Tweet CatalystCreator, We want AMD DX11 Multi-Threading support

Actually it wasn't a beta driver, it was an WHQL one that broke the fan on older cards:3eyes:

I don't think BF3 supports MT, or at least doesn't take advantage of it anyway, as i have yet to see any benchmarks that show Nvidia cards have any advantage.

But i hope you guys can pressure AMD into it, will be better for all of us in the long run.

 

BF3 Supports it, and its the only game I would need it for anyways.

This feature should be in the drivers, no and's if's or but's

That being said, I just tweeted CatalystCreator, I hope more people do the same. :nerd:

 

Repi said on a forum(forgot where, tho well known) that they ended up not using MTR...According to him, it worked slower than their current solution


I do wonder why AMD can't even implement MTR specifically for Civ5 though, like Nvidia did...At least it would show that they are trying to close the performance gap for that game

 

Wait, that doesn't make sense. You are utilizing more threads. IE more interaction from the CPU (the driver) to the GPU thus removing any bottlenecks. How is that slower? I like to see a link to that post if possible because that isn't making any sense. Also, it's been posted by them for a while now.
http://blogs.amd.com/play/2009/09/09/directx-11-–-what-to-expect/

Edit:
Also, it's suppose to elevate the draw call limitation when compared to consoles:
http://forums.guru3d.com/showpost.php?p=4165536&postcount=26

 

Found it, was on beyond3d
http://forum.beyond3d.com/showthread.php?t=54372&page=73

 

What he really meant was, it wasn't worth DICE's time to do something when AMD doesn't support it, so only half would benefit from it.

As said before, it doesn't make sense how DX11 MT would be slower. More draw calls, the better it is. There is no workaround that, other than DX11 MT for right now.

Notice what he says.

AMD's current implementation of DX11 MT without command lists is probably not making it really useable.

 

Yeah I read it but it needs some clarification as it doesn't make sense to me. Also he mentions that they want to reduce CPU overhead. How does that relate to MTR when you want to utilize the CPU more?

The person who asked the question about MTR mentioned the performance improvements from Civ 5 for nvidia, twice and I read no comment about it.

 

Why slower? More explained here: http://code4k.blogspot.com/2011/11/direct3d11-multithreading-micro.html Experienced developer wrote benchmark that shows how DX11 multithreading behave with hardware that have and have not such feature. Check out that article for results. All i can say is why you would need to have that so urgently?

EDIT: Now link is correct

All is explained in article. Under certain circumstances it may be slower though.

 

Your link doesn't work but the most rudimentary premise to MTR is that it allows for the driver to utilize more threads. If for whatever reason they are not needed then the performance should remain the same, not a drop.

 

Adding MT support is only viable if the increased speed outweighs the latency introduced.

When going from single threaded to multithreaded you have to take into account that multiple threads will usually have to keep synced with each other. If you have to introduce locks to do this, it can result in a slow down rather than a speedup.

 

That reads like a dependency in the driver itself not MTR though. An all or nothing approach. If the game doesn't benefit using it then MTR shouldn't be used.

 

Sorry, now link is ok. Check a benchmark

 

Ok, read it, but don't see the relevance. What game did they test? Because I'm not seeing one. And, they used a AMD mobile part vs 570. Also, what drivers are they using to do this? That left me with more questions then answers.

I understand the intent of the article but what would the result be if IHV worked with a developer to use MTR in game? Then post the results with and without MTR.

 

Yes, he used Mobile vs desktop, but non MTR fps/ms results both cards matches.
The whole point is in Analysis chapter, at the down of article: most important for us is that "Using multithreading and deferred context rendering is only relevant when the CPU is effectively used on each threads" which means we have a heavily Mutlithreaded game first to implement in it MTR. Second result is "The native support from driver of Direct3D11 multithreading doesn't seem to change so much (of performance)" - well dont argue, i just quote author here

 

But what game was used that can be measured? I have no idea where the cube program came from. What I like to see is an actual game developed using MTR and IHV supported. That tells us what the pros/cons are of the MTR. You based all of your post on just one program. That's simply not enough data to draw anything concrete other then just a talking point.

Also, the mobile gpu vs desktop gpu isn't really apples/apples. We don't know what drivers are used. As it stands now there is no public drivers that allow for MTR from AMD. If so, please link us so we can all use them.

 

Well, that aspect might be true, MTR might behave differently in real game. But do we have evidence that theese few actual MTR games benefits really because of that feature? Or maybe they are just optimised somehow for Nvidia hardware?

 

But that's the thing, we won't know until we have MTR support from the drivers. 1 step at a time IMO. Lets 1st have the MTR driver support then go from there. Your link doesn't indicate anything about the drivers actually supporting MTR or not. So how would I know with the data provided is a result of AMD's implementation of MTR?

 

Ah, forgot to ask to read article carefully Look we have at the begining two cards, with similar non MTR benchmark results. One with MTR driver support (nvidia) and second without MTR support in drivers (amd), at the end both cards in MTR tests permorm almost equal.. so result is that MTR in driver gives no benefit. Thats the point. Im not native english but thats how i read it.

 

I did . The point I was alluding to is what are the results of AMD's use of MTR vs Nvidia's use of MTR in a game (particularly a game that supports MTR vs without). Without that information it's difficult, if not impossible, to draw any of the conclusions you believe.

I understand you want to "tunnel" me to only view a certain aspect to the article. But I have a much wider view of this inwhich certain testing conditions need to be met in order to get a better picture of MTR and it's use.

As it stands the article shows:
-Nvidia with MTR support using a 570
compared to
-AMD mobile part without MTR support
That's not a real apples/applies comparison with a game that has MTR support vs without MTR support.

 

True, and they could implement it that way and have the profiles dictate which path to take. It would be easier and probably quicker to have 2 seperate files... one that supports MT and one that doesn't.

For that matter, you should be able to configure how many threads you wanted it to use in the profile.

I can see it now... "why does setting my profile to use 8 threads slower then if I set it to use 4 threads?"