3DS to be more powerful than Wii -- Not using Nvidia's Tegra

[Megadrive1988]

I'm starting this new thread for several reasons. New information (rumor) has just been reported about the 3DS, also we need a 3DS topic. The current DS2 topic is old and could be confusing to some people, not realising that it's 3DS now.

okay with that said:

How powerful is the 3DS?
Nintendo has not revealed any specs for the 3DS system, but expect it to well surpass the Nintendo DS in visual and processor capabilities. To provide stereoscopic 3D effects the system must have the ability to render each game field twice, one for each of the player's eyes, a technique that will require significant horsepower to produce.

Several developers that have experienced 3DS in its current form have reported, off the record, that it has processing capabilities that far exceed the Nintendo Wii and bring the device with abilities that are close to HD consoles such as PlayStation 3 and Xbox 360.

According to several developer sources, the 3DS device is not using the NVIDIA Tegra mobile chipset, a rumor that's been floating around since 2009.

http://ds.ign.com/articles/109/1094930p1.html



NVIDIA unconnected with 3DS
June 5th, 2010

Mobile industry insiders, speaking with Digital Foundry, have said that the much-rumoured tie-up between Nintendo and NVIDIA for the console-maker's auto-stereoscopic 3D handheld is not happening.

According to our two independent, unconnected sources, the Nintendo 3DS - almost certain to be revealed at E3 - features a design totally divorced from the NVIDIA Tegra SoC (system on chip) initially thought to have been powering the DS successor. It's now thought that Nintendo has instead chosen a Japanese partner for the 3D acceleration hardware within the 3DS.

Sources also confirmed that the 3DS' development codename is "Nintendo CTR", meaning that this motherboard picture we ran a couple of weeks back, sourced from the FCC website, is indeed something akin to a development or test station for the new handheld.

This strongly suggests that 3DS does feature a widescreen "glasses free" stereoscopic 3D display, along with a more conventional 4:3 2D display beneath it. Interestingly, it appears that the images of the board published on the FCC website were uploaded in error: they were supposed to have been made public 10 months after the submission in April this year, presumably after the 3DS itself ships.

Meanwhile, IGN corroborates the story that NVIDIA is out of the picture, quoting "off-the-record" developer sources as saying that the 3DS is up there in the power stakes with PS3 and Xbox 360. It's a statement that needs to be taken with barrel-loads of salt bearing in mind the enormous power draw such a chipset would require. If nothing else, Nintendo has a strong track record in excellent battery life with every one of its previous handhelds.

Indeed, even the four-core PowerVP chip said to be at the heart of the PSP2 offers a performance level some way between the original Xbox and the 360. A more modest GPU is therefore a much more realistic proposition, especially bearing in mind that even the iPhone 3GS with its PowerVR SGX535 architecture doesn't exactly command outstanding battery life in 3D gaming.

All of which is interesting tittle-tattle in the here and now, but let's hope that the Nintendo E3 conference on June 15, kicking off at 5pm UK time, will offer more concrete answers.


http://www.eurogamer.net/articles/nvidia-unconnected-with-3ds-blog-entry


So who could be powering the 3DS? My bet is PowerVR, manufactured in Japan.

[wsippel]

Imagination Technologies isn't a Japanese company, though. DMP might be a likely candidate - maybe something like this: http://www.dmprof.com/release/leaflet_SMAPH-S_en.pdf

[DeadlyNinja]

Who think IGN is talking out of their behind, or is misinterpreting what they've been told?

My guess is probably GC/Wii level with more advanced effects and shaders. It certainly would be very impressive for a handheld. I hope they keep the damn thing at 60fps.

[darkblu]

Who think IGN is talking out of their behind, or is misinterpreting what they've been told?
* raises right hand, left hand and a hind limb. waves those in the air, while desperately trying to keep balance on remaining grounded limb *

[brain_stew]

Maybe what IGN were really told was that it'll have a similar featureset to the PS3/360 but in terms of raw performance would be inline with the Wii? I could certainly understand how someone not so technically minded could get confused with such statements and come away with comments like they did. That's got to be around the sort of area we're expecting for this thing, right? Something like a SGX 535 would fit that sort of description and something of that calibre surely isn't out of the realms of possibility, no?

For reference with a high level of abstraction the iPad's pulling off some pretty impressive stuff @ 1024x768, like this:

http://www.youtube.com/watch?v=KqdvClxB-Zg&fmt=22

Now the 3DS is going to be using a much lower resolution screen (if it's the rumoured Sharp screen it'll be 800x480 in 2D mode, half that for 3D) and developers will be coding much closer to the metal so that level of hardware should be able to produce some pretty spectacular results. There's the issue of battery life of course so its unlikely that Nintendo will clock the thing quite as high as Apple did but perhaps they'll take a strategy similar to what Sony did with the PSP where they'll "unlock" the chip later in the platform's life to give it a new lease of life once new models with better battery life are released? What process node would Nintendo be able to use if they're launching this thing before the end of the calender year? 40nm?

Is DMP's chip a tiler? It looks a good candidate what with it being Japanese, incredibly scalable and (if their claims are to be believed) having the "industry lowest" power consumption. It coming from a more niche company may mean that Nintendo may be able to negotiate more favourable terms if they're desperate for their first big design win.

What are people's expectations when it comes to RAM? 128MB of LPDDR2 seems to be the minimum for even low end smartphones these days and is available as a single chip so is there any real good reason why Nintendo would go with less than this?

All in all I expect something with games that have graphics/complexity similar to Xbox games but for them not to be quite as RAM starved. Games that can't quite match the raw polycount of high end Wii games but make up for it through better effects/lighting and image quality is what I'd like to see, and games of that ilk should look very impressive in full stereo3D on a small 3/4" screen.

[Squilliam]

I don't see why not, given the effects possible on the iPad and the much smaller screen. This is especially true if they have made it possible to significantly lower the overhead when displaying 3D content which is a very plausible estimate given the fact that they designed the system for 3D display.

[MfA]

Is DMP's chip a tiler?
Unlikely. Broadcom's chip is probably, based on Gigapixel's Pixelsquirt.

Really though, everything out there in mobile seems a bit antique compared to SGX.

[Ailuros]

Imagination Technologies isn't a Japanese company, though. DMP might be a likely candidate - maybe something like this: http://www.dmprof.com/release/leaflet_SMAPH-S_en.pdf

DMP is circulating lately in the rumor mill, but hand in hand with the non programmable pipeline. Don't exclude something along the line of the PICA200 (OGL_ES1.1) just yet. The PICA200 Lite (geez another vendor that can't spell light) sounds too humble to me.

[Teasy]

Maybe what IGN were really told was that it'll have a similar featureset to the PS3/360 but in terms of raw performance would be inline with the Wii? I could certainly understand how someone not so technically minded could get confused with such statements and come away with comments like they did.

IGN have basically said its going to be much more powerful then Wii and with a similar feature set to 360/PS3. They didn't say it would be as powerful as 360 or PS3 though, Eurogamer misquoted them.

[Teasy]

Is there any firm info on the 3DS screen yet? I saw that Nintendo DS CTR development board with the widescreen on top and normal screen below, but is there any sort of confirmation that the top screen is a Sharp auto-stereoscopic 3D screen? (which is the rumour going around I believe).

[Lazy8s]

DMP have been a curiosity of mine for several years. With their continuing development roadmap and offerings of OpenGL ES seminars, they seemed to have had some assurance of at least one signicant licensee and/or investor. They were previously discussed here.

http://forum.beyond3d.com/showthread.php?t=32326

Gigapixel was of course the 3Dfx and, in turn, nVidia acquisition while Stellar Semi and their scan-line (scene-captured, deferred) renderer went to Broadcom.

That was way back in the Wild West days of the PC 3D accelerator market, and Broadcom's extremely low profile in the mobile SoC 3D market for all of these years made me think the Stellar assets had just been reassigned within Broadcom. However, the vague similarities to what I seem to recall learning about Nintendo's DS graphics processor makes me wonder whether Broadcom wasn't the mystery partner on that project and whether that was what had kept the former Stellar people busy for that time.

Also, lately, this Stellar architecture or something somewhat like it has seemingly resurfaced within Broadcom's current product line-up. With a similar cost to an MBX Lite but with significantly lower performance, it's not too exciting yet, though.

[sfried]

The DMP company is another small graphics firm, staffed by 29, which started as a university research project and turned into a business a few years ago through funding initiatives. Shinichi Okamoto, former Sony executive who headed up development of the PlayStation2, sits on the board of directors.Hmm...

[Rys]

Unlikely. Broadcom's chip is probably, based on Gigapixel's Pixelsquirt
OT, but wasn't Pixelsquirt a Stellar Semi thing, rather than Gigapixel? Edit: Ah, Lazy8s covered that above, ignore me.

[MfA]

Gah, too many sunk tiler manufacturers out there ...

[Ailuros]

Gah, too many sunk tiler manufacturers out there ...

Well there are more tilers integrated in the embedded market today on the other hand, than non tilers. SGX aside, Mali and Imageon are both tile based too.

[Arun]

I've thought for quite some time that talking of 'tilers' as a category has become too misleading (after all it's not as if tiles weren't omnipresent for memory access pattern reasons anyway). I'd much rather talk of 'binners'; how you use that binning data once you've got it is just an (important) implementation detail.

[DeadlyNinja]

Let me ask a silly question. Trust me, it's very silly. Let say Nintendo does intend to have the 3DS beat the Wii graphically, would an ultra crappy integrated GPU like the GeForce 6150SE be enough to do it? I've done some tests with a relatively modern game called Paraworld (2006) and the game runs relatively well at 1024 x 768 with a few things here and there at high.

This is what Paraworld looks like: http://image.com.com/gamespot/images/2006/135/920933_20060516_screen016.jpg

I also tested it on an older game called Jurassic Park Operation Genesis. It was a GC/Xbox/PS2 era game with some really nice looking water. That game ran like butter at 800 x 600.

Strangely enough, the GPU can barely handle Half Life 2. I think it was set to 480p. Looks horrible, yet the game is nowhere near as modern as Paraworld.

If Nintendo is aiming for 800 x 480 with the 3DS, would a GPU of that power be enough?

[darkblu]

I've thought for quite some time that talking of 'tilers' as a category has become too misleading (after all it's not as if tiles weren't omnipresent for memory access pattern reasons anyway). I'd much rather talk of 'binners'; how you use that binning data once you've got it is just an (important) implementation detail.
FWIW, I've been commonly referring to Imagination's designs as 'TBDR', and as 'tilers' to the rest TBDRs which don't do overdraw elimination. Clearly that's rather arbitrary (and misleading), and I, too, have been considering fixing my terms. I'm considering going with:

* TBDR/OE for tiling/binning scene-capturers that do rigid overdraw elimination, and
* TBDR for tiling/binning scene-capturers in general.

By that convention Imagination's designs would be classified as TBDR/OE, and everybody else (GMA (sans 500), Z4xx/Adreno, etc) as TBDR. I'm also considering using 'just DR' (deferred renderer) for scene capturers in general.

[darkblu]

If Nintendo is aiming for 800 x 480 with the 3DS, would a GPU of that power be enough?
I'd venture to guess that that particular 'ultra crappy' IGP draws between 3x to 7x (maybe even an order of magnitude) more power than Hollywood. Putting such a part into a handheld is not an option.

Also, I'd take the 'more powerful than wii' tidbits as utter nonsense ; )

[DeadlyNinja]

I'd venture to guess that that particular 'ultra crappy' IGP draws between 3x to 7x (maybe even an order of magnitude) more power than Hollywood. Putting such a part into a handheld is not an option.

Also, I'd take the 'more powerful than wii' tidbits as utter nonsense ; )

I don't mean sticking it in as it is. Of course it wouldn't work for a handheld, I mean if they hypothetically made the exact same thing but for handheld.

[Exophase]

FWIW, I've been commonly referring to Imagination's designs as 'TBDR', and as 'tilers' to the rest TBDRs which don't do overdraw elimination. Clearly that's rather arbitrary (and misleading), and I, too, have been considering fixing my terms. I'm considering going with:

* TBDR/OE for tiling/binning scene-capturers that do rigid overdraw elimination, and
* TBDR for tiling/binning scene-capturers in general.

By that convention Imagination's designs would be classified as TBDR/OE, and everybody else (GMA (sans 500), Z4xx/Adreno, etc) as TBDR. I'm also considering using 'just DR' (deferred renderer) for scene capturers in general.

I always thought the "deferred" part in TBDR referred to deferred shading/overdraw elimination approach by IMG (only ones I know of using the TBDR term), not deferred rendering necessitated by scene binning. I mean, now that I think about it I was obviously wrong, especially considering IMR is used as a contrasting term.. but it looks like we were using similar terminology.

Are tilers like Xenon, and by assumption Z4xx, really scene capturing? I mean, I might not understand the concept of operation completely, but it seems to me that instead of collecting all scene geometry then binning it it will render geometry immediately clipped against the tile boundaries, with the option to mark (or maybe remove entirely?) culled primitives and the requirement to render again for other tiles. It seems something in-between immediate and deferred, and also seems like something a modern IMR could mostly (entirely??) do, only lacking the explicit high speed tile memory, that might be serviced similarly with a framebuffer cache.

But I'm probably really showing my ignorance with this post:huh:

[Rys]

A 1VS, 2PS 6-series class GeForce isn't that far away from Tegra, functionally. The frequency is miles away, though.

[Rys]

Are tilers like Xenon, and by assumption Z4xx, really scene capturing? I mean, I might not understand the concept of operation completely, but it seems to me that instead of collecting all scene geometry then binning it it will render geometry immediately clipped against the tile boundaries, with the option to mark (or maybe remove entirely?) culled primitives and the requirement to render again for other tiles. It seems something in-between immediate and deferred, and also seems like something a modern IMR could mostly (entirely??) do, only lacking the explicit high speed tile memory, that might be serviced similarly with a framebuffer cache.
You're pretty much spot on.

[aaronspink]

I always thought the "deferred" part in TBDR referred to deferred shading/overdraw elimination approach by IMG (only ones I know of using the TBDR term), not deferred rendering necessitated by scene binning. I mean, now that I think about it I was obviously wrong, especially considering IMR is used as a contrasting term.. but it looks like we were using similar terminology.

There is a matrix.

DR
TBDR
TBR
IMR
etc

The deferred part specifically relates to deferring shading/color until order is known.

Are tilers like Xenon, and by assumption Z4xx, really scene capturing?

Xenon IS NOT a tiler. It is an IMR. The "tiling" comes from the limited framebuffer space due to the size of the embedded memory. The programmer is fully responsible for any tiling that takes place.

It seems something in-between immediate and deferred, and also seems like something a modern IMR could mostly (entirely??) do, only lacking the explicit high speed tile memory, that might be serviced similarly with a framebuffer cache.

No, xenon is fully an IMR.

[darkblu]

I always thought the "deferred" part in TBDR referred to deferred shading/overdraw elimination approach by IMG (only ones I know of using the TBDR term), not deferred rendering necessitated by scene binning. I mean, now that I think about it I was obviously wrong, especially considering IMR is used as a contrasting term.. but it looks like we were using similar terminology.
Well, I've always equated 'deferred' with 'capturing', whether it's for overdraw elimination or god-knows-what. Of course i'm not insisting that my side of the terminology is correct - on the contrary, i'm in search of the correct terminology.

Are tilers like Xenon, and by assumption Z4xx, really scene capturing? I mean, I might not understand the concept of operation completely, but it seems to me that instead of collecting all scene geometry then binning it it will render geometry immediately clipped against the tile boundaries, with the option to mark (or maybe remove entirely?) culled primitives and the requirement to render again for other tiles. It seems something in-between immediate and deferred, and also seems like something a modern IMR could mostly (entirely??) do, only lacking the explicit high speed tile memory, that might be serviced similarly with a framebuffer cache.
So what you suggest is clipping your primitives bin-by-bin without capturing? That spares you capturing but takes away from you on-tile blending and cheap FSAA resolves - hardly a good tradeoff, IMO. You could do that on a scene level on an IMR - ie. re-draw your entire scene and clip it bin-by-bin (see Xenos reference below), but that's hardly optimal vertex-wise.

AFAIK, Xenos and derivatives are not ovedraw-eliminating the way Imagination's are, but they are scene-capturing as much as Imagination's. Ok, in the case of the original Xenos - manual scene capturing ; ) ( or ; (, depending on whether you had to ship a product on that - I didn't). IIRC MS/ATI addressed that later into the lifecycle of the paltform with API augmentations.

Here's an ATI/AMD paper from some time ago discussing their idea of a TBR from back then: http://developer.amd.com/gpu_assets/gdc2008_ribble_maurice_TileBasedGpus.pdf - I believe it's quite relevant to Z4xx. Pay particular attention to the 'resolves' topic.

[MfA]

FWIW, I've been commonly referring to Imagination's designs as 'TBDR', and as 'tilers' to the rest TBDRs which don't do overdraw elimination. Clearly that's rather arbitrary (and misleading), and I, too, have been considering fixing my terms.
Pixelsquirt (and thus Broadcom) was also deferred I think.

[MfA]

Why would you try to capture your scene instead of rendering multiple times with different frustums?

[darkblu]

Why would you try to capture your scene instead doing two draw calls with different frustums?
Multiplicity of vertex work and related bandwidth expenditures.

[Exophase]

There is a matrix.

DR
TBDR
TBR
IMR
etc

The deferred part specifically relates to deferring shading/color until order is known.

If you consider it a TBDR to be a "tile based and deferred" renderer then yes, I agree. But if you consider it to be "tile based" and "deferred renderer" then I agree with darkblu. "Deferred" can logically apply to either deferred shading or scene capturing.

The problem with applying it only to deferred shading is that there's no opposite term for immediate mode rendering. Or maybe "DMR" should be used? You seem to be using the term "tiler" to include scene capturing for any sort, but that's incorrect because not all scene capturers do so with tiling.

Xenon IS NOT a tiler. It is an IMR. The "tiling" comes from the limited framebuffer space due to the size of the embedded memory. The programmer is fully responsible for any tiling that takes place.

Like here, you can see that you define tiler as being scene capturing/deferred rendering. By definition something that sorts a scene into tiles (with hardware) has to be, but is it authoritative that "tiler" has to mean that? Tiling could instead refer merely to the presence of high speed tile memory. I think "binning" is less ambiguous.

No, xenon is fully an IMR.

Xenon allows for the same basic data flow, in software, that a tile binning renderer would provide in hardware. I don't know precisely what it's clipping is (guard band only or if it can be done at the geometry level instead), which would say something about whether or not the manual rendering of geometry into separate tiles still counts as "rendering" when the geometry is off-tile. It seems at least slightly less black and white then you make it.

But ultimately all of this is just semantics, over terms that are not authoritatively governed. Seems like more focus should be spent on making sure it's understood how things operate and that people know what you're referring to, rather than arguing over what term should be used.

[aaronspink]

If you consider it a TBDR to be a "tile based and deferred" renderer then yes, I agree. But if you consider it to be "tile based" and "deferred renderer" then I agree with darkblu. "Deferred" can logically apply to either deferred shading or scene capturing.

It means exactly what a represents: a Tile Based Deferred Render. It defers the scene by binning it into sorted tiles for later rendering.



Like here, you can see that you define tiler as being scene capturing/deferred rendering. By definition something that sorts a scene into tiles (with hardware) has to be, but is it authoritative that "tiler" has to mean that? Tiling could instead refer merely to the presence of high speed tile memory. I think "binning" is less ambiguous.

Tiling itself simply refers to the hardware internally breaking down the scene into multiple smaller render regions.



Xenon allows for the same basic data flow, in software, that a tile binning renderer would provide in hardware. I don't know precisely what it's clipping is (guard band only or if it can be done at the geometry level instead), which would say something about whether or not the manual rendering of geometry into separate tiles still counts as "rendering" when the geometry is off-tile. It seems at least slightly less black and white then you make it.

By that argument ALL hardware renderers are tilers. ATI and Nvidia mainstream hardware have the functionality to clip outside the viewspace and you can set the viewspace to whatever you want.

I think it is fairly black and white. If you want to do any tiling with Xenon, you have to do it through software just like with an RV870 or GF100.

[thop]

It could be NEC too.

http://www2.renesas.com/mobile/en/emma_mobile/em_ev.html
http://www.imgtec.com/News/Release/index.asp?NewsID=546

[sfried]

It could be NEC too.

http://www2.renesas.com/mobile/en/emma_mobile/em_ev.html
http://www.imgtec.com/News/Release/index.asp?NewsID=546What's with the IMG link?

[Ailuros]

It could be NEC too.

http://www2.renesas.com/mobile/en/emma_mobile/em_ev.html
http://www.imgtec.com/News/Release/index.asp?NewsID=546

I thought EMMA is using SGX530?

[darkblu]

It means exactly what a represents: a Tile Based Deferred Render. It defers the scene by binning it into sorted tiles for later rendering.
And so do TB-non-DRs.

The issue with the term TBDR (as used in the context of Imagination's products) is that the 'deferred' portion of it is meant to signify 'overdraw elimination' whereas it says something else - 'scene capturing'. The logical relation between 'deferred' and 'overdraw elimination' is not one of equality - you can be deferred and non-OE at the same time. It all depends on whether an explicit step of occlusion resolution follows the deferring, or not.

By that argument ALL hardware renderers are tilers. ATI and Nvidia mainstream hardware have the functionality to clip outside the viewspace and you can set the viewspace to whatever you want.
Well, duh, of course all hardware renderes could be turned into tilers. It's how efficient they'd be at the task. The reverse is true too - a tiler turns into a non-tiling scene capturer as long as you limit your output to the confines of a single tile.

I think it is fairly black and white. If you want to do any tiling with Xenon, you have to do it through software just like with an RV870 or GF100.
HAL/drivers are software. And I'm yet to see a TB(D)R that does not use software assist for things like binning overflow/emergency flushes. Does that mean TBRs don't exist in your black and white view?

[tangey]

I thought EMMA is using SGX530?

it does indeed, SGX530 if i remember correctly

[Svensk Viking]

Must the hardware in the 3DS really be that strong to be along the leagues of the PS3/360 when we take the lower resolution into consideration??

[sfried]

Must the hardware in the 3DS really be that strong to be along the leagues of the PS3/360 when we take the lower resolution into consideration??I believe they mentioned something along the lines of "feature set". Whatever the case, hopefully the added flexibility would make it more appealing to developers than TEV (which, at the moment, only a few talented teams seem to grasp...most of them German).

In any case I still don't see how a rival chip supplier (IMG) could be a potential candidate for the Nintendo handheld, unless somebody could explain to me some business semantics that would allow them to do just that.

[Exophase]

In any case I still don't see how a rival chip supplier (IMG) could be a potential candidate for the Nintendo handheld, unless somebody could explain to me some business semantics that would allow them to do just that.

IMG is an IP supplier, not a chip supplier, although that's probably not what you meant. What are they rivaling exactly? The only other 3D product Nintendo would have on the market is Wii, which is very far removed from SGX, and I imagine so would Wii's successor.

Or are you saying this because of the rumor that PSP2 is using SGX? In that case, I think even iPhone having it would be considered competition. All three consoles are using CPUs made at least in part by IBM, so I don't really see what would make that sort of thing problematic for 3D.

[thop]

In any case I still don't see how a rival chip supplier (IMG) could be a potential candidate for the Nintendo handheld, unless somebody could explain to me some business semantics that would allow them to do just that.
Nintendo isn't in the chip business.

[DeanoC]

I think it is fairly black and white. If you want to do any tiling with Xenon, you have to do it through software just like with an RV870 or GF100.

Its not black and white though. Xenon GPU modifies the command buffer
on the first tile with screen space bounding areas so you don't need to use software tiling.
I.e. A valid Xenon GPU Command buffer is
Set Tile 0
Render *GPU calculates tile coverage*
Resolve
Set Tile 1
Render
Resolve

No CPU interaction is required for it to work.

Anything that is not in Tile 1 screen area will be skipped at the command buffer level (i.e. not submitted from the front end to the main GPU units).

So its somewhere in between... argueable the command buffer is a captured scene which the tile unit (actived when rendering Tile 0) writes back data about tile coverage for subsquent tiles.

[Exophase]

Its not black and white though. Xenon GPU modifies the command buffer
on the first tile with screen space bounding areas so you don't need to use software tiling.
I.e. A valid Xenon GPU Command buffer is
Set Tile 0
Render *GPU calculates tile coverage*
Resolve
Set Tile 1
Render
Resolve

No CPU interaction is required for it to work.

Anything that is not in Tile 1 screen area will be skipped at the command buffer level (i.e. not submitted from the front end to the main GPU units).

So its somewhere in between... argueable the command buffer is a captured scene which the tile unit (actived when rendering Tile 0) writes back data about tile coverage for subsquent tiles.

So is it culling, or is it really tiling? I mean, where you have "GPU calculates tile coverage", is it actually calculating which tile each GPU spans, or is it just calculating if it's inside tile 0 or not? Either way, it's still at least screen space culling done at the geometry level (don't know about z and w) and before triangle setup/rasterization. But I don't know if this is something modern IMRs do too; I was under the impression that guard band clipping was used during scanline setup, ie post triangle setup. Is the tile bounding in Xenon clipping as well, or just culling?

I would argue that just having an explicitly addressed intermediate render output buffer that can not act directly as framebuffer means you're "tiling", for a reasonable definition of the word. You output to tile memory, which is fast, then can burst data back to the target buffers in a single resolve. The only way this data flow would occur in an IMR is if the target buffer writes are being cached, but without explicit resolves there'd be no configuration available for avoiding writeback of temporary data like depth/stencil. I assume Xenon does have the ability to not resolve these buffers, which should be another tiling advantage over IMRs.

[Megadrive1988]

not long now until we know something, hopefully, later tomorrow (it's Monday here).

[sfried]

not long now until we know something, hopefully, later tomorrow (it's Monday here).I'm keeping my expectations fairly low. I don't expect Nintendo to go with cutting edge, but I'm willing to bet it is yet another ARM solution.

The big question would be concerning how they solve their piracy issues, but I don't think that subject matter something Beyond 3D discusses much...

[MfA]

We've discussed it plenty, I hold the position that their copy prevention implementation in the DS was simply weak and it shouldn't be too hard to make it nigh impossible to pirate cartridge and downloaded games. Everyone else disagrees with me.

[Megadrive1988]

3DS was just shown. Graphics look on par with PSP, maybe even Gamecube, but definitely not Wii-level.

oh and Kid Icarus was just revealed for 3DS.

[brain_stew]

3DS was just shown. Graphics look on par with PSP, maybe even Gamecube, but definitely not Wii-level.

oh and Kid Icarus was just revealed for 3DS.

The graphics seemed to display many of the traits of GCN games, I think the theory about a shrunken down Flipper may have been 100% correct. Looks above PSP level to me and plus its rendering in a much higher resolution than PSP (about 2x if its that Sharp screen) and then there's the extra overhead of stereo3D so the hardware must be significantly beyond what's in the PSP.

[sfried]

So 3DS has a gyro to much like iPhone4.

I'd say the graphics look better than PSP since there seems to be more shaders at work. Draw distance is also something to consider in the video, as it is very distant.

[brain_stew]

Some game screens:

http://i48.tinypic.com/2z72w61.jpg

http://i45.tinypic.com/281v34p.jpg

http://i46.tinypic.com/2zqfyo3.jpg

Specifications as well:


http://img2.pict.com/cc/eb/5f/3666316/0/untitled1.png

Looks about GCN level I guess but with lower vert. counts (expected considering S3D means processing triangles twice). Resolution is pretty impressive for a handheld machine. No AA though. :(

I seriously think those shrunken down Flipper rumours were spot on, its definitely in that sort of ball park imo.


Loads of third party screenshots here:

http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/

The Snake Eater port doesn't look to be compromised in any way at all, to run high end PS2 games at launch, in 3D and with a really nice resolution to boot is pretty impressive. It should be capable of some very decent stuff, definitely beyond PSP imo.

[DeadlyNinja]

Lots of normal mapping. I guess it's easier to normal map on this than Wii.

http://i45.tinypic.com/2rnto4n.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_01_ksd.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_02_osl.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_03_juf.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_04_ksu.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_05_qwe.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_10_mbv.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_09_dcv.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_08_ujn.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_07_nut.jpg
http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_06_gfh.jpg

[brain_stew]

Street Fighter shots are interesting as they seem to have most of the advanced shading of the PS3/360 version. Maybe this isn't Flipper based afterall.

http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/SUPER%20STREET%20FIGHTER%20IV%203D%20Edition/3DS_SSF4_04ss04_E3.png

Self shadowing as well, pretty impressive for a cheap port.

Edit: Yeah, the more screenshots I see, the more abundant nice shader effects look to be, so unlike before I don't think this is Flipper based at all, Nintendo's first party stuff just don't seem to be pushing too mnay advanced effects like third parties are.

[darkblu]

Looks about GCN level I guess but with lower vert. counts (expected considering S3D means processing triangles twice). Resolution is pretty impressive for a handheld machine. No AA though. :(
I noticed the lower geometry complexity in the kid icarus video too (below cube levels) - double rendition taxation on the pipeline indeed.

As re the total lack of AA - it's really curious how nintendo went from a free AA to no AA in a gen- must be something about stereoscopic display that allows them to do that. On that note, I can't wait to get my hands on the darn device!

[sfried]

I'm glad someone went ahead and uploaded Snake Eater.
Here's a pic of Peace Walker (http://www.psu.com/media/metal-gear-solid:-peace-walker/metal-gear-solid:-peace-walker-ss-11.jpg) for comparison. Note the lack of any shaders there.

[DeadlyNinja]

What is the PSP's resolution again? I forgot. I know it's not as high as the MGS pic posted there.

[sfried]

Hoping these are not FMV bullshots (Most likely not because of the low geometry models, 2D spalsh on boat), but still:

http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/RESIDENT%20EVIL/3DS_ResidentER_01ss01_E3.png
http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/RESIDENT%20EVIL/3DS_ResidentER_02ss02_E3.png
http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/RESIDENT%20EVIL/3DS_ResidentER_03ss03_E3.png
http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/RESIDENT%20EVIL/3DS_ResidentER_04ss04_E3.png
http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/RESIDENT%20EVIL/3DS_ResidentER_05ss05_E3.png
http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/RESIDENT%20EVIL/3DS_ResidentER_06ss06_E3.png

As someone said, theres more liberal use of normal maps. I'm guessing Kid Icarus doesn't use it because they reused some of the assests from SSBB and also kept the artstyle consistent.

Edit: Splinter Cell stuff
http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/Tom%20Clancy%27s%20Splinter%20Cell%20Chaos%20Theor y/3DS_SplinterC_04ss04_E3.png
http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/Tom%20Clancy%27s%20Splinter%20Cell%20Chaos%20Theor y/3DS_SplinterC_02ss02_E3.png
It's odd that some shots show the AA and some don't.

[DeadlyNinja]

Well Ubi Soft is known for bullshots on everything. They're release Red Steel screens at 720p with 999999999999999999x aa.

http://reformiklubi.fi/~mattu/Nintendo_3DS-3rd_Party_Assets/RESIDENT%20EVIL/3DS_ResidentER_01ss01_E3.png

You can see the polygons on her arm pretty clearly in this shot. They're probably realtime.

[sfried]

Here's some shots of the Star Fox 64 remake:
http://reformiklubi.fi/~mattu/Nintendo_3DS_1st_2nd_Party_Assets/Star%20Fox%2064%203D/3DS_Starfox64_02ss02_E3.png
http://reformiklubi.fi/~mattu/Nintendo_3DS_1st_2nd_Party_Assets/Star%20Fox%2064%203D/3DS_Starfox64_05ss05_E3.png
Looks like 3DS is capable of some good AA.

Which reminds me, is it nearly as necessary to have AA on stereoscopic vision?

[DeadlyNinja]

Oh man, I love Star Fox 64!!!

[Exophase]

What is the PSP's resolution again? I forgot. I know it's not as high as the MGS pic posted there.

480x272, higher than the screenshots posted which are all 400x240.

[DeadlyNinja]

480x272, higher than the screenshots posted which are all 400x240.

I was talking about the Peace Walker screen.

[Exophase]

I was talking about the Peace Walker screen.

Oh, sorry. Yeah, that one's much higher resolution.

[sfried]

One thing that have has been revealed is that there is dedicated video memory (http://e3.nintendo.com/interviews/#/?v=interview_slaczka). Don't know if this tidbit significant enough to be posted.

[gigadude]

One thing that have has been revealed is that there is dedicated video memory (http://e3.nintendo.com/interviews/#/?v=interview_slaczka). Don't know if this tidbit significant enough to be posted.

This would be a weird design decision if true. Why on earth would you have dedicated video memory in an SOC? I doubt they could have packed enough edram on-chip to be interesting, and why pay the cost of extra memory pins and then restrict what you can do with the extra bandwidth by segregating the usage?

[brain_stew]

One thing that have has been revealed is that there is dedicated video memory (http://e3.nintendo.com/interviews/#/?v=interview_slaczka). Don't know if this tidbit significant enough to be posted.

Good spot, most SOCs just have a single shared RAM pool so that would hint at a more custom design I guess.

[Exophase]

This would be a weird design decision if true. Why on earth would you have dedicated video memory in an SOC? I doubt they could have packed enough edram on-chip to be interesting, and why pay the cost of extra memory pins and then restrict what you can do with the extra bandwidth by segregating the usage?

How much would it need to be "interesting"? PSP had 2MB several years ago, 3DS can probably do better today. Depending on the design nearly any amount can be helpful. I strongly doubt there's a dedicated external bus for extra video memory.

I still haven't completely let go of the idea that the 3D graphics are an enhancement of the DS's, with especially more per-pixel capability and texture filtering. Maybe someone who is good at discerning polygon counts can give some estimates of geometric complexity on some of the screenshots. If an order of magnitude within DS's seems plausible (so say, ~20k triangles) then 3DS bumping the DS's internal RAM to accommodate that is not unrealistic.

Granted, I'm pretty impressed with Nintendo's offering. It's good that they're at least trying to deliver decent portable graphics instead of doing the bare minimum again, and the addition of an analog nub is a nice and somewhat unexpected decision (to me anyway). I do wonder what the CPU capabilities will be like. I also wonder exactly what implications externally controlled depth of field has, I would have figured that was a function of the depth of the actual 3D images. Maybe that's being scaled by the slider value.

[brain_stew]

A GAF user has confirmed the RE demo that has been reported to be running in real time is the same one depicted in those screenshots. So they're legit minus the super high level of AA:

http://www.neogaf.com/forum/showpost.php?p=21889029&postcount=931

Pretty impressive if you ask me, the poly count may be something approaching PSP levels but the lighting/shadowing and shaders seem to have much more in common with the 360 version and texturing looks to be a real step up above the PSP as well. I'm definitely satisfied with this level of hardware.

Edit: Oh and here's a off screen video of the MGS demo:

http://tinypic.com/usermedia.php?uo=zHAawG5Dm5Z5aQ0GYq45A4h4l5k2TGxc

[sfried]

Edit: Oh and here's a off screen video of the MGS demo:

http://tinypic.com/usermedia.php?uo=zHAawG5Dm5Z5aQ0GYq45A4h4l5k2TGxcT hanks. The framerate looks fantastic btw.

[ltcommander.data]

Any idea how this compares to what a SGX535 is capable of in say the iPhone 3G S or iPhone 4? Or how many cores a theoretical PSP2 with SGX543MP would need to match or overtake the 3DS?

[Exophase]

Any idea how this compares to what a SGX535 is capable of in say the iPhone 3G S or iPhone 4? Or how many cores a theoretical PSP2 with SGX543MP would need to match or overtake the 3DS?

The reality behind that is probably muddied by how much developers are (or aren't) pushing it and how talented they are at making games look good in spite of specs. iPhone 3GS hasn't exactly had a ton of top notch huge budget games made expressly for it.

I doubt anything we've now seen on 3DS can match or beat an SGX535 at decent clock speed, and a single SGX543 core is quite a bit more capable (2x the ALUs which are themselves wider and 2x the depth/stencil comparator rate)... but I don't think I'm a very good judge of capability from screenshots and videos. To me even some DS games look pretty good and I'm very acquainted with how weak that hardware is.

[thop]

Lack of dithering makes these graphics better than PSP graphics by default.

[Squeak]

480x272, higher than the screenshots posted which are all 400x240.

But then you have the whole screen for the game. Metres and counters can be put on the bottom screen. AND you can turn of the 3D to get very high horizontal resolution if need be.

[b3vcard]

Could any experienced users with good memory tell me;

Is there much difference between launch games on a handheld, compared to say game a year or two in ? I know on main consoles all the development makes the launch games look last gen.

What are the chances that these launch games are not using the 3DS to its full ability ? I know 'ability' is very subjective but I would assume, as time goes on, new algorithms and techniques would improve the graphics etc

[b3vcard]

http://www.playstationportable.de/images/db/22386.jpg

http://www.playstationportable.de/images/db/22385.jpg

http://www.playstationportable.de/images/db/22384.jpg

Some 3DS VS PSP shots. Not great resolution though..

[brain_stew]

What the hell is up with DF's take on the 3DS? Not at all in the same league as a Tegra device? Well past the PSP but maybe not quite as powerful as a Dreamcast? No sign that it has programmable shaders? Huh!? How on earth could they come to those conclusions? They're not even consistent with each other nevermind some of the screenshots we've seen.


Could any experienced users with good memory tell me;

Is there much difference between launch games on a handheld, compared to say game a year or two in ? I know on main consoles all the development makes the launch games look last gen.

What are the chances that these launch games are not using the 3DS to its full ability ? I know 'ability' is very subjective but I would assume, as time goes on, new algorithms and techniques would improve the graphics etc

All the same rules apply, that it is a handheld system changes very little. The PSP is a perfect example, God of War looks as though its produced on entirely different hardware than most launch titles.

[thop]

More MGS screenshots.

http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_01_ksd.jpg http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_02_osl.jpg

http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_03_juf.jpg http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_04_ksu.jpg

http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_05_qwe.jpg http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_06_gfh.jpg

[thop]

http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_07_nut.jpg http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_08_ujn.jpg

http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_09_dcv.jpg http://www.konami.jp/kojima_pro/e3_2010/mgs3d/en/imgs/pic_10_mbv.jpg

[Exophase]

Lack of dithering makes these graphics better than PSP graphics by default.

PSP has a 24bpp display and the means to display 24bpp graphics - I don't see how having dithering as a hardware feature is a disadvantage.

But then you have the whole screen for the game. Metres and counters can be put on the bottom screen. AND you can turn of the 3D to get very high horizontal resolution if need be.

I don't think "metres and counters" tend to take up a lot of screen area, except of course for DS (and now 3DS) games where they're looking to fill space on that second screen. As for being able to turn off 3D to double the horizontal resolution, I don't think we know for sure it actually works that way. There could be some obstacles involved.

[Josh128]

Check out this video of the 3DS MGS demo in action-- while the 3D is not discernable, the graphics look pretty outstanding. Certainly more impressive in my opinion than the still shots suggest.


http://tinypic.com/usermedia.php?uo=zHAawG5Dm5Z5aQ0GYq45A4h4l5k2TGxc

[Squeak]

I don't think "metres and counters" tend to take up a lot of screen area, except of course for DS (and now 3DS) games where they're looking to fill space on that second screen. As for being able to turn off 3D to double the horizontal resolution, I don't think we know for sure it actually works that way. There could be some obstacles involved.
Of course there are a few games where you don't need or by design leave out a HUD. But they are few and it would almost always be better to have one if not for anything else but a map. With the DS line you have a whole screen for that shit, if you need it.
Even a single 16x16 counter distracts and obfuscates much more than the rectangle it occupies.

[thop]

Unlike reported by DF, Miyamoto confirms shaders.

"We have shaders that people are able to take advantage of [...]."

http://www.joystiq.com/2010/06/16/interview-nintendos-shigeru-miyamoto-on-3ds-retro-zelda-and/

[RudeCurve]

I don't know who claimed the 3DS is capable of graphics close to X360/PS3 but they must have been smoking something hallucinogenic. :lol:

From the screenshots of RE and MGS, it's obvious there is some very good lighting but the polygons seem very low and textures quality is not very detailed. I'd say the graphics is somewhere inbetween Dreamcast and Xbox 1.

[sfried]

I don't know who claimed the 3DS is capable of graphics close to X360/PS3 but they must have been smoking something hallucinogenic. :lol:

From the screenshots of RE and MGS, it's obvious there is some very good lighting but the polygons seem very low and textures quality is not very detailed. I'd say the graphics is somewhere inbetween Dreamcast and Xbox 1.I'm pretty sure it's past Dreamcast levels. I'm noticing more shader effects being used liberally (such as the cliffside face in one of the MGS3D shots). Definitely beyond PS2/PSP at this point, that's for sure.

It's hard to "guage" performance given how little we know about the architecture. All we know right now is it has dedicated video memory.

[Fafalada]

This would be a weird design decision if true. Why on earth would you have dedicated video memory in an SOC? Both NDS (just under 500KB) and PSP (2+2MB) had embeded VRam. In case of the former, the design wouldn't function without it at all. And in case of PSP, the amount is certainly enough to be "interesting". I still haven't completely let go of the idea that the 3D graphics are an enhancement of the DS's, with especially more per-pixel capability and texture filtering After watching the media I kinda got the impression as well. And the cool things about DS rasterizer (like being overdraw free) would make that much more difference with the speculated pixel-shading upgrades in 3DS.

[Shifty Geezer]

BC also points towards DS extended hardware. And no AA on the Metal Gear screenshots suggests no improvements in that department, although that could just be this game.

[Fafalada]

And no AA on the Metal Gear screenshots suggests no improvements in that department, although that could just be this game.
One thing that really stands out in all the realtime media is the pretty bad PSP/PS2 esque texture aliasing (which points to either no-use of mipmaps in current demos, or hw having similar issues with mipmap filters as Sony's machines).
It's also why people probably put DC as nicer looking in motion (much like with PSP).

[fehu]

someone suggested that AA in 3D mode will not be necessary because the screen blurs a little the image sending the two images

one thing that i don't understand is why in 3d mode the resolution is a quarter and not full vertical and middle horizontal
and if it will be possible to use the screen in full resolution

[Shifty Geezer]

someone suggested that AA in 3D mode will not be necessary because the screen blurs a little the image sending the two imagesSony have been saying AA is more important in 3D. I've no personal experience to form an opinion.

[idsn6]

someone suggested that AA in 3D mode will not be necessary because the screen blurs a little the image sending the two images

That doesn't sound correct to me.

Think of "temporal AA," i.e. jitter'n'blend. That breaks down entirely when the two images aren't essentially the same. With 3D, this is the case all the time.

[Exophase]

And the cool things about DS rasterizer (like being overdraw free) would make that much more difference with the speculated pixel-shading upgrades in 3DS.

DS's rasterizer isn't deferred/overdraw free, it's just scanline ordered (so you could call it tiled). The fillrate of 30M pixel/s is 10x the screen resolution @ 60Hz, so it's probably pretty hard to exhaust it, but it's possible.

[darkblu]

One thing that really stands out in all the realtime media is the pretty bad PSP/PS2 esque texture aliasing (which points to either no-use of mipmaps in current demos, or hw having similar issues with mipmap filters as Sony's machines).
It's also why people probably put DC as nicer looking in motion (much like with PSP).
Check out Kid Icarus footage - plenty of oblique surfaces there (low flight scenes) and they all seem using trilinear to me. The reason I noticed that is that my video streaming here sucks and I see plenty of 'stills'.

[Squeak]

Both NDS (just under 500KB) and PSP (2+2MB) had embeded VRam. In case of the former, the design wouldn't function without it at all. And in case of PSP, the amount is certainly enough to be "interesting". After watching the media I kinda got the impression as well. And the cool things about DS rasterizer (like being overdraw free) would make that much more difference with the speculated pixel-shading upgrades in 3DS.
Of course! That would fit perfectly with relatively simple geometry, to keep the buffersize for that down. It would also fit with the need to keep fillrate requirements down. They already need to render the scene twice, having two z-buffers would make matters much worse.
It would also negate the cost of having to include the DS graphics chip for BC.

Scanline and the similar technologies, is really the only sensible thing for portable hardware. Both with regards to resources required/saved but also coubled with the geometry sublety you can actually discern on such a small screen.

[DeadlyNinja]

Unlike reported by DF, Miyamoto confirms shaders.

"We have shaders that people are able to take advantage of [...]."

http://www.joystiq.com/2010/06/16/interview-nintendos-shigeru-miyamoto-on-3ds-retro-zelda-and/

Yeah, I really don't understand how DF could come to the conclusion that the 3DS doesn't have shaders after seeing MGS and RER. Apparently, both Capcom and Konami hadaround 2 months to make those demos. I can't confirm if that's true, though.

[Fafalada]

DS's rasterizer isn't deferred/overdraw free, it's just scanline ordered
The way I remember it, scanline rendering time is fixed, and the restriction to number of polygons you can fit inside one is the primitive setup time(after the buffer size), not their size. It's been awhile since I looked at the thing though, and I didn't exactly test this in practice.

Check out Kid Icarus footage - plenty of oblique surfaces there (low flight scenes) and they all seem using trilinear to me. The reason I noticed that is that my video streaming here sucks and I see plenty of 'stills'.
That's the video where I first noticed texture aliasing (on those very oblique surfaces) - it's in motion that pixel shimmering is the most obvious (especially with streaming compression, I doubt I'd catch much in stills).

[Exophase]

The way I remember it, scanline rendering time is fixed, and the restriction to number of polygons you can fit inside one is the primitive setup time(after the buffer size), not their size. It's been awhile since I looked at the thing though, and I didn't exactly test this in practice.

Physical scanline time is fixed, but the 3D rasterizer is not actually synchronized with the LCD. There's a 48-line FIFO between the two, which should indicate to you that the scanline render time isn't fixed. The time necessary has a setup component but is mostly proportional to the number of pixels in the span.

Bear in mind that DS games also make heavy use of alpha test without penalty. Especially with 2D games utilized with 3D.

[thop]

That's the video where I first noticed texture aliasing (on those very oblique surfaces) - it's in motion that pixel shimmering is the most obvious (especially with streaming compression, I doubt I'd catch much in stills).

A high quality video is available from Nintendo here. (http://E32010:nintendo@press.nintendo.com/E32010/3DS/3DS_KidIcarus.zip)

[darkblu]

That's the video where I first noticed texture aliasing (on those very oblique surfaces) - it's in motion that pixel shimmering is the most obvious (especially with streaming compression, I doubt I'd catch much in stills).
Of course one can't catch the shimmering in stills, but one can see the degradation of texture detail due to the anisotropy of the oblique sufrace projection, when used with isotropic sampling filters (i.e. bi/trilinear). As for the video advantage, low-quality motion-estimating algos can contribute their own shimmer due to the 'subtle planning movement we'd rather skimp' logic in many of those.

ps: mucho gracias, thop.

ed: After watching the high-quality QT verstion of the trailer from nitendo's press material, I'm willing to agree with Faf - there's apparent and predominant shimmer in most of the oblique cases, and those instances which I thought showed trilinear are narrowed down to a single instance (the flight over the green field), and it's not so apparent either. OTOH, other cases where trilinear would have been easilly spottend (cobble-stone tiled floors, etc) show no traces of such. So what is mostly at work in that trailer is either billinear, or cheap AF. All IMO, of course.

[Alucardx23]

Can someone tell me what’s the native resolution for the 3DS screen? 800×240 or 400x240, I mean is Nintendo giving the 800X240 resolution because of the 3D effect, or the screen is true 800X240 native.

[sfried]

Can someone tell me what’s the native resolution for the 3DS screen? 800×240 or 400x240, I mean is Nintendo giving the 800X240 resolution because of the 3D effect, or the screen is true 800X240 native.400 of those 800 pixels are for the left eye, while the other 400 are for the right. The parallax barrier each separates them, so while it might be 800 native, you are really only seeing 400 in a screenshot.

[Alucardx23]

400 of those 800 pixels are for the left eye, while the other 400 are for the right. The parallax barrier each separates them, so while it might be 800 native, you are really only seeing 400 in a screenshot.

Thanks a lot.

[sfried]

I believe I read an IGN article somewhere comparing the 3D effects of each of the demos, and specifically mentioned one of the titles (Paper Mario 3DS) either had it on or off, suggesting that the 3D is perhaps driver controlled.

[brain_stew]

I believe I read an IGN article somewhere comparing the 3D effects of each of the demos, and specifically mentioned one of the titles (Paper Mario 3DS) either had it on or off, suggesting that the 3D is perhaps driver controlled.

Its user controlled. There's a slider that you can use to increase the intensity/depth of the 3D effect or dial it all the way down until it is completely disabled. Doesn't work with movies (unsurprisingly) but all the game demos let you change it on the fly.

[sfried]

Its user controlled. There's a slider that you can use to increase the intensity/depth of the 3D effect or dial it all the way down until it is completely disabled. Doesn't work with movies (unsurprisingly) but all the game demos let you change it on the fly.What I meant to say was software controlled. Edit: As opposed to hardware/analog similar to how the volume controls are in the DSLite. Yes, there's a slider, but it still controls by the numbers.

[Nightz]

The Street Fighter 4 shots almost guarantees this machine is more powerful than the Wii or Xbox 1. It looks like a low screen resolution version of the console game.

However surprisingly the Nintendo games are low on shaders and effects - they look just like Wii games.

How adequate is be the 800x240 screen resolution for gaming and watching movies?

[Svensk Viking]

I have one question...I believe that 3D takes quite a bit extra power to render? Would it be possible for the developers to have a scaling system, where the visuals get better the more you turn off the 3D effect?

[Exophase]

I have one question...I believe that 3D takes quite a bit extra power to render? Would it be possible for the developers to have a scaling system, where the visuals get better the more you turn off the 3D effect?

The amount you pay for the effect would be all or nothing, not something that would scale with "how 3D" something is.

The Street Fighter 4 shots almost guarantees this machine is more powerful than the Wii or Xbox 1. It looks like a low screen resolution version of the console game.

However surprisingly the Nintendo games are low on shaders and effects - they look just like Wii games.

The first party games definitely don't like Wii level. To me they don't quite look Gamecube level either.

Are there more SF4 shots? Because you can't glean very much from the one posted in this thread, but from what I can see the level of geometric complexity and texture detail is on a different order of magnitude, this isn't just rendered at a lower resolution. But the complexity information is really more in the backgrounds than in the fighters, and if you look at this one you see a few low polygon models for chairs and tables, low resolution textures for table shadows, a person that's a cardboard cutout, and a background behind him that's a flat and blurry skybox. It's just all taken at an angle that obscures this.

The lighting and shadowing on the console versions also looks better. Really I'm not sure what's surprising people about this screenshot - it has edge marking and shadowing, but these are two features DS had in hardware w/o shaders.

How adequate is be the 800x240 screen resolution for gaming and watching movies?

Can someone who understands the technology used in the screen remark on if this really becomes an effective 800x240 once the 3D is turned off?

[Squeak]

Can someone who understands the technology used in the screen remark on if this really becomes an effective 800x240 once the 3D is turned off?
It does, at least physically. Whether Nintendo chooses to support it is another question.
The parallax barrier is really just a fine LCD grille raised a tiny bit from the screen. When it's turned off it just becomes invisible.

[Npl]

The amount you pay for the effect would be all or nothing, not something that would scale with "how 3D" something is.Well, you could render the background (beyond a fixed distance d) only once for both eyes.
Now you can vary this distance d to get a more or less complete "3D Effect". I have a hard time imagining how much the transition from "Background" to "3D" would be noticeable, if its way worse than just a LOD transition or pop-in (could be coupled with LOD on the same boundary distance) .

[brain_stew]

It does, at least physically. Whether Nintendo chooses to support it is another question.
The parallax barrier is really just a fine LCD grille raised a tiny bit from the screen. When it's turned off it just becomes invisible.


In one of the many interviews I've read (can't remember which one) it was confirmed that games don't have to use the 3D effect. I'd assume these games would be able to take advantage of the extra resolution, although I guess Nintendo could still restrict it if they want to.

[thop]

Apparently leaked specs from a chinese message board. I'd advise a grain of salt.

CPU ARM946E-S (67 MHz) + ARM1176 (TSMC65LP 225-482MHz)
GPU PowerVR SGX543MP2
RAM 16MB + 48MB

http://forum.tgbus.com/viewthread.php?tid=1162347&extra=&page=1

[ltcommander.data]

Apparently leaked specs from a chinese message board. I'd advise a grain of salt.

CPU ARM946E-S (67 MHz) + ARM1176 (TSMC65LP 225-482MHz)
GPU PowerVR SGX543MP2
RAM 16MB + 48MB

http://forum.tgbus.com/viewthread.php?tid=1162347&extra=&page=1
Wouldn't that combination be very CPU bound? I wouldn't be surprised by the use of the ARM11 instead of a newer Cortex core, but I always thought such use would be more like a 45nm shrink of the original Tegra operating at 600MHz+.

[Exophase]

Wouldn't that combination be very CPU bound? I wouldn't be surprised by the use of the ARM11 instead of a newer Cortex core, but I always thought such use would be more like a 45nm shrink of the original Tegra operating at 600MHz+.

If those specs are true a big factor will be whether or not the ARM11 has L2 cache. This is the real differentiator in performance between say, iPhone 3G and 3GS CPU performance, much more so than the increased issue width between ARM11 and Cortex-A8. The typical difference in performance between 256KB of L2 and no L2 for an ARM11 at ~500MHz can be as much as 2x.

Unfortunately people lambasted Tegra for using ARM11 despite having L2 cache, without really looking to make a fair comparison.

[roninja]

There's more chance of England winning the World Cup than the 3DS featuring a PowerVR XT core.

[darkblu]

Apparently leaked specs from a chinese message board. I'd advise a grain of salt.

CPU ARM946E-S (67 MHz) + ARM1176 (TSMC65LP 225-482MHz)
GPU PowerVR SGX543MP2
RAM 16MB + 48MB

http://forum.tgbus.com/viewthread.php?tid=1162347&extra=&page=1
So an ARM9 slower than DSi's (what else would that ARM9 be doing there except BC?) and dual core SGX54x? Why, yes, makes total sense!

[b3vcard]

There's more chance of England winning the World Cup than the 3DS featuring a PowerVR XT core.

Very, very good sir ! :cool:

[wsippel]

So an ARM9 slower than DSi's (what else would that ARM9 be doing there except BC?) and dual core SGX54x? Why, yes, makes total sense!
I don't know. What's the Starlet doing all day? ;)

[darkblu]

I don't know. What's the Starlet doing all day? ;)
No clue. Rumor has it something along keeping a lone gate sticking in a field of butterflies.

[thop]

So an ARM9 slower than DSi's (what else would that ARM9 be doing there except BC?) and dual core SGX54x?
ARM11 and ARM9 are backwards compatible.

[Teasy]

So an ARM9 slower than DSi's (what else would that ARM9 be doing there except BC?) and dual core SGX54x? Why, yes, makes total sense!

Wii also has a low power low frequency ARM9 CPU (Starlet), not for backwards compatibility but for security tasks (OS?) and connect 24. 3DS also has connect 24 this time around so this may not be as strange as it sounds. Dual GPU's would make sense for 3D I suppose.

[darkblu]

ARM11 and ARM9 are backwards compatible.
I'm not sure backward compatibility on an ISA level is sufficient to safely mimic a heterogeneous system (particularly one with a massive sw catalog), as there's no such 'mode' you could put the ARM11 in, that would be per-clock BC with the ARM9. Of course, it's all subject to trade-offs, of which nintendo have the final word.

[DeadlyNinja]

Apparently leaked specs from a chinese message board. I'd advise a grain of salt.

CPU ARM946E-S (67 MHz) + ARM1176 (TSMC65LP 225-482MHz)
GPU PowerVR SGX543MP2
RAM 16MB + 48MB

http://forum.tgbus.com/viewthread.php?tid=1162347&extra=&page=1

Suppose this is real, is that a good thing or a bad thing?

[MfA]

It would be an order of magnitude faster than what I'm fearing is in the 3DS (being CPU limited doesn't come into the picture, the SGX is >DX10 level, for games targeted at it it really doesn't need that beefy a CPU). So yes, it would be a good thing.

[brain_stew]

Suppose this is real, is that a good thing or a bad thing?

A very good thing, "half" of that GPU was at the upper end of my estimates. I don't think anyone was predicting that Nintendo would be going with such a beefy graphics solution.

[DeadlyNinja]

Sounds like hell for the battery. I vote fake then.

[brain_stew]

Sounds like hell for the battery. I vote fake then.

The battery featured in that spec. list is nearly twice the capacity of the one that shipped with the original NDS and the CPU is still pretty conservative. They should be able to meet their goal of matching DSi battery life with that sort of setup imo.

[Svensk Viking]

Exactly how much of a hit does the 3D take on the visuals?
What can we expect from a 3DS game not using 3D?

[DeadlyNinja]

Considering the Metal Gear Solid and Resident Evil stuff looks great and still in 3D, who knows. Maybe the darn thing really is more powerful than the Wii.

[Teasy]

So if that rumoured spec is right how does the SGX543MP2 compare to the SGX535 in the IPhone 3Gs?

[brain_stew]

So if that rumoured spec is right how does the SGX543MP2 compare to the SGX535 in the IPhone 3Gs?

Even the single core version is faster than the SGX535 but this is the dual core chip so it'll be more than twice as fast (at the same clock speed). Add to that the fact that developers can target one specific hardware setup and they'll have much lower level hardware access to it, there's no heavy OS to deal with and a lower resolution screen to target and a really large gap in real world performance begins to develop.

There's obviously the issue of the S3D overhead to deal with (which may restrict the vertex count in some games) but it should make for a little pixel shading monster. I'd be stunned if those specifications proved real.

[Exophase]

So if that rumoured spec is right how does the SGX543MP2 compare to the SGX535 in the IPhone 3Gs?

I think this is the progression:

SGX530 - 1x TMU, 2x USSE,
SGX535 - 2x TMU, 2x USSE, doubles triangle setup rate over 530
SGX540 - 2x TMU, 4x USSE
SGX543 - 2x TMU, 4x USSE2, doubles depth/stencil (hidden surface removal rate) over 540, does some other stuff to improve bandwidth and increases triangle setup rate another 50%

The earlier cores were part of Series5, and 543 is part of Series5XT, so I think they consider it a bigger core change despite only increasing the product number by 3 (maybe because the number of TMUs/ALUs stays the same)

Now take two SGX543, assuming it scales linearly (should be close-ish, IMG says they scale very well) and you have something that's at least 4x better than SGX535, much more so in shader heavy applications. But this isn't taking into consideration clock speed differences.

[MfA]

There was a big gain in IMG stock. There was the SGX544 announcement, but AFAICS it's just SGX543 with an improved VXD for stereoscopic HD decoding ... a 25% increase in share price based on that seems a bit steep.

[obonicus]

From http://prw.kyodonews.jp/open/release.do?r=201006189844 via Neogaf

3DS Uses DMP's PICA200 GPU (http://www.neogaf.com/forum/showpost.php?p=21992392&postcount=1)

[sfried]

Edit: Beaten to it.