Is UE4 indicative of the sacrifices devs will have to make on consoles next gen?

Who is saying PS4 will reach 2x its theoretical performance? (Carmack's claim is in regards to dx9 and previous gen consoles)
 
Who is saying PS4 will reach 2x its theoretical performance? (Carmack's claim is in regards to dx9 and previous gen consoles)

And why exactly it is no longer valid for dx11 and PS4/x720 ? I am not saying 2 times, but let say 1,5? ;) PC architecture is such an overkill, you simply cannot put all this power to use... It's like Corvette on VW Golf tyres
 
Actually, the width of a tyre doesn't improve upon traction (but that's another story).

I'd say, it's not just DX9 who was the culprit here, though. The driver is also quite a thick layer, which has to be traveled through. The OS is also a factor. Coupled with virtually no optimization (as in "to the metal"), without any HAL etc... there sure will be enough performance to gain.
 
Question about SVOGI...does anyone know how sensitive that technique is to latency? Someone on GAF is suggesting the low latency of Durango's eSRAM may help it run UE 4's SVOGI better than PS4 since (he claims) SVOGI is highly sensitive to latency, of which GDDR5 is thought to be have plenty of compared to the eSRAM in Durango.

Any thoughts on such a claim from the tech heads here?

Here is the specific post I'm referring to, for reference:

Thraktor said:
By the way, I think those of you saying that this is a matter of the GPU simply not being powerful enough are somewhat off-base, as I would say this has a lot more to do with architecture than raw power.

The bulk of the computational work of SVOGI consists of cone traces over the octree representing direct and indirect light sources. Unlike more traditional rendering techniques, where the GPU simply wants as much bandwidth as possible, these cone traces are entirely latency-bound. This is an issue for the PS4, which has been designed around a big pool of GDDR5, because the latency for a GPU accessing GDDR5 memory is going to be crippling to something as latency-sensitive as SVOGI is. (PC GPUs, of course, have the same issue, but when you've got something as powerful as two 680s, efficiency doesn't matter that much.

The heirarchical memory architecture of Durango and Wii U is actually much more suited to SVOGI, as the embedded pool can be used as a very low latency buffer for the octree while the cone-traces are being performed. Of course the Wii U wouldn't have the GPU grunt necessary for a full SVOGI-based UE4, but Durango should potentially be much more capable of the rendering technique than PS4, despite the apparent gap in power.

In fact, I mentioned a while back that Durango looked very much like a system specifically designed to run UE4. While this might not be exactly the case, I'd feel comfortable in saying that Epic would have had a fair influence in the design of the console (as they did with the XBox360's RAM, and that was when they were much less influential in the industry than they are now). If Epic are actually dropping SVOGI altogether, though, this could prove a big issue for Microsoft. The big (in fact close to the only) difference between the PS4 and Durango's designs is MS's decision to put 32MB on-die with the GPU, which meant there was a bit less space on there for GPU logic, and that they could get away with using a DDR3 main pool, which (they assumed) would be larger than a GDDR5 pool. The PS4's design would be easier for developers, and more suited to current rendering techniques, but Durango's design would give it an advantage with newer, latency-bound rendering techniques like SVOGI. Therein lies the problem: if the biggest middleware engine doesn't use the technique that Durango has seemingly been designed around, where does that leave MS?

http://www.neogaf.com/forum/showthread.php?t=531771&page=4
 
And why exactly it is no longer valid for dx11 and PS4/x720

Because DX11 is much more efficient and all about multithreading. You can find examples here or in Repi presentation from BF 3 how they've managed to decrease amount of draw calls in DX 10+ in comparison to DX9 and how better multithreading works. There are more to DX11, like atomics or better use of shader cores etc
Why PS4/X720? Because they have almost identical architecture to current PC, where PS3 and X0 had completely different, especially PS3, so You had more way to optimize engine/assets around them.

I am not saying 2 times, but let say 1,5?
If You're saying 1.5 this means that 680 will be same or better in every instance of multiplatform game comparison, so basically lost Your 'bet' already.

---
Nvidia on their last presentation said that 8800GTX was only 40-50% faster than PS3/X0, yet in every case 8800GTX has much better performance in current gen games, up to twice as higher, so where is that magic optimization?
http://livedoor.blogimg.jp/hatima/imgs/b/3/b382e1f7.jpg
 
Last edited by a moderator:
Nvidia on their last presentation said that 8800GTX was only 40-50% faster than PS3/X0, yet in every case 8800GTX has much better performance in current gen games, up to twice as higher, so where is that magic optimization?
http://livedoor.blogimg.jp/hatima/imgs/b/3/b382e1f7.jpg

I wouldn't read too much in this slide, Nvidia is only speaking about gflops and it's not the best metric to compare GPU. Look at the comparison between NV20 vs NV2A it gives the impression that the later is two times more powerful and we know that it's far from the truth. A 8800 GTX is about 350GFlops (or 500 if you're counting the invisible dual issue MUL ;)) vs ~250GFlops for the Xenos/RSX but in terms of ROPs and texture units the gap is bigger.
 
I wouldn't read too much in this slide, Nvidia is only speaking about gflops and it's not the best metric to compare GPU. Look at the comparison between NV20 vs NV2A it gives the impression that the later is two times more powerful and we know that it's far from the truth. A 8800 GTX is about 350GFlops (or 500 if you're counting the invisible dual issue MUL ;)) vs ~250GFlops for the Xenos/RSX but in terms of ROPs and texture units the gap is bigger.

Yes i know, but even if we count every enhancements in G80 architecture and assume that 8800GTX is 150% faster, it still would be meaningful in discussion, because some people here think that coding to metal in closed architecture can even out 2 times performance gap, when with 8800GTX or even 8800GT that wasnt a case even after 7 years.
 
Can we have a conclusion? Since PS4 may reach 2x of its theoretical performance then we may see Unreal 4 Infiltrator runs on PS4 very close to PC demo. Someday, yes?
It's impossible to be faster than the theoretical performance. I think you mean traditionally console games have made more efficient use of their theoretical performance.
 
I wonder, if DX11 cards have hardware that supports tessellation then is it possible DX12 gpus will feature SVO GI? Nividia is going with stacked ram which should help with latency if that's one of the contributing factors to enable that feature. Would make pc games pretty awesome too.
 
I wonder, if DX11 cards have hardware that supports tessellation then is it possible DX12 gpus will feature SVO GI? Nividia is going with stacked ram which should help with latency if that's one of the contributing factors to enable that feature. Would make pc games pretty awesome too.
SVOGI acceleration strikes me as an oddly specific thing to include on a GPU. If developers want to use it, they can program shaders just like everyone else using every other lighting model out there.
 
Does it make any sense to fully jettison SVOGI? Is it the engineering expense/the additional engine upkeep? Make the right sacrifices and real-time GI is possible on next-gen consoles. I mean, discarding that it was announced as a PC exclusive and unless the fort-building is that much of a resource hog, shouldn't Fortnite with SVOGI be possible on the PS4?
 
http://www.videocardbenchmark.net/high_end_gpus.html

GeForce 680 vs Radeon 7970M

More like 32-35% slower with much more advanced memory subsytem and 8 core cpu. Its not very fast but show me PC game that usues 8 cores

Other than the fact Passmark is full of shit (showing that they do not even have SLI support with a GTX570 beating a GTX590 and a GTX670 beating a GTX690 and the clearly wrong CPU bench they have that shows just about 100% Hyperthreaded scaling)

Even from the Passmark numbers the GTX680 is 1.476991848540626 times faster, that is 47.7%, not 35%!

The GPU memory system is not more advanced!

Also if you knew anything about CPUs you should know that having 8 cores, 1.6 to 2.0Ghz with no more than half the IPC of an Ivy Bridge is not going to be faster than a 4 core (IVB) running at around 2x the clock speed due to not only the IPC difference but also non linear scaling!
 
Last edited by a moderator:
Other than the fact Passmark is full of shit (showing that they do not even have SLI support with a GTX570 beating a GTX590 and a GTX670 beating a GTX690 and the clearly wrong CPU bench they have that shows just about 100% Hyperthreaded scaling)

Even from the Passmark numbers the GTX680 is 1.476991848540626 times faster, that is 47.7%, not 35%!

The GPU memory system is not more advanced!

Also if you knew anything about CPUs you should know that an 8 core 1.6 to 2.0Ghz CPU with no more than half the IPC of an Ivy Bridge is not going to be faster than a 4 core IVB running at around 2x the clock speed!

Oh boy, I don't even know how to respond to that... If you take the 7850(7970M) as a base for your calculations (100%) GTX680 has 147% of its measured performance... But if you take GTX680 as base 100% - 7850 has something like 65% of its perf. Either way gtx680 IS NOT 2x faster.

Regarding passmark - I dont claim its one and only proper benchmark in the universe... But I can litterally "swamp" you with test results which will confirm relative performance of both GPUs - as they say - its all on the internet.

And I am not claiming that 8 Jaguar cores is faster than 4 IvyBridge, but show me PC game that has i5 as minimum requirements. Its ussually i3 or even core2duo and AMD equivalent, simultaneouslyy used for variuous tasks and services. Therefore what I am claiming is that Jaguar powered console (PS4/x720), albeit less powerful will be more capable than i5 powered PC .

Because DX11 is much more efficient and all about multithreading. You can find examples here or in Repi presentation from BF 3 how they've managed to decrease amount of draw calls in DX 10+ in comparison to DX9 and how better multithreading works. There are more to DX11, like atomics or better use of shader cores etc
Why PS4/X720? Because they have almost identical architecture to current PC, where PS3 and X0 had completely different, especially PS3, so You had more way to optimize engine/assets around them.

But there's still multiple configs, OS overhead, drivers overhead,gimped memory. And DX 11 evolution, however great, is not enough to bring PC optimisation anywhere near console level.

And this time there will be no main disadvantage of console enviroment - whch was not enough memory, which prevented multiplatforms parity more than anything else.

Offtopic
Actually, the width of a tyre doesn't improve upon traction (but that's another story).
Yes, in normal condtitions (dry weather) it certainly does.
 
Oh boy, I don't even know how to respond to that... If you take the 7850(7970M) as a base for your calculations (100%) GTX680 has 147% of its measured performance... But if you take GTX680 as base 100% - 7850 has something like 65% of its perf. Either way gtx680 IS NOT 2x faster.

Regarding passmark - I dont claim its one and only proper benchmark in the universe... But I can litterally "swamp" you with test results which will confirm relative performance of both GPUs - as they say - its all on the internet.

And I am not claiming that 8 Jaguar cores is faster than 4 IvyBridge, but show me PC game that has i5 as minimum requirements. Its ussually i3 or even core2duo and AMD equivalent, simultaneouslyy used for variuous tasks and services. Therefore what I am claiming is that Jaguar powered console (PS4/x720), albeit less powerful will be more capable than i5 powered PC .

The HD7970M is likely rated faster than the PS4 GPU, 2.17 TFLOPS vs 1.84 TFLOPS, less TMUs and ROPs (at least based on the VG leaks info).

It does not matter how many threads PC games have now, PC games are not going to hold back to 2 threads (a lot already use more).

Also funny how the background tasks of a Windows 7 PC only takes 0 to 1% of the CPU time even on an old low end 1.8Ghz C2D!
 
Last edited by a moderator:
And I am not claiming that 8 Jaguar cores is faster than 4 IvyBridge, but show me PC game that has i5 as minimum requirements. Its ussually i3 or even core2duo and AMD equivalent, simultaneouslyy used for variuous tasks and services. Therefore what I am claiming is that Jaguar powered console (PS4/x720), albeit less powerful will be more capable than i5 powered PC .

This makes no sense whatsoever. Utilisation in current games has no baring on utilisation in next generation games where requirements will be much higher. The minimum requirements today are so low because current generation games simply don't demand the power of an i5 because they're being held back by current gen consoles. Next generation demands will rise significantly and thus so will PC CPU requirements. If anything CPU requirements will be even higher because of the compute requirements in next gen games which may be required to run on the CPU side of the PCI-e interface on PC's.

An i5 is more powerful than 8 Jaguar cores, full stop. If that power is required by next generation games then it's there to be used.

But there's still multiple configs, OS overhead, drivers overhead,gimped memory. And DX 11 evolution, however great, is not enough to bring PC optimisation anywhere near console level.

No one's claiming PC's are going to be as efficient as consoles. The only claim being made is that DX11 improves the situation over DX9. And since Carmacks statement was made in relation to DX9, it can't be used as a measure of PC efficiency in future games.

And regarding OS overhead, exactly how much power do you think it takes to run Windows in the background? Bare in mind Windows 8 can run quite happily on a tablet. And don't forget that according to rumours the next generation consoles also dedicate a significant portion of their CPU resources to general OS and system level operations.

And this time there will be no main disadvantage of console enviroment - whch was not enough memory, which prevented multiplatforms parity more than anything else.

Well there's the disadvatage of launching at a main stream performance level rather than bleeding edge. And there's the seemingly much larger reservations being made for the OS and non gaming related operations this time. And while not a console disadvantage, the fact that both consoles have very PC like architectures will potentially mean more console optimisations being relevant to the PC this generation than they were in the current generation.
 
IMHO, the PC and console boundary will likely blur.

We may see Sony release Orbis-VAIO. :devilish:

Consoles may get more PC games. At which point, if the consoles form the most lucrative PC specs, developers will optimize their games to this class of devices. They may define the PC game graphics rather than holding them back.

If the average gamers' PC/Mac specs are lower than the nextgen consoles in the mean time, then they may be the ones holding back PC graphics -- until this average shifts higher and higher over time. The current gen consoles may also hold them back.

At the same time, there are other issues that may hold back PC games, like piracy and accessibility.
 
Question about SVOGI...does anyone know how sensitive that technique is to latency? Someone on GAF is suggesting the low latency of Durango's eSRAM may help it run UE 4's SVOGI better than PS4 since (he claims) SVOGI is highly sensitive to latency, of which GDDR5 is thought to be have plenty of compared to the eSRAM in Durango.

Any thoughts on such a claim from the tech heads here?

Here is the specific post I'm referring to, for reference:



http://www.neogaf.com/forum/showthread.php?t=531771&page=4


Interesting. Moreover, seems like the RAM quantity is very crucial for UE4's SVOGI, since I've read this fragment from Digital Foundry in June 2012:

Light doesn't just illuminate objects and cast shadows. Just as in real life, it bounces. Willard points to a red carpet in a new room, adjusting time of day so more sunlight enters the room - the net result being that the walls gradually become more illuminated with a red tone as more light bounces onto the surroundings. Epic is using a voxel-based approach to indirect lighting - somewhat intensive in terms of RAM but with decent performance, on high-end hardware, at least. It appears a little reminiscent of the light propogation volumes developed by Crytek for its own CryEngine 3 middleware.

Elemental and Infiltrator demos runs in 16gb ram machines with GTX 680. Maybe the 680's rawpower alleviates his latency issues but a massive amount of RAM is essential. Remember all these developers pushing for more ram in next-gen consoles?
 
No one's claiming PC's are going to be as efficient as consoles. The only claim being made is that DX11 improves the situation over DX9. And since Carmacks statement was made in relation to DX9, it can't be used as a measure of PC efficiency in future games.

This. Software does also evolve on PC for better efficiency and old statements might not be true anymore.

And regarding OS overhead, exactly how much power do you think it takes to run Windows in the background? Bare in mind Windows 8 can run quite happily on a tablet. And don't forget that according to rumours the next generation consoles also dedicate a significant portion of their CPU resources to general OS and system level operations.

Yeah no offence but sometimes I wonder if people are actually for real or just acting wacko. Like if the OS in the consoles wont need any CPU cycles at all and all the 8GB RAM will be exclusively for graphics!

Like previous consoles where a core(s) was dedicated to running the OS in the background. Then you have OS footprint in RAM and games having a decent amount of non graphics data taking up memory like engine, sound, caches and more. I wouldn't be surprised if realistically there will be top 3-4GB to be dedicated as VRAM. Btw wasn't there talk in the leaked documents about the "Xbox720" having 2 cores dedicated for OS/other non gaming tasks?
 
Back
Top