Equate what you expect from the revolution with a graphics card

[Shifty Geezer]

I don't know, but it's not just a single texture. And you can store a 3D scene in a single height map; you're just limited to what you can store (I believe that's how Novalogic's old voxel engine stored data).

Only a continuous surface. That is, you can't ever get behind an object.

Imagine a cellar in a dungeon RPG. Traditionally there would be numerous objects, barrels and racks of bottles and stone walls and floors. These would be modelled in 3D, individually textured, and lit. You could take these objects and perform offline GI calculations and bake the results onto textures. Now a character placed in the middle of the room is only presented with an individual face of each object. In essence where they are can be mapped onto a cube with each of directions. You could take that cubemap and project it onto a literal cube and place the viewpoint in the middle, and looking around it would be similar to standing in the 3D room. You can even create the geometry and texture the whole lot with a cube-map projection.

But if you want to walk around the room, and stand behind a barrel that you previously stood infront of, the 2D cube mapping won't have information on what the back of the barrel looks like.

A cubemapping for 3D data (a series of 6 2D maps) cannot store true 3D data. Only 2.5D, with a heightfield for displacement. It can create a continuous surface with various undulations, but not a full 3D scene; it cannot store data where one face obscure another. What Nintendo's patent was talking about AFAIK was creating 3D scenes with a cubemap. I clearly remember it talking about offline raytracing a super detailed scene and being able to use this data in realtime environments. Perhaps they were just patenting the already existing idea of bakes light and texture maps (wouldn't be a first!) but the idea of a depth channel is just confusing.

Anyhow, like everything else it's already been discussed here before. If you want to continue the discussion it'd be worth searching for the previous debate (only is it here on this forum yet?) and carrying on from there.

 

[jvd]

For starters:
1. NIntendo wants to be the cheapest out there when they come out.
-What is cheapest, 249 dollar or 199 dollar? Somewhere between these two, maybe one of them. Having the core pack att 299 dollars, well, I think that NIntendo wants the Rev to be sold with more than 50 dollars in between.

Well first off if we look at a graphics board there are a few things involved .

1) Pcb cost 2) ram cost 3) gpu cost 4) cooler 5) packaging

If you factor in that next year the x1800xt will have been on the market for about a year and the mid end chips will already be in mass production . I would be that a x1800 chip would only cost around 30$ at that point in time .

Put that with a dual core cpu that costs 50$ . Your at 80$ another 40$ for 512 megs of ram and your at 120$ . STill have 80$ left for pcb cost , northbridge / south bridge costs , casing , cooling , controller

 

[Urian]

I am starting to believe that is a RV515 (ATI X1300) with eDRAM and clocked at 600Mhz.

Taking the X1300 Hypermemory but clocked at the same speed of ATI X1300 and connected to an UMA configuration of the system.

 

[EndR]

Well first off if we look at a graphics board there are a few things involved .

1) Pcb cost 2) ram cost 3) gpu cost 4) cooler 5) packaging

If you factor in that next year the x1800xt will have been on the market for about a year and the mid end chips will already be in mass production . I would be that a x1800 chip would only cost around 30$ at that point in time .

Put that with a dual core cpu that costs 50$ . Your at 80$ another 40$ for 512 megs of ram and your at 120$ . STill have 80$ left for pcb cost , northbridge / south bridge costs , casing , cooling , controller

It all depends on when NIntendo wants to release Rev. Spring release with Sony? Autum 2006? The X1800 at 90nm is, in my mind a rather big component to put into that tiny case of Rev. It is not very likely that Nintendo will be using 65nm either, but then. it depends on when Nintendo wants to release Rev.

The question on when Nintendo wants to release Rev is also an important factor. Now that Next gen machines are going to be a little more expensive at launch and that games are a little more expensive also (some will be) then it could have an effect that people maybe will feel comfortable with just one machine for a long time. If Nintendo releases late 2006, then they will probably try to become the premier secondary machine, depending on which one mainstream buy as primary, 360 or PS3, thus making it more important to be selling Rev with a profit..

If a X1800 will cost around 30 bucks next year, imagine what the X800 or X850 will be at. Margin-wise, the X800-X850 tech/margins wise is the better choice. Nintendo will have much better margins on the hardware and could be selling it with a very good profit, even while being the cheapest of the three.

NIntendo have said that Rev will be the "weakest" of the three but the IQ will be "similar" (on SD-res). The X800-X850 offers a good deal of power for SD-screens. Not only that, Nintendo could put "more" eDRAM on a chipset based on x800-x850 than on a X1800 and get more benefits there.

Like I said, these are just my speculations. I strongly believe that NIntendo would want to, at launch, sell the machine with as good margins as possible, because they know that they will not be the "first choice" for mainstream. Nintendos philosophy with Rev isn´t about power either, it is about how you play the game. The x800-x850 offers a good deal of power, very cheaply and therefore, Nintendo can offer other things instead, maybe a PPU there coupled with a one OoO-core...

Nintendo are very sensitive with their margins.. look at GB Micro, cost to manufacture, around 44 bucks, they sell it for 99 dollars and that little thing sells. With Rev, it is more important to have the margins on their side...

 

[darkblu]

ok, my take. i believe the cost factor is more determinant to the revolution options than the size factor. so depending on cost ranges:

• 200-250USD:
  IIRC i read at that gadgets costs site jvd provieded the link to that the cost of the entry level mac mini was ~290USD. now, if we assume that correct, that's
  • a decent (desktop flavor 1.25G .5MB cache, altivec) powerpc
  • 256MB of 333MHz (166 MHz DDR) ram
  • an rv280 (i.e. ati mobilit 9200)
  • 40GB 2.5" HDD
  • a slick slot dvd rom/cdrw
  • a 100Mb eth and a 56Kb modem
  in less than half the volume of the GC. now drop the HDD (a pricey little bugger), the cdrw-ness, the modem, make the eth wifi and fast-forward the GPU production technology to H2 of next year and you'll get a beefed up but hdd-less mac-mini-style machine with at least an rv370 (i.e. x300) -cass video, hitting the respective cost range with ease. with the fancy controller on top of that.
  
  
• 150-200USD:
  same as the above, sans the upgraded video and possibly with slower memory.
  
  
• 100-150USD:
  cpu, memory and optical drive configuration along the mac mini line but memory reduced to ~128MB, video is a beefed up (i.e. clock, edram amount and pipes) flipper.

well, something along those lines.

 

[Ooh-videogames]

As long as the console can run the UE3 engine, then I'll be a happy camper.

 

[see colon]

i'm pretty sure revolution will have a GPU that's not 100% equal to anything ati has in the PC space. given what we know (or what's been infered) about revolution....

-not as "powerfull" as ps3/x360
-small form factor
-silent or nearly silent
-backwards compatable with GC
-GPU engeneered by the same team who designed flipper (artX)

...i think it's safe to bet on a custom design. if the x360 is having some issues with backwards compatibility through software emulation even with it's "high" system specs (compared to what's been assumed for rev.), and nintendo infering that they'll have 100% compatability with GC i would assume it's a mostly hardware solution. thus, Rev. would need to support most if not all of the features flipper does in hardware (TEV, embedded framebuffer, and embedded texture cache being the biggest things IMO).

i expect revolution to sport a "super flipper", maybe 2-3 times the performance of flipper if you limit yourself to it's (flippers) feature set, but i expect them to add a programable geometry engine, and at least some programability in the pixel pipeline.

for some reason i expect revolution to have pretty good geometry performance. about on par with x360 or ps3, but with less features overall.

As long as the console can run the UE3 engine, then I'll be a happy camper.

i'd love to see UE3 running on revolution as well, but i think the first big step would be to have renderware running first. renderware has the biggest market share for console middleware, and now that EA own criterion they've started moving most if not all of their internal studios onto the platform. it's a good idea for them, since they make so many multiplatform games, and renderware is on everything (all 3 current gen consoles, PC, and even the ngage).

 

[pc999]

Wouldnt be better if we say x number of transistors , at y speed arranged as they wanted

Only by fun I would guess ~ 150M-160M transistors, at ~ 500mhz or around 70M at ~1,5 ghz if Fast14 is used (altought I dont know if this would generate more heat).

 

[jvd]

If a X1800 will cost around 30 bucks next year, imagine what the X800 or X850 will be at. Margin-wise, the X800-X850 tech/margins wise is the better choice. Nintendo will have much better margins on the hardware and could be selling it with a very good profit, even while being the cheapest of the three.

Nintendo will most likely go with sm3.0 . I don't see them going with sm2.0 as sm3.0 is supposed to be easier to program for .

The x1800 may be a little expensive but they can allways use less rops . at 640x480 a 12rop x1800 will be a pretty damn nice gpu for 480p .

For the cpu i'm seeing dual core .


I don't see why they can't make almost as powerfull as the xbox 360 almost a year later esp when you take out the hardrive

 

[PC-Engine]

I am starting to believe that is a RV515 (ATI X1300) with eDRAM and clocked at 600Mhz.

Taking the X1300 Hypermemory but clocked at the same speed of ATI X1300 and connected to an UMA configuration of the system.

Very nice... Though I think it's fairly underpowered with only 4PS and 2VS. I think the X1600 would be a good choice with 12PS and 5VS and only 160 million transistors. Maybe a custom version with 8PS + 8VS@600MHz since it's targeted at SDTV resolutions. That's roughly the same VS and PS power as PS3/Xbox 360 at 1/3 the resolution and about 1/2 the transistors.

Oh btw SM3.0 is even supported by the X1300.

 

[Urian]

Very nice... Though I think it's fairly underpowered with only 4PS and 2VS. I think the X1600 would be a good choice with 12PS and 5VS and only 160 million transistors. Maybe a custom version with 8PS + 8VS@600MHz since it's targeted at SDTV resolutions. That's roughly the same VS and PS power as PS3/Xbox 360 at 1/3 the resolution and about 1/2 the transistors.

Oh btw SM3.0 is even supported by the X1300.

I don´t know if the AVIVO is in the X1300 but we can put the AVIVO out for more Pixel and Vertex Shaders.

But I talked about the X1300 and not the X1600 because the chinese market is now an important market for Nintendo and iQue and iQue DS (Nintendo DS) are two great hits in China.

 

[PC-Engine]

I don´t know if the AVIVO is in the X1300 but we can put the AVIVO out for more Pixel and Vertex Shaders.

But I talked about the X1300 and not the X1600 because the chinese market is now an important market for Nintendo and iQue and iQue DS (Nintendo DS) are two great hits in China.

Yes it does. Check out the specs for the three GPUs.

http://www.ati.com/products/RadeonX1300/specs.html

http://www.ati.com/products/RadeonX1600/specs.html

http://www.ati.com/products/RadeonX1800/specs.html

 

[Urian]

Yes it does. Check out the specs for the three GPUs.

http://www.ati.com/products/RadeonX1300/specs.html

http://www.ati.com/products/RadeonX1600/specs.html

http://www.ati.com/products/RadeonX1800/specs.html

Good.

In other words, we can get a more powerful GPU than X1300 Pro taking out the AVIVO from the core and adding more Shader units.

The Flipper has 25 milions transistors in the core and 26 milion transistors in the eDRAM.

The X1300 has 100 milions transistors in the core with AVIVO included and in 90nm process we can put 204 milions transistors in the same area of the entire Flipper.

In other words, 100 milions transistors for the memory and 100 milion transistors for the eDRAM but in this scenario the eDRAM will be 12MB but we can reduce it to 8MB and adding the transistor count to the main core for adding more Vertex and Pixel Shaders.

 

[pc999]

Why so many transistores in/edram , even a fuction "like Xenus" that would be too big as you would need much less than XB360 (1/2?, about 5MB).These way they would be able to put more ALUs or whatever they want and in the same die which would give more BW and reduce costs.

 

[Shifty Geezer]

Why so many transistores in/edram , even a fuction "like Xenus" that would be too big as you would need much less than XB360 (1/2?, about 5MB).These way they would be able to put more ALUs or whatever they want and in the same die which would give more BW and reduce costs.

If they want to avoid tile rendering they could whack on as much as needed for full floating point HDR buiffers.

 

[pc999]

If they want to avoid tile rendering they could whack on as much as needed for full floating point HDR buiffers.

Why would they want to avoid tile rendering , it seems to have a great result in Xenus, plus they could get higher BW if they put it on the same die, and it seems that the last gen of gfx chips (both X1xxx and xenus) o have great performance with HDR so I do not see not using it.

 

[Shifty Geezer]

Why would they want to avoid tile rendering.

Dunno. It's something KK wanted to avoid though. Or at least he said they didn't want eDRAM if they couldn't fit a whole buffer in.

 

[see colon]

i'm still going with my "it's not going to be a PC part" theory. all modern ati graphics parts (everything from the rage 128 up) have what amounts to almost full hardware DVD playback support. if nintendo is making you buy an add-on to play back DVD content, whet would the add on be for? the controller is already a familiar remote shape and wireless and hardware support for playback would be in the box.

 

[pc999]

Dunno. It's something KK wanted to avoid though. Or at least he said they didn't want eDRAM if they couldn't fit a whole buffer in.

Didnt know that , anyway if it is the second 5MB should be enought and it would be ~40M of transistores, so using Urian maths they should still have about 164M to spend in logic (good for the theory of a X1600 ).

i'm still going with my "it's not going to be a PC part" theory.

I do think too that they will not use a PC part, at least not if it isnt a very modified/costum one, they are used to using their own ideas for their own targets, like fliper wich is pretty good at their job when a game is made to it, we all love to see a 60 FPS MP.

 

[Shifty Geezer]

if nintendo is making you buy an add-on to play back DVD content, whet would the add on be for?

I believe the optional extra playback is to cover the licensing fees. Like XB, the inclusion of DVD playback seems rather daft as most people will already have a DVD player by now, and so MS losing $20 or whatever it is, or charging an extra $20 licensing fee per console for a feature most people won't want is pointless extra cost. If like XB you move the license fee onto a peripheral, only those who actually want that feature have to pay for it.