Mantle - a new low-level API for AMD GPUs
category: code [glöplog]
Yes! Another graphics API! Just what the world has been waiting for! Not that drivers do not already suck enough to make people go insane...
Fuckings to AMD: Get you fucking drivers fixed already. And while you're at it build a proper GPU switching system like Nvidia has with Optimus. Oh, yeah, and fix your Linux driver too.
kthxbye.
Fuckings to AMD: Get you fucking drivers fixed already. And while you're at it build a proper GPU switching system like Nvidia has with Optimus. Oh, yeah, and fix your Linux driver too.
kthxbye.
Here's another angle: in the end, it all boils down to a single CPU core having to fill some command buffer.
AMD's per-core performance is well behind that of Intel, ESPECIALLY on those underpowered Jaguar APUs used in the Xbox and PS4.
For AMD it is far more relevant to reduce CPU overhead in the GPU drivers than for Intel (in case you haven't noticed, in recent years AMD always has their GPUs reviewed with Intel CPUs, because their own CPUs would not be able to show the GPU's full potential).
AMD's per-core performance is well behind that of Intel, ESPECIALLY on those underpowered Jaguar APUs used in the Xbox and PS4.
For AMD it is far more relevant to reduce CPU overhead in the GPU drivers than for Intel (in case you haven't noticed, in recent years AMD always has their GPUs reviewed with Intel CPUs, because their own CPUs would not be able to show the GPU's full potential).
I think it's a big step in the right direction. The lowest accessible layer should be as close to the hardware and simple as possible, while the standards should be open and organic implementations, maintained by the hardware vendors and people like Khronos and Microsoft and contributed to by the public. As long as the hardware is working correctly, everyone would be able to fix "driver" bugs instead of waiting for fixes or creating ugly workarounds.
Quote:
... while you're at it build a proper GPU switching system like Nvidia has with Optimus. ...
sorry i caught this. but is this joke?
where i am living this optimus is a shit.
Quote:
big step in the right direction
Might be, but it's useless for most people, who don't have the resources to optimize for specific architectures.
xernobyl - what do you mean by "most people"? If you mean "human beings", then you are definitely right - most don't care about Mantle, or GL, or pixels... If you mean developers professionally doing graphics, then I would argue most do put in the effort of optimizing for specific platforms... quite a few even for multiple vendors on PC. If you're talking about hobbyists, then they already have a plethora of APIs to choose from, so no reason in developing another ease-over-performance API...
I for one welcome back our bluescreening overlords...
I for one welcome back our bluescreening overlords...
http://www.frostbite.com/connect/#battlefield-4-frostbite-mantle
Unmantled
Dismantled
Mantlepiece
PC sux.
Mickey Mantle
Maybe this is a direct effect of:
[url=http://www.zdnet.com/amd-sees-the-era-of-moores-law-coming-to-a-close-7000013413[/url]
When you cant push the envelope by building faster hardware as easily anymore then you'll just have to bite down and start to look at the code. Either that or people will see no point in keep upgrading to new hardware for pointless performance increments.
Thats why you will probably see some dumb shit going in with this to make sure it only works properly with newest hardware.
[url=http://www.zdnet.com/amd-sees-the-era-of-moores-law-coming-to-a-close-7000013413[/url]
When you cant push the envelope by building faster hardware as easily anymore then you'll just have to bite down and start to look at the code. Either that or people will see no point in keep upgrading to new hardware for pointless performance increments.
Thats why you will probably see some dumb shit going in with this to make sure it only works properly with newest hardware.
I'd say that if Moore's law would indeed come to a close, that's all the more reason to concentrate on optimizing the architecture. But we're still a ways off, Intel is already working with 14 nm, so AMD has at least two more GPU nodes to go until they get there, even if that's the last one.
So I'd say Mantle is about 5-10 years too early for that. Moore's law has to come to a close first, and then the hardware architectures have to be optimized in full.
So I'd say Mantle is about 5-10 years too early for that. Moore's law has to come to a close first, and then the hardware architectures have to be optimized in full.
"in recent years AMD always has their GPUs reviewed with Intel CPUs, because their own CPUs would not be able to show the GPU's full potential)."
"So I'd say Mantle is about 5-10 years too early for that"
Could it be that AMD need to give an edge to its Kaveri architecture?
AMD cant use Intel architecture in its APU, its stuck with its own.
Mantle give AMD an edge in term of driver efficiency. This way they can unlock more of the GPU with a lower end CPU.
See review of AMD mobile discreet graphics with AMD CPU / VS Intel. even AMD fastest mobile cpu is the bottleneck.
For AMD Mantle is 10 year late...
"So I'd say Mantle is about 5-10 years too early for that"
Could it be that AMD need to give an edge to its Kaveri architecture?
AMD cant use Intel architecture in its APU, its stuck with its own.
Mantle give AMD an edge in term of driver efficiency. This way they can unlock more of the GPU with a lower end CPU.
See review of AMD mobile discreet graphics with AMD CPU / VS Intel. even AMD fastest mobile cpu is the bottleneck.
For AMD Mantle is 10 year late...
Did someone called me?
Quote:
Could it be that AMD need to give an edge to its Kaveri architecture?
Yes, that is a possibility.
And then Mantle is indeed very late, since Intel's integrated GPUs have all but closed the gap in terms of performance and features. I wonder what they'll bring with Broadwell in a few months.