Digital Foundry Article Technical Discussion Archive [2016 - 2017]

Status
Not open for further replies.
And this confirms that there is no magic about the improvement of the XB1 version...

According to DICE they've got some more out out X1's highly customised command processor. I'd be inclined to take them at their word.

Figures from DF seem to indicate they may have got a little more out of X1 than we would have expected.
 
According to DICE they've got some more out out X1's highly customised command processor. I'd be inclined to take them at their word.

Figures from DF seem to indicate they may have got a little more out of X1 than we would have expected.

And what was the special trick for the PS4 ?

Also, i have a hard to imagine that several developers would miss such an advantage during many years... and how can you know it wasn't used in Battlefront (DX12) ?
 
Last edited:
And what was the special trick for the PS4 ?

Also, i have a hard to imagine that several developers would miss such an advantage during many years... and how can you know it wasn't used in Battlefront (DX12) ?
And again my issue about you in the UC4 thread, you aim to win the argument instead of having a discussion.

PS4 and XBO have access to the same features. It was DICE who made the move to ExecuteIndirect for DX12 which has equivalents in OpenGL and in GNM. So it benefits all of them. Hence why we saw both copies likely move up into performance.

How do you know it's not in SWBF? We don't know, but the presentation came after BF was released so it makes sense they invested more into it.

If ExecuteIndirect and it's equivalent names are implemented more and more, expect to see better performance from the consoles, but XBO performs well in handling EI but it still suffers from a lack of overall power.

How is this basic optimization ? Rewriting the rendering pipeline is far from basic.
 
If anything DICE have been behind the curve optimizing for consoles if you ask me, with so many titles sticking to 720p for such a long time on XBox one and 900p on PS4 with seemingly very few changes from one title to the next. Of course 60fps does take its toll, but you'd have expected dynamic resolution to show up sooner. I am guessing part of it had to do with the engine becoming so widely used within EA?
 
If anything DICE have been behind the curve optimizing for consoles if you ask me, with so many titles sticking to 720p for such a long time on XBox one and 900p on PS4 with seemingly very few changes from one title to the next. Of course 60fps does take its toll, but you'd have expected dynamic resolution to show up sooner. I am guessing part of it had to do with the engine becoming so widely used within EA?

Even with dynamic resolution there's no reason for either console to gain a burst of 30% resolution at 60fps. They moved mountains I think. The hold up may have been waiting for DX12 to be released. But someone will point out that DX11 PC benchmarks run faster than DX12. But then again we know that Dx12 features are being exposed/hacked into DX11 variants through drivers.

Assuming I'm wrong entirely, then I'd look closely at how they are doing dynamic resolution. Perhaps they are doing lower resolutions on the top and bottom and keeping it higher in the middle?

Hard to say. IMO, BF1 is among the best looking titles to date and it's surprising to me that they can get that much out of either console at 60fps.
 
Last edited:
If anything DICE have been behind the curve optimizing for consoles if you ask me, with so many titles sticking to 720p for such a long time on XBox one and 900p on PS4 with seemingly very few changes from one title to the next. Of course 60fps does take its toll, but you'd have expected dynamic resolution to show up sooner. I am guessing part of it had to do with the engine becoming so widely used within EA?

Also probably a time thing, EA Dice has been going full steam the last few years and are now seem to be on 1 year release cycles like COD. Not to mention games like BF:4 were a complete mess on launch and required a lot of fixes. Also BF:4 and Battlefront have received a fair bit of extra content...
 
If ExecuteIndirect and it's equivalent names are implemented more and more, expect to see better performance from the consoles, but XBO performs well in handling EI but it still suffers from a lack of overall power.

Better performances, yet you have a worse framerate compared to Battlefront... i'm not here to win a discussion, but you can't base a discussion on unsupported claims.

Let me ask you this question : how much better is BF1 compared to Battlefront ?

You can't know...
 
Better performances, yet you have a worse framerate compared to Battlefront... i'm not here to win a discussion, but you can't base a discussion on unsupported claims.

Let me ask you this question : how much better is BF1 compared to Battlefront ?

You can't know...

The claims about Execute Indirect and the X1's enhanced support for it are not unsupported, and as of DICE's presentation for GDC March 2016 they say they will "definitely be using it going forward". It is logical that they are using these features to increase performance.

So iroboto's position - that we can't know for sure this is it's first outing for DICE but it may well be - is entirely reasonable.

And yes, DICE did have some very positive things to say about X1's command processor extensions. Turns out X1 does indeed have a dash of secret sauce locked away, though I'm sure that sauce will taste sour to some.
 
Chances are it was locked away due to the higher level API bullshit MS was pulling early on that hurt them so badly. Was probably used primarily by the OS early on.

How much help could this give, though? I know things aren't as simple as X thing equals Y% speedup, but...
 
The claims about Execute Indirect and the X1's enhanced support for it are not unsupported, and as of DICE's presentation for GDC March 2016 they say they will "definitely be using it going forward". It is logical that they are using these features to increase performance.

So iroboto's position - that we can't know for sure this is it's first outing for DICE but it may well be - is entirely reasonable.

And yes, DICE did have some very positive things to say about X1's command processor extensions. Turns out X1 does indeed have a dash of secret sauce locked away, though I'm sure that sauce will taste sour to some.

So, it's an awesome advantage, yet none developers used it so far ? It's against basic logic...

So far, only bold speculations from a GDC paper... sorry, but this is not serious...
 
Better performances, yet you have a worse framerate compared to Battlefront... i'm not here to win a discussion, but you can't base a discussion on unsupported claims.

Let me ask you this question : how much better is BF1 compared to Battlefront ?

You can't know...
I can't know when DICE started their implementation period of ExecuteIndirect and equivalent. True, nor do I know the impact. But I do know that
PS4 plays a majority of it's play in the 95% near 1080p resolution
XBO plays a majority of it's play in the near 900p resolution.
Numbers were provided earlier for it's min/max, but for the absolute sake of discussion we can round to both 1080 and 900 (for PS4) and 900 and 720 (for XBO).

You say I can't know how much better BF1 is compared to battlefront, visually no. Technically we may be able to deduce this.
SWBF was 900p on PS4, not always locked 60, but pretty close to locked 60.
Dynamic resolution aims to preserve frame rate, in this case, it will reduce resolution until the frame rate is met. We know that if performance in frame rate in worse on XBO and PS4 for BF1 (which isn't nearly as bad as you make it out to be, it looks to be pretty close to 1080p and 900p respectively for a majority of the time).. but if frame rate is tanking in BF1, then the resolution will be also at it's minimum, which is 900 and 720p respectively.

If framerate is tanking at these resolutions, we know that BF1 has a larger workload than SWBF. What causes the frame rate hitch can very well be a discussion point, perhaps it's CPU, perhaps GPU.
 
So, it's an awesome advantage, yet none developers used it so far ? It's against basic logic...

So far, only bold speculations from a GDC paper... sorry, but this is not serious...

It's an advantage in dealing with executeIndirect _overhead_. Workload is still resolved over the CUs.
You haven't been reading the forums have you. ExecuteIndirect is where we are headed. Developers haven't been using it because as we've stated many times over, it's not simple to integrate any technology into your engine and rebuild. It costs time and money, and it happens over a longer period of time.
 
It's an advantage in dealing with executeIndirect _overhead_. Workload is still resolved over the CUs.
You haven't been reading the forums have you. ExecuteIndirect is where we are headed. Developers haven't been using it because as we've stated many times over, it's not simple to integrate any technology into your engine and rebuild. It costs time and money, and it happens over a longer period of time.

The main point of this GDC slide was compute... you're acting like if excecuteindirect was the main point of their presentation while there are just some words on it...

http://www.slideshare.net/gwihlidal...peline-with-compute-gdc-2016?from_action=save

Here is the summary of their work : http://image.slidesharecdn.com/gdc2...ith-compute-gdc-2016-94-638.jpg?cb=1458345423

Also, can you provide evidence supporting the claim that other developers don't use Execute ?
 
Last edited:
If framerate is tanking at these resolutions, we know that BF1 has a larger workload than SWBF. What causes the frame rate hitch can very well be a discussion point, perhaps it's CPU, perhaps GPU.

Given it does have a dynamic resolution and it tanks that seems a sign it's more likely CPU related. Then the tanking is worse on the console with less CPU resources, coincidence?

It would be interesting to pixel count during frame drops to further implicate the CPU.

Either way it seems a massive jump in visuals and performance from BF4. I wonder what the ps4 pro will turn in, that could be pretty special and make the wait till Scorpio quite difficult.
 
Given it does have a dynamic resolution and it tanks that seems a sign it's more likely CPU related. Then the tanking is worse on the console with less CPU resources, coincidence?

It would be interesting to pixel count during frame drops to further implicate the CPU.

Either way it seems a massive jump in visuals and performance from BF4. I wonder what the ps4 pro will turn in, that could be pretty special and make the wait till Scorpio quite difficult.
I was thinking this as well. And to prove it exactly in the same manner
 
The main point of this GDC slide was compute... you're acting like if excecuteindirect was the main point of their presentation while there are just some words on it...

http://www.slideshare.net/gwihlidal...peline-with-compute-gdc-2016?from_action=save

Here is the summary of their work : http://image.slidesharecdn.com/gdc2...ith-compute-gdc-2016-94-638.jpg?cb=1458345423

Also, can you provide evidence supporting the claim that other developers don't use Execute ?
I didn't say no other developer uses it, I just wanted to be clear that not everyone can immediately and for different reasons. We are generalizing words, soI know the point in which you were trying to get at.
tldr;
It's never as simple as just integrating it in just because its available. We've waited a very very long time for Tiled resources to appear in games after it's announcement in 2012.

I certainly trust our resident developer word's on that topic this forum:

#484
#482

Article about execute Indirect:
http://www.dsogaming.com/news/direc...proves-performance-greatly-reduces-cpu-usage/
Our resident moderator Andrew Lauritzen was the one who coded that demo so he can verify the performance gains written about in the article.
I was there in person when Max McCullen presented that demo at Build, so I can verify that the DSO wasn't pulling numbers out of thin air.
 
The main point of this GDC slide was compute... you're acting like if excecuteindirect was the main point of their presentation while there are just some words on it...

http://www.slideshare.net/gwihlidal...peline-with-compute-gdc-2016?from_action=save

Here is the summary of their work : http://image.slidesharecdn.com/gdc2...ith-compute-gdc-2016-94-638.jpg?cb=1458345423

Also, can you provide evidence supporting the claim that other developers don't use Execute ?

You missed a few slides especially on the Future work. It's obvious that the XB1 has hardware features that are still not used in today's games. To support these features the console needs a lot of stuff to happen.

1. Xbox One running Windows 10 which happened in November 2015 about 11 months ago.
2. The XB1 SDK must be updated to support the feature.
3. The game engine must be updated to support the feature.
4. The games must then use the feature.
5. DX12 is *still* not finished. It's still waiting for Model Shader 6 which is expected to release in Holiday 2016 (this year). You can watch the presentation as they're using WARP12 and XB1 backend to test the software.


But most importantly, the game engine MUST be rewritten from the ground up for DX12. You can port DX11 to DX12 but performance gains may be minimal. To give you a perspective only recently that Unreal Engine 4.12 support the following on XB1: async compute and hardware compression. In fact, experimental DX12 and Fast Semantics were only added in 4.11. Gears of War 4 is one of the first to use UE4.11 with great visual and performance results.

Can you provide evidence that other developers are using ExecuteIndirect for XB1 games, prior or current?
 

Attachments

  • diceGDC2016.png
    diceGDC2016.png
    623 KB · Views: 14
  • SM6onXB1.PNG
    SM6onXB1.PNG
    118.8 KB · Views: 15
We are focusing on executeindirect and this feature has been already shown running on PC. When it comes to XBOX ONE what must be highlighted from GDC slides are those features that, at that time, were exclusive to XBOX ONE. No other GPU were able to make use of this features, GCN 1.4 might make use of them today but in the slides they told that it couldn´t be used by any other gpu apart from Durango.

Those features are:
EWmRYnx.png

cjqzd19.png

JTuvPcW.png
 
Previous frostbite games had shown a huge difference between ps4 and xbox one.

If we do the maths:
(1600*900)/(1280/720)=+56%

In battlefield if we trust NXgamer we have:

Ps4
GVADfMx.jpg


XBOX ONE
UNW1IkT.jpg


If we dod the maths again:
(1780*1000)/(1624*945) = +15%

We can clearly see an advantage of about 56% in previous games, now this advantages has been shortened and is about 15%.

This is a huge boost on XBOX ONE.

Obviously it depends on how acurate are the figures provided by NXgamer
 
Status
Not open for further replies.
Back
Top