Yeah I also thought this was a very interesting test. Wonder if someone more technical than me could interpret the results.
-
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
Don't forget
"The fixed profiles that devs are talking about are only used in ps5 dev kits and in retail games their code will tap into the variable clocks for extra performance "
I don't think this was addressed at you, I read it that way: "[He] shut down that console war crap argument pretty quickly."
He literally said AGAIN that both frequencies can be reached simultaneously, after he previously explained this already in his talk. There is absolutely no confusion. Are you trolling?
Nope, this is how PS5 retail consoles will work. It's in the video.
90Mb/s in 1/100th of a second is ridiculous. That is near instant you wouldn't even notice a delay....
Well, I still don't really see the point of variable clocks, unless it it's for scaling them down. If it was always possible to have both CPU and GPU at max clock speed, why bother with that kind of trick ?
Every CPU/GPU on the market already adapt its frequencies to the workload, so is it just some kind of better optimization on that subject ?
Every CPU/GPU on the market already adapt its frequencies to the workload, so is it just some kind of better optimization on that subject ?
Nope. The "argument" being thrown around was that it was impossible for the PS5 to run both at max, and that has been refuted.
Okay I missed that thanks. I am very curious if the "boost" will be used that often.
Posting my comments from the other thread:
This is also an important point. Some Intel desktop CPUs run faster than their integrated graphics-less counterparts because those idle graphics cores actually add as a thermal spreader. If you had uneven thermal density, they'd be contributing to each other's hot spots.
This is important to show they're not gaming the metric.
This is a cubic relationship. 0.9^3 = 72.9%. That means drastic reductions in power. Downclocks should indeed be minor.
This is also an important point. Some Intel desktop CPUs run faster than their integrated graphics-less counterparts because those idle graphics cores actually add as a thermal spreader. If you had uneven thermal density, they'd be contributing to each other's hot spots.
Yes, developers already see dropped clocks, but it's intentional. Some PC benchmarking sites tear their hair out trying to interpret benchmark results because of unpredictable boost behavior. This eliminates that.
This is the subject of several of Cerny's patents.
This is very important. It gives us an upper bound for how much memory bandwidth we would want on a per CU basis. Since the clock is the same as the GPU, PS5's number is 720GB/s.
This shows just how little of a system's teraflops can be realistically used. 40% VALU usage!
Both in the article and in the video it is clearly stated that clocks can potentially vary depending on workload.
No, it's separate hardware. Video mentions that it consumes memory bandwidth but that's to be expected since everything is stored in the shared RAM including sounds.
I suspect the peak power thing is a momentary peak, not sustained. Processors are all about that these days, especially when power limits are around.
that's why geekbench results for iOS devices are deceptive. You won't see those results in real world applications.
even the i9-9900k is supposed to have a MUCH lower sustained power limit than what mobo manufacturers run it at.
this is why smartshift is still needed and integral.
that's why geekbench results for iOS devices are deceptive. You won't see those results in real world applications.
even the i9-9900k is supposed to have a MUCH lower sustained power limit than what mobo manufacturers run it at.
this is why smartshift is still needed and integral.
The variable clocks might simply be a technique to use the power headroom more efficiently by making sure every frame is at it's optimum power consumption.
I expect those development profiles are pretty conservative with clocks - in order to perhaps guarantee that 'in the wild' the performance will only ever be better?
Something to bear in mind, if and when those profiles ever leak. I expect they'd be grabbed onto in some quarters as the 'real' PS5 clockspeeds, but I think they'd necessarily be quite conservative vs actual performance on retail PS5s, if they need to be able to guarantee what devs see on their dev kits is the minimum that users will see.
Something to bear in mind, if and when those profiles ever leak. I expect they'd be grabbed onto in some quarters as the 'real' PS5 clockspeeds, but I think they'd necessarily be quite conservative vs actual performance on retail PS5s, if they need to be able to guarantee what devs see on their dev kits is the minimum that users will see.
I disagree. I've followed most discussions on this subject and the question has always been how often will the clocks stick to these frequencies in real-world conditions.
It is silly for anyone to claim that the PS5 isn't powerful and won't have some crazy good next generation games developed on it, but outside of the SSD itself there are clear compromises compared to the Series X that developers have to contend with that will have some impact that may be most pronounced when we get true next generation games that really push the limits of what is possible with next gen hardware.
- Variable frequency approach is definitely a way to probably have it "punch above its weight" in some cases at least, but this is still something that developers have to be aware of which could impact their full vision by not having fixed known power available for CPU and GPU at all times. A game that wants to do some insane things with AI and physics may have to throttle down the GPU more than desired which would impact available GPU compute performance for ray tracing and other graphics capabilities and vice versa.
- 36 CUs is pretty much guaranteed to make ray tracing less performant on the PS5 compared to 52 CUs on the Series X based on how AMD does things, and the fact that Cerny didn't even mention it again in the deeper tech talk with DF is likely an indication that he knows where he should be selling his platform which is smart
- Memory bandwidth could definitely end up being a potential challenge with the mention of the audio 20 GB/s where even Cerny admits that you have to be careful not to impact graphics processing, and that is on top of the 9 GB/s decompressed bandwidth required by the SSD
OP
OP
VRS is purely a software implementation waiting to happen on the PS5 at this point (maybe it's already in the latest Sony Graphics API?) being based on RDNA2 and all.
so in retail their game will perform different than in the dev kit environment? i dont understand this
It makes business sense. Games are stupidly expensive to make and these consoles may be priced higher.
RDNA2 has VRS. PS5 uses RDNA2. I don't understand where this idea that VRS is absent is coming from.
Yes? What is your point? The "argument" some people brought up, was that the CPU / GPU can not reach max clocks at the same time. And Cerny confirmed again, yes they can. That the clocks will vary depending on workload was always the case and never disputed.
Yes the 448 GB/s of bandwidth looks very little. This is the only things worrying me on PS5.
Shame Cerny doesn't comment on the DF findings or give no specific answers.
No answer for the lower clock bounds, no follow up on his claims that low amount of CUs with a higher clock beating more CUs but slower not being seen in practice. No follow up on how Rdna 1 (and all gpus basically) does not scale very well with frequency when you don't increase the bandwidth (and for a 5700 going over 1.9ghz is already non linear).
So basically this video just raises many questions about the actual ps5 performance but unfortunately still can't answer them, and I think that it will be unable to until the console launches.
Though that sounds a bit troubling.
No answer for the lower clock bounds, no follow up on his claims that low amount of CUs with a higher clock beating more CUs but slower not being seen in practice. No follow up on how Rdna 1 (and all gpus basically) does not scale very well with frequency when you don't increase the bandwidth (and for a 5700 going over 1.9ghz is already non linear).
So basically this video just raises many questions about the actual ps5 performance but unfortunately still can't answer them, and I think that it will be unable to until the console launches.
They asked Cerny about this and Cerny response was that these fixed clocks are only temporary because the dynamic setup is still not enabled on devkits.
Though that sounds a bit troubling.
Did you just completely ignore what was said right after that?
I am shocked!
At higher clocks, at some times, perhaps yes.
If the game is fixed framerate you won't see it. So this won't 'break' games with fixed framerates or fixed cpu frametimes for gameplay logic purposes.
If the game has a variable framerate, the bits where framerate dips on a dev kit might be smoother on retail if the machine finds room to move power to the bound. I don't think any developer will mind that - but this is why I have to presume those profiles are set in such a way that they can basically guarantee 'in the wild' clocks will be as good or better than on the dev kit.
Because not every instruction uses the same amount of power.
The Series X which is constant frequency does not run at constant power. Simple instructions will cause the series X to run lower power. Complex instructions will cause the system to use more power. Despite the fact that the system is always running at a constant speed.
The PS5 is the opposite. It runs at constant power. More power intensive instructions will tend to cause the system to downclock. Less power intensive instructions will tend to cause the system to run at 2.23 GHz.
I don't know anything about this, but was Cerny lying? Because that was a really common take on the PS5 deep dive thread.
Yeah it's a very Sony thing to do: recognize the importance of certain types of tech and if they can't flat out license or deal with patents they come up with their own solution. Hopefully it's easy for devs to implement if that's the case, but I have little doubt it will be.
On the second bolded- it's indeed a very MS thing to try and put their own initiatives front and center and hype them up as much as possible. That includes giving them flashy names to try and market them or hype them up. I don't doubt that they have some useful or interesting things they patent like with the tier 2 VRS stuff but I agree that a lot of the times these kinds of things may be of little consequence.
EDIT: just saw the additional posts about VRS T1 and T2. very interesting! hopefully it's something that isn't some DX/DXR exclusive and/or Sony has their own implementation of that feature with those same kinds of gains.
Last edited:
As long as the PS5 has exclusives that take advantage of the system, that's fine with me. If all the PS5 will do is play prettier PS4 games, though, then it's pointless.
Haven't posted in these threads in a while but it's pretty funny seeing the same old people doing their best lol
thanks, and do we know how severe is the bandwidth consumation? and can i disable this thing if i'm not interested in 3d audio with my crappy tv speaker?
LoL I was about to say just that. It's so fucking transparent. As usual this thread is going to get bogged down with presumably a mod member coming to tell the kids to play nice.
PS5 threads quickly become quite the garbage fire.
Is that said in the video?
Because the text makes it sound like optional modes for debugging/optimising purposes rather than the dynamic clocking being unavailable on dev kits.
I think PS5 games will look fine, but what variable frequency means is it will punch BELOW its weight when it comes to real world performance. Just like with iOS. Just like if you limited an i9-9900k to its 5ms of peak performance rather than humming along at over 200W with AVX operations as mobo manufactures run the thing way out of """"spec"""".
but we don't have any decent measure of weight now anyways. SSD will be a huge factor, all these other innovations in the hardware will be a huge factor. You simply cannot compare even PC components on paper anymore, which is even more true for consoles.
Not a bad question, i wondered if maybe there is a newer dev kit coming in the summer that handles this more closely to the retail description? maybe we'll find out later.