[eastmen]

Originally Posted by Shifty Geezer

What is DX11 and 12 hardware going to offer us? Is there going to be hardware acceleration of features, or will DX11 and 12 be focussed on more open, programmable hardware?

The biggest thing may be the tessellator

Now don't let this technical assessment of fixed function tessellation make you think we aren't interested in reaping the benefits of the tessellator. Currently, artists need to create different versions of their objects for different LODs (Level of Detail -- reducing or increasing complexity as the object moves further or nearer the viewer), and geometry simulation through texturing at each LOD needs to be done by pixel shaders. This requires extra work from both artists and programmers and costs a good bit in terms of performance. There are also some effects than can only be done with more geometry.

Tessellation is a great way to get that geometry in there for more detail, shadowing, and smooth edges. High geometry also allows really cool displacement mapping effects. Currently, much geometry is simulated through textures and techniques like bump mapping or parallax occlusion mapping or some other technique. Even with high geometry, we will want to have large normal maps for our lighting algorithms to use, but we won't need to do so much work to make things like cracks, bumps, ridges, and small detail geometry appear to be there when it isn't because we can just tessellate and displace in a single pass through the pipeline. This is fast, efficient, and can produce very detailed effects while freeing up pixel shader resources for other uses. With tessellation, artists can create one sub division surface that can have a dynamic LOD free of charge with a simple hull shader and a displacement map applied in the domain shader will save a lot of work, increase quality and improve performance quite a bit.

Sounds like the cost of art could actually go down from this generation to next generation if these features are taken advantage of. I know the 360 has one of tehse units in it , however the dx 11 unit seems to have been expanded and of course a 2011 varient would be much more powerfull than the 2005 verison in the 360.

I read this as allowing for more destructable enviroments also. If you shoot a rocket at a wall the tessellator should be able to deform the wall and remove parts of it that aren't there anymore.


vs


http://www.anandtech.com/video/showdoc.aspx?i=3507&p=8

[Nano]

Its a shame the industry hasn't thrown its weight behind tessellation already considering the architecture we've been seeing since Xbox 360 in 2005.

[eastmen]

Originally Posted by Nano

Its a shame the industry hasn't thrown its weight behind tessellation already considering the architecture we've been seeing since Xbox 360 in 2005.

I allways thought that the 360 had a very basic one like the r600. I would also bet that the lack of exclusive engines for the 360 si one of the reasons its gone unused. I think viva pinata is one of the titles that uses the feature and they use it for the grass if i'm nto mistaken

[scificube]

Originally Posted by pc999

Good tools are really needed, but they will vary a lot depending on the HW.

They can but this is a clear case of better to have than have not. Tools, libraries, examples, infrastructure (farms,mo-cap,network,servers,etc), standards, delayed and immediate support paths, etc all need to be addressed. The hardware is only one piece of the puzzle and there is great deal of technology that is just as critically important. This gen was just a sampling of many things that must mature and quickly or there will be a rather nasty price to pay.

[scificube]

Originally Posted by Nano

Its a shame the industry hasn't thrown its weight behind tessellation already considering the architecture we've been seeing since Xbox 360 in 2005.

It is a tall task when most tool chains aren't close to ready for it and the common skill set of artists and programmers alike does not map well to using it.

[zed]

this happens, every few years

... tesselation ... is gonna happen now ... its the next new big thing ...
but every single time it doesnt occur, ppl just are not interested,
you would of thought ppl would of learnt from history (oh wait maybe not..)

like those cults proclaiming the end of the world

[liolio]

Originally Posted by pc999

That would be a real minor update to the CPU, no? specially given how weak each core is, probably many todays dual cores (at least if costum) would beat it...
and would keep all of the problems.

Without some rework it's more than likely. I remember Capcom (early on) stating that they were able to achieve (with the xenon) the same level of performance as a dual core P4.
The P4 was crushed per cycle per Athlon who got crushed by Intel's core architecture the upcoming nehalem will raise the bar again( likely not by the same amount nor on every workload).
I'm just trying to enphasis your "how weak the cores are" but it's also to show that there is room for improvement. I also think that MS may not need (and most likely could reach) the level of performances Intel or AMD will be able to deliver by this time.
Ms needs something cheaper/tinier than what intel or AMD has to offer and with lower power requirements. Xenon is weak in so many in regard to top of the line x86 CPU and can be improved in so many ways that I think that MS/IBM have room to make substantial gains in perfs without breaking their power/die space budget.
How much? I think some members may come with proper estimates, by watching at AMD/Intel progress (witch came at a cost) I would say that MS/IBM should aim at an overall 30% improvement per cyle.
Overall that wouldn't be a great jump in power but basically really good serial performances as found in the PC realm is likely out of reach of what MS can afford.

Maybe MS on top of offloading a lot of calculations to the gpu could consider the use fixed function untis enslaved to the cpu cores to free CPU resources. I think especcially of units handling decompression, maybe network acceleration too.
I remember seeing average cpu utilisation for PGR3 and basically more than one core was dedicated to decompression speak about a waste of silicon.
Using ~30millions transistors running 3.2GHz to achieve what a dsp would handle faster while being way tinier... not too mention power efficience.
Shortly

[MfA]

Originally Posted by liolio

Using ~30millions transistors running 3.2GHz to achieve what a dsp would handle faster while being way tinier... not too mention power efficience.

Even with state of the art efficiency per transistor per cycle for their superscalar cores this won't stop being true. This is more an argument for spending more area on SPUs/Larrabee than anything else.

[Nano]

Originally Posted by scificube

It is a tall task when most tool chains aren't close to ready for it and the common skill set of artists and programmers alike does not map well to using it.

Originally Posted by eastmen

I allways thought that the 360 had a very basic one like the r600. I would also bet that the lack of exclusive engines for the 360 si one of the reasons its gone unused.

Yes. Unfortunately technologies and tools haven't been built around it. As 360 is a closed platform there was a lot of potential there from the early days, but we haven't seen many adaptations several years later. As mentioned, exclusive engines are lacking for the console. Even Alan Wake (also for PC) which will agressively use tessellation appears to be doing so on the CPU and I don't know if it would be practical development to support it in hardware. PS3 as well with the Cell processor has a lot of potential for tessellation. And of course DX10 GPUs with the geometry shader, but that architecture still isn't the baseline for development in PC titles ^.^

I think viva pinata is one of the titles that uses the feature and they use it for the grass if i'm nto mistaken

Yes. In fact I believe most of the terrain makes use of h/w tessellation. Impressive but one in a million. I hoped for Rare to come up with a bigger more ambitious game using that sort of rendering techique, like a new Perfect Dark or something.

[eastmen]

I'm c orrect in assume that every gpu from ati has this feature in them since the r600 correct ?

Perhaps it be in ati's best intrest to get a small team together and start going to all the diffrent companys and adding the tesselator support to them. It seems to me that the only thing that would change is that instead of using the lod model it would dynamicly create its own best case one on the fly.

That may help them alot in current games.


Hopefully next gen though with it being better supported in dx11 we get to see alot of it in th next round of consoles

[shiznit]

I don't think that MS necessarily has to go the Sony route next gen and make a console based on tech from the current gen. While Sony spent billions developing Cell and have to try to milk it for all they can, I think MS has more room to experiment as long as they can maintain backwards compatibility and release before PS4 at a decent price. I'm not sure how it is in Europe now, but here in North America the Xbox has replaced the Playstation as the hardcore gamer's go-to console, and I don't see that changing any time soon. Almost all my gamer friends in highschool and college had PS2s, now only two have a PS3 and almost all of them have 360s and some have more than one. When the next gen comes along, the hardcore crowd that Xbox now has in the pocket will expect great games to keep coming and they will not care or even know if MS went a different route with the hardware and publishers/developers would just have to live with it. That said, as much as I would like to see a Larrabee Xbox it's probably highly unlikely.

[3dcgi]

Originally Posted by eastmen

I allways thought that the 360 had a very basic one like the r600.

Don't know what's very basic about it. DX11's implementation is just some evolutionary changes and a new shader stage.

Originally Posted by eastmen

I'm c orrect in assume that every gpu from ati has this feature in them since the r600 correct ?

Correct.

[pc999]

Originally Posted by scificube

They can but this is a clear case of better to have than have not. Tools, libraries, examples, infrastructure (farms,mo-cap,network,servers,etc), standards, delayed and immediate support paths, etc all need to be addressed. The hardware is only one piece of the puzzle and there is great deal of technology that is just as critically important. This gen was just a sampling of many things that must mature and quickly or there will be a rather nasty price to pay.

True, but it is the initial thing, they can only build around the HW, anyway, like pointed by the tesselator, there is the question if there is a real point in having to much specialized HW.

Originally Posted by liolio

Without some rework it's more than likely. I remember Capcom (early on) stating that they were able to achieve (with the xenon) the same level of performance as a dual core P4.
The P4 was crushed per cycle per Athlon who got crushed by Intel's core architecture the upcoming nehalem will raise the bar again( likely not by the same amount nor on every workload).
I'm just trying to enphasis your "how weak the cores are" but it's also to show that there is room for improvement. I also think that MS may not need (and most likely could reach) the level of performances Intel or AMD will be able to deliver by this time.
Ms needs something cheaper/tinier than what intel or AMD has to offer and with lower power requirements. Xenon is weak in so many in regard to top of the line x86 CPU and can be improved in so many ways that I think that MS/IBM have room to make substantial gains in perfs without breaking their power/die space budget.
How much? I think some members may come with proper estimates, by watching at AMD/Intel progress (witch came at a cost) I would say that MS/IBM should aim at an overall 30% improvement per cyle.
Overall that wouldn't be a great jump in power but basically really good serial performances as found in the PC realm is likely out of reach of what MS can afford.

Why spend more time and money with a "flawed" CPU , in 2010 any dual core soluction (or a cut down one) would beat it in every single aspect, many of them offering OOE, branch prediction, good latency, power effeciency... Only for very easy BC.

Anyway dont take to hard on Xenom in some things it is probably way better tahn PC CPUs (raw fp or dot products)

Maybe MS on top of offloading a lot of calculations to the gpu could consider the use fixed function untis enslaved to the cpu cores to free CPU resources. I think especcially of units handling decompression, maybe network acceleration too.
I remember seeing average cpu utilisation for PGR3 and basically more than one core was dedicated to decompression speak about a waste of silicon.
Using ~30millions transistors running 3.2GHz to achieve what a dsp would handle faster while being way tinier... not too mention power efficience.
Shortly

That is the beauty of Fusion like designs, althought there may be a problem of such units going undersude (likethe tesselator), depending of how much it brings and how easy it is to use.

Anyway it seems to me that a FUsion design is perfect for a console...

[Sound_Card]

I was reading some articles today and thought for a minute. It seems perfectly logical if indeed Valhalla is in the works, to take the same chip, scale it upwards while shrinking the process and using that as the main CPU in the nextbox. What a clever and cost effective solution that does not sacrifice on anything while providing the needed "umph" next gen. For those that don't know what Valhalla is, it's Xenon and Xenos combined on a single 65nm die for the xbox 360 that indication points towards around a 2011/2012 release (eh what do you know? around the approximal time the next xbox is predicted to hit. What better time to offer a cost reduced... dare I say.... "360 mini?").

Taking Valhalla and scaling it upwards. Perhaps doubling L2, twice the threads, and increased clock speeds, along with minor tweaks and keeping the Xenos portion intact and unchanged. Developer kits can effectivly scale accordinly and the learning curve is just simply carried over from last genertion. Backwards compatibility would not be sacerficed either which to me is the golden egg in all this. though to pull something like this off, I of course imagine them sitting down with IBM and AMD very closely and to perhaps hit 45nm with it as well. Now fitting what we know about Valhalla, I was also hit in the head on this good old thread here at B3D underling this important news link on the very subject by newyork times. Please, just never mind the close proximity architecture thing with Sun and leave that out of this.

For more than two decades, Microsoft’s software and Intel’s processors were so wedded that the pairing came to be known as Wintel. But as that computing era wanes, Microsoft is turning to a new source of chip design: its own labs.
The design effort will initially be split between research labs at the company’s headquarters in Redmond, Wash., and its Silicon Valley campus here. Tentatively named the Computer Architecture Group, the project underscores sweeping changes in the industry.

They have the personal, budget, and the right partners to pull this off too. This to me is the project under Microsofts Architecture group.

[RudeCurve]

That sounds too logical.

[Carl B]

Originally Posted by Sound_Card

Taking Valhalla and scaling it upwards. Perhaps doubling L2, twice the threads, and increased clock speeds, along with minor tweaks and keeping the Xenos portion intact and unchanged. Developer kits can effectivly scale accordinly and the learning curve is just simply carried over from last genertion. Backwards compatibility would not be sacerficed either which to me is the golden egg in all this. though to pull something like this off, I of course imagine them sitting down with IBM and AMD very closely and to perhaps hit 45nm with it as well. Now fitting what we know about Valhalla, I was also hit in the head on this good old thread here at B3D underling this important news link on the very subject by newyork times. Please, just never mind the close proximity architecture thing with Sun and leave that out of this

Perhaps not, because the resultant chip might be overly large and the thermals, yields, and expense could thus be thrown off. The Valhalla project makes sense and is sensible because it takes small disparate chips size wise and combines them into a single larger package with decent thermal bounds and workable yields. If you were to take that resultant chip, however, and scale it up - granted to what extent would matter quite a bit - you would find yourself back at the overly large result that might rule out the combination of such chips outright from an economic (or potentially even engineering) point of view.

On 45nm in theory you could double the size of the chip logic and end up at the die size of just one of the original 360 chips... not bad at first glance perhaps, but there's several variables at play.

Beyond any of that also, I do expect there to be an architectural shift for the 720 to reflect the work taking place on MS' DX evolutions - it just makes sense for them to bring that to the console side as it benefits their overall ecosystem. So I would expect modernized, rather than just expanded, graphics hardware in that regard.

[sebbbi]

Originally Posted by eastmen

I allways thought that the 360 had a very basic one like the r600. I would also bet that the lack of exclusive engines for the 360 si one of the reasons its gone unused. I think viva pinata is one of the titles that uses the feature and they use it for the grass if i'm nto mistaken

The Xbox 360 tesselator might be simple, but it's very flexible. Combined with random address fully programmable vfetch (data input) and random address fully programmable memexport (data output) you can basically do almost any kind of tesselation, geometry manipulation and creation purely on the shaders, and even store the results for reuse later on.

Sadly the PC hardware seems to be lagging a bit behind on this department. Even DX11 does not seem to support memexport from the pixel shaders (for tricks like outputting several pixels simulaneously, reusing the blur kernel sampled data, etc). Compute shaders should help a lot (by providing proper random addressable memory input and output).

[Sound_Card]

Originally Posted by Carl B

Beyond any of that also, I do expect there to be an architectural shift for the 720 to reflect the work taking place on MS' DX evolutions - it just makes sense for them to bring that to the console side as it benefits their overall ecosystem. So I would expect modernized, rather than just expanded, graphics hardware in that regard.

Well, I'm not saying they should expand the Xenos portion. Just the Xenon portion. Expand, tweak, increase clocks etc on the CPU portion. They could add another GPU outside of the combined chip. A "modernized" 2011/2012 GPU.

As to what you said above, I think a 45nm Valhalla scaled upwards (to what degree?) could maintain thermals to a perfectly acceptable thresh hold.

[scificube]

Originally Posted by Sound_Card

Well, I'm not saying they should expand the Xenos portion. Just the Xenon portion. Expand, tweak, increase clocks etc on the CPU portion. They could add another GPU outside of the combined chip. A "modernized" 2011/2012 GPU.

As to what you said above, I think a 45nm Valhalla scaled upwards (to what degree?) could maintain thermals to a perfectly acceptable thresh hold.

There is little to no benefit in doing that. Xenos and RSX alike will be absolutely thrashed by GPU/VPU tech in 20112. If there is another GPU in the system then Xenos should be shown the door and thanked vociferously for its past service to the cause. There is no need for 2 GPUs to compete in the system especially when one of them would be severely over matched.

Carl B has already pointed out why the next CPU/GPU combo is very unlikely to appear on the same die. I would add if you scale up the clocks on Valhalla then you also need to scale up the cooling system for it in addition to fabrication concerns. Cooling is not cheap on the BOM.

The Xenon core and the PPE core of Cell need to be dropped into a very deep pit never to return. It is unlikely MS or Sony has much desire to see them return when better cores can be had quite easily from IBM smoothing BC concerns considerably.

[Sound_Card]

Originally Posted by scificube

There is little to no benefit in doing that. Xenos and RSX alike will be absolutely thrashed by GPU/VPU tech in 20112. If there is another GPU in the system then Xenos should be shown the door and thanked vociferously for its past service to the cause. There is no need for 2 GPUs to compete in the system especially when one of them would be severely over matched.

Carl B has already pointed out why the next CPU/GPU combo is very unlikely to appear on the same die. I would add if you scale up the clocks on Valhalla then you also need to scale up the cooling system for it in addition to fabrication concerns. Cooling is not cheap on the BOM.

The Xenon core and the PPE core of Cell need to be dropped into a very deep pit never to return. It is unlikely MS or Sony has much desire to see them return when better cores can be had quite easily from IBM smoothing BC concerns considerably.

So their was absolutely no nobility in adding the PS2's logic in the original PS3's? Or am I mistaken here in that was only the PS2's CPU? I can see the merits to the idea, but as well as the short comings. I think the fundamental issue here is that I'm still very much ignorant to the overall scheme of things.

Why would Xenos have to compete with the new graphics hardware? Would you not need Xenos to
properly support full BC? How easy do you think it can be mapped out on a new graphics arc.? Then again, I'm not keen on the idea of building next gen systems around old hardware, but merely thought of this is possibility. I do believe to a rather good extent that Valhalla is the project or one of the projects in MS's computer Architecture group. Not some iteration of Suns close proximity tech CPU.

Also, what kind of better cores are we talking about from IBM?

Thanks.

[pc999]

Originally Posted by Sound_Card

So their was absolutely no nobility in adding the PS2's logic in the original PS3's? Or am I mistaken here in that was only the PS2's CPU? I can see the merits to the idea, but as well as the short comings. I think the fundamental issue here is that I'm still very much ignorant to the overall scheme of things.

Why would Xenos have to compete with the new graphics hardware? Would you not need Xenos to
properly support full BC? How easy do you think it can be mapped out on a new graphics arc.? Then again, I'm not keen on the idea of building next gen systems around old hardware, but merely thought of this is possibility. I do believe to a rather good extent that Valhalla is the project or one of the projects in MS's computer Architecture group. Not some iteration of Suns close proximity tech CPU.

Also, what kind of better cores are we talking about from IBM?

Thanks.

I think that PS2 HW is just in the first models then BC is made only on software, even not being the slightest similar in any way.

Anyway if you have to diferent GPUs it will only bring problems on the programing side, add to the cost, create resource problems, add to the heat/power, making bigger...

Plus DX10/11 is almost fully compatible it Xenos (MemoExport, or going directly to CPU cache is not), so BC in the GPU side is a easy thing.

On the CPU is a bit harder but there is SW soluction for it (even at the 360 released there was one very popular because some seid it would bring BC to 360, dont recal the name...). Xenon could now die a agonizing death for all I care, if they go for a small upgrade, just costumize a already inexpensive X2 or a Pentium C2D meybe even a cut down from Power 7 (a dual or tri core, if cheap enought), BC would be harder but they probably would save money doing the tools for it.

PLus some problems will only go worst (eg, latency with faster RAM).

[Carl B]

Originally Posted by Sound_Card

So their was absolutely no nobility in adding the PS2's logic in the original PS3's? Or am I mistaken here in that was only the PS2's CPU? I can see the merits to the idea, but as well as the short comings. I think the fundamental issue here is that I'm still very much ignorant to the overall scheme of things.

Why would Xenos have to compete with the new graphics hardware? Would you not need Xenos to
properly support full BC? How easy do you think it can be mapped out on a new graphics arc.? Then again, I'm not keen on the idea of building next gen systems around old hardware, but merely thought of this is possibility. I do believe to a rather good extent that Valhalla is the project or one of the projects in MS's computer Architecture group. Not some iteration of Suns close proximity tech CPU.

Also, what kind of better cores are we talking about from IBM?

Thanks.

The new GPU, if it originated with ATI/AMD, would likely ensure full B/C simply simply by being a forward extension of the architecture. Throw in some specific instruction supports and it can be derived from whatever family (I do think DX11 though) they choose to branch off of when the time comes.

In terms of the CPU, well... we'll see what they do. If it's IBM again then a scaling of the XeCPU or a new branch off a more current architecture with some instructional supports again should leave them just fine on B/C.

For PS2 on PS3, the RSX from the Graphics Synthesizer was just too dramatic a break architecturally, so yes they needed that chip in there to make it work out. They put the EE in there at first as well until they could get emulation up to par on Cell, but they did so in fairly short order.

Thinking that XBox 720 might be a scaling of the 360 architecture is a fine theory; the Sun architectural stuff I don't buy into in the least anyway, so that one's not even a factor. But for myself, I do think that the hardware will be modernized rather than simply scaled, and if scaled, probably not from a single-chip solution (unless truly truly the next gen becomes one of incrementalism).

[scificube]

Originally Posted by Sound_Card

So their was absolutely no nobility in adding the PS2's logic in the original PS3's? Or am I mistaken here in that was only the PS2's CPU? I can see the merits to the idea, but as well as the short comings. I think the fundamental issue here is that I'm still very much ignorant to the overall scheme of things.

Why would Xenos have to compete with the new graphics hardware? Would you not need Xenos to
properly support full BC? How easy do you think it can be mapped out on a new graphics arc.? Then again, I'm not keen on the idea of building next gen systems around old hardware, but merely thought of this is possibility. I do believe to a rather good extent that Valhalla is the project or one of the projects in MS's computer Architecture group. Not some iteration of Suns close proximity tech CPU.

Also, what kind of better cores are we talking about from IBM?

Thanks.

For the next Xbox just about any Power core from 5 on with proper tweaking would be a real improvement and still make BC relatively easy due to a similar ISA to Xenon.

Keeping the CPU and GPU seperate is likely to offer better yields and more flexibility in taking advantage of fabrication advancements during the generation.

Any D3D 11 GPU will be able cover Xenos's feature set in 2012 barring a few eccentricities here and there. If D3D 12 is an option then emulating Xenos will only be that much easier.

BC should be much easier to implement for MS and Sony alike without any need to place the previous generations chips on the BOM.

[Sound_Card]

Originally Posted by Carl B

The new GPU, if it originated with ATI/AMD, would likely ensure full B/C simply simply by being a forward extension of the architecture. Throw in some specific instruction supports and it can be derived from whatever family (I do think DX11 though) they choose to branch off of when the time comes.

In terms of the CPU, well... we'll see what they do. If it's IBM again then a scaling of the XeCPU or a new branch off a more current architecture with some instructional supports again should leave them just fine on B/C.

For PS2 on PS3, the RSX from the Graphics Synthesizer was just too dramatic a break architecturally, so yes they needed that chip in there to make it work out. They put the EE in there at first as well until they could get emulation up to par on Cell, but they did so in fairly short order.

Thinking that XBox 720 might be a scaling of the 360 architecture is a fine theory; the Sun architectural stuff I don't buy into in the least anyway, so that one's not even a factor. But for myself, I do think that the hardware will be modernized rather than simply scaled, and if scaled, probably not from a single-chip solution (unless truly truly the next gen becomes one of incrementalism).

Originally Posted by scificube

For the next Xbox just about any Power core from 5 on with proper tweaking would be a real improvement and still make BC relatively easy due to a similar ISA to Xenon.

Keeping the CPU and GPU seperate is likely to offer better yields and more flexibility in taking advantage of fabrication advancements during the generation.

Any D3D 11 GPU will be able cover Xenos's feature set in 2012 barring a few eccentricities here and there. If D3D 12 is an option then emulating Xenos will only be that much easier.

BC should be much easier to implement for MS and Sony alike without any need to place the previous generations chips on the BOM.

Thanks for the much better understanding.

[AlStrong]

Originally Posted by Carl B

In terms of the CPU, well... we'll see what they do. If it's IBM again then a scaling of the XeCPU or a new branch off a more current architecture with some instructional supports again should leave them just fine on B/C.

Just a few questions to throw out there...

a) Would there be much sense in taking VMX128 any further And if so, in what way ?
b) Considering the closed-system development environment, what might be an optimal setup for cache hierarchy (considering L2 size or scope of L2 or existence of L3). I was thinking along the lines of how L2 is different between Core 2 and Nehalem i.e. L2 is large and shared in Core 2, but small and per core in Nehalem. I suppose it is a question of how large is too large to be fast enough... Does a higher level cache for the SPE's make any sense?