AMD Mantle API [updating]

[Alexko]

If true, then that pretty much confirms Kepler and GCN are forward compatible with DX12. Pretty good news I think.

Also I think that confirms DX12 is just a direct response to Mantle.

I think it pretty much confirms that GCN 1.1 is compatible with DX12, but implies little about GCN 1.0 and Kepler.

 

[DavidGraham]

I think it pretty much confirms that GCN 1.1 is compatible with DX12, but implies little about GCN 1.0 and Kepler.

I thought consoles are GCN, not GCN 1.1?!

 

[dobwal]

That there is one fine chronicle of corporate bullshit speak.

If Xbox didn't exist and MS was trying to push Windows against Playstation then I imagine things would be a bit more "focused" for PC gaming. It would be interesting if Xbox was running regular Windows 8, blurring the lines between platforms. But obviously they prefer total control and lockdown on Xbox.

If the Xbox didn't exist, the PC market would probably be no different than it is today.

The reality is that PC is where it is today because its relatively a niche market. The surge in console sales came at a time where there existed an even larger disparity between IQ produced by consoles versus PC than exists today. A PS2 was hardly a match for 9700 pro.

Console simply went mainstream and became more attractive to consumers, developers and manufacturers. Its not a coincidence that most of the major players in the console space used to major players in the PC gaming space. And it has nothing to do with MS and their less than enthusiastic approach to moving DirectX forward.

MS's effort with DX is a sign of the times not a driver of it.

 

[3dilettante]

I think it pretty much confirms that GCN 1.1 is compatible with DX12, but implies little about GCN 1.0 and Kepler.

I think Microsoft has an interest in not making DX12 so lopsided. Either it has comparable backwards compatibility for the major graphics vendors, or it tweaks things enough so that there's at least a fighting chance for them all in an upcoming product cycle.

I thought consoles are GCN, not GCN 1.1?!

Both console GPUs have features found in the Sea Islands ISA document and in driver features mentioned for Sea Islands, although it isn't entirely clear how much of that pool each one has and how much is tweaked specifically for each console.
It's not even clear all the time what features the discrete GPU products have from the pool versus the APUs, or what each specific product may or may not have.
Everyone (out here) wants to have a coherent way of discussing the IP set of these architectures, but AMD and its customers are not interested.

The volatile bit, for example, has been mooted as something Sony wanted specifically, even though the instructions for using it are in the Sea Islands document.
The separately programmable DMA engines echo the first two move engines in Durango. Orbis might have this facility, but I haven't seen it disclosed.

 

[eastmen]

I hope there is a lot of house keeping going on. They need to get power draw down on their mobile products

 

[DmitryKo]

If true, then that pretty much confirms Kepler and GCN are forward compatible with DX12

Strictly speaking, forward compatibility means there is no need for new display driver, since existing DDI works without changes. This is how it happened with feature levels 9_1, 9_2 and 9_3 in Direct3D 11 runtime, which directly calls DDI9.

I'd rather think Direct3D 12 will take feature level 11_0 as a baseline, but will require a new driver model.

I think it pretty much confirms that GCN 1.1 is compatible with DX12, but implies little about GCN 1.0 and Kepler.

I thought consoles are GCN, not GCN 1.1?!

In terms of D3D feature levels, both of them are level 11_1 AFAIK, but I believe GCN 1.1 supports [post=1780981]tiled resources Tier 2[/post] as an optional feature on levels 11_0 and 11_1, while GCN 1.0 only supports Tier 1.

Microsoft could as well devise an entirely new capability grouping scheme in D3D12.

Both console GPUs have features found in the Sea Islands ISA document and in driver features mentioned for Sea Islands, although it isn't entirely clear how much of that pool each one has and how much is tweaked specifically for each console.

Is this document available on http://developer.amd.com/resources/documentation-articles/developer-guides-manuals/ ?

 

[Alexko]

I think Microsoft has an interest in not making DX12 so lopsided. Either it has comparable backwards compatibility for the major graphics vendors, or it tweaks things enough so that there's at least a fighting chance for them all in an upcoming product cycle.

Then again, Maxwell is already here, while DX12 will probably take some time, so Kepler compatibility is not necessarily essential. Ditto for GCN 1.0.

 

[3dilettante]

We'd have to inquire with those more familiar with the low-level driver interfaces with the GCN architecture.
Mantle doesn't make a distinction between GCN products, so what minor tweak is going to keep a DX12 that has full support for Bonaire or Hawaii from having the same support on the older cards?

 

[Alexko]

We'd have to inquire with those more familiar with the low-level driver interfaces with the GCN architecture.
Mantle doesn't make a distinction between GCN products, so what minor tweak is going to keep a DX12 that has full support for Bonaire or Hawaii from having the same support on the older cards?

Actually, Mantle was originally announced with only limited support for GCN 1.0. I don't know if that has changed since then.

 

[WaltC]

If the Xbox didn't exist, the PC market would probably be no different than it is today.

The reality is that PC is where it is today because its relatively a niche market. The surge in console sales came at a time where there existed an even larger disparity between IQ produced by consoles versus PC than exists today. A PS2 was hardly a match for 9700 pro.

Console simply went mainstream and became more attractive to consumers, developers and manufacturers. Its not a coincidence that most of the major players in the console space used to major players in the PC gaming space. And it has nothing to do with MS and their less than enthusiastic approach to moving DirectX forward.

MS's effort with DX is a sign of the times not a driver of it.

I think you pretty much have it backwards--except for the line about the 9700P blowing away the PS2. The new consoles from Microsoft/Sony *are* PCs--100% "PC" according to AMD's definition of a PC, if you haven't noticed. That's because the PC, far from being "niche", is ubiquitous. The custom console is the endangered (if not already extinct) species. IE, the "PC" has "absorbed" the gaming console as it was formerly known. Even in an off year, PCs still sell at the rate of ~30M a month--whereas it took Sony 3 months to move 6M PS4's (outpacing xBone.) The PC is in the best place it's ever been as far as gaming is concerned, imo--and I think the fly-away success of Kickstarter proves it even to diehard skeptics (custom console exclusives could never support KS.) Economies of scale are huge in the PC market (which is why Microsoft and Sony went with AMD's PCs), and the advantage to consumers in the areas of component upgradability and self-service maintenance is legion (unless you buy into 1980's-style surface-mounted motherboard tech like the new consoles and many Apple products use--you know, upgrade the cpu or add more ram?-Get a new motherboard, etc.) There simply isn't a better buy out there for consumers than a user-serviceable, user-updatable PC.

So what if AMD's Mantle initiative lights a fire under Microsoft in terms of DX development? What Microsoft says isn't important anyway--it's what the company does that matters. If Microsoft won't get the job done, someone else will, you can bet on it. It may yet come down to AMD, after all, to move things ahead--as it has so many times before (thinking here of what it took to get Intel off of Rdram and Itanium-for-64-bits, etc.)

 

[Kaotik]

Actually, Mantle was originally announced with only limited support for GCN 1.0. I don't know if that has changed since then.

Where exactly was it mentioned "limited support for GCN 1.0"? It's been just "GCN" since day 1, not "GCN 1.1" or "GCN 1.0" or anything, BF4 optimizations being "work in progress" for 1.0's, though.

 

[homerdog]

It doesn't do much for most GCN cards including "APUs" is maybe what he meant. I think AMD intends to fix this at some point.

Edit or perhaps this is just a BF4 thing. Hard to say at this point.

 

[Kaotik]

It doesn't do much for most GCN cards including "APUs" is maybe what he meant. I think AMD intends to fix this at some point.

Edit or perhaps this is just a BF4 thing. Hard to say at this point.

It works just fine in StarSwarm, and AFAIK the BF4-issues aren't that much performance related itself, but memory usage (reportedly it goes way past 3GB on HD7970 for example, which of course causes issues, but 'till then the performance upgrade Mantle gives is just fine)

 

[3dilettante]

Actually, Mantle was originally announced with only limited support for GCN 1.0. I don't know if that has changed since then.

AMD has been consistent about saying it worked with GCN.

Deviations from that tend to be more consistent with AMD's failure to provide a consistent platform.
The 290 and 280 series cover the original and refreshed version of GCN, and work well with Mantle.
BF4 has problematic GCN support below the high end cards, but that includes Bonaire.

 

[Blazkowicz]

Good point, the "front-end" stuff in GPUs, number of Asynchronous Compute Engine and whatever stuff probably play a good part in Mantle performance so there Tahiti is good despite being 1.0.

If that logic holds then Kaveri should do well (?) but you would have to run it at 1366x768 with whatever lowish detail level and try to estimate results in multiplayer maybe. (where you need CPU power the most)

Other possibility the drivers, game patches are optimized for the big boys first anyway i.e. AMD wants big benchmark bars for "flagship" and "mindshare" foremost.

 

[3dilettante]

Bonaire, Pitcairn, and Tahiti have a single command processor and two ACEs, but Mantle so far only works fine with Tahiti in BF4.

The ACEs have more bearing for GPU compute, and potentially more so for HSA in particular. However, scattered statements were made to the effect that Mantle is very much focused on graphics. It's not clear what Mantle can do with the extra ACEs in a card like Hawaii, if it can do anything.
The numbers for BF4 don't give Hawaii a disproportionate improvement with Mantle compared to Tahiti.

 

[Blazkowicz]

Thanks (and sorry about my laziness on this)

 

[Alexko]

AMD has been consistent about saying it worked with GCN.

Deviations from that tend to be more consistent with AMD's failure to provide a consistent platform.
The 290 and 280 series cover the original and refreshed version of GCN, and work well with Mantle.
BF4 has problematic GCN support below the high end cards, but that includes Bonaire.

Pour rappel, les GPU les plus anciens en GCN 1.0 n'obtiendront que des gains limités avec cette première version sous Battlefield 4, en attendant des optimisations promises par AMD. Ce sont donc les Radeon HD 7790, Radeon R7 intégrés dans les APU, les R7 260, 260X et surtout R9 290 et 290X qui profiteront le plus de gains, plus particulièrement quand la limite de performance se situe au niveau du CPU.

http://www.hardware.fr/news/13553/amd-catalyst-14-1-beta-avec-mantle-ligne.html

I'll try to find a source in English later. Obviously, GCN 1.0 is still supported, but it required more work, presumably significantly different codepaths here and there. I'm not saying GCN 1.0 and/or Kepler is likely to lack DX12 support, just that nothing is certain.

And while having broad initial support for a new API is a good thing, there's also something to be said for trying to design the best possible API by ignoring what hardware features are currently available and aiming for what ought to be available. Sometimes, this kind of approach pays off in the long term, even if it can mean slow initial adoption.

 

[zupallinere]

Nothing is known about the time frame for DirectX12. Will it be something that will only be available for the next version of windows ( Win 9 or whatever ) to help drive it's adoption ?? I mean I read that it will most likely be something that will make cross-platform Windows graphics programming easier but in terms of Windows on the PC will any major performance changes be used to drive folks to a new OS ?

 

[DmitryKo]

Will it be something that will only be available for the next version of windows ( Win 9 or whatever )

It's Xbox One and most likely Windows 9 (and please, please don't start yet another [post=1830752]"but Mantle works on Windows 7" [/post] argument or ask [post=1830789] "why they can't backport D3D12 to Windows 7, 8, or 8.1"[/post] )

will any major performance changes be used to drive folks to a new OS ?

When Windows 9 will be released next Spring, Windows 7 will be almost six years old. This fact alone should make users consider a newer OS, not to mention things like SATA Express/NVMe for SSDs (since Windows 8.1 FYI), Direct3D 12 (and WDDM 2.x?), and other yet undisclosed improvements.

Look, Microsoft is on a new yearly OS release schedule for a foreseeable future. There won't be major service packs that add significant features. Tablet/phone consumers don't understand it, they want simple numbers in the product name, not mumbo-jumbo like "Windows Vista SP2 with Platform Update OEI DSP" etc. which is decipherable only by tech-support junkies. The casual users see Android going from 2.2 to 4.4 or Chrome going from 7 to 22 in just two years and think "hey, that's probably cool". So Microsoft will play this game too. No more beating the dead horse, no backporting - if you want new features, get a newer OS.