-
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 System Architecture Deep Dive |OT| Secret Agent Cerny
- Thread starter vestan
- Start date
- Status
I've heard PSVR2 is codenamed Caesar.
Edit: realised this might have sounded insidery, not an insider, I think I heard that from BGs on the other place.
A GPU's job is to output what you see on screen. It is resolution based for the most part and all tied to what the power and efficiency of the GPU.
lmao at your avatar.
For Sony's sake, i hope that Cerny's secret sauce in the gpu overcomes that tflops advantage because i am pretty sure its going to cost them sales. i used to visit gamespot forums back in the day and almost everyone switched from sony to ms that gen. they did come back later. but the initial power delta really did have an effect.
of course MS had some amazing exclusives in the first two years like Bioshock, Gears, Oblivion, Halo 3 and Mass Effect so that made it easier for them, if Sony can show off some exclusives on par with those classics, it might slow the tide.
thats true. but i think MS knows that too which is why they revived the lockhart. if sony is $100 (by some miracle i should add) then lockhart could be too. at which point, it becomes a whole lot more difficult for sony to sell the more cheaper console.
It depends on how the GPU is scaled. It's very clear RDNA is a lot more flexible when it comes to scaling its resources than GCN. If they make an 80 CU GPU, then they most likely will go with 8 SAs, which means 10CUs/SA and equivalent to two geometry units of a 40CU GPU. That's very different from scaling to a 56 CU GPU though. The most common building block is the SA, which contains the a primitive unit/rasterizer/ZROPs/ROPs and CUs. They need to scale that block into a descret number, which means MS either went with 14CUs/SA, 8CUs/SA (if RDNA supports odd numbers of SAs) or 7CUs/SA. There is no magic about this. They most likely went with 14CUs/SA as they still need to keep the chip area small.
I agree with you that SONY will lose a lot of hardcore gamers, which are important to them, since they are very passionate/vocal and work as ambassadors/influencers for the platform. IMO, that's unfortunate but that's no reason to be overly pessemistic about the performance delta. Let's not forget that we don't know the price of the machines and that can reverse the sentiment entirely. Also, I don't doubt Sony's first party output will be superior to MS.
Not really being the middle ground might end being the best thing for them.
There is whole bunch of sales and marketing data that show a lot of the time the middle price item end up selling the best .
Ariel, Gonzalo, Flute, Oberon (and supposedly Prospero) were already GPU and platform codenames.
Yes, but GPU performance is not are not strictly = to CUs x clock
No one is making the argument that the PS5 GPU is overall more powerful than the SX GPU. Just that you can't just look at TFs as the only measure of performance. Parts outside the CU will scale with clocks, not CU count.
I thought this during the presentation but thought the other names were an AMD thing and not a Sony thing.
That (or Project Night's Dream) would go great since PSVR was Project Morpheus (the god of dreams).
Also, now I want Midsommar VR.
Now I'm hyped to see what's in store for Project Titus or Project Andronicus.
You play the game as a bear :D
Vega 56 x Vega 64 is a good example of that: iso-clock they performed mostly the same, despite 64 being considerably bigger. Not saying that is going to be the case here, but going wider virtually always has some loss in performance.
The increased efficiency comes from architectural improvements, which is something completely different from what we are talking about here. As long as all the resources of the GPU are scaled correspondingly there is no efficiency loss when scaling up/down a GPU. A GPU is divided into sub-blocks and those sub-blocks need to be a discret number, which means it's not possible to scale all the resources proportionally for every possible CU configuration. A 56 CU configuration gives an indication which resources have been scaled and which haven't.
That same argument was made in favor of XBox One/ XBS vs PS4 when Microsoft up the gpu clocks. PS4s overall CUs advantage and memory bandwidth still produce higher resolutions and graphical performance despite XB1/S clock speed advantage.
Good, because there's nothing to follow.
Eventually when the hot takes die down, more info is known, the consoles finally get released we will see why Sony went in on the SSD.
I'm just waiting on the OS, features to be revealed.
In a separate debate yesterday, I asked why these gains being talked about here did not help the base Xbox One with 12CU's@853Mhz or The Xbox One S with 12CU at 913Mhz compared to an 18CU@800Mhz PS4.Yes, but GPU performance is not are not strictly = to CUs x clock
No one is making the argument that the PS5 GPU is overall more powerful than the SX GPU. Just that you can't just look at TFs as the only measure of performance. Parts outside the CU will scale with clocks, not CU count.
You are limited by power, and bandwidth, and if you want to sustain high frame rates, the CPU has to be competent. The laws around which GPU's perform are not changing.
Yes, that's a good example. When only the CU count is scaled, only the operations that are shader bound will be affected, which is typically long shaders applied during the main render pass I.E. when rendering the frame-buffer. There are a lot of other types of rendering going on though. For instance, when rendering shadow maps or depth pre-pass, the shaders are typically very short and the operations are entirely geometry/rasterizer bound. Scaling up the CUs will have no effect on those kind of operations. It's possible to run compute or async compute workloads during those operations, but it's difficult to find suitable workloads at that point of the frame since the main render pass is yet to be executed.
Who says the didn't help? By how much? who knows. The numbers are much different this time around, as noted below.
I agree. BUT the TF gap is much smaller this time, and the Clock gap is much greater this time, and the bandwidth of memory (considering all 16GBs average speed) is smaller MUCH smaller as well. PS4 had nearly 3X the bandwidth.
Again, not saying PS5 is magically out performing the SX on the basis of clocks here. I have no idea how impactful it will actually be. Just trying to understand the capabilities of the machine, and it sure isn't boiled down to TF only.
I hate to get in on the system wars, especially in a dedicated thread so I'll only make one post on the matter..
We seem to be going round in circles going from one extreme to another. The answer will no doubt lie somewhere in between
Will the XSX have a compute advantage over PS5? Yes..
Will the 16% to ~20% difference between the system's TF counts represent the real world performance differential? No..
When the entire GPU is clocking much higher everything outside the CUs is running faster; in addition you can make slightly more efficient use of those CUs.
The CPU difference is ~3x smaller than the difference between the OG X1 and PS4.
The SSD won't magically give you more frames but it does remove the most pressing bottleneck games have been facing for years.
I expect the real-world performance differential to be about half the GPU TF differential. So perhaps 8-10% in resolution as a rough example. Think 2052p vs 2160p.
The question you need to ask yourself as a developer or consumer is, is that tiny real-world difference worth the gains in I/O and the possibilities that brings in changing the fundamentals of the games we get to play or develop? I without a doubt think it is.
We seem to be going round in circles going from one extreme to another. The answer will no doubt lie somewhere in between
Will the XSX have a compute advantage over PS5? Yes..
Will the 16% to ~20% difference between the system's TF counts represent the real world performance differential? No..
When the entire GPU is clocking much higher everything outside the CUs is running faster; in addition you can make slightly more efficient use of those CUs.
The CPU difference is ~3x smaller than the difference between the OG X1 and PS4.
The SSD won't magically give you more frames but it does remove the most pressing bottleneck games have been facing for years.
I expect the real-world performance differential to be about half the GPU TF differential. So perhaps 8-10% in resolution as a rough example. Think 2052p vs 2160p.
The question you need to ask yourself as a developer or consumer is, is that tiny real-world difference worth the gains in I/O and the possibilities that brings in changing the fundamentals of the games we get to play or develop? I without a doubt think it is.
Of course it helped, but it's not reasonable to expect a 6% clock frequency advantage will negate a 50% CU advantage, especially when the ESRAM of the XB1 was way too small, which resulted in a lot of unnecessary copying of data. XB1 also only had half the amount of ROPs.
In the same way, it's not reasonable to expect that a 22% clock frequency advantage will negate a 44% CU advantage, but it will make the performance delta less than what the TF delta implies.
When matt revealed xbo he focus on ESRam and praised it as the revolution... But it flop
Now cerny is praising CFC as its the revolution...
Will it cause ps5 to flop
I see y'all are still acting up. Meanwhile I've got the big brain master plan. Getting the Switch this summer, the PS5 for exclusives. Then building a maxed out PC with an RTX 3080Ti AND one of the PS5 certified SSDs when available since I can play Xbox exclusives there.
Checkmate gamers😏
Checkmate gamers😏
Yeah, it seems to me that a lot of the new features need a show-not-tell type of presentation, which is why I think they said they were planning on doing hundreds of events in the coming year. That way people can experience everything -- such as the new controller and 3d audio -- firsthand, but the coronavirus outbreak kind of threw a wrench in those plans.
XSX does not have a fast SSD?
Why would XSX games design not change too?
The Xbox One didn't "flop."
The 360s ownership numbers were hugely inflated by a shit ton of casual owners that jumped in late gen because of Kinect. Something insane like 25 million kinects and copies of Kinect adventures got sold.
These people bought almost nothing else, but they DID watch a lot of TV- which explains the kinect integration and TV centric approach of the launch XBO.
Zhuge mentioned on twitter that the XBO has hit 50 million units and its not unreasonable to think it will sell a few million more before MS ends the platform.
That being the case, the majority of the sales gulf between the 360 and XBO will be chalked up to Kinect owners who didn't come back. They retained most of their core audience.
This is actually the only correct response to this (and also my general plan).
PlayStation 5 vs Xbox Series X: Console Face-Off
We reviewed both systems. Here is how they compare.
Don't they still need to take into consideration the base Xbox One? They won't be making next-gen exclusives for a year or so. Unless that stays throughout the generation.
There's no reason to suspect it will stay for the generation, surely if it was going to Booty wouldn't have said "year, two years" in his response to the question.
If this is the case, only 1st party titles really taking advantage of the tech, then I would assume Sony has a bunch of 1st party titles ready for launch to show it off. Otherwise, what is their plan? Let people imagine all these supposed game changers?
Ps5 will likely continue where the PRO left off and again be bandwidth limited.
Cerny said inner SSD has 6 priority levels and NVME SSD spec has only 2 priority levels.
This means games that make use of more than 2 priority levels (most 1st party) HAVE to run from the built in SSD, not directly from the NVME.
When you want to game, the game will be copied from the external NVME to the internal drive.
Fortunately, at 5GB/s, this would take mere seconds to accomplish, and can be hidden from the player.
Thus the external NVMEs should have 0 overheating problems.
Didn't Cerny say that the I/O cluster would arbitrate the extra priority levels from the expansion SSDs?
Yeah. And is why there needs to be some extra speed on the expansion ssd to compensate. Validating ssds for expansion will be about testing they're up to the task of being >= the internal ssd performance and behaviour for running games.
MS have a very fast SSD and a from everything we've been told a great I/O stack and set of optimisations; in itself it deserves praise (particularly their texture solution), I don't question whether their solution will change game design, it will. But the systems have clear differences in priorities and Sony's solution has simply gone further. When Cerny refers to "the dream" from a developer perspective, it's referring to lifting the bottleneck and reaching a greater threshold where you can just load what you want on the fly; they seem to have surpassed that threshold.
I have been thinking about this.
Tempest is a AMD GPU CU modified to work like an SPU right?
...
Tempest is the second GPU!
Jk but it's funny that the theory has come true on a technical level, albeit obviously for a different purpose.
- Status