I'm still not sure why something that eliminates the whole CPU vs GPU hasn't been designed... I know OSX does some tricks in software to use GPU cycles, but why not a 196 core multi-purpose processor?
There are, it's just not general usage.
Mostly for HPC applications or for projects such as Folding@Home or the World Health Grid by IBM.
If I'm playing a game on an iPad, I sure as hell will notice a difference between a game running at 20 FPS VS a game running at 60 FPS. And I bet that many other people will too. Most people with two eyeballs will notice.
There are things where it would be useful beyond gaming, although gaming is arguably the largest market that takes advantage of it. I still want to see them support OpenCL as opposed to trying (and often failing miserably) to write things in shader code.
Quote:
Originally Posted by BeyondYourFrontDoor
I'm still not sure why something that eliminates the whole CPU vs GPU hasn't been designed... I know OSX does some tricks in software to use GPU cycles, but why not a 196 core multi-purpose processor?
I think we're approaching the threshold of usefulness... Doing 4000 fps on a 600dpi screen is no better than 60fps on a 300dpi screen...
There aren't a lot of transparent libraries that use the gpu wherever possible, or at least none that I have come across. I don't know about private APIs, but so far I haven't found any evidence of this. Most of the claims are just conjecture. There are certain tasks that naturally use the gpu, but OpenCL has to be explicitly used. You would also have to write additional code for situations where it is not available, but those will dwindle on currently supported hardware. Mavericks actually added support for the 2012 integrated graphics models. You also seem to forget the difference in complexity between gpu cores and typical ARM or X86 cores.
So, how does the new Imagination GPU compared to any of those water-cooled desktop gamer cards??? Enquiring minds want to know.
Apple currently uses a quad-core PowerVR, which is on the level of an Intel HD3000. There is a 6-core version of the PowerVR6, which should be 50% faster and this 6-core XT version should be 50% faster again so 2.25x faster than an Intel HD3000, which is maybe 640M performance. That's not bad considering it'll likely be so low power as to be passively cooled.
It's more about what they do with the power than the raw power itself. In the Unreal 4 engine, they've made physically accurate rendering demos and others with very realistic lighting on the new Tegra K1:
[VIDEO]
NVidia has a spec comparison with the PS3 and 360 and claims games can be ported over ok:
Carmack suggests it may not turn out to be the case but isn't dismissing the claims. The demos at least show significant visual quality and as APIs develop, they find faster ways to get the same quality. NVidia for example has their own anti-aliasing technology that is much faster than competing platforms.
Current Apple A7 devices run games at visual quality comparable with the consoles. The last-gen consoles have the latest titles like GTA 5, Call of Duty Ghosts, Tomb Raider, Assassin's Creed 4. Skip to 6:13 in the following video to see quality on the A7 already:
[VIDEO]
Games publishers are struggling to stay afloat. The studio behind Bioshock is now gone, the one behind Remember Me struggling too, the Tomb Raider and Hitman developer Square Enix isn't making enough sales:
They need to focus more on what the problem is. The reason they don't sell enough profitable copies is that not every game is a first day hit. The ones that are like Call of Duty Ghosts and GTA 5 make massive profit in the first week so sales of used games don't affect them as much and they get the hype from the initial sales. By the time the franchises with lower marketing budgets get any ground, there are used copies and sales floating around everywhere. Every mobile purchase is a first time sale and the only distribution cost is the 30% fee of digital stores.
The console audience is about 300 million people. Most games don't sell more than 20 million copies but GTA 5 is at 32.5m just now. Apple's audience rivals the console audience. If a game costs $50m to make and they sell it at $9.99 on the App Store and make 70%, then it needs to sell about 7m copies but that's a lot easier to do with a $10 price than a $50-60 price.
Apple needs to help them out though. They need a better controller setup than they have now. I'd rather that they make their own and build it properly at a reasonable price. Then they can commission ports from these studios - $1m per port and they'll easily make it back from their 30% cut. Then they put the games in a dedicated section of the store so buyers can find them easily. They need the faster devices so it drives people to buy the higher priced products with the fast GPUs.
am I the only one around here that thinks Imagination is better buy than Tesla? am I the only one who wants PowerVR to be Apple exclusive?
Tesla would still be the better buy. Wall Street has already stated that Apple's hardware business is completely ruined due to low-cost Android devices. Apple needs a completely new revenue stream that doesn't involve smartphones or tablets. This new graphics chip would be like adding whipped cream on top of a stale cake (as far as Wall Street is concerned). Tesla would bring a whole slew of new technologies to Apple and Elon Musk's foresight. Apple isn't short on tech. The main problem is that Apple has run short of target customers to buy their tech with Samsung flooding their Android devices in every available niche. The smartphone market is almost totally saturated. Tesla's electric car market looks to be wide open.
There are things where it would be useful beyond gaming, although gaming is arguably the largest market that takes advantage of it. I still want to see them support OpenCL as opposed to trying (and often failing miserably) to write things in shader code.
Information is hard to come by. The lack of OpenCL has me wondering it the GPU architecture simply doesn't support it well. I've never much like the idea that these companies think it is wise to keep documentation locked up behind NDAs.
There aren't a lot of transparent libraries that use the gpu wherever possible, or at least none that I have come across. I don't know about private APIs, but so far I haven't found any evidence of this.
To the best of my knowledge Apple makes use of NEON and other ARM features for most of the math libraries. Obviously the GPU is used extensively in iOS for graphical acceleration but as far as I know nothing beyond that.
Most of the claims are just conjecture. There are certain tasks that naturally use the gpu, but OpenCL has to be explicitly used. You would also have to write additional code for situations where it is not available, but those will dwindle on currently supported hardware. Mavericks actually added support for the 2012 integrated graphics models. You also seem to forget the difference in complexity between gpu cores and typical ARM or X86 cores.
In some ways the cores in a GPU are simple in that each ALU has a limited instruction set compared to an Intel i86 core. It is tying all those cores together that gets to be complicated. Mavericks has very good OpenCL support even for Intel hardware (finally). This makes me wonder what the next step will be for Mac OS.
Information is hard to come by. The lack of OpenCL has me wondering it the GPU architecture simply doesn't support it well. I've never much like the idea that these companies think it is wise to keep documentation locked up behind NDAs.
Some of the graphics hardware components used by Apple have seen OpenCL support from other vendors. I looked it up some time ago. The reason I don't expect it soon is that they allowed GLSL compute shaders some time ago. I'm not sure they would have allowed that if a more readable method was being finalized. The difference in performance is enough that people still use that method, but it's definitely more difficult. GLSL isn't that pretty, given the need (to the best of my knowledge) to pass everything as a VBO.
edit: I was thinking of PowerVR. They do support OpenCL implementations on much of their hardware.
Comments
I'm still not sure why something that eliminates the whole CPU vs GPU hasn't been designed... I know OSX does some tricks in software to use GPU cycles, but why not a 196 core multi-purpose processor?
There are, it's just not general usage.Mostly for HPC applications or for projects such as Folding@Home or the World Health Grid by IBM.
Wow my iPad Air already feels crazy fast I can't imagine where we'll be in terms of power in a few years. Seems like we're doubling every year!
I disagree.
If I'm playing a game on an iPad, I sure as hell will notice a difference between a game running at 20 FPS VS a game running at 60 FPS. And I bet that many other people will too. Most people with two eyeballs will notice.
There are things where it would be useful beyond gaming, although gaming is arguably the largest market that takes advantage of it. I still want to see them support OpenCL as opposed to trying (and often failing miserably) to write things in shader code.
I'm still not sure why something that eliminates the whole CPU vs GPU hasn't been designed... I know OSX does some tricks in software to use GPU cycles, but why not a 196 core multi-purpose processor?
I think we're approaching the threshold of usefulness... Doing 4000 fps on a 600dpi screen is no better than 60fps on a 300dpi screen...
There aren't a lot of transparent libraries that use the gpu wherever possible, or at least none that I have come across. I don't know about private APIs, but so far I haven't found any evidence of this. Most of the claims are just conjecture. There are certain tasks that naturally use the gpu, but OpenCL has to be explicitly used. You would also have to write additional code for situations where it is not available, but those will dwindle on currently supported hardware. Mavericks actually added support for the 2012 integrated graphics models. You also seem to forget the difference in complexity between gpu cores and typical ARM or X86 cores.
Apple currently uses a quad-core PowerVR, which is on the level of an Intel HD3000. There is a 6-core version of the PowerVR6, which should be 50% faster and this 6-core XT version should be 50% faster again so 2.25x faster than an Intel HD3000, which is maybe 640M performance. That's not bad considering it'll likely be so low power as to be passively cooled.
It's more about what they do with the power than the raw power itself. In the Unreal 4 engine, they've made physically accurate rendering demos and others with very realistic lighting on the new Tegra K1:
[VIDEO]
NVidia has a spec comparison with the PS3 and 360 and claims games can be ported over ok:
http://www.extremetech.com/gaming/177002-john-carmack-suspicious-of-nvidias-outlandish-tegra-k1-claims
Carmack suggests it may not turn out to be the case but isn't dismissing the claims. The demos at least show significant visual quality and as APIs develop, they find faster ways to get the same quality. NVidia for example has their own anti-aliasing technology that is much faster than competing platforms.
http://www.slashgear.com/nvidia-tegra-k1-out-performs-intel-haswell-in-early-benchmarks-13312939/
Current Apple A7 devices run games at visual quality comparable with the consoles. The last-gen consoles have the latest titles like GTA 5, Call of Duty Ghosts, Tomb Raider, Assassin's Creed 4. Skip to 6:13 in the following video to see quality on the A7 already:
[VIDEO]
Games publishers are struggling to stay afloat. The studio behind Bioshock is now gone, the one behind Remember Me struggling too, the Tomb Raider and Hitman developer Square Enix isn't making enough sales:
http://www.polygon.com/2014/2/21/5431332/what-irrational-games-closure-means-for-the-industry
http://www.polygon.com/2014/1/31/5364452/report-remember-me-studio-dontnod-filing-for-bankruptcy
http://www.eurogamer.net/articles/2013-04-09-square-enix-thought-tomb-raider-could-sell-nearly-double-its-3-4-million-first-month-sales
http://www.vg247.com/2013/03/26/square-financials-tomb-raider-hitman-sleeping-dogs-fail-to-hit-sales-targets/
They need to focus more on what the problem is. The reason they don't sell enough profitable copies is that not every game is a first day hit. The ones that are like Call of Duty Ghosts and GTA 5 make massive profit in the first week so sales of used games don't affect them as much and they get the hype from the initial sales. By the time the franchises with lower marketing budgets get any ground, there are used copies and sales floating around everywhere. Every mobile purchase is a first time sale and the only distribution cost is the 30% fee of digital stores.
The console audience is about 300 million people. Most games don't sell more than 20 million copies but GTA 5 is at 32.5m just now. Apple's audience rivals the console audience. If a game costs $50m to make and they sell it at $9.99 on the App Store and make 70%, then it needs to sell about 7m copies but that's a lot easier to do with a $10 price than a $50-60 price.
Apple needs to help them out though. They need a better controller setup than they have now. I'd rather that they make their own and build it properly at a reasonable price. Then they can commission ports from these studios - $1m per port and they'll easily make it back from their 30% cut. Then they put the games in a dedicated section of the store so buyers can find them easily. They need the faster devices so it drives people to buy the higher priced products with the fast GPUs.
am I the only one around here that thinks Imagination is better buy than Tesla? am I the only one who wants PowerVR to be Apple exclusive?
Tesla would still be the better buy. Wall Street has already stated that Apple's hardware business is completely ruined due to low-cost Android devices. Apple needs a completely new revenue stream that doesn't involve smartphones or tablets. This new graphics chip would be like adding whipped cream on top of a stale cake (as far as Wall Street is concerned). Tesla would bring a whole slew of new technologies to Apple and Elon Musk's foresight. Apple isn't short on tech. The main problem is that Apple has run short of target customers to buy their tech with Samsung flooding their Android devices in every available niche. The smartphone market is almost totally saturated. Tesla's electric car market looks to be wide open.
To the best of my knowledge Apple makes use of NEON and other ARM features for most of the math libraries. Obviously the GPU is used extensively in iOS for graphical acceleration but as far as I know nothing beyond that.
In some ways the cores in a GPU are simple in that each ALU has a limited instruction set compared to an Intel i86 core. It is tying all those cores together that gets to be complicated. Mavericks has very good OpenCL support even for Intel hardware (finally). This makes me wonder what the next step will be for Mac OS.
Information is hard to come by. The lack of OpenCL has me wondering it the GPU architecture simply doesn't support it well. I've never much like the idea that these companies think it is wise to keep documentation locked up behind NDAs.
Some of the graphics hardware components used by Apple have seen OpenCL support from other vendors. I looked it up some time ago. The reason I don't expect it soon is that they allowed GLSL compute shaders some time ago. I'm not sure they would have allowed that if a more readable method was being finalized. The difference in performance is enough that people still use that method, but it's definitely more difficult. GLSL isn't that pretty, given the need (to the best of my knowledge) to pass everything as a VBO.
edit: I was thinking of PowerVR. They do support OpenCL implementations on much of their hardware.