Well Doom 3 will have no less then 6 texture layers and upto 8, Carmack has already said that, also on price I think allot of people are thinking the Kyro II is going to be allot more expensive then it really is, the 64mb Herc 4500 has a recommended price of $149 but try looking for the recommended price of a 64mb MX and you'll see its similar or more, over here in the U.K were prices for the Herc 4500 64mb are already in a lot of online stores, its listed as £110 inc VAT with free delivery in most cases so I expect $110 or even less in the U.S, then if you look here in the U.K for a Kyro II 32mb you'll see the Videologic Vivid!XS 32mb TV-Out and this card is listed at only £85 inc VAT with free delivery, the cheapest 32mb MX I can find here in the U.K is £75 and thats without Tv-Out so the Kyro II looks to have the same price as the MX in both 32mb and 64mb configurations and there's really no comparison in performance between the Kyro II and MX, the Kyro II dominates the MX, I expect the 32mb Kyro II's to sell in the U.S for $85 or $90 at most, and if they sell some non Tv-Out 32mb Kyro II's over in the U.S we could see some $80 Kyro II's over there, usually things work on a 1-1 ratio with dollars and pounds even though it shouldn't be like that really because a pound is worth about 1.4 dollars, but for some reason if something sells for say £80 in the U.K you can usually find it for $80 or less in the U.S, which partly the fualt of VAT.
noko
Distinguished
- Jan 22, 2001
- 2,413
- 1
- 19,785
Excellent, excellent!!! Maybe I will buy a KyroII and a cheap pci MX or other Linux capable card for my second machine and rip out the pci card as soon as Linux drivers are available for the Kyro. Still I would want to keep it simple. There is no competition for the Kyro and the MX. The mx is dead and should be a sub $50 card as far as I see it. Now DoomIII is slotted for next year right? Doesn't it use the programmable vertex shaders and pixel shaders? Meaning KyroII will be out to lunch on that game regardless of 8 texture capability as well as many other cards.
noko
Distinguished
- Jan 22, 2001
- 2,413
- 1
- 19,785
Do believe that the KyroII is the killer of nVidia MX line of cards but not nVidia. Will nVidia and ATI convert over to a Tile Base Rendener chip? I believe it depends on how successful the KyroII/III is. Looks like the KyroII cost will always be lower than the Radeon or GF line of cards so cheaper for us to buy, cheaper to manufacture but yet more profitable to sell. Good times are ahead indeed.
No I don't think Doom3 will be using pixel shaders actually or if it is it'll only be in a small way and won't be neccesary to play the game, Carmack said he was more impressed with the Vertex shaders in DX8 so I think thats what he'll be using, and also obviously they can be done in software so the Kyro II or MX or any card can play a game with vertex shaders thats probably the other reason he's using them, if he used pixel shaders allot then only the Geforce 3 would be able to play the game, infact I think thats why he's using so many texture layers too, instead of pixel shaders, you can make similar pixel shader style effects with multi-texturing and also every card can use this, even though most cards will be forced into lots of passes, still having to do lots of passes and at least being able to play the game is better then not being able to play at all when pixel shaders are used.
I'm hearing from allot of people close to the industry that Nvidia would rather eat there own crap then make a tile based renderer, the reason for this seems to be that they don't want to use 3dfx's tech to bail themselves out of the memory bandwith hole they find themsevles in, there too proud for that, they'll move onto QDR next which will ensure an incredibly expensive card from Nvidia, PowerVR will move onto cheap DDR and still have loads more bandwidth then Nvidia's QDR cards, after QDR I don't think Nvidia have anywhere to go, its either a different design or bust, they can't keep bolting on faster and faster ram, I don't think the Kyro II will kill Nvidia that would be impossible but I certainly think the MX is now useless with the Kyro II being just as cheap to manufacture (if not cheaper, hercules said thats why there using Kyro II because its cheaper to make then a MX and much much faster) and a far supperior card, the only thing it doesn't have that the MX has is HW T&L and I think everyone knows that were it matters (at least 1024x768) HW T&L can't help the MX, the Kyro just has to much fillrate and memory bandwidth for the MX, well infact it has less fillrate and the same memory bandwidth but it uses those resources so much more efficiently then the MX that it seems as if it has more. Appart from HW T&L the Kyro actually has allot more features then the MX, I think the Kyro II will also take allot of GTS sales away, in the U.K the GTS can't be found for under £150 for a £32mb version (I can't even find one for that price) and the Kyro II 32mb is £85 with TV-out and comparable performance (better at high res or FSAA)
<P ID="edit"><FONT SIZE=-1><EM>Edited by Teasy on 04/14/01 12:24 PM.</EM></FONT></P>
I'm hearing from allot of people close to the industry that Nvidia would rather eat there own crap then make a tile based renderer, the reason for this seems to be that they don't want to use 3dfx's tech to bail themselves out of the memory bandwith hole they find themsevles in, there too proud for that, they'll move onto QDR next which will ensure an incredibly expensive card from Nvidia, PowerVR will move onto cheap DDR and still have loads more bandwidth then Nvidia's QDR cards, after QDR I don't think Nvidia have anywhere to go, its either a different design or bust, they can't keep bolting on faster and faster ram, I don't think the Kyro II will kill Nvidia that would be impossible but I certainly think the MX is now useless with the Kyro II being just as cheap to manufacture (if not cheaper, hercules said thats why there using Kyro II because its cheaper to make then a MX and much much faster) and a far supperior card, the only thing it doesn't have that the MX has is HW T&L and I think everyone knows that were it matters (at least 1024x768) HW T&L can't help the MX, the Kyro just has to much fillrate and memory bandwidth for the MX, well infact it has less fillrate and the same memory bandwidth but it uses those resources so much more efficiently then the MX that it seems as if it has more. Appart from HW T&L the Kyro actually has allot more features then the MX, I think the Kyro II will also take allot of GTS sales away, in the U.K the GTS can't be found for under £150 for a £32mb version (I can't even find one for that price) and the Kyro II 32mb is £85 with TV-out and comparable performance (better at high res or FSAA)
<P ID="edit"><FONT SIZE=-1><EM>Edited by Teasy on 04/14/01 12:24 PM.</EM></FONT></P>
noko
Distinguished
- Jan 22, 2001
- 2,413
- 1
- 19,785
How about 3d textures? Which is a spec for OpenGL1.2 which the Radeon supports in hardware. Is Carmack using OpenGL or DX8 for Doom3? Anyways the Radeon 3d texture is used in the Ark Demo which is on the installation disk. A very impressive demo in which the Radeon Girl has eyes that relistically move (3d texture). Anyways does KyroII support 3d textures and is it fully OpenGL1.2 compliant?
noko
Distinguished
- Jan 22, 2001
- 2,413
- 1
- 19,785
Well nVidia may have too if the market dictates that it does. ATI as well. It is becoming apparent that Pixel and Vertex Shaders, T&L engine and other features on the ATI and nVidia cards are there to conserve bandwidth because of OverDraw. So the problem is being worked upon by both companies in a rather round about way. ATI expects to improve significantly on HyperZ technology meaning perserving there R&D and manufacturing money on previous designs on there next chipset. Still it is kinda funny that a 15 million transitor chip can outdo the competition in a number of benchmarks, I love it. It is becoming apparent who is working the problem of overdraw in a more efficient manner. ArtX which ATI acquired also which designed the GameCube chip for Nintendo uses a graphic chip with onboard ram. Much faster solution then DDR ram alone, maybe like the cache on a typical processor. This would allow the graphic chip to break away from the memory limitations since most of the caculations needed can be done on the GPU without waiting for slower DDR ram to catch up. Once again it would perserve the status quo and allow the GPU to run at 300mhz and up with no clear limitation in sight. So now cost becomes the chooser between these technologies in the market place. Looks like it will be hard to beat the KyroII extremely low cost but these advance features could topple the tile base rendering solution. Unless these same features maybe incorporated into the Kyro chips. I just wonder if STmicro/Imagination tech will have the R&D money necessary to incorporate this new technology.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/14/01 04:52 PM.</EM></FONT></P>
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/14/01 04:52 PM.</EM></FONT></P>
Teasy... great posts !!!!
I would like to know what Holygrenade,Warden and Negaverse think about those “new” issues that you have raised...
If you have changed your mind I don't blame you !!!
😉
I would like to know what Holygrenade,Warden and Negaverse think about those “new” issues that you have raised...
If you have changed your mind I don't blame you !!!
😉
<<<<<<It is becoming apparent that Pixel and Vertex Shaders, T&L engine and other features on the ATI and nVidia cards are there to conserve bandwidth because of OverDraw.>>>>>>
I don't understand what your saying there because those feature really have nothing to do with conserving bandwidth unless I'm missing something?
Yeah ATI and Nvidia could use a large amount of on-chip ram (the Kyro II already uses on-chip ram, though only a small amount), and this could solve there bandwidtn problems but there still playing catch up, the Kyro II gets 100% of its theoretical fillrate even at 1024x768x32 because its just not bandwidth limited at all until you go into much higher res, in FSAA mode its not limited by bandwidth in the slightest, so the Kyro II has already solved the bandwidth problem, Nvidia may try on-chip ram like the GameCube but will they get it right first time?, even if they do that still leaves them having to draw 3 times the amount of pixels in todays games then the Kyro II to get the same effect which is a massive waste of fillrate, games coming out now are starting to have more like a 4-5 overdraw, which means that a normal card will have to render 5 times the amount of pixels then the Kyro II which is an even bigger waste of fillrate, Nvidia may try a large amount of on-chip ram and deferred rendering but not tile based rendering, as in they render the entire scene in an on-chip framebuffer (the Kyro II does this but 1 tile at a time) and do HSF before rendering but if they use this complex design how many products will it take then to get all the bugs out? since the start Nvidia has been using a traditional design, can they just change to a radically different design without any glitches?, its taken IMGTEC about 5 years to finally release a tile based renderer that has no problems (they've release loads of tile based rendering cards but they all had glitches until Kyro), so I don't think Nvidia can just smoothly move onto a totally different design without a hitch.
For an indication at how well a Kyro II uses its available bandwidth I'll put some numbers up, the Kyro II uses SDR ram running at 175mhz with a 128bit mem bus, thats a mem bandwidth of 2.8gb/s, the GTS uses DDR ram running at 200mhz with a 128bit mem bus, thats a mem bandwidth of 6.4bg/s, so the GTS has more then double the memory bandwidth of the Kyro II, yet look at the benchmarks on any site and the Kyro II beats the GTS when memory bandwidth becomes the limiting factor at high res or with FSAA, so the Kyro II somehow has more memory bandidth in real world tests even though the GTS actually has more then double the amount of physical memory bandwidth then the Kyro II, now I think everyone can agree that is cool😱), its incredible to think that the Kyro II must be the only new high performance card that really wouldn't benefit from using DDR ram because its SDR ram already allows its fillrate to be the cards limiting factor, if Nvidia use QDR on there next card then since there current cards are already so limited by there bandwidth with DDR ram even if they doubled the cards fillrate the QDR ram wouldn't be good enough to see much of that extra fillrate, imagine how many pixel pipes the Kyro II design could use with QDR ram and still not be memory bandwidth limited, the Kyro II design could have 8 pixel pipes and still not be at all bandwidth limited with QDR ram (based on the fact that its not bandwidth limited with 2 pixel pipes and SDR ram and that QDR ram is 4 times the speed of SDR ram), now imagine the card this would be, 8 pixel pipe Kyro II running at 250mhz with 250mhz QDR ram, it would have a fillrate of 2000mpixels/s (the current Kyro II has a fillrate of 350mpixels/s), and it would be able to use all of that 2000mpixels/s because of the QDR ram and tile based rendering, so the Kyro II design has the potential to be able to be 6 times faster then the current Kyro II (and I mean truly 6 times faster because its not bandwidth limited) with technology that can be made today, the GTS could be maybe 2 times the speed it is right now if it used the same number of pipes and QDR ram, why because even with 8 pipes its architecture wouldn't allow it to use more them 4 of those pipes without being totally and unterly memory bandwidth limited, I'm not saying IMGTEC are going to do this but I'm just trying to show how IMGTEC's design still has loads of space to grow, while Nvidia's architecture has hit a brick wall.
I'm not sure about 3d textures, I don't think the current Kyro II supports it, on what Carmack is going to use for Doom3 I don't really know but he is always talking about the Dx8 vertex shaders so.......
I don't understand what your saying there because those feature really have nothing to do with conserving bandwidth unless I'm missing something?
Yeah ATI and Nvidia could use a large amount of on-chip ram (the Kyro II already uses on-chip ram, though only a small amount), and this could solve there bandwidtn problems but there still playing catch up, the Kyro II gets 100% of its theoretical fillrate even at 1024x768x32 because its just not bandwidth limited at all until you go into much higher res, in FSAA mode its not limited by bandwidth in the slightest, so the Kyro II has already solved the bandwidth problem, Nvidia may try on-chip ram like the GameCube but will they get it right first time?, even if they do that still leaves them having to draw 3 times the amount of pixels in todays games then the Kyro II to get the same effect which is a massive waste of fillrate, games coming out now are starting to have more like a 4-5 overdraw, which means that a normal card will have to render 5 times the amount of pixels then the Kyro II which is an even bigger waste of fillrate, Nvidia may try a large amount of on-chip ram and deferred rendering but not tile based rendering, as in they render the entire scene in an on-chip framebuffer (the Kyro II does this but 1 tile at a time) and do HSF before rendering but if they use this complex design how many products will it take then to get all the bugs out? since the start Nvidia has been using a traditional design, can they just change to a radically different design without any glitches?, its taken IMGTEC about 5 years to finally release a tile based renderer that has no problems (they've release loads of tile based rendering cards but they all had glitches until Kyro), so I don't think Nvidia can just smoothly move onto a totally different design without a hitch.
For an indication at how well a Kyro II uses its available bandwidth I'll put some numbers up, the Kyro II uses SDR ram running at 175mhz with a 128bit mem bus, thats a mem bandwidth of 2.8gb/s, the GTS uses DDR ram running at 200mhz with a 128bit mem bus, thats a mem bandwidth of 6.4bg/s, so the GTS has more then double the memory bandwidth of the Kyro II, yet look at the benchmarks on any site and the Kyro II beats the GTS when memory bandwidth becomes the limiting factor at high res or with FSAA, so the Kyro II somehow has more memory bandidth in real world tests even though the GTS actually has more then double the amount of physical memory bandwidth then the Kyro II, now I think everyone can agree that is cool😱), its incredible to think that the Kyro II must be the only new high performance card that really wouldn't benefit from using DDR ram because its SDR ram already allows its fillrate to be the cards limiting factor, if Nvidia use QDR on there next card then since there current cards are already so limited by there bandwidth with DDR ram even if they doubled the cards fillrate the QDR ram wouldn't be good enough to see much of that extra fillrate, imagine how many pixel pipes the Kyro II design could use with QDR ram and still not be memory bandwidth limited, the Kyro II design could have 8 pixel pipes and still not be at all bandwidth limited with QDR ram (based on the fact that its not bandwidth limited with 2 pixel pipes and SDR ram and that QDR ram is 4 times the speed of SDR ram), now imagine the card this would be, 8 pixel pipe Kyro II running at 250mhz with 250mhz QDR ram, it would have a fillrate of 2000mpixels/s (the current Kyro II has a fillrate of 350mpixels/s), and it would be able to use all of that 2000mpixels/s because of the QDR ram and tile based rendering, so the Kyro II design has the potential to be able to be 6 times faster then the current Kyro II (and I mean truly 6 times faster because its not bandwidth limited) with technology that can be made today, the GTS could be maybe 2 times the speed it is right now if it used the same number of pipes and QDR ram, why because even with 8 pipes its architecture wouldn't allow it to use more them 4 of those pipes without being totally and unterly memory bandwidth limited, I'm not saying IMGTEC are going to do this but I'm just trying to show how IMGTEC's design still has loads of space to grow, while Nvidia's architecture has hit a brick wall.
I'm not sure about 3d textures, I don't think the current Kyro II supports it, on what Carmack is going to use for Doom3 I don't really know but he is always talking about the Dx8 vertex shaders so.......
noko
Distinguished
- Jan 22, 2001
- 2,413
- 1
- 19,785
Using pixel shaders means less need to use multifple textures, which means you don't have to load as many textures (saving ram space) plus when you do have overdraw your dealing with less textures rendering on a pixel. I did say it is a round about solution.
Using vertex shaders means less loading of polygons from cpu to DDR ram to graphics chip for animations of polygons. Meaning a saving on bandwidth of both system resources and graphics card memory.
No reason why programmable vertex shaders can't be used in OpenGL all nVidia has to do is to have an OpenGL extension for them. Nvidia has many custom OpenGL extensions which developers can use to enhance their games.
<b>Now looking at AnAndTech benchmarks on Fillrates has a shocking reality: 😱 </b>
<A HREF="http://www.anandtech.com/showdoc.html?i=1435&p=13" target="_new">http://www.anandtech.com/showdoc.html?i=1435&p=13</A>
1. On the <b>Single Texture Test</b> the Kyro did surprisenly well compared to the competition on measureing MPixels/sec. It had 352.89 MP/s. The Radeon had only 280.33 while the Ultra just squeek by with 374.95 MP/s.
2. Now doubling the textures to just two, <b>Multitexture Fillrate Test</b> the Kyro jumps down to 175.46MP/s while the Radeon goes down a little to 269.09MP/s. In fact even the Radeon SDR beats out the Kyro in fillrate when using two textures at 177.02MP/s!!! How about 3 textures which the Radeon supports in hardware?
I hate to think what would happen if you throw 8 textures per pixel at a KyroII card on fill rates. Maybe I am missing something here so please fill me in and set me straight. Why did the Kyro drop down to the near bottom when the textures where doubled? that is in fill rates. Why 1/2 the previous fillrate from 1 to 2 textures? would this mean going from 1 texture to 8 textures it would drop down to 1/16 of its 1 texture fillrate? 22MP/s fillrate!!! With that kind of fillrate a commodore 64 may even look good
(I am exagerating here). Please explain. Something is not right here.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/14/01 03:15 PM.</EM></FONT></P>
Using vertex shaders means less loading of polygons from cpu to DDR ram to graphics chip for animations of polygons. Meaning a saving on bandwidth of both system resources and graphics card memory.
No reason why programmable vertex shaders can't be used in OpenGL all nVidia has to do is to have an OpenGL extension for them. Nvidia has many custom OpenGL extensions which developers can use to enhance their games.
<b>Now looking at AnAndTech benchmarks on Fillrates has a shocking reality: 😱 </b>
<A HREF="http://www.anandtech.com/showdoc.html?i=1435&p=13" target="_new">http://www.anandtech.com/showdoc.html?i=1435&p=13</A>
1. On the <b>Single Texture Test</b> the Kyro did surprisenly well compared to the competition on measureing MPixels/sec. It had 352.89 MP/s. The Radeon had only 280.33 while the Ultra just squeek by with 374.95 MP/s.
2. Now doubling the textures to just two, <b>Multitexture Fillrate Test</b> the Kyro jumps down to 175.46MP/s while the Radeon goes down a little to 269.09MP/s. In fact even the Radeon SDR beats out the Kyro in fillrate when using two textures at 177.02MP/s!!! How about 3 textures which the Radeon supports in hardware?
I hate to think what would happen if you throw 8 textures per pixel at a KyroII card on fill rates. Maybe I am missing something here so please fill me in and set me straight. Why did the Kyro drop down to the near bottom when the textures where doubled? that is in fill rates. Why 1/2 the previous fillrate from 1 to 2 textures? would this mean going from 1 texture to 8 textures it would drop down to 1/16 of its 1 texture fillrate? 22MP/s fillrate!!! With that kind of fillrate a commodore 64 may even look good
(I am exagerating here). Please explain. Something is not right here.<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/14/01 03:15 PM.</EM></FONT></P>
noko
Distinguished
- Jan 22, 2001
- 2,413
- 1
- 19,785
Right now orders can be taken for the KyroII card Hercules 3D Prophet 4500 64MB w/o TV-OUT OEM for $149, with TV-OUT it is going for $159. The Kyro board from Hercules 3D Prophet 4000 32MB w/o TV-OUT is going for $89, with $99.
<A HREF="http://207.181.248.118/" target="_new">http://207.181.248.118/</A>
These are the cheapest prices on pricewatch.com for the above listed cards. Which means the Kyro2 will initially be competing against GF2/pro and Radeon 64. The Kyro 3d Prophet 4000 will be competing against the Radeon LE and MX.
Obviously prices will come down for both chipsets in a couple of months but then again so will the Radeons and GF2s.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/14/01 04:21 PM.</EM></FONT></P>
<A HREF="http://207.181.248.118/" target="_new">http://207.181.248.118/</A>
These are the cheapest prices on pricewatch.com for the above listed cards. Which means the Kyro2 will initially be competing against GF2/pro and Radeon 64. The Kyro 3d Prophet 4000 will be competing against the Radeon LE and MX.
Obviously prices will come down for both chipsets in a couple of months but then again so will the Radeons and GF2s.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/14/01 04:21 PM.</EM></FONT></P>
think hat Teasy could explain this better, but I think that it's like this:
kyro have 2 pipelines so 2x175=350 theoretical fill rate in single textures (the true fill rate is higher than it's theoretical, I wonder why) 😉
geforce and radeon have 4 pipelines they should do better yet they didn´t (they should do 4 times the clock of geforce and radeon)
if we went to 2 textures then yes we will have 175 theoretical fill rate on kyro 2 (and 4 times the clock of geforce/radeon, they have 2/3 TMU per pipeline)
if we went to 4 textures this geforces and radeons unlike kyro will drop fast on their true fill rate why because they are optimized to 2 textures but theoretically they should do better than kyro 2 ...(they are clocked at higher frequencies that is why...and have 4 pipelines)
we will get only 90 theoretically for kyro 2(we already now that this theoretically means real
what 8 layer pass support means is that those textures will be loaded only once and get blended on a local (inside the chip) tile buffer .
that happens with kyro with 3,4,..8 textures unlike the others... (they have to load the textures more than only one time, I may be wrong on these please enlight us Teasy 😉 )
If you saw the benchmark of that game (serious sam) you would see who wins with 3 or 4 textures layers…
😉
It's because of this that the imagination guys are giving us that quake 3 level with 4 textures layers to show off their's kyro ....
😉
sorry about my english
<P ID="edit"><FONT SIZE=-1><EM>Edited by powervr2 on 04/14/01 10:49 PM.</EM></FONT></P>
kyro have 2 pipelines so 2x175=350 theoretical fill rate in single textures (the true fill rate is higher than it's theoretical, I wonder why) 😉
geforce and radeon have 4 pipelines they should do better yet they didn´t (they should do 4 times the clock of geforce and radeon)
if we went to 2 textures then yes we will have 175 theoretical fill rate on kyro 2 (and 4 times the clock of geforce/radeon, they have 2/3 TMU per pipeline)
if we went to 4 textures this geforces and radeons unlike kyro will drop fast on their true fill rate why because they are optimized to 2 textures but theoretically they should do better than kyro 2 ...(they are clocked at higher frequencies that is why...and have 4 pipelines)
we will get only 90 theoretically for kyro 2(we already now that this theoretically means real
what 8 layer pass support means is that those textures will be loaded only once and get blended on a local (inside the chip) tile buffer .
that happens with kyro with 3,4,..8 textures unlike the others... (they have to load the textures more than only one time, I may be wrong on these please enlight us Teasy 😉 )
If you saw the benchmark of that game (serious sam) you would see who wins with 3 or 4 textures layers…
😉
It's because of this that the imagination guys are giving us that quake 3 level with 4 textures layers to show off their's kyro ....
😉
sorry about my english
<P ID="edit"><FONT SIZE=-1><EM>Edited by powervr2 on 04/14/01 10:49 PM.</EM></FONT></P>
hum...
let see something...
in a 4 layer texture game
if we get 90 million pixel per second what can these kyro card do ???
let do some math:
in a 1024x768 game that should be enough to a 114.441 fps game
90000000/(1024x768)=114.441 of course this is theoretical
😉
90000000/(1600x1200)=46 fps that is enough for me...
there will be other factors that would slow down the game... and the card must do others things as well, but we are talking about a smart kyro that is more efficient than the rest of the bunch
😉
<P ID="edit"><FONT SIZE=-1><EM>Edited by powervr2 on 04/14/01 05:59 PM.</EM></FONT></P>
let see something...
in a 4 layer texture game
if we get 90 million pixel per second what can these kyro card do ???
let do some math:
in a 1024x768 game that should be enough to a 114.441 fps game
90000000/(1024x768)=114.441 of course this is theoretical
😉
90000000/(1600x1200)=46 fps that is enough for me...
there will be other factors that would slow down the game... and the card must do others things as well, but we are talking about a smart kyro that is more efficient than the rest of the bunch
😉
<P ID="edit"><FONT SIZE=-1><EM>Edited by powervr2 on 04/14/01 05:59 PM.</EM></FONT></P>
Yes it has to lose fillrate with more then 1 texture layer because it takes more cycles to add more texture layers, it only has 1 TMU per pipe, thats why for 2 texture layers it loses half its fillrate, the other cards there don't lose half there fillrate when doing 2 textures because they have 2 TMU's per pipe, but anymore then 2 textures and all those cards fillrate will drop in half and the memory badnwidth will more then drop in half because it needs 2 passes with more then 2 textures which wastes allot of mem bandwidth. The Kyro doing 8 texture layers will need 8 cycles so its fillrate will drop 8 times to 44mpixels/s (352mpixels/s taken from that Anand single texturing test divided by 8), but remember this is after the 8 texture are taken care of so to think of the performance you'd get you have to think of it as if its a 44mpixel/s card playing a single texture game (and also it doesn't render hidden pixel too so 44mpixels/s is more then it looks), but also you have to remember that it won't need to pass the textures to the frambuffer 8 times and it won't have to reload or retransform the polys 8 times, it'll only have to do all that once even with 8 texture layers, so its bandwidth won't drop anywhere near as much as the fillrate, now with say an MX your looking at 88mpixels/s theoretically because it has 2 TMU's per pipe (its fillrate drops 4 times from the single texture score of 142mpixel/s in that Anand test you posted), but in reality since it has to make 4 passes to the framebuffer for each pixel and has to reupload the polys 4 times its already limited bandwidth will die completely, its bandwidth will drop faster then its fillrate, its bandwidth would already have droped 8 times because of 8 times more textures, now take 4 passes for every pixel and 4 times the poly bandwidth needed and its bandwidth would be completely gone, in that single texturing test the MX only gets about 40% of its theoretical peak fillrate, now thats single texturing with only 1 texture layer 1 pass and no poly reloading so imagine how much of its fillrate it'll actually be able to use after all those extra passes and after having to reload the polys 4 times, I'd be supprised if it gets 10% of its fillrate in these circumstances (when I say fillrate I mean the 88mpixel/s fillrate it would be able to reach with the normal 1 pass in single texturing devided by 8 times the amount of textures not its peak theoretical fillrate), meaning it'll be able to use about 11mpixels/s of its fillrate, yes this is just a very very rough estimate, there's no way I can sit down and work out exactly how efficient its bandwidth will be, it may be able to use 22mpixel/s fillrate, maybe 33mpixels/s but I would say that being optimistic considering the terrible mem bandwidth problems it has even in single texturing with only 1 pass and with all those extra passes its bandwidth will be far far less efficient then it is in a single or dual texturing tests (40% efficiency in the Anand test). One of the reasons why the Kyro II is so dominant in the Serious Sam performance tests in that Anand review is because Serious Sam uses 4 texture layers (it only uses 1 and 2 texture layers in the fillrate tests though), now that theoretically should take the Kyro II's fillrate down to 88mpixels/s in Serious Sam (352 devided by 4 texture layers), but look at those fps results, the Kyro II gets 61.5fps at 1024x768x32 while the MX is getting 31fps, now add 4 more texture layers to that (yes I know this is simplistic), and you'd get something like this: Kyro II 31fps, MX 16fps, GTS 24fps, Ultra 29fps, so the Kyro II won't be super fast in a 8 texture layer game but it should be playable, while with the MX its unplayable, again this is all estimates using the benchmarks we have now but at the moment thats the best I can do, phew thank good thats over😱), it was tough writing that, too much thinking, I need a drink😱), I hope you can understand what I was trying to say.
And of course all of this is fotgetting that if you make multiple passes to the framebuffer your going to get horrible colour innacuracies, so forget 16bit colour on the GTS with games that use lots of texture layers, while on the Kyro II you could actually use 16bit and it'll be just as nice looking as it is in a single texture games (it looks almost as nice as 32bit colour), and I'm pretty sure that even 32bit colour will deteriate if loads of passes are needed, like 4 passes if the GTS was playing a 8 texture layered game.
BTW incase you didn't know the reason the Kyro II can do 8 texture layers in 1 pass it because it keeps the tile its working on inside the chip until its finished adding everything to the tile, so after its added all 8 textures only then does it send the tile to ram once so appart from having 8 times the amount of textures the memory bandwidth it uneffected with the Kyro II.
And of course all of this is fotgetting that if you make multiple passes to the framebuffer your going to get horrible colour innacuracies, so forget 16bit colour on the GTS with games that use lots of texture layers, while on the Kyro II you could actually use 16bit and it'll be just as nice looking as it is in a single texture games (it looks almost as nice as 32bit colour), and I'm pretty sure that even 32bit colour will deteriate if loads of passes are needed, like 4 passes if the GTS was playing a 8 texture layered game.
BTW incase you didn't know the reason the Kyro II can do 8 texture layers in 1 pass it because it keeps the tile its working on inside the chip until its finished adding everything to the tile, so after its added all 8 textures only then does it send the tile to ram once so appart from having 8 times the amount of textures the memory bandwidth it uneffected with the Kyro II.
Negaverse23
Distinguished
- Dec 31, 2007
- 449
- 0
- 18,780
The texture compression capabilities of the Kyro II isn't really that impressive to me. Even though the Kyro II can do a 6:1 compression ratio in DXT1 format, it lacks support of DXT2-5. I believe all of the Geforce 2 lines supports all 5 texture compression formats. If I remember correctly, the textre compression ratio in the Geforce 2 is 4:1 for all 5 compression formats.
=
<font color=green>Ran out of bullsh!t to feed to the flies.</font color=green>
=
<font color=green>Ran out of bullsh!t to feed to the flies.</font color=green>
But does the GTS have the option of forcing compression of large textures in any game (even games that don't support TC) in both D3D and OpenGL?, also the Kyro II doesn't have the same bad sky problem in Quake 3 with TC enabled, DXT1 is the most used TC format by a long way too.
So we can use 6:1 texture compression in any game ???
nice !!!
6:1 is better than 1:5 or 1:4 ´
I think 😉
nice !!!
6:1 is better than 1:5 or 1:4 ´
I think 😉
noko
Distinguished
- Jan 22, 2001
- 2,413
- 1
- 19,785
Thanks for you reply and the research you must have done, I appreciate that very much. Also thanks for correcting my math error I don't know where I got that 1/16 value when it should have been 1/8. So I like to get something in perspective here; 44mp/sec (million pixels per sec) fillrate on the KyroII when doing 8 textures per pixel. Especially when you hear such large numbers of 800mp/sec for the GF2 and 1000mp/sec on the GF2 Ultra.
1. <b>So how many pixels are on a screen of a 1024x768 display?</b>
. . a. Take 1024 pixels across times the 768 pixels down
. . . <font color=blue>1024</font color=blue> <font color=red>x</font color=red> <font color=blue>768</font color=blue> = <b>786432 pixels total</b>
2. <b>How many FPS could you achieve with 44mp/s fillrate at 1024x768 resolution?</b>
. . a. Take the number of pixels rendered divided by number of pixels in a frame
. . . 44000000/786432 = <b>55.96FPS*</b>
3. <b>How many FPS could you achieve with 44mp/s fillrate at 1600x1200 resolution?</b>
. . a. Pixels at 1600x1200 = 1920000 pixels
. . b. Frames per second, 44000000/1920000 = <b>22.9FPS*</b>
<b>*</b>This would be max FPS assuming no stalls in the pixel pipe line in the KyroII, that the pipeline is being fed everytime the pixel pipeline is ready to receive the next vertex information and no overdraw of pixels not being displayed. Now that would mean the KyroII would be 100% efficient in everything, which no card is realistically. Still on the fillrate test the KyroII indicates 100% efficiency (amazing :redface: !). Still the KyroII depends on the cpu for vertex transform and lighting so the texel pipeline will be stalled during those moments of calculation if cpu is not speedy enough, contention with AGP bandwidth or moments of high complexity and action going on.
So it appears that the KyroII could be fillrate limited in this extraordinary 8 texture scenario. At four textures per pixel just double the above values for maximum FPS. At 6 textures per pixel the fill rate would be 58 million pixels/sec, Max FPS at 1024x768 would be 73.8FPS. At 1600x1200, 30.2 FPS. Really not much room for any stalls on the card or from receiving vertex and lighting data from the cpu. So the actual frame rate I would think would be less than the maximum best.
So I am concerned about the fillrate ability of the KyroII cards, it can maybe render 8 textures per pass in the texture pipeline but it takes many clock cycles to accomplish that. Even at 6 textures per pixel there is a apparent fillrate bottle neck. What all this means when there are no 6 or greater texture games I don't know and it may just be unimportant anyways.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/15/01 02:57 AM.</EM></FONT></P>
1. <b>So how many pixels are on a screen of a 1024x768 display?</b>
. . a. Take 1024 pixels across times the 768 pixels down
. . . <font color=blue>1024</font color=blue> <font color=red>x</font color=red> <font color=blue>768</font color=blue> = <b>786432 pixels total</b>
2. <b>How many FPS could you achieve with 44mp/s fillrate at 1024x768 resolution?</b>
. . a. Take the number of pixels rendered divided by number of pixels in a frame
. . . 44000000/786432 = <b>55.96FPS*</b>
3. <b>How many FPS could you achieve with 44mp/s fillrate at 1600x1200 resolution?</b>
. . a. Pixels at 1600x1200 = 1920000 pixels
. . b. Frames per second, 44000000/1920000 = <b>22.9FPS*</b>
<b>*</b>This would be max FPS assuming no stalls in the pixel pipe line in the KyroII, that the pipeline is being fed everytime the pixel pipeline is ready to receive the next vertex information and no overdraw of pixels not being displayed. Now that would mean the KyroII would be 100% efficient in everything, which no card is realistically. Still on the fillrate test the KyroII indicates 100% efficiency (amazing :redface: !). Still the KyroII depends on the cpu for vertex transform and lighting so the texel pipeline will be stalled during those moments of calculation if cpu is not speedy enough, contention with AGP bandwidth or moments of high complexity and action going on.
So it appears that the KyroII could be fillrate limited in this extraordinary 8 texture scenario. At four textures per pixel just double the above values for maximum FPS. At 6 textures per pixel the fill rate would be 58 million pixels/sec, Max FPS at 1024x768 would be 73.8FPS. At 1600x1200, 30.2 FPS. Really not much room for any stalls on the card or from receiving vertex and lighting data from the cpu. So the actual frame rate I would think would be less than the maximum best.
So I am concerned about the fillrate ability of the KyroII cards, it can maybe render 8 textures per pass in the texture pipeline but it takes many clock cycles to accomplish that. Even at 6 textures per pixel there is a apparent fillrate bottle neck. What all this means when there are no 6 or greater texture games I don't know and it may just be unimportant anyways.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/15/01 02:57 AM.</EM></FONT></P>
noko
Distinguished
- Jan 22, 2001
- 2,413
- 1
- 19,785
How cheasy of me, I don't know How I missed it earlier but what you said ealier was right and pretty much on what I said above. At 6-8 textures I tend to think there will be a fillrate problem affecting the FPS on the KyroII. Would like to run the 5 texture demo on my Radeon to compare results. I did install the 4 texture map for quakeIII, it ran just fine. I did have some textures not rendered as the text document says since I am using 127h not the 117 version of QuakeIII. Will try to do a benchmark on the demo latter. Plus choosing a 5 texture demo is interesting, for one it requires the GF2 to make 3 passes with on pass only using one half of the pixel pipeline texture abiltiy and the Radeon has to make two passes with the last pass not using its 3rd texture ability. It would be better if the demo was 6 textures vice 5 for comparisons.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/14/01 11:59 PM.</EM></FONT></P>
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/14/01 11:59 PM.</EM></FONT></P>
You are missing the point noko, because of the "FAT" bandwidth of the kyro the theoretical fill rate is close to the REAL fill rate... and this is showing on games in high resolutions or with those that uses more than 2 textures...
so the maximum theoretical fps in a game will be close to the real one on a kyro card.
I went here:
http://www.croteam.com/engine_technology.html
the site of serious sam.
and I saw this:
- Fill rate
Serious Sam can be very fill rate dependant. There are some cases when one polygon can have even 5 passes (main texture + hyper map + detail texture + shadow map + haze or fog)!!! If you have older boards (TNT1/V2 or even slower) I advise you to play in 640x480x16. 32-bit color tends to slow down 3d accelerator performance, so if you're playing in high res, my advice is to stay away from 32bits color depth!
5 textures ?
close to that 6 layer texture that you had mention...
Now You know why kyro beats a geforce gts with ease in these game...
😉
so the maximum theoretical fps in a game will be close to the real one on a kyro card.
I went here:
http://www.croteam.com/engine_technology.html
the site of serious sam.
and I saw this:
- Fill rate
Serious Sam can be very fill rate dependant. There are some cases when one polygon can have even 5 passes (main texture + hyper map + detail texture + shadow map + haze or fog)!!! If you have older boards (TNT1/V2 or even slower) I advise you to play in 640x480x16. 32-bit color tends to slow down 3d accelerator performance, so if you're playing in high res, my advice is to stay away from 32bits color depth!
5 textures ?
close to that 6 layer texture that you had mention...
Now You know why kyro beats a geforce gts with ease in these game...
😉
noko
Distinguished
- Jan 22, 2001
- 2,413
- 1
- 19,785
Agreed, the theoretical fillrate is indeed close to the actual fill rate. Now the theoretical fillrate calculations shows limitations when it is rendering 6-8 textures per pixel at the higher resolutions. Meaning the fillrate will be limiting not the bandwidth of the memory at the higher resolutions. In other words if your max FPS is 30 FPS based on fillrate at a given resolution then your average FPS will probably be much less giving no headway for stalls in the the rendering process. I hope this makes some sense.
Also your reference indicate that the 5 passes are limited, as in not all objects will have 5 textures in Serious Sam. Which means Serious Sam is mostly a 3 texture game with exceptions on some objects.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/15/01 01:42 AM.</EM></FONT></P>
Also your reference indicate that the 5 passes are limited, as in not all objects will have 5 textures in Serious Sam. Which means Serious Sam is mostly a 3 texture game with exceptions on some objects.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 04/15/01 01:42 AM.</EM></FONT></P>
The 800mpixels/s and 1000mpixels/s numbers of the Geforce 2 GTS and Geforce 2 Ultra really mean nothing, take the GTS, its theoretical peak fillrate will have to drop 4 times when using 8 texture layers to 200mpixels/s, now consider that in those Serious Sam tests the GTS could only reach a true fillrate of 257mpixels/s in single texturing, now drop that by 4 times (4 clock cycles) and you get 64mpixels/s, now this is assuming that the card is just as efficient memory bandwidth wise as it is in single texturing but with its fillrate cut by the 4 clock cycles it takes to put the 8 textures on each pixel, but in reality all the extra passes to the framebuffer in ram and having to resend the polys 4 times is going to make the card even more inefficient in the mem bandwidth department (while the Kyro II's memory bandwidth efficieny will stay the same), so that 64mpixels/s number will be more like 40-50mpixels/s at most, so the GTS which usually has so much more raw fillrate then the Kyro II (and the kyro II still keeps up and beats it allot of the time because of overdraw) now has about the same fillrate, now factor in overdraw and the Kyro II will easily beat the GTS, look at those Serious Sam benchmarks in the Anand Kyro II review, the Kyro II beats every other card, especially in FSAA, the Kyro II is upto 50% faster then the GTS in those benchmarks, is it any coincidence that Serious Sam uses 4 texture layers per pixel and the Kyro II beats every card in the game?, I don't think so.
Also to be honest I'm not even saying the Kyro II will be great in 8 texture layer games, I think it'll be adequate looking at the Seroius Sam scores (that being a 4 texture game), but what I am saying is with this feature and deferred rendering its better equiped to deal with future games then the GTS or MX are with there HW T&L unit, when both these cards are playing an 8 texture layered game (or even 5-6 textures) and they both have a fillrate of around 50mpixels/s and the GTS's memory bandwidth is totally clogged what is the HW T&L unit going to do to help it?, especially when the card is having to resend all the polys 4 times and the AGP bus is also clogged, a Kyro II with a good CPU will end up better because the kyro II won't need to send polys more then once (and lets not forget that system CPU's are getting faster and faster), so to sum up, hardwired HW T&L is a good feature but its not the magic thing that'll mean your cards feature proof, it doesn't help you with the cards fillrate, it doesn't help you with memory bandwidth problems, it only helps you when the graphics card is CPU limited at lower res with single or dual textures when the cards fillrate and memory bandwidth aren't being taxed, so its not a magical feature thats the be all and end all of 3d graphics so don't beleive the Nvidia propaganda about it.
BTW I'm glad you appretiate my posts, I also appretiate that your open to new idea's and your not a closed minded Nvidia drone like some (no effence to any Nvidia drones out there😱))
Also to be honest I'm not even saying the Kyro II will be great in 8 texture layer games, I think it'll be adequate looking at the Seroius Sam scores (that being a 4 texture game), but what I am saying is with this feature and deferred rendering its better equiped to deal with future games then the GTS or MX are with there HW T&L unit, when both these cards are playing an 8 texture layered game (or even 5-6 textures) and they both have a fillrate of around 50mpixels/s and the GTS's memory bandwidth is totally clogged what is the HW T&L unit going to do to help it?, especially when the card is having to resend all the polys 4 times and the AGP bus is also clogged, a Kyro II with a good CPU will end up better because the kyro II won't need to send polys more then once (and lets not forget that system CPU's are getting faster and faster), so to sum up, hardwired HW T&L is a good feature but its not the magic thing that'll mean your cards feature proof, it doesn't help you with the cards fillrate, it doesn't help you with memory bandwidth problems, it only helps you when the graphics card is CPU limited at lower res with single or dual textures when the cards fillrate and memory bandwidth aren't being taxed, so its not a magical feature thats the be all and end all of 3d graphics so don't beleive the Nvidia propaganda about it.
BTW I'm glad you appretiate my posts, I also appretiate that your open to new idea's and your not a closed minded Nvidia drone like some (no effence to any Nvidia drones out there😱))
OzzieBloke
Distinguished
- Mar 29, 2001
- 167
- 0
- 18,680
Mate, your posts are a breath of fresh air compared to some of the other stuff that gets posted in this place Technically well thought out too... but, could you please put carriage returns between every two or three sentences? Solid text is harder to read than I thought 😉
Kyro II is certainly looking good... now if only it would consider putting in the extras that nVidia and Radeon are putting into their cards, and you would have a card NASA would go ga-ga over
Australian PMs are like steer horns; a point here, a point there, & a heap of bull in the middle.
Technically well thought out too... but, could you please put carriage returns between every two or three sentences? Solid text is harder to read than I thought 😉Kyro II is certainly looking good... now if only it would consider putting in the extras that nVidia and Radeon are putting into their cards, and you would have a card NASA would go ga-ga over
Australian PMs are like steer horns; a point here, a point there, & a heap of bull in the middle.
TRENDING THREADS
-
-
-
-
-
Question Asrock B450 Steel Legend motherboard is not loading the OS installer from pendrive ?
- Started by surya_sapui
- Replies: 9
-
-
Question RAID 10 disk failure and reallocated sectors count ?
- Started by rcfant89
- Replies: 5
Facebook
Twitter
Instagram