Is Turing the first step that nvidia took to bring their GPUs to consoles in the future ?

[Vya Domus]

No one in their right mind is doing 4K gaming on a RX 580 on PC.

But if you would , the performance you'd get is similar to what you get on something like the Xbox One X.

That's just the reality of things, no matter how much you'd trust developers to make miracles happen just because it's a console platform.

Don't underestimate the power of optimizations on a closed platform.

I would like to see that in action. What game out there runs flawlessly on the X and doesn't on a similarly equipped PC with the same IQ ?

I mean you are all saying something like the Xbox One X can do so much more than an a PC with an RX580 simply because of the difference in platform, show me an instance where that's true.

 

[TheoneandonlyMrK]

No one in their right mind is doing 4K gaming on a RX 580 on PC. That's the console (strict platform) difference.

two works fine on most games , for the rest dx12 makes it better then you would expect and better then ps4 pro or scorpio on the same title , An Rx580 is close to an xbox in shader count ,not the pro but no where near its power budget so it does better in typical enthusiast hands .

But yeah not good enough ,,your not getting ultra on every game with one card in any mode, two yeah but just dx 11.

I moved on but anyway they are too dissimilar.

 

[Aquinus]

I mean you are all saying something like the Xbox One X can do so much more than an a PC with an RX580 simply because of the difference in platform, show me an instance where that's true.

Are we going to test it with a CPU running at 2.3Ghz too?

 

[FordGT90Concept]

I would like to see that in action. What game out there runs flawlessly on the X and doesn't on a similarly equipped PC with the same IQ ?

Forza Motorsport 7. Microsoft markets 4K@60 and they deliever 4K@60 on hardware that's used predominantly for 1920x1080 @ 60 Hz on Windows.
https://www.eurogamer.net/articles/...orza-motorsport-7-xbox-one-x-true-4k-showcase

 

[Vya Domus]

Are we going to test it with a CPU running at 2.3Ghz too?

Considering that most games on the X run at 30, you'd surprised how little CPU power you need for that.

Look I am not battling to prove that these two platforms are identical and there is no differentiation. There is, clearly, but it's not earth shattering.

Forza 7. Microsoft markets 4K@60 and they deliever 4K@60 on hardware that's used predominantly for 1920x1080 @ 60 Hz on Windows.

You know, as it happens I do have that game and believe it or not it runs mostly at a locked 60 at 4K with dynamic rendering enabled, admittedly. And it's damn close in terms of IQ too just looking at the two side by side, not pixel peeping.

But Turn 10 really pulled all the stops in this game, I'll give you that.

 

[Aquinus]

Look I am not battling to prove that these two platforms are identical and there is no differentiation. There is, clearly, but it's not earth shattering.

Of course not. It's not about the hardware, it's about knowing what the software games will be running on. You optimize for what it will be running on and it's easy to optimize for one or two similar platforms. It's a lot harder when you start dealing with combinations of hardware or even operating systems with different characteristics. It's simply easier to extract performance out of a closed system. It just is, because the more that needs to be handled is time not processing your software game.

 

[lexluthermiester]

RTX 2080 Ti can barely manage 1080p30 using Direct Raytracing.

That is completely incorrect. The disinformation is strong with this one..

 

[ShurikN]

This. After all of this time, Microsoft finally has their console on x86.

Same can be said for Sony as well. After the coding hell that was Cell.
Damn I rhymed that better than 2Pac.
And not only was it hard to code for, most of the times the hardware was underutilized. First party games worked better of course, as more attention was given to them, but third party games were a framerate mess more often than not.

 

[silentbogo]

Look I am not battling to prove that these two platforms are identical

But from reading all of the above that's what it looks like.
There are major, and I mean MAJOR differences between PC and modern consoles. Some is superficial and don't do jack for performance, but most are just for that.
* First of all, you have unified memory. No need to move data between system RAM and VRAM, like you do on PCs. Just loading textures into memory saves up at least 2 time-consuming steps.
* Second, as already mentioned, is the fact that you don't have a gazillion layers of abstraction within those systems. It's much simpler and as a result - faster, on consoles.
* Third and most important - it's a fixed configuration of hardware. If you follow the news, you should know that PC releases are always followed by media headlines w/ titles like "XXX game receives the patch optimizing performance on Green/Red GPUs", or "NVidia/AMD releases a driver, which improves performance of recent XXX and YYY titles". You don't have that in consoles, cause the hardware is always the same. The only patches you get are general performance improvements and bug fixes, and even those are rare due to strict quality control on both sides (at least for AAA titles).
* Lastly, no bloatware, no unnecessary background services, minimal amount of running system processes etc. Just eliminating the legacy stuff from Windows or Linux would make them run much faster, but there will be a bunch of angry people who's stuff will stop working.

Simply put - the spec may look the same on paper, but architecturally the difference is huge: both in hardware and in software.

 

[Vayra86]

Most of the improvements for console are about efficiency. But what Vya is saying is certainly true, you cant get around the raw processing power limitations. You can only optimize the pipeline better on a console. Beyond that, the 'optimization' that remains has a direct impact on what the GPU is actually drawing and is just a very fine grained quality setting adjustment.

The days that consoles could magically do things differently are past history. There are no special render techniques, there is no customized hardware to cater to such a technique. The current console crop is in no way comparable to the PS2 or earlier consoles. Current engines do the same old tricks that PCs use and use their hardware more efficiently, thats it. On top of that, they use tricks to fake higher native resolutions, and those are always visible. Native 4K on a console can still use a lower internal res, and does so most of the time on a console.

The days of magic console bullets are over.

 

[Vya Domus]

* First of all, you have unified memory. No need to move data between system RAM and VRAM, like you do on PCs. Just loading textures into memory saves up at least 2 time-consuming steps.

That is a double edged sword , paradoxically you trade ease of use with performance. Sharing memory between the CPU and GPU isn't the best case scenario for performance by any stretch of the imagination, it's the exact opposite. Again you don't have to believe me, see Microsoft and how well that DDR3 & ESRAM (guess as to why they had to include this) combination turned out. Hurray for unified memory huh ?

Unified memory = cheaper to implement and easier to mange but not better. Believe me if they had a better budget they wouldn't have went for that.

* Second, as already mentioned, is the fact that you don't have a gazillion layers of abstraction within those systems. It's much simpler and as a result - faster, on consoles.

You don't have a million levels of abstraction, I don't know where you people got this idea from. Consoles have an OS and an API through which your software interfaces with the hardware in the same way it does on a windows PC.

* Lastly, no bloatware, no unnecessary background services, minimal amount of running system processes etc.

Agreed.

Just eliminating the legacy stuff from Windows or Linux would make them run much faster

This I do not believe, what legacy stuff hinders performance that much ? You do realize all those services that are running and are mandatory were specifically designed to sip memory and CPU cycles. These things were written by people who knew very well what they are doing.

The only patches you get are general performance improvements and bug fixes, and even those are rare due to strict quality control on both sides (at least for AAA titles).

You just know that isn't true at all, you can botch performance on a console game just as much as you can on a PC port.

Also Fallout 76 says hi :

Sub 20 fps and frame spikes that last several seconds ? Where's the console optimization here ?

 

[EarthDog]

This discussion is still going on? Seemed pretty wrapped up already. Have fun men.

 

[Aquinus]

You don't have a million levels of abstraction, I don't know where you people got this idea from. Consoles have an OS and an API through which your software interfaces with the hardware in the same way it does on a PC.

Actually, they are different to some degree. Don't get me wrong, there are a lot of similar APIs, but when it comes to CPU and GPU scheduling and stuff like that, it's a very different animal because you're not doing the same kinds of things on a console as you are on a PC, such as switching between any number of tasks with unknown complexity and latency requirements. You also don't need the same kind of hardware abstraction because hardware isn't going to be changing either and if it does, it's a similar variant based off what was already built. These are all things that reduce the amount of OS overhead for the device. So no, it doesn't do it the exact same way as a PC. Just because it's x86 now doesn't mean that every OS design decision happens to come along for the ride just because of that fact.

 

[Vya Domus]

You also don't need the same kind of hardware abstraction because hardware isn't going to be changing

That's not the point and nor have I said anything about that. We are talking here about a supposed number of software layers (in huge numbers according to some of you) that are inherent and unavoidable to something like a Windows platform and that also affect performance to a great extent. That's just not the case.

You want to render something , you make a function call through an API which in return talks to what is essentially a graphics driver that has direct control over the GPU hardware. All of those levels of inter communication are present on either platforms, you don't get more or fewer layers of software in between those. I simply do not understand why some of you believe it is any different than that.

Here is an actual example of an additional abstraction layer that does come with a performance penalty: https://moltengl.com/press/moltenvk-brings-vulkan-to-ios-and-macos/

Maybe we are not talking about and referencing the same things.

 

[ShurikN]

Sub 20 fps and frame spikes that last several seconds ? Where's the console optimization here ?

That game runs like crap even on high end desktop hardware, that's at least 2 times the price of an X1X. Not really a showcase example if it runs horribly everywhere.

 

[Vya Domus]

That game runs like crap even on high end desktop hardware, that's three (or more) times the price of an X1X. Not really a showcase example if it runs horrible everywhere.

Thing is it's not the only game, Digital Foundry's channel is an eye opener, you'd never guess how many AAA games cannot maintain a locked 30 or whatever performance target they have. Not only that but this was always the case, any of you played GTA 4 on 360/PS3 ? That game ran horrendously bad and ironically enough back then the hardware that those consoles used was a novelty that allowed for all sorts of optimizations.

 

[Vayra86]

Actually, they are different to some degree. Don't get me wrong, there are a lot of similar APIs, but when it comes to CPU and GPU scheduling and stuff like that, it's a very different animal because you're not doing the same kinds of things on a console as you are on a PC, such as switching between any number of tasks with unknown complexity and latency requirements. You also don't need the same kind of hardware abstraction because hardware isn't going to be changing either and if it does, it's a similar variant based off what was already built. These are all things that reduce the amount of OS overhead for the device. So no, it doesn't do it the exact same way as a PC. Just because it's x86 now doesn't mean that every OS design decision happens to come along for the ride just because of that fact.

What you are describing now, the PC already fixes simply by having a much faster CPU. All of this is CPU overhead and not GPU load.

The optimization on consoles happens mostly on the CPU side too, they have to, if they want the weak jaguar cores to do enough to keep the GPU going.

None if this makes the console better at rendering.

In addition, proof is in the pudding. You can find as many very well optimized PC titles as you can find smoothly running console games. They are exceptions to the norm and irrespective of the platform they run on, that optimization costs a lot of dev time. For consoles more work is needed to optimize each scene, specific assets etc. and for PC the work is needed to have a range of quality settings run well on varying configurations.

TL DR none of the optimizing done on consoles has the ability to make the GPU push more pixels. And on the PC, stronger CPUs also manage to push GPUs to (near) max utilization. What remains is optimization through quality adjustments.

 

[silentbogo]

Maybe we are not talking about and referencing the same things.

What we are talking about is fundamental differences between PC and consoles, which you seem to be unable to grasp. You still cling to the fact that CPU is the same, GPU is the same, and oh-my it even has and OS, just like a PC... But the difference is in how it all works together. No amount of highlighted bold words can change that.
In XBOne X and PS4 Pro you have a unified memory pool accessible by both CPU and GPU. You don't have to move or copy data between locations for loading/processing/rendering. Just give it a pointer - done.
On PCs your CPU can only access system memory, and your GPU can only access video memory. That's why your iGPU, even though located on the same die as CPU, is unable to load stuff directly. Good example is a chinese Subor Z which has that traditional split between RAM and vRAM, cause it's inherently a PC, not a console.
ESRAM is an example of how an additional type of memory can complicate things to the worse. Even though it's fast, you still have to explicitly copy stuff to this high-speed buffer in order to use it, and it takes time and resources. Devs simply had no use for this stuff, so it died as a bad idea in the next iteration of XBOX.

I think I should take a break from this conversation, cause it's going nowhere.

 

[Vayra86]

What we are talking about is fundamental differences between PC and consoles, which you seem to be unable to grasp. You still cling to the fact that CPU is the same, GPU is the same, and oh-my it even has and OS, just like a PC... But the difference is in how it all works together. No amount of highlighted bold words can change that.
In XBOne X and PS4 Pro you have a unified memory pool accessible by both CPU and GPU. You don't have to move or copy data between locations for loading/processing/rendering. Just give it a pointer - done.
On PCs your CPU can only access system memory, and your GPU can only access video memory. That's why your iGPU, even though located on the same die as CPU, is unable to load stuff directly. Good example is a chinese Subor Z which has that traditional split between RAM and vRAM, cause it's inherently a PC, not a console.
ESRAM is an example of how an additional type of memory can complicate things to the worse. Even though it's fast, you still have to explicitly copy stuff to this high-speed buffer in order to use it, and it takes time and resources. Devs simply had no use for this stuff, so it died as a bad idea in the next iteration of XBOX.

I think I should take a break from this conversation, cause it's going nowhere.

You gotta turn it around to see it: are GPUs on PC idling because they have to move all the data around? You know the answer; with weaker CPUs they actually are. With fast ones that gap is virtually gone.

 

[silentbogo]

You gotta turn it around to see it: are GPUs on PC idling because they have to move all the data around? You know the answer; with weaker CPUs they actually are. With fast ones that gap is virtually gone.

And? That's the whole fucking point I'm trying to make : consoles don't move data around, it's all in one big shared pool.

 

[lexluthermiester]

On PCs your CPU can only access system memory, and your GPU can only access video memory.

That is patently incorrect. In a PC(any PC) the CPU can directly access and address VRAM and the GPU can directly access and address system ram. While the RAM pools are in separate physical locations, they are mapped by the system as one large pool. While that is an over simplification of the dynamic involved, it is the actual effect. Architecturally, the PS4/XBO systems are 99.8% identical to a PC. That's what makes them so appealing to develop for.

But hey we're off topic here, let's rope it in..

 

[Vya Domus]

That's why your iGPU, even though located on the same die as CPU, is unable to load stuff directly.

That is just incorrect : https://www.khronos.org/registry/OpenGL/extensions/INTEL/INTEL_map_texture.txt Not only that this isn't true on an API level, this is simply not true in general, the CPU can access memory on any device addressed through it's IO ports and vice versa. How else would all of this even work ? Of course you can directly access memory from anywhere. If it's slow, not secure, etc is a different matter.

But the difference is in how it all works together.

Elaborate please, how does it all work together such that it's so vastly different ? When you do a draw call does it work differently compared to a Windows machine running DirectX ? Give me an actual example of how something is so fundamentally different at a software level on a current generation console.

That's why this discussion doesn't go anywhere because of all you are just telling me that I am wrong and when asked why all I get are incredibly vague and elusive claims such as "there is a million abstraction layers" and "legacy stuff makes things slower".

Excuse me but, I simply cannot take those as serious arguments and explanations. With the advent of these new consoles all they need is an API , a graphics driver and an OS and they all interface in the same way they do on PC. Just like everything else out there nowadays, there is no reason to believe it all somehow works differently despite them being nearly identical platforms, hardware wise.

If you really want to see a console in action who works differently and does benefit from optimizations unseen on PC take a look at the PS2 which had a vector co-processor and other DSP-like components who did actually need a completely different framework. Those days are simply gone.

 

[silentbogo]

Jesus fucking christ, do I really have to explain to you the difference between the actual physical memory(chips located on PCBs) and memory addressing physical or virtual (the thing you both meant in your posts)?
I'm outta here.

 

[Vayra86]

And? That's the whole fucking point I'm trying to make : consoles don't move data around, it's all in one big shared pool.

Yes and my point is that this wont magically increase a GPUs capability, it just makes sure that its being used fully. When you look at PC GPU utilization especially at high res, that is not much different. There is no extra performance here, not moving data just saves on Cpu cycles these consoles cant spare anyway.

No need to get all emotional over it... perhaps you need to take more effort to convey why you believe this creates a major performance gap. Im not seeing it - not in theory or in practice.

 

[Vya Domus]

do I really have to explain to you the difference between the actual physical memory(chips located on PCBs) and memory addressing physical or virtual (the thing you both meant in your posts)?

Not only that you didn't mention one thing about different memory chips or whatever but "accessing memory" (your quote, your words) can only be done in one way : through physical addresses, whether you try to read/write data on the same physical chip or from another device it doesn't matter, the only difference is you do it via DMAs (but not necessarily). And those are still addresses that can be mapped by the CPU.

Absolutely anything connected to the IO ports can be mapped onto the address space that the CPU has.

We knew very well what we meant to say. So no, no explanation needed.