people a car like that is impractical or just doesn't meet all their needs. It's not a good *value* even if it's an incredible performer at a low price.
What's more interesting to me looking at both these consoles is to figure out what conditions you would get a casual PlayStation or Xbox fan to switch to the other side. This is where I think things like price and back compat come into play. I'm not sure if either console has presented a vision that would get the other guy to come over (especially now that it's been clarified you'll get 1000's of titles over back compat).
On the question of cooling - the cooling solution is not a top driver of BOM costs on it's own. The top drivers are SOC, Memory, Storage, and ODD. Then the game controller (can't forget that in the total price of the system!). Cooling however does play a role in the overall costs and I've talked about cooling before because that determines form factor, SOC size, power supply size, etc. So even though the individual costs of the cooling solution aren't huge from a BOM perspective, they do impact console costs a lot from an OVERALL cost perspective.
The person who talked about value above I wholeheartedly agree - customers (and I'm talking outside the core) tend to want a good value which is a combination of price, quality, and performance - on top of content and services that people enjoy. They don't generally want the top performer nor the cheapest thing either (although there are customers that care about these outer boundaries). There are other reasons to have "halo" product or low entry pricepoints so long as they are a good value.
I used to make a joke to my team that if you made a Lamborghini cost the same as a Camry, you still wouldn't outsell the Camry. For some people a car like that is impractical or just doesn't meet all their needs. It's not a good *value* even if it's an incredible performer at a low price.
What's more interesting to me looking at both these consoles is to figure out what conditions you would get a casual PlayStation or Xbox fan to switch to the other side. This is where I think things like price and back compat come into play. I'm not sure if either console has presented a vision that would get the other guy to come over (especially now that it's been clarified you'll get 1000's of titles over back compat).
On the question of cooling - the cooling solution is not a top driver of BOM costs on it's own. The top drivers are SOC, Memory, Storage, and ODD. Then the game controller (can't forget that in the total price of the system!). Cooling however does play a role in the overall costs and I've talked about cooling before because that determines form factor, SOC size, power supply size, etc. So even though the individual costs of the cooling solution aren't huge from a BOM perspective, they do impact console costs a lot from an OVERALL cost perspective.
Your perspective is appreciated.
One thing I'd like your opinion on: yields for PS5's APU v. XSX's APU. Folks have noted that the degree to which Sony is clocking their APU, at least on the GPU side, will affect yields. This of course makes sense, as the higher you lock the less chips you will have that can reach that level of performance. But XSX's chip is quite a bit larger - likely around 50-60 mm^2 larger. And the larger the die area the lower the yield.
Given these competing tradeoffs, do you see one APU as clearly giving higher yields - and thus lower cost - than the other?
Oh that's a good question. I'm not sure I have a strong opinion. Since yields affect the cost, it's sort of a circular conversation. It also has a time-horizon perspective since the yields for the first, say, million consoles are somewhat irrelevant to the yields long-term.
I can't imagine either company making a call that would hurt the long term ability to get good yields. So long term almost certainly a wash.
As to the short term, so long as the shipping chips are what was originally spec'd with AMD then I can't see a meaningful difference.
I would guess, all things equal, that a larger or more complex chip would have more opportunities for failure than a smaller but faster chip on the manufacturing side but I'm NOT a silicon development expert so my POV shouldn't be weighed more heavily than anyone else.
I watched Mark Cerny's conference today. While I can understand people being upset with the focus of it, I do think it was nice to use as a form of ASMR.
Based on what we know so far this is what the power curve on the PS5 could look like.
How did you estimate this?Based on what we know so far this is what the power curve on the PS5 could look like.
We have 3 point on how the frequency and power relates so we can approximate an exponential curve from there. As for power I used data from the RX 5700 and included theoretical performance per watt improvements from RDNA2. This last part only changes the values at different points (coefficient of the function) and doesn't affect the trend of the curve.
What is the graph depicting though? A power figure for 100% workload scenario? 50%? 75%? Is it max power workload or max efficiency workload? 160 watts is really ok if it is under max power workload.
That is a real stretch for me. If load times are 2 seconds instead of 1 then I doubt that adds any perceived value over higher resolution or framerates. And I agree that the vast majority cannot tell the difference, but there is more to that story if current-gen games are upscaled to 4k via back compat.
I don't think we'll ever see the GPU drop as low as 2GHz if this chart is any indication, considering that a 4900HS has a 35W TDP for a base clock of 3GHz. I think a shared 200W power budget for the APU would make sense:
-It's not crazy high for a console
-If you just add the ~160W from the GPU and the ~50W that I'd guess the CPU need at 3.5GHz, you'd go above budget. But not by much.
-Therefore, you'd only need minor downclocks to get back on budget for most cases, which fits with Cerny's expectations.
-I don't think AMD (or Intel for that matter) really include AVX etc for their TDP numbers, so there would still be the possibility of larger downclocks for extensive use of those instructions. This is something that already happens on PC with motherboards offering AVX clock offset options.
Edit: do note that I'm not including SSD, RAM, etc for those 200W - this is just for the APU.
-It's not crazy high for a console
-If you just add the ~160W from the GPU and the ~50W that I'd guess the CPU need at 3.5GHz, you'd go above budget. But not by much.
-Therefore, you'd only need minor downclocks to get back on budget for most cases, which fits with Cerny's expectations.
-I don't think AMD (or Intel for that matter) really include AVX etc for their TDP numbers, so there would still be the possibility of larger downclocks for extensive use of those instructions. This is something that already happens on PC with motherboards offering AVX clock offset options.
Edit: do note that I'm not including SSD, RAM, etc for those 200W - this is just for the APU.
Should be average power and I agree it looks like a really good estimate, very much in line with what the XSX is estimated at. Nice work ThatNerdGUI
Based on their demo a current gen game already needs 8 seconds to load on XSX. I doubt many next gen games will load in 2 seconds on that machine.
Out of all of those the only thing that matters to me is form factor.
Games above everything then how it will look on my entertainment centre. Followed closely by the controller.
After that everything else is irrelevant.
I don't know if it's a problem. But it's highter than One X when it on somes games can't reach 4k/60fps bc memory cant be so fast.
So maybe 448GB/s is okay for 4k/60fps. Even, I think we see much games using reconstruction or checkboarding.
I don't think is the same scenario like DDR3 on One Fat
But if it's problematic that's smell the "PS5 Pro" in the next 5 years... that clear. (8k / VRS / 7mm+ / Bandwitch / SSD Size x2)
It has to share with tempest engine as well. Even MC say's it going to be tight:
It is going to be an issue. The XSX has a similar issue. RAM prices just aren't where we need them to be for consoles. Also, MC had been experimenting with much faster ram in Oberon B0, so we know it is something that was considered.
Once you get the exponential trend line you can plot for any frequency values. I added those points because that's the optimal range the data suggest.
The power required at a certain frequency based on the details Cerny has provided. And yeah the power consumption changes depending on the workload. I assume Cerny was referring to just average gaming workloads when he mentioned these figures.
the 4900HS is a highly binned silicon so I wouldn't expect similar TDP for any of the two consoles CPUs. I expect them to be more in the 4800H range at best.
Last edited:
You are right, there's more than power/perfomance. By the way, I didn't mean to speak about TF (that seems to be the metric that most people like to use) but perfomance when I talked about power.
I'd like to point one more thing: we are going into a new generation. I don't think we are into an X1X vs Pro situation. We need to take this from various perspectives:
- The buyer already have an extensive library in one of the ecosystems.
- The buyer don't have a current gen console.
- The majority of games are 3rd party, this way the best perfomance in these games does matter. (This factor loses its relevance when/if games run pretty much the same.)
- Exclusives are exclusives. I don't deny the existence of system sellers.
From my own experience, when I talked about Lockhart it wasn't thinking about power/performance, but the perceived cost benefit. That's why I put that these were my own opinions when shopping. I generally found the system that have the price in the middle to have a great balance between perfomance and price.
Like I typed above, I don't have a doubt that Sony and MS can lock people in with their games and features. Specially when Sony's 1st parties will push some customised hardware that isn't readily available on PC and the console itself will be compatible with PSVR(1).
But if you take an approach of potential buyer, someone who doesn't have a current gen library/console, this things become niche. One may not like the genres that Sony studios do or simply is not interested in VR. Like you and gothmog said, there's more to it.
Either way, I'm not disagreeing with your points, guys. It does make me curious about decisions that will be made for the price of these machines, but this isn't the topic of the thread. I'll try to stay on topic.
