<!--quoteo--><div class='quotetop'>QUOTE </div><div class='quotemain'><!--quotec-->Abrash said that Larrabee isn’t likely to be as fast at raw graphics performance as other graphics chips, but it is power-efficient and flexible<!--QuoteEnd--></div><!--QuoteEEnd-->
you have to think very very tactical in this case!
why did intel puplish information about an uber gc? because nvidia tried to enter the cpu segment with his cuda and ati followed. intel had to send them a signal that this wouldnt work out this way. the problem is, if intel puts 64 or more cores under a die with 2-3 ghz pro core it would rock the ###### out of everything and it would instantly destroy everything on the market. they cant do that. so they are limeting their gc in a way it can hold up with nvidia and ati at the grafic segment but stops immediatley their attempt to enter a cpu controlled marked. why? because the cpus on larrabee are x86 processors. sure they got an extended instructionset and code needs a lilte bit refining but not whole rewriting in a new programming- languge.
so, thats one point. the other point is, that there are two ways to use such a chip. std easy way use directX
hard but ownage way: code directly for the card and get incredible results.
why? : the card is not a gpu its a mass rig built of x86 cpu cores (pentium p45c ? dunno atm) it has half funktional scalar and vector units for every die but they are operating mostly softwarebased, wich slows it down, compared to acutall gc's, about 40 times. and this is the advantage you can get out of it if you code directly for the larrabee chip.
sure a problem, yet known, is that the whole infrastructur of larrabee will change with version/generation three, but there will be an interpreter for it. but seriously, the whole code must be adapted to the new design.
i only want to know if there are any plans of implementing such support :) i think we will know more after the idf, i think its 23-26 sept.
atm there are roumors, that intel will delay larrabee two yeahrs. in this case i think it allready have made clear, what they want to made clear. "dont try to enter our territory, you got your own playfield!"
ps.: sorry about my english skills :/// but i try ^^
No sane game developer bases his actions on vapourwave. Larrabee is vapourwave.
Intel is hyping larrabee because they first and foremost want Nvidia and AMD out of business, not because they want to empower people ! Intel is very keen to preserve its monopoly, keep prices high and innovation low.
More cores != better. We are years after introduction of multicore CPUs to the mainstream public, and nothing particularly exciting happened ! It is VERY rare to see a game which gets close to double performance in 2 core mode compared to single core. This slow adoption of multicore usage is because multithreaded programming is ... hard. And error prone. And it is unclear whether something can actually be parallelized. Many try and fail. Some things, like raytracing, are easy to parallelize and you get close to linear improvement with extra cores. Raytracing is a solved problem. But pure raytracing gets you nowhere - a lot of research has been put into rasterized graphics and engines, while it may be easy to <i>display</i> something in raytracing, combining it with collision detection, and other stuff I don't remember well (problems typical for game development) is going to require hybrid approach. Read some interviews with John Carmack. ( google +carmack +rayracing )
puzlThe Old FirmJoin Date: 2003-02-26Member: 14029Retired Developer, NS1 Playtester, Forum Moderators, Constellation
edited August 2009
<!--quoteo(post=1725569:date=Aug 30 2009, 04:30 PM:name=RobB)--><div class='quotetop'>QUOTE (RobB @ Aug 30 2009, 04:30 PM) <a href="index.php?act=findpost&pid=1725569"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->Some guys already predicted that CPUs and graphics will be combined in one unit (again?).
I don't like that idea, i dont want to upgrade every year because my old unit cant keep up with newer games.<!--QuoteEnd--></div><!--QuoteEEnd-->
That 'Some Guy' would be Tim Sweeney, one of the leading innovators in 3D rendering during the boom that led to 3D acceleration in desktop pcs.
edit: in retrospect, this interview with sweeney is probably a better read: <a href="http://arstechnica.com/gaming/news/2008/09/gpu-sweeney-interview.ars" target="_blank">http://arstechnica.com/gaming/news/2008/09...y-interview.ars</a> Here's the paper he wrote. Recommended reading for anyone interested in this stuff <a href="http://graphics.cs.williams.edu/archive/SweeneyHPG2009/TimHPG2009.pdf" target="_blank">http://graphics.cs.williams.edu/archive/Sw.../TimHPG2009.pdf</a>
The TL;DR version:
Hardware acceleration provided a performance advantage at the cost of flexibility. 3D acceleration provides a limited amount of operations and shaders can combine these to give some flexibility. Moores law keeps marching on and now CPU power is coming close enough to support fully generalised graphics routines. Want real time ray tracing? Volumetric rendering? All of that will be possible soon. Think 16 core architectures and well developed systems for distributing operations among hundreds of hardware assisted threads.
Comments
<a href="http://games.venturebeat.com/2009/03/27/intels-larrabee-graphics-processor-draws-a-crowd-at-game-developers-conference/" target="_blank">http://games.venturebeat.com/2009/03/27/in...ers-conference/</a>
why did intel puplish information about an uber gc?
because nvidia tried to enter the cpu segment with his cuda and ati followed.
intel had to send them a signal that this wouldnt work out this way.
the problem is, if intel puts 64 or more cores under a die
with 2-3 ghz pro core it would rock the ###### out of everything and
it would instantly destroy everything on the market.
they cant do that. so they are limeting their gc in a way
it can hold up with nvidia and ati at the grafic segment
but stops immediatley their attempt to enter a cpu controlled
marked. why? because the cpus on larrabee are x86 processors.
sure they got an extended instructionset and code
needs a lilte bit refining but not whole rewriting in a new programming-
languge.
so, thats one point.
the other point is, that there are two ways to use such a chip.
std easy way use directX
hard but ownage way: code directly for the card and get incredible
results.
why? : the card is not a gpu its a mass rig built of x86 cpu cores
(pentium p45c ? dunno atm) it has half funktional scalar and vector units
for every die but they are operating mostly softwarebased, wich slows
it down, compared to acutall gc's, about 40 times.
and this is the advantage you can get out of it if you code directly
for the larrabee chip.
sure a problem, yet known, is that the whole infrastructur of larrabee
will change with version/generation three, but there will be an interpreter
for it. but seriously, the whole code must be adapted to the new design.
i only want to know if there are any plans of implementing such support :)
i think we will know more after the idf, i think its 23-26 sept.
atm there are roumors, that intel will delay larrabee two yeahrs.
in this case i think it allready have made clear, what they want to made clear.
"dont try to enter our territory, you got your own playfield!"
ps.: sorry about my english skills :/// but i try ^^
I don't like that idea, i dont want to upgrade every year because my old unit cant keep up with newer games.
Intel is hyping larrabee because they first and foremost want Nvidia and AMD out of business, not because they want to empower people ! Intel is very keen to preserve its monopoly, keep prices high and innovation low.
More cores != better. We are years after introduction of multicore CPUs to the mainstream public, and nothing particularly exciting happened ! It is VERY rare to see a game which gets close to double performance in 2 core mode compared to single core. This slow adoption of multicore usage is because multithreaded programming is ... hard. And error prone. And it is unclear whether something can actually be parallelized. Many try and fail. Some things, like raytracing, are easy to parallelize and you get close to linear improvement with extra cores. Raytracing is a solved problem. But pure raytracing gets you nowhere - a lot of research has been put into rasterized graphics and engines, while it may be easy to <i>display</i> something in raytracing, combining it with collision detection, and other stuff I don't remember well (problems typical for game development) is going to require hybrid approach. Read some interviews with John Carmack. ( google +carmack +rayracing )
j/k
anyway, most people would agree to "raytracing or not, larabee will be a flop".
<!--quoteo--><div class='quotetop'>QUOTE </div><div class='quotemain'><!--quotec-->and nothing particularly exciting happened<!--QuoteEnd--></div><!--QuoteEEnd-->
besides the obvious addition of each core's speed.
I don't like that idea, i dont want to upgrade every year because my old unit cant keep up with newer games.<!--QuoteEnd--></div><!--QuoteEEnd-->
That 'Some Guy' would be Tim Sweeney, one of the leading innovators in 3D rendering during the boom that led to 3D acceleration in desktop pcs.
edit: in retrospect, this interview with sweeney is probably a better read: <a href="http://arstechnica.com/gaming/news/2008/09/gpu-sweeney-interview.ars" target="_blank">http://arstechnica.com/gaming/news/2008/09...y-interview.ars</a>
Here's the paper he wrote. Recommended reading for anyone interested in this stuff <a href="http://graphics.cs.williams.edu/archive/SweeneyHPG2009/TimHPG2009.pdf" target="_blank">http://graphics.cs.williams.edu/archive/Sw.../TimHPG2009.pdf</a>
The TL;DR version:
Hardware acceleration provided a performance advantage at the cost of flexibility. 3D acceleration provides a limited amount of operations and shaders can combine these to give some flexibility. Moores law keeps marching on and now CPU power is coming close enough to support fully generalised graphics routines. Want real time ray tracing? Volumetric rendering? All of that will be possible soon. Think 16 core architectures and well developed systems for distributing operations among hundreds of hardware assisted threads.
To me it sounds like CPU Power will outweigh GPUs and make them superflous since they came up because CPUs couldn't keep up anymore.
While it sounds good, it sounds too good.