I sense a dark force hovering over the "close thread" button ... maybe we can defeat it by going back to topic?
@ Carfax83
In DX11, rendering commands (aka batch/draw call) are still not thread safe.
That's why Dan call DX's attempt to MT a "failure". I would call it a "not there yet".
TC's Civ5 benchmark for the HD 7970 seems to be from when the HD 7000 series was released. The HD 7000 series got big performance improvements as the drivers matured.
This more recent test by Anandtech shows that Nvidia still has the advantage in Civ5, but it's less than it was before
http://www.anandtech.com/show/6774/nvidias-geforce-gtx-titan-part-2-titans-performance-unveiled/13
In the case of Civ5, it's also possible that there were alot more driver optimizations done than just adding support for DX11 multithreaded rendering.
There's nothing wrong with the technology. It works. Project C.A.R.S also uses it, and so did AC III. The only reason why more games don't support it is because it took awhile before the driver quality was up to snuff for it to result in significant performance increases, and AMD has never implemented it in their drivers at all.. The latter in particular has been a major obstacle to it's adoption I'd wager, because for DX11 multithreading to be fully effective, the engine has to be designed with it in mind.IMO, considering the few confirmed games supporting it, you're giving DX11 multithreaded rendering far too much credit. It's a warning sign when there are so many games that DON'T support it.
It would be a good move by Microsoft is DX12 would be a superset of DX11, with extended benefits and backward compatibility for DX11 cards..I sure as hell will be extremely happy if the next DirectX and OpenGL even can get remotely close to closing the gap with Mantle, but that's probably happening first with Windows 9. And it'll probably require new GPUs as well. I hope DirectX 12 will be backwards compatible with current cards though, like how the DX11 API brought improvements to DX10.x as well
Really?
This forum just hit a low for me ... gl hf!
Yea it must be everyone else, not the fact that your argument is completely flawed.I swear, no matter how many times I cover this, people still just don't get it.
I swear, no matter how many times I cover this, people still just don't get it.
I guess I'll repeat myself again for the 10th time.. The Civilization 5 benchmark can be either GPU, or CPU limited to varying degrees based on the hardware and the settings.
The one you just linked to right now, is more GPU bound than the one I had on the first page of this thread. Therefore, the 7970 is faster than the GTX 580.
One of the benchmarks I linked to in the first page ran at 1680x1050, making it more CPU limited, which is why the GTX 580 was able to pull ahead of the 7970. It underscores the fact that DX11 multithreading does work rather well at resolving CPU limitations:
There's nothing wrong with the technology. It works. Project C.A.R.S also uses it, and so did AC III. The only reason why more games don't support it is because it took awhile before the driver quality was up to snuff for it to result in significant performance increases, and AMD has never implemented it in their drivers at all.. The latter in particular has been a major obstacle to it's adoption I'd wager, because for DX11 multithreading to be fully effective, the engine has to be designed with it in mind.
Most game engines opt to use manual threading, which runs on both NVidia and AMD hardware with no issues, but I don't think is nearly as effective at reducing CPU bottlenecks...
It would be a good move by Microsoft is DX12 would be a superset of DX11, with extended benefits and backward compatibility for DX11 cards..
TC's Civ5 benchmark for the HD 7970 seems to be from when the HD 7000 series was released. The HD 7000 series got big performance improvements as the drivers matured.
This more recent test by Anandtech shows that Nvidia still has the advantage in Civ5, but it's less than it was before
http://www.anandtech.com/show/6774/nvidias-geforce-gtx-titan-part-2-titans-performance-unveiled/13
In the case of Civ5, it's also possible that there were alot more driver optimizations done than just adding support for DX11 multithreaded rendering.
I swear, no matter how many times I cover this, people still just don't get it.
I guess I'll repeat myself again for the 10th time.. The Civilization 5 benchmark can be either GPU, or CPU limited to varying degrees based on the hardware and the settings.
The one you just linked to right now, is more GPU bound than the one I had on the first page of this thread. Therefore, the 7970 is faster than the GTX 580.
One of the benchmarks I linked to in the first page ran at 1680x1050, making it more CPU limited, which is why the GTX 580 was able to pull ahead of the 7970. It underscores the fact that DX11 multithreading does work rather well at resolving CPU limitations:
There's nothing wrong with the technology. It works. Project C.A.R.S also uses it, and so did AC III. The only reason why more games don't support it is because it took awhile before the driver quality was up to snuff for it to result in significant performance increases, and AMD has never implemented it in their drivers at all.. The latter in particular has been a major obstacle to it's adoption I'd wager, because for DX11 multithreading to be fully effective, the engine has to be designed with it in mind.
Most game engines opt to use manual threading, which runs on both NVidia and AMD hardware with no issues, but I don't think is nearly as effective at reducing CPU bottlenecks...
It would be a good move by Microsoft is DX12 would be a superset of DX11, with extended benefits and backward compatibility for DX11 cards..
Now that's very interesting. Thanks for posting zlatan.
The cynic in me really has to wonder though at some of his statements. In the interview, Baker claims that DirectX made an attempt to be threaded, and failed. Yet the Lore engine, something which he himself had a big part in creating, obviously benefits tremendously from DX11 multithreading. In fact, it helped NVidia dominate AMD in that game for years.
So why was he extolling the benefits of driver command lists in the Lore presentation I linked to on the first page, and now he seems dead set against it?
And benchmarks some of our members have posted on the forum of the Star Swarm benchmark shows that it can benefit from DX11 multithreading in a big way.
The Star Swarm benchmark is seemingly contrived to make Mantle look good. It doesn't use instancing, one of the technologies in DX11 that can drastically reduce the amount of draw calls necessary by using the same geometry to draw multiple objects, and so the amount of draw calls is inflated relative to what it would be in an actual game.
And CPU and GPU utilization are poor on non Mantle systems, which really begs the question of how well it's optimized or DX..
Everybody should read this. It's from Dan Baker who made Lore Engine for Civ5.
Exactly like hardware PhysX - lower level API, exposes extra performance, only works with one vendor with specific cards, and only in rare sponsored titles.
If someone isn't consistent with Mantle and hardware PhysX, they have a vendor bias.
...
If know trolls why not to ban them?
I can't do that with Starswarm. Not that Starswarm is a game, but with the level of rhetoric being bandied around, it seems like there are definitely people who wish that it were. With a title like that, if those programming techniques are (allowed?, Encouraged?), there would be a clear bisection of people who could play the game, and people who effectively couldn't, all just based on which brand video card I happen to have. To make it clear, I know that the game still will run on a quad core with an NVIDIA card, but I consider 6 FPS with only one pixelated ship on screen and no actual commands being issued 'Unplayable'.
So, you would rather have less options in the gaming world, so you don't have to make the choice?
Even better would be to just do away with multiple hardware companies, so that way games only have to build for that company. Imagine how awesome the would would be if AMD or Nvidia were to just stop making hardware. Devs would then spend way less time making games for multiple platforms and oh that would be great right?
If Mantle cannot open possibilities, then Devs wont want it even if AMD pays them. If Mantle does allow for new things in games, then great. Why be worried about new things when those new things offer up cool games?
Further more, people already have to make choices. Want to play new AAA games on 1080 with full eye candy? Guess what, you have to buy a card that is more than $100. Want to play the same game at 4k, now you have to buy more. The idea that innovation or new ideas will fracture the market and hold back people from having access is very very short sighted.
If you don't like something, don't buy it. If you do, then support it.
The entire point of the post was that I want more choice, that I am leery of a scheme that locks people out of performance, or worse, consigns people that buy one brand of hardware to a crippled, inferior product. The whole point was that vendor specific functions are fine when they are minor, insignificant things, but that when the entire code base of a game might depend on a vendor specific feature we're treading in very dangerous waters.
The problem with your benchmark is that it's done with very early drivers. The GCN architecture was new, and the HD 7000 series has gotten significant improvements as the drivers have gotten better.
The biggest problem with DX11 MT is that it's performance is wholly dependent on the drivers. The time it took for NVidia to come to grips with DX11 MT turned off a lot of developers, and AMD of course still to this day haven't implemented it.Would DX11 multithreaded rendering be better supported if AMD supported it too? Possibly. But with only one to three games currently benefiting from it, it's a huge generalization to say there's nothing wrong with it.
SourceConsidering the current GPU drivers (Nvidia, AMD) don't support deferred contexts, we have had to give up on the idea of multithreading.
Dice would have used it if it had been available when BF3 came out, as Repi lamented. As for Crytek, I doubt it would give them much benefit as Crytek games are always heavily GPU bound, and they have exceptional manual threading implementation anyway with CryEngine 3. Crysis 3 is the most GPU bound game I've ever played personally speaking, but it uses the CPU very well...Each game is different, and studios like Dice and Crytek aren't using it.
Using DX11 MT doesn't automatically guarantee a large performance increase, much the same as with Mantle. The performance increase is proportional to how CPU bound the game is..And let's be real now. If Dice or Crytek or any other studio did use DX11 multithreaded rendering and show a big performance advantage for Nvidia hardware, AMD wouldn't be in a position to say no.
The BF3 slides said that they were still waiting for the drivers. And after NVidia came out with their driver, I think he said it wasn't effective.Dice was very clear already when they were developing BF3 that it doesn't work for their use cases. And they keep contact with both AMD's and Nvidia's driver teams.
Was DX multi threading showing any real advantages before Win 8.1, most recently? Even now, after 5 years, it seems to work in only 2 games. People are complaining about the lack of adoption Mantle has, ans it's gonna far outstrip what DX multi thread has accomplished in it's first few months to a year.
Repi credits instancing with major CPU performance increases in BF3 and BF4.. There's nothing wrong with instancing, as long as it doesn't curtail the artists' vision.Extensive use of instancing is one of the major peeves that Dan has w/DX. Instancing is a dumbing down of the game to make it fit inside the limits of DX. It's not a good thing from the developers perspective. It's more of a necessary evil to reduce CPU load. Actually, it's not all bad. It is good for things like grass, for example.
Quoting this again so people actually read it.
I find it odd, regardless of the CPU/GPU being the possible bottleneck in the benchmarks, that people are using Civ 5 as some kind of evidence that DX can compete with Mantle. The DEVELOPER for Civ 5 is saying that DX multithreading cannot hang in today's hardware environment let alone the future.
The driver date has some impact, but not as much as you think.. I posted the latest benchmark of Civ 5 on Anandtech in the old Mantle thread:
Here you see the same thing. The GTX 580 is only 1% slower than the 7970. If the benchmark had used a lower resolution, the GTX 580 would be ahead I do not doubt.
That's what DX11 multithreading does, it removes CPU bottlenecks. The 7970 is a faster card than the GTX 580, but because it doesn't use DX11 MT, it's handicapped.
The biggest problem with DX11 MT is that it's performance is wholly dependent on the drivers. The time it took for NVidia to come to grips with DX11 MT turned off a lot of developers, and AMD of course still to this day haven't implemented it.
For example, here is what a Capcom developer said back in 2010 about DX11 MT:
Source
So much of the frustration by developers towards DX11 MT came from it not even being available for a long time. And even when it's available, it may not necessarily work as it's such a difficult thing for IHVs to do properly and fine tune.
Dice would have used it if it had been available when BF3 came out, as Repi lamented. As for Crytek, I doubt it would give them much benefit as Crytek games are always heavily GPU bound, and they have exceptional manual threading implementation anyway with CryEngine 3. Crysis 3 is the most GPU bound game I've ever played personally speaking, but it uses the CPU very well...
Using DX11 MT doesn't automatically guarantee a large performance increase, much the same as with Mantle. The performance increase is proportional to how CPU bound the game is..
That's why Civ 5 gives such a hefty increase with DX11 MT. BF3 MP likely would have benefited from it, but we'll never know I suppose.
The BF3 slides said that they were still waiting for the drivers. And after NVidia came out with their driver, I think he said it wasn't effective.
That was years ago though. NVidia's DX11 MT algorithm has surely improved considerably since then..
AC III also uses it. The 650 Ti Boost is a much slower card than the HD 7870, but here it's pulling even in this CPU dependent benchmark. Same thing with the 7950 boost and the GTX 660:
I'm sort of confused with this statement. It's not like artists want to be repetitive in outdoor enviroments. And it isn't even the only way to do it in a cost effective manner, in fact something like this here could be fairly simple for mapmakers to handle (think brush tool).Repi credits instancing with major CPU performance increases in BF3 and BF4.. There's nothing wrong with instancing, as long as it doesn't curtail the artists' vision.
Much of the artwork in games has always been repetitive anyway..
Direct X involves bigger costs for developers and it will be dropped naturally