• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Intel GPU in the PS4?

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Originally posted by: Nemesis 1
More food for thought . I like the 6 game part.

<a target=_blank class=ftalternatingbarlinklarge href="http://
<b">
http://news.softpedia....s-Lar...s-Larrabee-90268.shtml]http://news.softpedia.com/news/Summer-2009-Intel-Launches-Larrabee-90268.shtml[/L]

Heres one about whats actually taking place,

http://news.softpedia.com/news...XP-Drivers-97402.shtml

Your first link is broken, here is the fixed link: http://news.softpedia.com/news...s-Larrabee-90268.shtml
 
Originally posted by: Nemesis 1
I like this one the best. There are just to many high profile people saying the same thing.

Thats alot of egg on face with all the hype.

http://news.softpedia.com/news...-to-an-End-82315.shtml

I'm not so convinced of Intel's prognostication capabilities when it comes to hearkening the time of death of anything in this industry.

They can't even get the time of death nailed down for their x86 business, as EPIC was supposed to supplant x86 and not become a parallel CPU architecture like an in-grown hair on the rump of a dodo bird. x86 wasn't supposed to live long enough to see 64bit, let alone be the fundamental basis of an entire new GPU product generation. And now look where x86 ended up, king of the world, and next year they will be king of the universe.

Calling time of death on a segment of the industry they have yet to really set foot in (discreet GPU's) is definitely not confidence building IMO.

This kind of stuff, these pontifications and prognostications from executive management, it's all talk and no action.

Show me results from the first battle and then you can talk about all you want about how it is a foregone conclusion that you will win the war.

(by "you" I mean Intel, not "you" as in you, Nemesis you...)
 
Are ya beginning to see why I keep saying. When giving advice about what platiform to buy. You should be informed. Nehalem and Larrabee are a team apart not so tough . together they are NEMESIS! LOL . SO If your buying a complete settup anything but X58 and I7 is foolish. Once Larrabee is booted set back and enjoy the game graphics and interface. It should blow ya away . Oh. Since Tru 3D is part of intels package I guese this is what Larrabbee has to offer along with Nehalem . This stuff has been in the works for years.

Heres what Larrabee brings.

Gcpu $ cpu work as =

raytracing rays can also be used in Collisiom detection its almost free. Than you have Havok which will use that collision detection .Than Ya have TRu 3D . ect ect ect.

Than ya got Pat here practicly screaming at ya saying. YA larrabee isn't working with anthing but I7 and X58 on the high end . Leaving NV ati to the middle to low end.

Now some make think Pat is cocking off here. I don't I think Pat is smarter than that . Pat is FLAT out Crowing here . He is acting and talking like he is the Rooster in the Yard .

He has me convinced.

Ya got this part right?
Gelsinger also demonstrated two extremely graphics-intensive titles slated for release later this year: FarCry 2 and Quake 4 RT. The Larrabee project might be regarded as an important leap towards delivering photorealistic games that will be available to the masses

What is Quake 4 RT. To be released???
 
Originally posted by: Idontcare
Originally posted by: Nemesis 1
I like this one the best. There are just to many high profile people saying the same thing.

Thats alot of egg on face with all the hype.

http://news.softpedia.com/news...-to-an-End-82315.shtml

I'm not so convinced of Intel's prognostication capabilities when it comes to hearkening the time of death of anything in this industry.

They can't even get the time of death nailed down for their x86 business, as EPIC was supposed to supplant x86 and not become a parallel CPU architecture like an in-grown hair on the rump of a dodo bird. x86 wasn't supposed to live long enough to see 64bit, let alone be the fundamental basis of an entire new GPU product generation. And now look where x86 ended up, king of the world, and next year they will be king of the universe.

Calling time of death on a segment of the industry they have yet to really set foot in (discreet GPU's) is definitely not confidence building IMO.

This kind of stuff, these pontifications and prognostications from executive management, it's all talk and no action.

Show me results from the first battle and then you can talk about all you want about how it is a foregone conclusion that you will win the war.

(by "you" I mean Intel, not "you" as in you, Nemesis you...)

Ya can thank AMD for the prolonged X86. Intel wanted Epic . But MS and AMD conspired against intel and co developed AMD 64.

Its fitting tho how it all turned out . Intel is going to take that very same X86 and choke the life out of those who conspired against them . Look at larrabee it doesn't need MS, at all yet intel can run everthing they can through at them . Gaming has been a big part of MS success. Suddenly Apple looks very good for gamers . Does it not. Also I bet those I macs with Nehalem can all run larrabee.

Intel is on a mission and its gonna get ugly.

 
Originally posted by: Nemesis 1
Now some make think Pat is cocking off here. I don't I think Pat is smarter than that .

Settle. Setttllllle. There'll be less talk of Pat cocking off versus being too dumb to know how to...this is a family site :laugh:

Originally posted by: Nemesis 1
What is Quake 4 RT. To be released???

http://www.q4rt.de/

According to wiki Quake 4 RT was released at the Spring 2007 IDF. Was it really released then or was that merely the public demo?

Have you seen this one: http://isdlibrary.intel-dispat...ars_NOcovers_final.pdf

About the Author
Daniel Pohl started researching real-time ray tracing for games in 2004 during his study of computer science at the
Erlangen-Nuernberg University in Germany. As his master?s thesis, he developed a ray-traced version of Quake 4.
In 2007, he joined Intel?s ray-tracing group. In 2008, he moved from Germany to sunny California where he
continues to research game-related ray tracing.
 

Very interesting...


Drawing on the Pentium heritage The report also claims that one of Larrabee's purported strengths--its x86 heritage that taps into a large existing software infrastructure--is a weakness, too, because it is based on the Pentium, a design Intel launched in 1993. "Larrabee proposes to compete by fielding a couple of dozen x86 cores on a single chip. Each core, by dint of its (Pentium) heritage...is carrying all of the baggage of the full x86 instruction set," Kumar writes, adding that the Pentium design is "antiquated."

This is from the link. I remember that Intel deviced a little trick with the pentium 4. Instead of storing x86 instructions in the trace cache they store the decoded instructions called micro ops in the trace cache. Maybe Larrabee has a huge front end decoding al instructions into easy uops and then issue the uops to the different cores ? That would make executing a lot easier. Then we have lot's of in order cores who are fed decoded instructions all the time, when a pipeline flush would happen , the pipeline is fed again with decoded instructions from the trace cache . I wonder if that would be possible.

trace cache P4

This statement from the blog is interesting when thinking of the PS4.

The first Larrabee product will be "targeted at the personal computer market," according to Intel.

Not just an ordinary Pentium core in the larrabee.

The individual cores in Larrabee are derived from the Intel Pentium processor and "then we added 64-bit instructions and multi-threading," Seiler said. Each core has 256 kilobytes of level-2 cache allowing the size of the cache to scale with the total number of cores, according to Seiler.

Development tools anyone :

He explained that Intel is also helping DreamWorks to redesign its animation tools. "Our animation tools are all proprietary here. Intel is rearchitecting our software tools...to take advantage of multicore and make our renderer highly scalable as well as making our character animation tools highly scalable."

Intel is not going to make the same mistake Sony and IBM did with the PS3. The software support will be there, the hardware will be there. It seems like a well thought out strategy this time. AMD lost the contract while designing similair products. The money is in the movie industry, It is prophecised that sooner or later a large part of future movies are going to be completely CGI. And we know that the games industry always keeps an eye out to the movie industry.
 
Larrabee is not at all a replacement for Cell, it is a replacement for RSX, or whatever nVidia/ATI GPU Sony would otherwise put in the PS4. The comparison isn't between Cell and Larrabee; by functionality there simply is no comparison there. The parts do not perform the same tasks. The rumor is that the PS4 would be Cell+Larrabee.

That is exactly my point. Larrabee would be trying to do what Cell already does, only Cell is better suited for it. It would be akin to have a 295GTX in your system then throwing in a 9800GT for when you want to game, it just doesn't make a lot of sense. Sony went with nV because they realized that there goals weren't going to work, that rasterizers were still significantly stronger then what they could offer using alternate rendering techniques. This hasn't changed.

come on, a 64 core cell @ 5ghz in the PS4? You don't think they learned something from 599 US dollars?

Cell is cheap to produce. I would wager a 64 core Cell @5GHZ would be less expensive then Larrabee when looking at options for the PS4.

Doesn't sony do movies . Even 3D movies. There are alot of reasons for sony to play nice with intel or AMD. Not so much for NV tho. I think Sony feels the same about NV as Microsoft did with Xbox. So it will be AMD ATI in all the consols. Because I just don't see NV getting Sony back.

Why would Sony care about Intel or AMD for 3D movies?

Would you agree that the XBOX was quite effective in disrupting Sony's revenue stream and presumed dominance in the console market?

Not at all, in any way whatsoever really. For time on market Dreamcast was a bigger impact in the global market.

Surely this momentum and confidence carried forward into assisting them with their XBOX 360 contracts, roadmaps/timelines, as well going some distance to destabilizing Sony's presumed path to dominance and had a little something to do with their profit situation.

And then Nintendo comes along and does everything in the opposite direction of MS and dominates the market. When looked at from a business perspective that shouldn't be underestimated. MS has had a year longer on the market and is cheaper then the Wii, and the Wii has almost doubled its installed base. If you look at the XB vs the PS2 the PS2 had a year lead and a comparable price point and outsold the XBox 4:1, this time the 360 has a year lead and costs half as much and has managed a 40% edge. I understand what you are saying, the XBox was meant to get their foot in the door, I just think that they are realizing they went in the wrong door. While Nintendo has Olympic sized swimming pools of cash they have already made this generation, MS is still writing off costs. MS is a business first and foremost, and Nintendo is currently showing that the 'non core' market is at least as large if not larger then the core one.

The logic tree you lay out is quite logical, and had Sony reported profits instead of losses this past quarter and had the management team that brought PS3 to the market not started talking about old vs. new Sony then I would be inclined to subscribe to the logic you laid out.

For the PS3 in particular Sony used it to win the High Def format war, and it worked. Their market position in relation to Nintendo is something that certainly will have an impact going forward, much as it will for MS, that brings us back around to Intel having little to gain by landing in the PS4. How many ports are there from the Wii to the PC?

I don't understand why people think there is a lot of tension between Sony and nVidia atm, where does this come from? Sony is licensing IP from nVidia, nothing more. There isn't even the potential for issues like there with MS trying to back out of its contracts for the XBox, Sony pays a royalty fee per chip and fabs it themselves. I'm assuming people think this is the case as Charlie said so, the same guy that thinks Sony will go x86 for the PS4(that would be a very cold day in hell).
 
Originally posted by: BenSkywalker
Would you agree that the XBOX was quite effective in disrupting Sony's revenue stream and presumed dominance in the console market?

Not at all, in any way whatsoever really. For time on market Dreamcast was a bigger impact in the global market.

Surely this momentum and confidence carried forward into assisting them with their XBOX 360 contracts, roadmaps/timelines, as well going some distance to destabilizing Sony's presumed path to dominance and had a little something to do with their profit situation.

And then Nintendo comes along and does everything in the opposite direction of MS and dominates the market. When looked at from a business perspective that shouldn't be underestimated. MS has had a year longer on the market and is cheaper then the Wii, and the Wii has almost doubled its installed base. If you look at the XB vs the PS2 the PS2 had a year lead and a comparable price point and outsold the XBox 4:1, this time the 360 has a year lead and costs half as much and has managed a 40% edge. I understand what you are saying, the XBox was meant to get their foot in the door, I just think that they are realizing they went in the wrong door. While Nintendo has Olympic sized swimming pools of cash they have already made this generation, MS is still writing off costs. MS is a business first and foremost, and Nintendo is currently showing that the 'non core' market is at least as large if not larger then the core one.

You appear to be making the argument that just because someone did better than MS that MS's efforts were a failure. Dreamcast can be the biggest impact on the market but that fact doesn't mean it was the only console having an impact on the market.

I'm not saying XBOX was THE best approach to destabilizing Sony's console aspirations, I am merely saying XBOX was effective (albeit less effective) and labeling it an absolute failure seems a tad extreme and unnecessary.

That Nintendo did better than MS certainly suggests there was a superior strategy to MS's XBOX plan, but that does not mean MS's XBOX was a failure. Being second does not mean you are last, this isn't Talladega Nights.

At any rate the quote of relevance here is:

Originally posted by: BenSkywalker
The XBox, no matter your personal views on how it was as a gaming platform, was an absolute failure in a business sense due to MS buying parts off of other people.

I feel I provided a viable example of how XBOX provided value to MS's business plans for breaking into the console market, which I don't think you have refuted.

XBOX was a loss-leader, a necessary painful first-step for any business attempting to break-into an established market space such as the console gaming market.

IMO no different than what Intel must do with Larrabee, no different than what they have been doing with Itanium.

That makes Intel a wild-card darkhorse in the graphics world, console markets included. Combined with the wounded animal instincts that Sony is currently feeling and I say all bets are off and any traditional linear thinking regarding Sony's presumed evolutionary path for PS3 -> PS4 ought to be challenged.
 
I found this information from a post at realworld technologies.


Cell's raw massive FLOPage was what made that possible. You can do a lot with 155.5 GFLOPS on a single chip (single precision Linpack benchmark) ? oops, that's the first release. The one in RoadRunner is the new and improved PowerXCell, which peaks at 204.8 GFLOPS single precision, 102.4 GFLOPS double, and you double that for a QS22 blade, each of which has two PowerXCell chips. (Peak quoted because I couldn't find per-chip or -blade Linpack numbers. Sorry.) (Decoder ring: GFLOPS = GigaFLOPS is a mere thousand million per second.) Will IBM continue development of Cell without Playstation volumes? HPC volumes are a nice frosting on the console-volume cake for IBM, but hardly a full meal for a fab. Further development will be required to stay competitive, since a widely-leaked Intel Larrabee presentation, given in universities, indicates (with a little arithmetic) that an initial 16-processor Larrabee is supposed to reach 640 GFLOPS single precision peak, half that for double. That's supposed to see the light of day in late '09 or early 2010, and maybe 24-way, with 32-way and 48-way versions following, providing proportionally larger peak numbers (that's 1,920 GFLOPS for 48-way); and possibly even more with a process shrink. Those are really big numbers. And Cray is planning on using it, so there will be big RoadRunner-ish non-Cell future systems, maybe with a PS4-derived development system. On the other hand, the newer PowerXCell isn't used in the PS3, and development always knew it wouldn't be a high-volume product. Will IBM continue to fund it, perhaps using some of their pile of profits from Services and Software? (In IBM, Hardware has lagged those two for many years in the money department.)


http://perilsofparallel.blogsp...ts-future-of-ibms.html




 
Originally posted by: William Gaatjes
I found this information from a post at realworld technologies.


Cell's raw massive FLOPage was what made that possible. You can do a lot with 155.5 GFLOPS on a single chip (single precision Linpack benchmark) ? oops, that's the first release. The one in RoadRunner is the new and improved PowerXCell, which peaks at 204.8 GFLOPS single precision, 102.4 GFLOPS double, and you double that for a QS22 blade, each of which has two PowerXCell chips. (Peak quoted because I couldn't find per-chip or -blade Linpack numbers. Sorry.) (Decoder ring: GFLOPS = GigaFLOPS is a mere thousand million per second.) Will IBM continue development of Cell without Playstation volumes? HPC volumes are a nice frosting on the console-volume cake for IBM, but hardly a full meal for a fab. Further development will be required to stay competitive, since a widely-leaked Intel Larrabee presentation, given in universities, indicates (with a little arithmetic) that an initial 16-processor Larrabee is supposed to reach 640 GFLOPS single precision peak, half that for double. That's supposed to see the light of day in late '09 or early 2010, and maybe 24-way, with 32-way and 48-way versions following, providing proportionally larger peak numbers (that's 1,920 GFLOPS for 48-way); and possibly even more with a process shrink. Those are really big numbers. And Cray is planning on using it, so there will be big RoadRunner-ish non-Cell future systems, maybe with a PS4-derived development system. On the other hand, the newer PowerXCell isn't used in the PS3, and development always knew it wouldn't be a high-volume product. Will IBM continue to fund it, perhaps using some of their pile of profits from Services and Software? (In IBM, Hardware has lagged those two for many years in the money department.)


http://perilsofparallel.blogsp...ts-future-of-ibms.html

Those are some very eye-opening numbers.

It begs the question though - why hasn't anyone else produced chips capable of scaling to these numbers that Intel is projecting? What is the secret sauce for Intel that makes Intel suddenly capable of doing this while the existing folks (IBM/Sony, Nvidia, ATi) have slowly struggled to increase their computational processing power (relative to where Intel is expected to start at)?

Is it the process node (45nm) enabling them to just stack enough parallel cores on the chip? Or is it specifically to do with the chip's architecture, being x86, that has kept the non-x86 licensees from doing what Intel is doing?

I'm not saying I believe the numbers that Intel is putting out, nor am I saying I disagree with them.

What I am wondering is what makes Intel's approach here so special and revolutionary that it hasn't been done before and won't readily be duplicated in the immediate future as well?
 
Originally posted by: Idontcare
Those are some very eye-opening numbers.

It begs the question though - why hasn't anyone else produced chips capable of scaling to these numbers that Intel is projecting? What is the secret sauce for Intel that makes Intel suddenly capable of doing this while the existing folks (IBM/Sony, Nvidia, ATi) have slowly struggled to increase their computational processing power (relative to where Intel is expected to start at)?

Is it the process node (45nm) enabling them to just stack enough parallel cores on the chip? Or is it specifically to do with the chip's architecture, being x86, that has kept the non-x86 licensees from doing what Intel is doing?

I'm not saying I believe the numbers that Intel is putting out, nor am I saying I disagree with them.

What I am wondering is what makes Intel's approach here so special and revolutionary that it hasn't been done before and won't readily be duplicated in the immediate future as well?
I'm not sure how much stock you should be putting in quotes like these, and I'd certainly caution against becoming enamored when Intel throws around terms like "cores" when it comes to Larrabee. Keep in mind, each Larrabee "core" is about the equivalent of a Pentium processor and is comparable in performance to a single SP on a current-gen ATI/NV GPU at worst, and perhaps a SP TPC cluster at best. When you see the Nvidia and ATI current-gen have 240 and 800SP, respectively, you'll see the 20-40 numbers being thrown around by Intel aren't so impressive. Here's a link from AT that discussed similar numbers and scaling only a few months ago:

  • AT: How many cores on a Larrabee?

    Remember the design experiment? Intel was able to fit a 10-core Larrabee into the space of a Core 2 Duo die. Given the specs of the Core 2 Duo Intel used (4MB L2 cache), it appears to be a 65nm Conroe/Merom based Core 2 Duo - with a 143 mm^2 die size.

    At 143 mm^2, Intel could fit 10 Larrabee-like cores so let's double that. Now we're at 286mm^2 (still smaller than GT200 and about the size of AMD's RV770) and 20-cores. Double that once more and we've got 40-cores and have a 572mm^2 die, virtually the same size as NVIDIA's GT200 but on a 65nm process.

    The move to 45nm could scale as well as 50%, but chances are we'll see something closer to 60 - 70% of the die size simply by moving to 45nm (which is the node that Larrabee will be built on). Our 40-core Larrabee is now at ~370mm^2 on 45nm. If Intel wanted to push for a NVIDIA-like die size we could easily see a 64-core Larrabee at launch for the high end, with 24 or 32-core versions aiming at the mainstream. Update: One thing we did not consider here is power limitations. So while Intel may be able to produce a 64-core Larrabee with a GT200-like die-size, such a chip may exceed physical power limitations. It's far more likely that we'll see something in the 16 - 32 core range at 45nm due to power constraints rather than die size constraints.

    This is all purely speculation but it's a discussion that was worth having publicly.
 
If would be easier now because Intel is already working with the 32nm process for next Nehalem versions, but it seems that it will happen in the next update for Larrabee, but it's all pure speculation.
 
Dreamcast can be the biggest impact on the market but that fact doesn't mean it was the only console having an impact on the market.

Dreamcast had pretty much no impact, that was the point I was making.

I'm not saying XBOX was THE best approach to destabilizing Sony's console aspirations, I am merely saying XBOX was effective (albeit less effective) and labeling it an absolute failure seems a tad extreme and unnecessary.

What did it destabilize? Sony didn't change any of their plans at all due to the XBox, and the PS2 ended up more popular then the PSX by a considerable margin. The original XBox didn't impact Sony in any meaningful way.

I feel I provided a viable example of how XBOX provided value to MS's business plans for breaking into the console market, which I don't think you have refuted.

I completely understand what MS was trying to do, what I have not seen demonstrated is that it has been viable in the least. They took a loss leader approach for two generations in a row both times being soundly beaten by inferior technology and value. If their goal was to disrupt Sony then they failed before they took any action at all, not only because they had no real chance at disruption, but also due to the fact that playing into another companies' strategy is simply setting yourself up for failure. Yes, MS has much deeper pockets then Sony, but why waste billions of dollars fighting over the smaller slice of the pie for two generations in a row? If MS had learned something with the original XBox you could make reasonable claims that it was worth it, it appears they did not however.

IMO no different than what Intel must do with Larrabee, no different than what they have been doing with Itanium.

I have to agree with you there, Larrabee is looking very much like Itanium.

That makes Intel a wild-card darkhorse in the graphics world, console markets included. Combined with the wounded animal instincts that Sony is currently feeling and I say all bets are off and any traditional linear thinking regarding Sony's presumed evolutionary path for PS3 -> PS4 ought to be challenged.

I don't see how Sony's financials are going to make them lose any sense they still have, that is the issue I'm seeing. Larrabee is a flat out bad product for the market, just like Itanium. The big difference between Larrabee and Itanium is that Intel dominated the CPU market and still Itanium was laughed out of the mass market segment before it ever got started, they don't have a viable presence in perfromance graphics and haven't since the i740 first hit.

It begs the question though - why hasn't anyone else produced chips capable of scaling to these numbers that Intel is projecting? What is the secret sauce for Intel that makes Intel suddenly capable of doing this while the existing folks (IBM/Sony, Nvidia, ATi) have slowly struggled to increase their computational processing power (relative to where Intel is expected to start at)?

The speculative numbers don't really look that good at all, not sure what you mean exactly. By the end of '09 or early '10 Intel is supposed to have a chip that offers 64% of the raw computational power of something nVidia and ATi offered in Q3 of '08. That is somehow good? What's far worse, Ray Tracing will require significantly more raw floating point computational power then either of those offerings, a lot of fixed function interger ops on the rasterizer are going to need to be handled by FPU on Larrabee(hello ray tracing), ATi and nV are packing TFLOPS power just for their shader hardware and doing it at least a year before Intel is supposed to have this 640GFLOPS ray trace hardware out.

What I am wondering is what makes Intel's approach here so special and revolutionary that it hasn't been done before and won't readily be duplicated in the immediate future as well?

It isn't revolutionary at all, it is them making their own version of Cell. The BCU-100 is a single Cell(chip) setup pushing 230GFLOPS and consuming under 350Watts of power for the entire system, that was launched last year(still has a RSX to handle graphics). Not only is Larrabee not revolutionary, it isn't even evolutionary from what we have seen, it is a fairly blatant knock off of what Cell has been for years. Will Larrabee end up being ahead of Cell when it comes out? Not sure honestly, I imagine that Intel would be pushing hard to make sure it is, but then there is the issue with how superior using a MIPS style processor is in general compared to x87 FPU for this type of computing anyway. Also we have power issues, again- Cell is the most powerful general purpose chip per watt by a considerable margin atm, and Intel's main limitation with Larrabee is going to be power.

Intel isn't being creative, forward thinking or anything remotely close to that with Larrabee. They are copying an idea Sony/IBM brought to the mass market years ago only Sony had enough sense to realize even before it hit it couldn't handle its' initial goals(which seems to be where Intel is having a bit of trouble). They need to up their projected computational power by more then an order of manitude to hope to keep up with a rasterized, today's rasterizers, and that is only if we ignore basic functionality like AA(raytraced AA increases computationaly demands perfectly linear with samples utilized, rasterizers are a very small fraction of that).
 
Originally posted by: Cookie Monster
I think people are failing to see the bigger picture here. Intel has a big mountain to climb, and yet some people make it out like its the end of everything within the graphics market because its intel. It is going to be interesting fosho, but i dont have a high expectation of intel's first attempt at what they call a video card.
Technically, their first vid card was the i740 😉
 
An interview with Tim Sweeney from Epic .


Tim Sweeney interview

Some excerpts from the interview with Tim Sweeney :


TS: You could write a software rasterizer that uses the traditional SGI rendering approach; you could write a software renderer that generates a scene using a tiled rendering approach. Take for instance Unreal Engine 1, which was one of the industry's last really great software renderers. Back in 1996, it was doing real-time colored lighting, volumetric fog, and filtered texture mapping, all in real-time on a 90MHz Pentium. The prospect now is that we can do that quality of rendering on a multi-teraflop computing device, and whether that device calls a CPU or GPU its ancestor is really quite irrelevant. Once you look at rendering that way, you're just writing code to generate pixels. So you could use any possible rendering scheme; you could render triangles, you could use the REYES micropolygon tesselation scheme and render sub-pixel triangles with flat shading but really high-quality anti-aliasing ? a lot of off-line movie renderers do that?or you could represent your scene as voxels and raycast through it to generate data. Remember all the old voxel-based games in the software rendering era?

TS: And what else could you do? You could do a ray tracing-based renderer. You could do a volumetric primitive-based renderer; one idea is to divide your scene into a bunch of little tiny spherical primitives and then just render the spheres with anti-aliasing. So you can get really efficiently rendered scenes like forests and vegetation. There are really countless possibilities. I think you'll see an explosion of new game types and looks and feels once rendering is freed from the old SGI rendering model. Remember, the model we're using now with DirectX was defined 25 years ago by SGI with the first OpenGL API. It's a very, very restrictive model that assumes you're going to generate all the pixels in your scene by submitting a bunch of triangles in a fixed order to be blended into some frame buffer using some blending equation, and the fact that we have these ultra-complex programmable pixel shaders running on each pixel?that part of the pipeline has been made extensible, but it's still the same back-end rendering approach underneath.


JS: So to follow up with that, I hear that Larrabee will be more general-purpose than whatever NVIDIA has out at the time, because NVIDIA is still gonna have some hardware blocks that support whatever parts of the standard rasterization pipeline. TS: That's kind of irrelevant, right? If you have a completely programmable GPU core, the fact that you have some fixed-function stuff off to the side doesn't hurt you. Even if you're not utilizing it at all in a 100 percent software-based renderer, there are economic arguments that say it might be worthwhile to have that hardware even if it goes unused during a lot of the game, for instance, if it consumes far less power when you're running old DirectX applications, or if it can perform better for legacy usage cases. Because, one important thing in moving to future hardware models is that they can't afford to suddenly lose all the current benchmarks. So DirectX remains relevant even after the majority of games shipping are using 100 percent software-based rendering techniques, just because those benchmarks can't be ignored. So I think you'll see some degree of fixed-function hardware in everybody's architectures for the foreseeable future, and it doesn't matter.

TS: From my point of view, the ideal software layer is just to have a vectorizing C++ compiler for every architecture?NVIDIA, Intel, AMD, whoever. Let us write code in C++ to run on the GPU, including shaders in C++, and rendering algorithms in C++, where programmers explicitly create threads, hand work off to them, synchronize shared data, and so on. And then use what NVIDIA calls "pixel pipelines" and Intel calls "vector registers" by means of a vectorizing compiler that takes loops, unrolls them, and runs it on the wide vector units.

JS: What do you think about the next generation of consoles. It sounds like, instead of the standard CPU plus GPU configuration, we may just have many-core CPUs... or, sorry, not many-core general-purpose CPUs, because that would be silly, but something that's like... nevermind, I can't think of how to put this question. TS: No, I see exactly where you're heading. In the next console generation you could have consoles consist of a single non-commodity chip. It could be a general processor, whether it evolved from a past CPU architecture or GPU architecture, and it could potentially run everything?the graphics, the AI, sound, and all these systems in an entirely homogeneous manner. That's a very interesting prospect, because it could dramatically simplify the toolset and the processes for creating software.



We have more and more people in the graphics business agreeing that the software renderer comes back. Only this time the software renderer runs on highly parallel hardware like larrabee and more or less on the current generation of gfx chips from nvidia and ati. But to be honest, Nvidia has already seen that this would happen and that is where cuda comes from. And AMD has ATI stream that does essentially the same. The next generation gfx chips from amd and nvidia might as well be fully programmable like larrabee.

And i do not think larrabee is just lot's of atom's together. If they are "just" atom's they sure had an anabolic steriod diet.

 
1 more interesting quote:

JS: Now I have to ask the obligatory performance question, but I don't want to ask, "which is going to be faster, Intel or NVIDIA," because it sounds like it's going to depend heavily on the software. In other words, [the performance question] is a different question than it is when you have a fixed-function GPU and everybody's doing things the same way, vs. asking the performance question when things are fully programmable. Does this make sense? TS: Do you mean, what are the factors that determine performance of these new processors? JS: Yeah. TS: These processors don't entirely exist yet, so it's a challenge to say even what aspects of performance are relevant. So you have a certain number of cores running at a certain clock rate, and each core has a vector instruction set of some sort with a certain width?Intel has said that their width is 16, and NVIDIA's public presentations indicate their GPUs run with 16- to 32-wide pixel pipelines or vectors. So there are all those parameters, and for code that's perfectly parallel, you can have performance potentially be determined by number of cores times clockrate times vector width. That's powerful scaling; that gets you up into the Teraflop range with currently-manufacturable chips. And then on top of that, you have a lot of ancillary issues that could be significant but are hard to analyze without having a complete engine written for a next-gen processor like that. One is the tradeoff between the cache and memory systems. How much memory bandwidth do you have? Because DRAM buses are going to be fundamentally constrained, so you might be able to get 100 or 200 GBs of bandwidth, but will that be enough to power several Teraflops of computing power? Typically you've wanted about 1 byte for every FLOP; the further away you get from that, the more likely you are to be bottlenecked by memory. So clearly the way to achieve that is with some degree of caches on these chips. Caches can provide far more bandwidth and lower latency for memory accesses. So how much cache do you have, how is it organized, what's the latency, what's the cache architecture?those are the key dimensions of performance.

Intel is it seems the prefetching king, hiding bottlenecks. And have a lot of experience with large low latency caches. I think when it really comes to this point Intel will start to flex their muscles on this territory. If i am correct, from a programmer's point of view they will also start to use (and don't they already start doing that with the latest SSE iteration) three operand instructions creating denser and thus faster executing code.
( When you can spend more cycles on arithmetic instructions then on swapping data from 1 register to another register you can get more work done).

 
Originally posted by: Wreckage
http://techreport.com/discussions.x/14538

Huang believes Intel's promise of ten-times-greater integrated graphics performance by 2010 will yield hardware barely at the level of current mainstream Nvidia GPUs


I agree on this one...

Nvidia's vision is that consumers will need relatively powerful GPUs as raw processing devices in the future, too. Later in the conference, software firm Elemental Technologies gave a demonstration of its H.264 video transcoding software?the kind you need to convert videos for an iPod?running about 19 times faster on an Nvidia GPU than on a quad-core processor. Huang thinks video transcoding in such instances ought to be "instantaneous" in the future.

And i think Microsoft is as usuall a sleep again. Instead of 3d rendering a word document ( blah aero) , they should start to make a cuda/ati stream / larrabee api for programmers and promoting it more. Microsoft is in a position to create a solid standard making sure that when the gpu-cpu combo starts to become mainstream for the customers it is also mainstream under windows to use the gpu as just another cpu. This is very important because so much calculation power we have and microsoft is just watching how we have 3 different api's in the future. Then microsoft will come with their version which runs on top and will take a performance hit.

And for as far as i know, Huang is still looking for the can opener...


 
Originally posted by: BenSkywalker
Larrabee is not at all a replacement for Cell, it is a replacement for RSX, or whatever nVidia/ATI GPU Sony would otherwise put in the PS4. The comparison isn't between Cell and Larrabee; by functionality there simply is no comparison there. The parts do not perform the same tasks. The rumor is that the PS4 would be Cell+Larrabee.

That is exactly my point. Larrabee would be trying to do what Cell already does, only Cell is better suited for it. It would be akin to have a 295GTX in your system then throwing in a 9800GT for when you want to game, it just doesn't make a lot of sense. Sony went with nV because they realized that there goals weren't going to work, that rasterizers were still significantly stronger then what they could offer using alternate rendering techniques. This hasn't changed.

You seem to have been colonized by a marketing infection. Larrabee won't usher in a raytracing revolution. To succeed, it will have to be capable of running contemporary software, and Intel knows this.

If Cell was a better GPU than what Intel would want to put in the PS4, it would be better than RSX, and Sony would have put two Cells in the PS3.

come on, a 64 core cell @ 5ghz in the PS4? You don't think they learned something from 599 US dollars?

Cell is cheap to produce. I would wager a 64 core Cell @5GHZ would be less expensive then Larrabee when looking at options for the PS4.

Where's your source for that? How is Sony still losing money on the PS3 with a 7-core 3.2Ghz Cell? You sound like Ken Kutaragi. How's life in 4-D?

The entire point of Larrabee in the PS4 is that Intel would give it away for free. If they didn't do that, Sony would have no reason to consider it. The only upside of Larrabee is that Intel would be willing to throw money at it to get a foot in the door, a la Microsoft.
 
Originally posted by: mmnno
Originally posted by: BenSkywalker
Larrabee is not at all a replacement for Cell, it is a replacement for RSX, or whatever nVidia/ATI GPU Sony would otherwise put in the PS4. The comparison isn't between Cell and Larrabee; by functionality there simply is no comparison there. The parts do not perform the same tasks. The rumor is that the PS4 would be Cell+Larrabee.

That is exactly my point. Larrabee would be trying to do what Cell already does, only Cell is better suited for it. It would be akin to have a 295GTX in your system then throwing in a 9800GT for when you want to game, it just doesn't make a lot of sense. Sony went with nV because they realized that there goals weren't going to work, that rasterizers were still significantly stronger then what they could offer using alternate rendering techniques. This hasn't changed.

You seem to have been colonized by a marketing infection. Larrabee won't usher in a raytracing revolution. To succeed, it will have to be capable of running contemporary software, and Intel knows this.


Excerpt from the link you provided :

Tom Forsyth is currently a software engineer working for Intel on Larrabee. He previously worked at Rad Game Tools on Pixomatic (a software rasterizer) and Granny3D, as well as Microprose, 3Dlabs, and most notably Muckyfoot Productions (RIP). He is well respected throughout the industry for the high quality insight on graphics programming techniques he posts on his blog. Last Friday, though, his post's subject was quite different:

"I've been trying to keep quiet, but I need to get one thing very clear. Larrabee is going to render DirectX and OpenGL games through rasterisation, not through raytracing.


that fit's with what Tim Sweeney thinks...

JS: So to follow up with that, I hear that Larrabee will be more general-purpose than whatever NVIDIA has out at the time, because NVIDIA is still gonna have some hardware blocks that support whatever parts of the standard rasterization pipeline. TS: That's kind of irrelevant, right? If you have a completely programmable GPU core, the fact that you have some fixed-function stuff off to the side doesn't hurt you. Even if you're not utilizing it at all in a 100 percent software-based renderer, there are economic arguments that say it might be worthwhile to have that hardware even if it goes unused during a lot of the game, for instance, if it consumes far less power when you're running old DirectX applications, or if it can perform better for legacy usage cases. Because, one important thing in moving to future hardware models is that they can't afford to suddenly lose all the current benchmarks. So DirectX remains relevant even after the majority of games shipping are using 100 percent software-based rendering techniques, just because those benchmarks can't be ignored. So I think you'll see some degree of fixed-function hardware in everybody's architectures for the foreseeable future, and it doesn't matter.





 
Originally posted by: nosfe
Originally posted by: taltamir
they have WORKING DX10 drivers for their IGP...
And the larb will include fixed function parts, it is not one hundred percent x86 emulation...
i've used intel igp's, the drivers are "working" but they're still crap and i've yet to meet someone who likes them

i totally agree, but they don't have to write a new driver from scratch, they have working drivers that need improving, which is certainly within intel's power to do.
 
Originally posted by: BenSkywalker
The speculative numbers don't really look that good at all, not sure what you mean exactly. By the end of '09 or early '10 Intel is supposed to have a chip that offers 64% of the raw computational power of something nVidia and ATi offered in Q3 of '08. That is somehow good? What's far worse, Ray Tracing will require significantly more raw floating point computational power then either of those offerings, a lot of fixed function interger ops on the rasterizer are going to need to be handled by FPU on Larrabee(hello ray tracing), ATi and nV are packing TFLOPS power just for their shader hardware and doing it at least a year before Intel is supposed to have this 640GFLOPS ray trace hardware out.
Exactly, which is why all this sudden enthusiasm about Larrabee is a bit surprisng. Besides the lower raw computational power, I don't think they've even firmed up their final rasterizer solution. Last I checked they were still planning on a tile-based software renderer.

Personally I think Intel is going to make Lucid and Hydra an integral part of Larrabee's success or failure. At this point, I don't think a single discrete Larrabee GPU will be competitive with current-gen DX10 parts, much less the DX11 parts scheduled to launch around the time of Larrabee's debut. If Hydra is able to combine and scale multiple Larrabee well, then it has a chance, otherwise it'll just be a GPU about as powerful as a 9400M about a year after the 9400M was already released.
 
Back
Top