-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Intel: DirectX 12 yields big gains on tablets, other thermally constrained devices
- Thread starter jh30uk
- Start date
Soldato
- Joined
- 25 Nov 2011
- Posts
- 20,679
- Location
- The KOP
Coming out of beta should be the next stage for mantle.
And within Time it will... I believe this year is when the public can get a hold of the code.
Associate
- Joined
- 28 Feb 2013
- Posts
- 2,468
- Location
- Birmingham
The Compute Shader Pipeline in DirectX 11 already does this (to an extent). During scenes where the GPU has relatively little work to do, traditional CPU calculated computations such as physics and AI are moved over to the GPU for calculating.
I fully expect an expansion of the CS Pipeline in functionality and efficiency in DirectX 12.
53 Developers at the last count involved with the Mantle closed BETA. who knows what they are doing and testing, they are no doubt under a strict NDA.
With the potential programing levels of the API and things like hUMA there is so much tasty potential.
We have seen nothing yet.
I would love to know what they are upto.
CUDA already offers this on the desktop.
http://devblogs.nvidia.com/parallelforall/unified-memory-in-cuda-6/
I wonder how much of an advantage it will offer to games when the GPU and CPU are discreet parts separated over a relatively slow bus.
Soldato
- Joined
- 5 Sep 2011
- Posts
- 12,883
- Location
- Surrey
lol
We need Mantle 1.0 first. DX12 fundamentally does the same things (and more), whether it sparked it or not is any ones guess although I'd say it's pretty clear it had a large part in pushing it forward. Mantle was actually comprised by a few OpenCL enthusiasts that were picked up by AMD. This would make sense as prior to any of this AMD wouldn't of had a division capable of comprising something like an API. Basically, they're still very new to it. Currently it would appear to a lot of people as a marketing snow job - thanks to implementations in games that frankly do not need it. It's got a long way to go, and developers need to really clamp down on memory paging as so far what we've seen has left people including myself unimpressed.
But then I have no interest in incremental gain as I don't have a middle of the road system.
Last edited:
To an extent, but in DX12 will the serial calc have access to the GPU's memory pool, that's 300 Gflops of Floating Point performance vs 30. to quote rounded averages.
Its already here for serial to parallel productivity calculation.
Last edited:
Soldato
- Joined
- 5 Sep 2011
- Posts
- 12,883
- Location
- Surrey
There is documentation somewhere that contains all the new pipeline features, will see if I can find it.
I'd say it's a given considering the goals. Developers are limited in where and how resources can be allocated, dispatched and managed. This is largely controlled auto-magically by the DX driver with the developer only being able to 'hint' at what they wanted the hardware to do in an ideal world. DX12 changes that. Taken to the extreme, the developer can (effectively) control that resource handling themselves. By having control of memory allocation, memory paging and various other low-level accessor-like functionality, they have much more control over how resources are handled. In conjunction with other changes to the API (which are numerous and completely rewritten in most cases) this is where a significant performance gain can be made. Of course, on the flip side, it also is the place where you could make a complete hash of it and degrade performance too through poor handling of said resources. A lot like with how we've seen thus far with Mantle.
I suppose a lot of uptake will rely heavily on what tools are easier/universal to use. I know a lot of engines including UE4.0 has apparently had a lot of it's roots completely stripped. Given what we've heard (albeit from closed doors participants) that Mantle is extremely easy to translate to, still no idea if this is the case with DX12. I'd like to think if it is actually more of a challenge, then there is a lot more fundamentally improved than there is with Mantle. Time will tell
It seems that the X-Box One did not get HSA - hUMA while the PS4 did.
Which is understandable as its impossible to implement through something like DX11, Hence Mantle.
I don't know what API the X-Box One has at the moment but it looks to me like something derived from DX11.
DX12 should enable hUMA, one reason its Getting it to compete with the PS4.
Looks to me like DX12 is a new API to make proper use of the Architecture powering it in Consoles.
PS4 will use its unified architecture to pound Xbox One into the dust – eventually
Which is understandable as its impossible to implement through something like DX11, Hence Mantle.
I don't know what API the X-Box One has at the moment but it looks to me like something derived from DX11.
DX12 should enable hUMA, one reason its Getting it to compete with the PS4.
Looks to me like DX12 is a new API to make proper use of the Architecture powering it in Consoles.
PS4 will use its unified architecture to pound Xbox One into the dust – eventually
Last edited:
Soldato
- Joined
- 5 Sep 2011
- Posts
- 12,883
- Location
- Surrey
There was a bit hoo-har about how the technology demonstration was fake, the ESRAM has no direct access to the CPU it uses a cache or memory pool type before it can access it. Which is why it's a pile of crap essentially compared to the PS4, which is a true direct connection between CPU/GPU
+1, The industry cannot move forward if advanced technology is monopolised by one vendor.
The question was once asked if Mantle was to be available to all vendors who has control?
It has to be Microsoft, willing or by necessity upon themselves
Why didn't the X-Box One not get or take up hUMA while the PS4 did?
Just new hardware for Microsoft, they didn't have time, Skill or willingness to create an API for?
Microsoft not cooperating with AMD demanding that such an API also be available on the Desktop in exchange for it, is that the reason for AMD springing what is called Mantle to the Desktop out of the blue, to pile on the pressure?
Whatever was going on behind closed doors I think AMD very much will keep Mantle alive and keep developing it as a stick, or carrot to Microsoft.
Soldato
- Joined
- 5 Sep 2011
- Posts
- 12,883
- Location
- Surrey
You know what you were saying about being naive earlier? 
Well you'd be swinging that way if you thought AMD took on a project like this which currently only has support for GCN if they were doing it to 'pile on pressure'
Well you'd be swinging that way if you thought AMD took on a project like this which currently only has support for GCN if they were doing it to 'pile on pressure'
You know what you were saying about being naive earlier?
Well you'd be swinging that way if you thought AMD took on a project like this which currently only has support for GCN if they were doing it to 'pile on pressure'
You misunderstand what I'm driving at, it stands a much better chance of making it if it has universal architectural support, not just GCN. IE: as DX12.
Soldato
- Joined
- 5 Sep 2011
- Posts
- 12,883
- Location
- Surrey
Well unfortunately that won't happen if the statement that Nvidia released last week is anything to go on, but that is in no way surprising.
Nvidia are not going for DX12 compatibility? since when?
Soldato
- Joined
- 5 Sep 2011
- Posts
- 12,883
- Location
- Surrey
Lol, No. Are you still philosophising that Mantle is DX12? If so I'm out.
Oh it will be - but it'll be DX12. Just like Nvidia won't touch mantle (they'll work on reducing the gap in drivers as usual) MS won't touch mantle. They've no vested interest in not being at the centre of things - not just for Windows but for XBOX too (which has commonality with DX). I'm not sure if DX12 is a direct product of Mantle being released but either way DX12 is shaping up to be very good. From a developer perspective i'd also point out that DX does an awful lot (and generically) that you'd otherwise have to implement yourself. If you look at DX and it's history it's slowly been moving toward being more per-formant anyway. Just like Phsyx I hope mantle goes the way of the dodo and that DX simply absorbs the the best bits. Mantle smells of Glide - those were not great days..