It would have been interesting to see just what PVR could have come up with for 2000 if they could have started from scratch without having to bolt together DC parts from the previous decade (100mHz SDRAM was hardly high end for a graphics chip by then!). Something with Elan's abilities and multiple pipelines, all from a common memory pool shouldn't have been out of the question, surely.
Consoles that provide(d) the best performance increase from 1stgen to last gen titles
- Thread starter Butta
- Start date
Many details were lost from the Arcade game - and it was a good comparision - launch Model 3 game converted as launch DC game.
Maybe comparing VF3 on DC against Tekken Tag on PS2 would be better as they were both launch games.
but that's a different topic completely ( that's available via seach )
rubbish - PS2 is a very elegant 4 chip design ( 2 ram chips, EE and GS ) which was shrunk down to 3 chips.
Naomi 2 was a very powerfull arcade board with far more components , and many seperate memory pools , and had it's own custom designed chip (elan) as well as reusing the PowerVR DC chips ... ( When it came out that was surprising, as there were more powerfull PVR chips such as Kyro available , but I guess they must have had a huge stock of unsold dreamcast components
)I wonder if elan would have been a choice - maybe in 2000 it could have been a faster part ( Kyro ) tied with a faster SH4 and more memory...
32MB main + 400MHz SH4 + 200MHZ Kyro2 + 32MB VRAM would have been pretty reasonable
...as well as reusing the PowerVR DC chips ... ( When it came out that was surprising, as there were more powerfull PVR chips such as Kyro available , but I guess they must have had a huge stock of unsold dreamcast components
Not surprising at all, surely?
I believe Naomi 2 was designed to be backwards compatible with Naomi 1. Being able to re-use existing code and experience was also probably a factor. When Elan was commissioned in 98 or 99 Sega were ramping up for a global release, so using up surplus parts from 2000 and 2001 probably wasn't a factor in their decision making!
You're probably right - Elan may never have existed but for Sega's cost-insensitive bolting together of old hardware.
I read somewhere (maybe here actually) that Kyro lost some of the cooler stuff from Naomi to keep die size down. Can't remember what though. I always got the impression that Naomi and DC were mostly held back by polygon processing on the CPU - in this theoretical 2000 system, perhaps a pair of fast SH4s (if Sega could overcome their Saturn memories) might have proved a cost effective and flexible alternative to one SH4 plus Elan...
I was under the impression that the 200 MHz SH4 was pretty beastly. It had a great FPU, I believe. I guess if you look at it compared to PS2, it's really sad. But if you compare it to N64, things look a lot different.
There's always something better, but they have to balance costs obviously. SH4 was a great CPU for Dreamcast, considering when it was released.
There's always something better, but they have to balance costs obviously. SH4 was a great CPU for Dreamcast, considering when it was released.
I was under the impression that the 200 MHz SH4 was pretty beastly. It had a great FPU, I believe. I guess if you look at it compared to PS2, it's really sad. But if you compare it to N64, things look a lot different.
There's always something better, but they have to balance costs obviously. SH4 was a great CPU for Dreamcast, considering when it was released.
Yet compared to more modern systems, I'd venture to say it was the overall weakpoint of the system.
I'd say Series 2 is at least in the same league as the immediate systems following it, but I'm not so sure the SH-4 is.
SH-4 held its own against the PS2's MIPS CPU while NAOMI2's addition of a T&L co-processor led to far more performance and a far lower silicon cost than PS2. Still needing only SDR DRAM, Elan did make for a cost effective addition to the system.
Targeted more closely at PC APIs, the Kyros wouldn't have been able to make good use of custom functionalities that Series 2 had like the hardware translucency sorting and the modifier volumes (which could be replaced by stencils anyway) and therefore didn't include them.
SH-4 iterations couldn't clock much higher for several years, so 400 MHz wasn't feasible until more recently.
Being spread across fewer chips than NAOMI2 didn't reduce the PS2's one-third higher die area costs, nor prevent its RAM from costing multiple times the price per space.
Missing some details from the arcade like DC did in its Virtua Fighter reproduction was nowhere near the inadequacy of missing half the polygon count on top of reduced lighting, worsened image quality, and worsened texturing like the PS2 did in its Virtua Fighter reproduction.
The DC's ~250% disadvantage in die area, 1+ year disadvantage in release, and amount/type/brand disadvantage in RAM relative to PS2 makes any direct comparison between the two systems that can be entertained, like with DOA2 versus Tekken Tag, compelling evidence of a massive DC architectural advantage.
Targeted more closely at PC APIs, the Kyros wouldn't have been able to make good use of custom functionalities that Series 2 had like the hardware translucency sorting and the modifier volumes (which could be replaced by stencils anyway) and therefore didn't include them.
SH-4 iterations couldn't clock much higher for several years, so 400 MHz wasn't feasible until more recently.
Being spread across fewer chips than NAOMI2 didn't reduce the PS2's one-third higher die area costs, nor prevent its RAM from costing multiple times the price per space.
Missing some details from the arcade like DC did in its Virtua Fighter reproduction was nowhere near the inadequacy of missing half the polygon count on top of reduced lighting, worsened image quality, and worsened texturing like the PS2 did in its Virtua Fighter reproduction.
The DC's ~250% disadvantage in die area, 1+ year disadvantage in release, and amount/type/brand disadvantage in RAM relative to PS2 makes any direct comparison between the two systems that can be entertained, like with DOA2 versus Tekken Tag, compelling evidence of a massive DC architectural advantage.
silicon cost is a bit of a red herring, by the time Naomi 2 shipped the PS2 chip size had shrunk dramatically.
In what way was it cost effective - Naomi 2 was priced way above the PS2?
It made up for that with the improved fillrate, but that's a valid point, although maybe only 1 Kyro would have been needed.
Which makes the EE more impressive for the time it came out
Again - this is rubbish, PS2 did not have 1/3 higher die area costs - and the ram did not cost multiple times the price per space. But it's a wierd argument as Naomi 2 was not a consumer chip, it was an arcade board
I dont know - there were more complaints about VF3 at the DC launch .. ( and dont get me started on Sega Rally 2 )
For VF4 the image quality problem was fixed in the evo release, and the reduced texturing and polycount were more due to memory constraints in general. It was a pretty good job for a first attempt on the platform in my opinion.
No. The complete failure of the DC as a platform , and the relegation of it's graphics technology to non performance critical mobile platforms, compared to the success of the PS2 and the integration of it's main graphics feature ( edram ) into the xbox360 can be taken as compelling evidence of it's architectural advantage.
( Take that as a joke - it's merely to provide a counterpoint to your rhetoric )
Comparing the 'silicon' cost of GS to the CLX + the SDRAM shows that there was never a '250% disadvantage', merely a complete difference in design.
Being left with almost no room for textures and/or an image, nor the ability to process or store the vertex part of a graphics workload, makes the GS by itself not comparable to the CLX2+RAM and Elan.
PowerVR's superior bandwidth efficiency affords it more RAM for the same cost and was part of the architectural advantage over PS2. Elan's real-world, sustained speed with general lights was also far beyond the EE's vector units, and this was highlighted by the usage of search lights in certain stages of VF4 and also in the NAOMI2 Virtua Fighter tech demo shown at the board's unveiling.
http://media.dreamcast.ign.com/articles/085/085311/vids_1.html
NAOMI2 rendered unconditionally with floating-point, ~32-bit Z accuracy and displayed Virtua Fighter 4 in proscan and at the full 640x480 image resolution, so the improvement in image quality in the Evo revision of PS2's VF4 still wasn't a match.
Any simple check of price comparisons between SDR and DDR SDRAM variants of a product from early 2000 clearly shows that a whole multiple pricing difference would've separated Rambus's newly introduced DDR from generic SDR.
http://arstechnica.com/reviews/1q00/geforces/gfru-1.html
A CLX2 was estimated at not much more than 100 mm^2 of die, Elan was supposedly comparable, and SH-4 was 42.25 while PS2's GS and EE combined for around 520.
Needing a lot more silicon than an SH-4 and an Elan combined while performing much worse does not make the EE impressive at all.
The only real difference between Kyro 1 and 2 was the clock speed, so a 200 MHz Kyro would be a Kyro 2+.
Kyros would have also been a fine choice for a 2000-era PowerVR system; the discussion about the missing Series 2 custom features was just a clarification of why they weren't included.
NAOMI2 was priced way above PS2 like System246 was priced way above a DC: not even remotely comparable business models. NAOMI2 cost similarly to NAOMI at their respective launches, and NAOMI was just the arcade variant of a very cost effective home console.
NAOMI2 was released to SEGA by May 2000 and had had almost all of its chips in volume production for almost two years. Demonstrations and location tests of VF4 followed within a few months, but arcade boards aren't rushed to release with "launch" software like consoles and are instead held back for a full software development cycle of their debut game.
Not only had PS2 not been shrunk dramatically by mid 2000, but die shrinks aren't an architectural advantage anyway. Had Sony had access to NAOMI2's design, they could've applied their fabs and manufacturing technology to its higher performance and lower cost architecture.
Cost is directly proportional to silicon area: twice the die area means half the production per time means twice the cost.
PowerVR's superior bandwidth efficiency affords it more RAM for the same cost and was part of the architectural advantage over PS2. Elan's real-world, sustained speed with general lights was also far beyond the EE's vector units, and this was highlighted by the usage of search lights in certain stages of VF4 and also in the NAOMI2 Virtua Fighter tech demo shown at the board's unveiling.
http://media.dreamcast.ign.com/articles/085/085311/vids_1.html
NAOMI2 rendered unconditionally with floating-point, ~32-bit Z accuracy and displayed Virtua Fighter 4 in proscan and at the full 640x480 image resolution, so the improvement in image quality in the Evo revision of PS2's VF4 still wasn't a match.
Any simple check of price comparisons between SDR and DDR SDRAM variants of a product from early 2000 clearly shows that a whole multiple pricing difference would've separated Rambus's newly introduced DDR from generic SDR.
http://arstechnica.com/reviews/1q00/geforces/gfru-1.html
A CLX2 was estimated at not much more than 100 mm^2 of die, Elan was supposedly comparable, and SH-4 was 42.25 while PS2's GS and EE combined for around 520.
Needing a lot more silicon than an SH-4 and an Elan combined while performing much worse does not make the EE impressive at all.
The only real difference between Kyro 1 and 2 was the clock speed, so a 200 MHz Kyro would be a Kyro 2+.
Kyros would have also been a fine choice for a 2000-era PowerVR system; the discussion about the missing Series 2 custom features was just a clarification of why they weren't included.
NAOMI2 was priced way above PS2 like System246 was priced way above a DC: not even remotely comparable business models. NAOMI2 cost similarly to NAOMI at their respective launches, and NAOMI was just the arcade variant of a very cost effective home console.
NAOMI2 was released to SEGA by May 2000 and had had almost all of its chips in volume production for almost two years. Demonstrations and location tests of VF4 followed within a few months, but arcade boards aren't rushed to release with "launch" software like consoles and are instead held back for a full software development cycle of their debut game.
Not only had PS2 not been shrunk dramatically by mid 2000, but die shrinks aren't an architectural advantage anyway. Had Sony had access to NAOMI2's design, they could've applied their fabs and manufacturing technology to its higher performance and lower cost architecture.
Cost is directly proportional to silicon area: twice the die area means half the production per time means twice the cost.
A simplistic and stupid comparision - the GS was designed from the outset to work with the EE, and support streaming of textures - something not used by the first titles. And with 96MB of combined ram in the graphics pipeline ( as well as the 32MB for the SH4 ) you would expect more textures
A fallacy - GS had amazing bandwidth efficiency, due to not actually using any external memory
10M polys with 6 lights was possible with the EE VU1, but most games would put more effort into animation and more advanced effects than simple n lights per poly. ( God of War, and also Spartan Total Warrior looked very good ) and the EE also contained VU0 for physics, and the IPU for movie decoding to textures as well as the MIPS core
I haven't seen those for ages, thanks.
One thing shown was how poor the Naomi2 was at transparencies - the feather demo looked way simpler than the launch demo for PS2.
Obviously not, given that the PS2 was outputing to a TV. But the evo revision was a major improvement ( As close as NTSC would be ) and most differences were due to the textures ( memory limitations again )
I dont think that you really would notice the difference between 32bit floating Z and 24 bit Z - it seems to have been good enough for pretty much all PC cards
I think that you must at least admit that the PS2 VF4evo was a good conversion, given that the arcade machine had nearly 4x the memory footprint.
I'm not sure that this is relevant at all to the rambus in PS2, the big advantage was the smaller number of traces on the board, as well as the very large bandwidth and good latency ( not to be confused with the problems rambus had in the PC space )
Again GS didnt need any external memory chips - and many people considered the EE impressive. It was quite easy to transform and light >30Million polygons at the start, something Naomi 2 was not capable of. ( With trivial lighting, but for some situations that was enough )
If you want to look at what the VU1 on EE was capable of look at
http://playstation2-linux.com/files/vudemocontest/vu_coding_contest_2003_scea_1.1.tgz
This shows geometry shader techniques that would be difficult to implement efficiently using DX10!
(It's explained in vu_coding_contest_2003_scea_1.1\linux_for_ps2\demosrc\vuniverse\VUniverse.doc)
Ok - I'm hazy on powervr, last time I really cared was when DC was out
( although Omap2 is interesting )System246 was priced high for profits - and I can be sure that Naomi2 was far more expensive than Naomi1 to build, no matter what the cost was at launch. Anyway, there's no point debating this, there was no consumer hardware version of Naomi2.
how long was Elan in volume production? the rest was old tech ( "IOP and SPU on PS2 had been in volume for over 5 years at it's launch" - it's nonsense )
The EE and GS had an architectural advantage at their conception. The die shrinks showed the 'learning' curve of implementing their own designs. ( CLX was a progression of PowerVRs designs honed over many revisions.. and the SH4 was a ( superb ) evolution of hitachi's SH series )
... and in 2000 at Siggraph the GS cube showed 1080x1920 60Hz rendering of bits of Final Fantasy ( much smoother than the geforce demo shown later ) - with a GS chip containing 32MByte of memory rather than 4MB. ( and I cant remember if it was E3 that year or not, but there was a demo of a PS2 playing 1080p@60 movies - something that my PC sometimes has trouble with even today )
I ve been looking for videos and images of the GScube demonstration for ages and couldnt find anything at all except from the GScube itself.
Do you have anything to share with us?? I d appreciate it a lot.
I remember reading that there were other demos as well.
The dance scene in Antz, with thousands of ants dancing on screen
A chace scene of the Matrix rendered in real time, looking almost indistinguishable from the movie itself
And I think there were a few others too, but cant remember right now
The GSCube showed 4 demos, the Antz scene, the Final Fantasy scene, the Matrix Scene, and the FlightSim.
I think the final fantasy scene was the most impressive in terms of animation and source material, the Antz scene showed millions of polys at 60Hz - but was quite simple in it's animation and lighting, the Matrix demo was more about using film assets (textures).
The flight sim was very impressive as it streamed in high res assets from a huge SGI server ( some sim model of part of france )
It looked really cool ( like a minimalist Borg cube ) but was more research than product I guess.
I think the final fantasy scene was the most impressive in terms of animation and source material, the Antz scene showed millions of polys at 60Hz - but was quite simple in it's animation and lighting, the Matrix demo was more about using film assets (textures).
The flight sim was very impressive as it streamed in high res assets from a huge SGI server ( some sim model of part of france )
It looked really cool ( like a minimalist Borg cube ) but was more research than product I guess.
Probally, though I haven't seen much on the internet - I think it escaped the notice of the normal game sites because it wasn't E3 or GDC, and it wasn't a game
It was impressive because it ran the demos at 1080p 60 - the NV demos at the next Siggraph were still 'realtime' but at a lower framerate.. ( However they were still really cool, because they were running on a single video card )
It was impressive because it ran the demos at 1080p 60 - the NV demos at the next Siggraph were still 'realtime' but at a lower framerate.. ( However they were still really cool, because they were running on a single video card )
Flexible functionality would be the purpose of doing geometry or vertex shading on a general purpose processor like a CPU, so having advanced geometry shading algorithms be better suited to PS2 vector unit processing than that of a DirectX 10 processor is only obvious.
Being capable of T&Ling tens of millions of polygons per second while having only the throughput in VF4's game conditions to model its characters at half detail (though still bearing a relatively close resemblance, making PS2 VF4 impressive for just that feat of modeling by Sega's artists), even after dropping the lighting complexity, shows the gross inefficiency of PS2's design. It's like the triangle set-up limit of 75M/s: a better balanced system would've used that excess of die area to improve real world performance which would then be a higher percentage of its theoretical peak.
Like the Emotion Engine, some people consider Cell impressive, too, just not the actual engineers at top CE companies like Sharp, NEC, and Apple who use their evaluation and benchmarking tools to select the best technology available to include in their products and have chosen PowerVR video display technology.
No chips in isolation are 1:1 in functionality between the PS2 and the NAOMI2, so the only 1:1 comparison is between the whole systems of each. There, NAOMI2's processors use less silicon while achieving their much higher performance.
Rambus charged a premium on top of the already expensive, generic DDR type DRAM, swallowing up cost savings from its fewer traces. The difference from 100 MHz SDRAM, which was cheaper than even cheap SDR and being sold at practically clearance prices, would be much larger than even the great price difference shown in that GeForce SDR versus DDR comparison.
The silicon used by the EE for frivolities like its IPU would've been better appreciated in enhancing the system's general texturing capabilities, or its texture LOD, or obviously the very-weak-for-its-era CPU which was the processing bound of most PS2 games.
The last thing Sega would've wanted to do in rebuilding VF4 for PS2 would be to remodel every character and some background objects, and certainly not with half the geometry budget, especially after reducing the lighting. Fast T&L was what Sega needed in VF4, and only NAOMI2 could do that, a reason Sega chose it over PS2's architecture in 1999 when they evaluated both systems for the design of their then-forthcoming arcade system.
GS wasted bandwidth calculating/texturing overdrawn pixels, and the system as a whole's need for RDRAM left PS2 with a small amount of total RAM for its era. Had it been able to perform adequately with SDR like NAOMI2, it too would've been better equipped.
eDRAM sacrifices more die space than a tiler and still ends up more memory constrained for frame buffers and image quality.
The term "naive IMR" is basically the result of the influence PowerVR's TBDR algorithm has had on the graphics field. Features like decoupled vertex and pixel processing give it superior prevention of pipeline stalls. More uniform, contiguous accesses to memory, like with framebuffer writes, give it better efficiency. Better visible surface determination makes it more effective with fillrate.
Beating comparable competitors by margins not seen since... well... Kyro 2 plowed a GeForce2 MX and NAOMI2 lapped PS2, MBX and especially its OMAP2 implementations have run over its challengers so far, yet IMRs weren't supposed to be "naive" anymore...
http://jbenchmark.com/result.jsp?benchmark=hd
Being capable of T&Ling tens of millions of polygons per second while having only the throughput in VF4's game conditions to model its characters at half detail (though still bearing a relatively close resemblance, making PS2 VF4 impressive for just that feat of modeling by Sega's artists), even after dropping the lighting complexity, shows the gross inefficiency of PS2's design. It's like the triangle set-up limit of 75M/s: a better balanced system would've used that excess of die area to improve real world performance which would then be a higher percentage of its theoretical peak.
Like the Emotion Engine, some people consider Cell impressive, too, just not the actual engineers at top CE companies like Sharp, NEC, and Apple who use their evaluation and benchmarking tools to select the best technology available to include in their products and have chosen PowerVR video display technology.
No chips in isolation are 1:1 in functionality between the PS2 and the NAOMI2, so the only 1:1 comparison is between the whole systems of each. There, NAOMI2's processors use less silicon while achieving their much higher performance.
Rambus charged a premium on top of the already expensive, generic DDR type DRAM, swallowing up cost savings from its fewer traces. The difference from 100 MHz SDRAM, which was cheaper than even cheap SDR and being sold at practically clearance prices, would be much larger than even the great price difference shown in that GeForce SDR versus DDR comparison.
The silicon used by the EE for frivolities like its IPU would've been better appreciated in enhancing the system's general texturing capabilities, or its texture LOD, or obviously the very-weak-for-its-era CPU which was the processing bound of most PS2 games.
The last thing Sega would've wanted to do in rebuilding VF4 for PS2 would be to remodel every character and some background objects, and certainly not with half the geometry budget, especially after reducing the lighting. Fast T&L was what Sega needed in VF4, and only NAOMI2 could do that, a reason Sega chose it over PS2's architecture in 1999 when they evaluated both systems for the design of their then-forthcoming arcade system.
GS wasted bandwidth calculating/texturing overdrawn pixels, and the system as a whole's need for RDRAM left PS2 with a small amount of total RAM for its era. Had it been able to perform adequately with SDR like NAOMI2, it too would've been better equipped.
eDRAM sacrifices more die space than a tiler and still ends up more memory constrained for frame buffers and image quality.
The term "naive IMR" is basically the result of the influence PowerVR's TBDR algorithm has had on the graphics field. Features like decoupled vertex and pixel processing give it superior prevention of pipeline stalls. More uniform, contiguous accesses to memory, like with framebuffer writes, give it better efficiency. Better visible surface determination makes it more effective with fillrate.
Beating comparable competitors by margins not seen since... well... Kyro 2 plowed a GeForce2 MX and NAOMI2 lapped PS2, MBX and especially its OMAP2 implementations have run over its challengers so far, yet IMRs weren't supposed to be "naive" anymore...
http://jbenchmark.com/result.jsp?benchmark=hd
So effectively you are agreeing that VU1 was way more capable than Elan here
Maybe this shows the gross inefficiency of Sega's AM2 programmers
- You really are talking complete rubbish here.With the 75M figure I think you are stupidly reading spec. sheets with no understanding. The original GS press release gave multiple 'realistic' figures, as well as the hyperbole.
I actually still strongly favour that memory ( and of course time ) was a major problem when porting from Naomi 2 to PS2.
Sorry, what has this got to do with the discussion? ( Or even our sidetracking into comparing Naomi2 and PS2 )
Rubbish, again you fail to count the total silicon cost - and I think PS2 still offers higher performance in certain cases than Naomi2, and lower performance in other cases.
You have no real evidence of this, and it is not relevant anyway, as over the entire lifetime of the console the RDRAM cost would drop - there is little 'silicon cost' in RDRAM over SDRAM
The EE was quite strong for when it came out - it would not match the top end intel chips, but how could it do that and still be priced low enough for a console. It definitely was faster than the SH-4 afterall.
And the IPU was not a friviolity, but an important part of the console.
Maybe more memory was important for Sega
- or even control over their own hardware. PS2 was a fixed console after all - so adding more video memory would be difficult.. .... and they still had all of those unsold chips left from the failure of dreamcast.GS didn't waste any bandwidth , you speak as if bandwidth was a liquid, freely available wherever needed.
And the PS2 had way more memory than DC - comparable memory to Gamecube , but less memory than Xbox ( and certainly less memory than Naomi2 - which was expected really )
rubbish as usual... You harp on about a single difference ( eDRAM and tiling ) as if it was the most important thing. Without any external memory a tiler is completely useless - the GS functions as a single chip
Decoupled vertex and pixel processing has nothing to do with TBDR - I think you may need to read up on the underlying technology a bit more rather than blindly quoting PR out of context. Those quotes are relevant to old PC graphics cards, but not to the GS.
That benchmark is completely irrelevant to PS2/Naomi2/DC - so why are you bringing it up? ( I have no problems admiring Omap2 as a chipset...)
Even Kyro2 against Geforce is pointless - why do you bring that up?
As for PS2 vs Naomi2 I think the memory is the biggest advantage - having 64MB for framebuffer and textures is a great deal more than PS2
- For things like transparencies and frame buffer effects the Naomi2 was really very weak.I'm curious as to what developers think about the 360 in terms of what we are going to see in improvements from this point on graphically. My feeing is that we've pretty much seen what we are going to see and there will only be subtle increases.Definately not the dramtic increases that we saw on PS2.
Reason being the system is easy to develop for to begin with,and the devs working on the 360 are used to creating very advanced graphics already.Devs like Epic,Monolith,Valve and Bioware are PC devs and used to dealing with high resolution graphics and advanced effects and graphics in general. They didn't have to learn that stuff,just the system.
Not only that but we are already seeing the best looking games with unstable framerates and games being made at lower than 720p.
Reason being the system is easy to develop for to begin with,and the devs working on the 360 are used to creating very advanced graphics already.Devs like Epic,Monolith,Valve and Bioware are PC devs and used to dealing with high resolution graphics and advanced effects and graphics in general. They didn't have to learn that stuff,just the system.
Not only that but we are already seeing the best looking games with unstable framerates and games being made at lower than 720p.
Similar threads
- Locked
