Larrabee Dead
|
Prev: SMT exploiting 21264-like clustering?
Next: Batched_Instruction_Windowing_for_Single_Threaded_Execution
From: Brett Davis on 28 May 2010 01:12 Larrabee Dead http://www.semiaccurate.com/2010/05/27/larrabee-alive-and-well/ >'Note: For what it's worth, the leading candidate at Sony right now >is an internal design that several describe as "Emotion Engine based". >Sony should have waited for Larrabee three.....' Without a console win Larrabee is dead, no matter what they say. It becomes a hobby project with little hope. Avatar and the other 3D movies have changed the equation from back when I posted that Larrabee won PS4. Two years from now every new TV not in the bargain bin will do 3D. The bandwidth demands of a second frame buffer to do 120Hz updates likely makes things like ray tracing fall by the wayside. Economics has dictated that the new console will not intro at such high prices as the last generation, that nearly killed the PS3. A scaled up PS3 with more Cell processors can render polys to tiles without breaking a sweat. Hits the 3D cost/performance window, unlike Larrabee. And of course I could be wrong again, this is all a SWAG. Brett
From: Terje Mathisen "terje.mathisen at on 28 May 2010 01:57 Brett Davis wrote: > Larrabee Dead > http://www.semiaccurate.com/2010/05/27/larrabee-alive-and-well/ > >> 'Note: For what it's worth, the leading candidate at Sony right now >> is an internal design that several describe as "Emotion Engine based". >> Sony should have waited for Larrabee three.....' > > Without a console win Larrabee is dead, no matter what they say. > It becomes a hobby project with little hope. Without a console win, LRB has lost the X million guaranteed production run, which I agree is significant, but not a real showstopper since those X million chips would probably generate close to zero real income. I.e. Intel would have to price the chips at a point where they gave compelling cost advantages. Having a single chip which could do both the core game logic and all the 3D stuff at the same time still seems like a very good idea to me for a mass market, rock bottom production cost, product. > > Avatar and the other 3D movies have changed the equation from back > when I posted that Larrabee won PS4. Not at all. > Two years from now every new TV not in the bargain bin will do 3D. > The bandwidth demands of a second frame buffer to do 120Hz updates > likely makes things like ray tracing fall by the wayside. A factor of two is the smallest factor you even notice in this business. > Economics has dictated that the new console will not intro at such > high prices as the last generation, that nearly killed the PS3. > > A scaled up PS3 with more Cell processors can render polys to > tiles without breaking a sweat. Hits the 3D cost/performance > window, unlike Larrabee. The main advantage to this is of course that it is almost completely known territory, with much smaller risks. The original Cell/PS3 design was in fact more of a leap into the unknown than LRB would have been. Terje PS. Those of my friends who have worked on LRB for some years now, are all still doing it. :-) -- - <Terje.Mathisen at tmsw.no> "almost all programming can be viewed as an exercise in caching"
From: Brett Davis on 29 May 2010 01:47 In article <ei24d7-85o.ln1(a)ntp.tmsw.no>, Terje Mathisen <"terje.mathisen at tmsw.no"> wrote: > Brett Davis wrote: > > Larrabee Dead > > http://www.semiaccurate.com/2010/05/27/larrabee-alive-and-well/ > > > >> 'Note: For what it's worth, the leading candidate at Sony right now > >> is an internal design that several describe as "Emotion Engine based". > >> Sony should have waited for Larrabee three.....' > > > > Without a console win Larrabee is dead, no matter what they say. > > It becomes a hobby project with little hope. > > Without a console win, LRB has lost the X million guaranteed production > run, which I agree is significant, but not a real showstopper since > those X million chips would probably generate close to zero real income. > > > Two years from now every new TV not in the bargain bin will do 3D. > > The bandwidth demands of a second frame buffer to do 120Hz updates > > likely makes things like ray tracing fall by the wayside. > > A factor of two is the smallest factor you even notice in this business. Each of the Playstation generations was 10 times faster, including ten times more memory bandwidth. The PS3 has two memory buses, XDR and GDDR3. The PS4 will have roughly twice the bandwidth, cost and heat reasons have presented a brick wall for external RAM. Todays graphics chips are now largely limited in performance by bandwidth. We a facing a temporary end of scaling, and that pathetic two X increase might be largely eaten up by the double frame buffer for real 3D. Yes the CPU will be 10 times faster, but what you see besides 3D is not much better than today. Will the user pay $600 for a PS4 with the PS3 at $200, and looking the same on the 2D TV the user may have? Bear in mind that the Wii_HD-3D will cost $250, with the bare minimum hardware to meet the users needs. Just like the original Wii. For Sony the phrase is "We live in interesting times." Brett RRAM will get us an easy 10 times bandwidth speedup, but that is looking like a PS5 issue. I am betting against an embedded frame buffer, as it would have to be HUGE and costly. Tiling looks like the way forward, due to the small parallax difference it makes sense to render the same tiles for both screens at a time, to benefit from nearly identical texture fetch, and nearly identical verts, etc. Best use of precious bandwidth. Cell processors are a good match for compute rendering tiles.
From: Jeremy Linton on 3 Jun 2010 18:03
On 5/29/2010 12:47 AM, Brett Davis wrote: > Todays graphics chips are now largely limited in performance by bandwidth. > We a facing a temporary end of scaling, and that pathetic two X increase > might be largely eaten up by the double frame buffer for real 3D. But is that going to cause an image quality wall? I was under the impression that tessellation was fairly localized and provided a nice image boost without requiring a large increase in geometry/texture/etc bandwidth. My basic understanding was that there is still a fair number of other "tricks" just awaiting more processing power. I've seen the nivida 3d vision demos, and from what I understand they are just utilizing the cards ability to render frames at rates > 30fps. Tweak the camera angle every other frame and you have 3d. The game detail doesn't seem to suffer because of the 3d. |