What could have been for Dreamcast....

[Crazyace]

Since it seemed such a passionate discussion that derailed the previous topic ( and it was quite fun ) I thought I'd give a 'realistic' (imo) view of what Sega could have brought to market a year later directly against PS2..

DC: ( Original )
SH4@200Mhz with 16MB ram
CLX graphics chip @100Mhz with 8MB ram.

Peak performance 7M polys/s , CPU transform 10M polys/s

Naomi 2 ( Some suggestions that this might have been DC a year later.. )
SH4@200Mhz with 32MB ram
Elan T&L chip @100MHz(?) with 32MB ram
2 CLX graphics chips @100Mhz with 32MB each

Peak performance 14M polys/s, Elan transform 10M with 6 lights..

This is twice the raw performance with a powerfull lighting solution but I dont think it would have been a suitable home release. The wasteful duplication of 32MB of video ram would be a bit much for a home machine.
Also the seperate large memory pool for the elan chip might have been too expensive..

Instead Sega might have pulled out the stops to match Sony's clockspeed..

DC (Possible)
SH4@300Mhz with 32MB ram
CLX graphics chip @150Mhz with 8MB ram

Peak performance 10.5M polys/s, CPU transform 15M polys/s.

I think this would have been a more realistic 'update' than the Naomi 2 options, as engineering a higher clockspeed would increase the performance without involving extra chips, and the main memory would be increased to allow larger programs.

Another alternative might have been..

DC (2nd alternative)
SH4@200MHz with 16MB ram
Elan@100MHz with 16MB ram
CLX graphics chip @100MHz with 8MB ram

Peak performance 7M polys/s, Transform 10M polys/s with 6 lights


All of these could have been options ( I favour the increase clock speed myself as it seems a bit more elegant.. ) but would they have been enough to compete with the PS2?

(Note I keep the video memory at 8MB as it was formed from 4 2MB chips, with a 5th chip for the sound.. and it was already more than PS2 so may not have been a target for improvement)

comments..

 

[cybamerc]

It wasn't the hardware that killed the Dreamcast.

 

[hey69]

I think this has to been the most discussed topic of all the consoles history

the force is strong with Sega Fans

 

[Sonic]

I don't see where in this thread that it indicates hardware killed the Dreamcast.

I think an increase in memory would also be a nice straightforward leap for the Dreamcast if it was released a year later. It would probably even cost less than the DC did initially. There are some other things I would do to increase hardware.

Exactly when was the original Kyro released? If it was in mid 2000 then I do think that would have been a nice chip instead of the CLX.

 

[cybamerc]

Sonic:

> I don't see where in this thread that it indicates hardware killed the Dreamcast.

The premise is that postponing the launch for a year would have enabled Sega to release a more competitive product. But what would the point be? The DC didn't fail because it was underpowered. If it had launched directly against the PS2 it would just have done even worse than was the case.

 

[thop]

I think this has to been the most discussed topic of all the consoles history

the force is strong with Sega Fans

Seconded and thirted

 

[Lazy8s]

If the DC hardware partners had instead been contracted to develop parts for a system releasing at the same time and price as PS2, the graphics part from Img Tech would've certainly been a lot more powerful.

Judging from Series 2 releases for the console and arcade versus its releases for PC, PowerVR's roadmap for embedded technology was well ahead of that for the PC. The part I'd expect would basically have at least Kyro functionality with embedded-specific optimizations like hardware translucency sorting and modifier volumes. As a parallel to what was demostrated in the example of DC versus Neon 250 and with the larger budget this part would be getting, I would expect all of this at a few times the rasterization performance of Kyro within its console environment.

Seeing as March 2000 was into the era of hardware T&L and into PowerVR's development of ELAN, I would expect specific support for vertex processing. Img Tech had not believed in multiple chip solutions for the PC, yet this hypothetical DC would be a custom system at a non-budget pricepoint. So, a graphics part would go with either a separate ELAN-like co-processor or with on-die vertex acceleration. ELAN was something like a 10 million transistor chip at 100 MHz, so that kind of T&L performance would be reasonable.

PowerVR tech manages with inexpensive RAM that got very affordable in 2000, so keeping to the same expense margins and just scaling up for pricepoint and release date would yeild allocations of around 32 MB of video RAM (divided with T&L if separate) and 32 MB of main RAM.

As DC's real-world performance is about 2.5 million polygons per second with four lights, texture mipmapping and bilinear filtering with limited usage of trilinear and anisotropic, and a scene-average 275 Mpixel effective fillrate, I'd expect this newer, higher priced system of the T&L era to sustain around 12 million polygons per second with six lights, texture mipmapping and trilinear filtering with limited usage of ansio+bilinear, and a scene-average 600 Mpixel effective fillrate.

edit: fillrate guess

 

[Sonic]

Cybamerc, the point of discussion is not why the Dreamcast failed, please understand that. If you cannot reply to the author's topic with any sort of relevancy then why post at all? You came up with a statement that had nothing to do with the topic of discussion.

 

[jvd]

Sonic . They say that the kyro and the neon 250 was late because of the dc . So perhaps if the dc and naomi 1 never happened they could have indeed gone with a kyro .

The original kyro
was

12 million transitors on 250nm . 2x1 at 115 mhz. So it was already more than twice as fast as the powervr dc (100x1) . It used a 128bit bus with sdram. So the cost of ram wouldn't have gone up.

This was released q3 2000.

The powervr dc was 100x1 with 64 bit buss for ram .

If you read the beyond3d article on the neon250 they say the dreamcast caused a year gap. Using that figure they could have most likely launch the kyro in 1999 if there wasn't a dc in 1998.


So

sh4 at 250mhz with 16 megs of ram
elan @ 100mhz with 32 megs of ram
kyro @ 100mhz with 32 megs of ram

same sound with 8 megs of ram

Or

sh4 at 250mhz with 32 megs of ram
kyro at 125 mhz with 32 megs of ram



The first set up would have giving you 10million polygons with 6 lights instead of about 4 million with 2-4 lights. Twice the fillrate , twice the memory bandwidth. Twice the memory and a more advanced hardware features.


The second one would keep the ploygon power about the same but would add 2.5 times the fillrate and you'd have advantage of more ram , twice the bandwidth and a better hardware feature set .



I never understood why sega pulled the plug on the dreamcast.

They just needed to drop down to 50$ before the gc and xbox launched.

Keep creating games for the system. Throw on ports from old sega systems , fighting games , sports games . Perhaps even make a dreamcast mini .

In 2000 they must have been able to shrink the dc more so than it was .

They already had a 10 million or so user base . Not to far from what nintendo and ms are sitting at.

If they kept supporting it and were selling at 50$ they would have at least doubled the installed base.


Its sad that they didn't. They used to make great games with 1 system to concentrate on . Now they are sadly not doing as well.

I mean if nintendo can make due at 18 million systems why couldn't sega do well with 10 million +

 

[Fafalada]

DC (Possible)
SH4@300Mhz with 32MB ram
CLX graphics chip @150Mhz with 8MB ram

Add a reasonably sized scratchpad to SH4 that maybe allows DMAing straight to PVR mem - it's not a terribly complicated architectural change and it would go a long way to free up more CPU from T&L load.
The FP instruction set could use some upgrading too but that's probably pushing it too far. Although at the very least they could put in stuff like FP instructions for clamping etc.

(Note I keep the video memory at 8MB as it was formed from 4 2MB chips, with a 5th chip for the sound.. and it was already more than PS2 so may not have been a target for improvement)

Wouldn't 24/16MB work better in that regard? I was under impression that both texture and polygon use on CLX are mostly dependant on available video memory.
I don't want to make comparisons that people will jump me for, but with 8MB you'd need to keep your polycount Very low to match texture assets of highend PS2 games (or vice versa).

As DC's real-world performance is about 2.5 million polygons per second with four lights

SH4 is a nice little chip but that's stretching it pretty far, even if you just used directional lights.
With actual local lights(to which you've compared later on with Elan) you're not even close - even if you only run transforms on the CPU and nothing else.

 

[Lazy8s]

Fafalada:

SH4 is a nice little chip but that's stretching it pretty far, even if you just used directional lights.

Dead or Alive 2 lists 3 Mpps with two local (point) lights and two infinite.

With actual local lights(to which you've compared later on with Elan)

I wasn't drawing a complete parallel between the set of attributes for the old DC to the new DC - just a rough comparison as the amount of areas of advancement would be quite numerous. Yeah, areas where quantity saw an increase could also see an increase in quality. I also didn't mean to imply that my hypothetical system necessarily included a separate ELAN-like co-processor, so my figures at the end weren't a complete association with ELAN performance attributes.

 

[jvd]

Fafalada , why not toss out what u would have wanted sega to release at the time. I know u gave some hints but a complete speced out machine would be really cool

 

[Crazyace]

The premise is that postponing the launch for a year would have enabled Sega to release a more competitive product. But what would the point be? The DC didn't fail because it was underpowered. If it had launched directly against the PS2 it would just have done even worse than was the case.

That may be so, or it may have eaten some market share from Gamecube and Xbox - as many Sega titles would not have appeared on the other platforms..
At least at the technical level I can make a 'realistic' (imo) guess at the capabilities.

The part I'd expect would basically have at least Kyro functionality with embedded-specific optimizations like hardware translucency sorting and modifier volumes.

ELAN was something like a 10 million transistor chip at 100 MHz, so that kind of T&L performance would be reasonable.
32 MB of video RAM (divided with T&L if separate) and 32 MB of main RAM.
I'd expect this newer, higher priced system of the T&L era to sustain around 12 million polygons per second with six lights, texture mipmapping and trilinear filtering with limited usage of ansio+bilinear, and a scene-average 600 Mpixel effective fillrate.

This is an expensive option - I think that if Kyro had been picked, that it would not be paired with a T&L chip - the SH4 already had 3D transform support added, and an extra chip and the memory may have made the product unfeasable. ( This would be a good competitor for Xbox in 2001 )

sh4 at 250mhz with 16 megs of ram
elan @ 100mhz with 32 megs of ram
kyro @ 100mhz with 32 megs of ram

same sound with 8 megs of ram

Or

sh4 at 250mhz with 32 megs of ram
kyro at 125 mhz with 32 megs of ram

...

I never understood why sega pulled the plug on the dreamcast.

They just needed to drop down to 50$ before the gc and xbox launched.

Kyro would be interesting - I think it may have been more like 32/16 or 16/16 for the memory though.

I dont think Sega could have dropped to $50 except for stock clearance/writeoffs as they were losing too much per unit at the higher prices already. ( No bottomless pockets for Sega )

Wouldn't 24/16MB work better in that regard? I was under impression that both texture and polygon use on CLX are mostly dependant on available video memory.
I don't want to make comparisons that people will jump me for, but with 8MB you'd need to keep your polycount Very low to match texture assets of highend PS2 games (or vice versa).

I'm assuming that faster memory would be more expensive .. and 32MB main would be more important for games - the faster path to the CLX would allow allow more dynamic uploads per frame.

I'm not thinking about designing a 1999/2000 DC using hindsight from 2004, but thinking about how a 1999/2000 design at that time. So a 50% increase in capabilities might not map to a 50% increase in peak polycount, but instead a 50% increase in lighting.
I didnt think about Kyro , but a modified realistic D+1year might be..

SH4@250Mhz with 16MB ram ( maybe bigger D$ - Hitachi tend to offer lockable D$, and the write buffers were considered a good solution to direct cpu to gpu communictation )
Kyro@250Mhz with 16MB ram

 

[MODEL 3]

Sega, made a lot mistakes from a hardware standpoint and market planning even on dreamcast design:
Their plan was to compete with the first playstation in the short term hopping that the small price difference (Dreamcast launched at 199$ and the price for playstation at that time was 99$-Surely someone that was buying for the first time a new game console would think that 199$ is not a high price certainly in relation with monthly or 60 days feas of 50$(new game))
Also they were late since 0,35nm SH4 200Mhz and power VR2 35nm 100Mhz (DirectX5 based) was perfectly possible on Q1 98 (ofcource a Q1 98 launch was possible also for the 12X GD-ROM and for the memory (16Mb and 8Mb 100Mhz 64bit SDR) and for the 56K modem and for everything with the same cost price).The problem was that all the industry wanted to exploit the huge installed (and growing rapidly) base of playstation and 90% of their resources (game design,marketing...) was for playstation.Also Sony was informing the industry that would launch PS2 in Q4 99 (only a year after DC launch) with staggering specs (250Mhz CPU and 125Mhz graphics chip 60million pps...).
Since the software the Dreamcast had was superb (despite the limited resources the industry devoting to it) in relation with what Sony had for PS2 the first two years of each console lifetime In order to 1)force the industry to devote resources to Dreamcast and 2)also compete effectively with PS2 on hardware terms they should have done this:
1) in order to force the industry (3rd party) to devote more resources to dreamcast (effectively simultaniously devoting less resources on PS1 and PS2) was to sell phenomenally good the first year so the developers had a sufficient installed base and faith in the SEGA brand for the future in order to devote resource to DC.That was going to be achieved if Dreamcast had a must buy feature that was going to lure millions into buying it.That feature was DVD-ROM.It was perfectly possible in Q1 99 (only five months after the original DC launch) a redesigned DC at 299$ with DVD (DVD-ROM 2X/CD-ROM 16X, speed) unit, at that time it would sell massive at 299$ becauce it was right on the beggining of the DVD era with no DVD hi-fi even nearly at that price.I also believe that the primary reason PS2 sold so well within the first two years was that it was the first 299$ DVD unit on the market.Ofcource PS2 had superb specs for the time of launch and also phenomenal marketing but it also had crap games.So someone was buying PS2 with the added feeling that it cheated SONY since it could not bought a DVD unit at 299$ elseware so what if it had crap games the value was there.Imagine that dreamcast could have invoke the same feeling to the consumers a whole year before PS2.I want to clarify that SONY had in Q1 2000 at 299$ a game console with a DVD unit with DVD-ROM 4X/CD-ROM 24X speed I assure you that in Q1 99 SEGA could have launched a redesigned DC at 299$ with a DVD unit with DVD-ROM 2X/CD-ROM 16X speed without making loss (in contrast with PS2).In fact maybe it would have made more money per unit in relation with the originall DC (199$).
2) in order to compete effectively with PS2 in hardware terms* they should have made the following adjustments:
a)original DC 200MHz 0,35nm Hitachi SH4 360mips/1,4Gflops -- redesigned DC 400MHz 0,25nm redesigned SH4 800mips/3,1 Gflops (10% per clock performace increase due to 6month more time to SH4 to redesigned more efficiently).Perfectly possible nearly at the same cost in Q1 99.
b)Original DC 100Mhz 0,35nm Power VR2 7,8 million transistors(actually NEC supplied 0,25nm CLX1 due to higher yields and increased profits for NEC) -- redesigned DC 133Mhz 0,25nm Kyro level of design 12million transistors.The actual PC based Kyro I launched in Q1 00 but it was actually a Q1 99 design and due to the late Neon 250 launch and the internal problems Imagination technologies was facing from design stage to actually launching a product delayed Kyro I.The fact is that in Q1 99 DC could have a complete DirextX 6+ based design like Kyro I.This would elevate dreamcast polygon and fillrate performance 2,67 times.Also it would elevate the feature set suppling better quality graphics possibility in relation to DC.If a first generation DC achieved 1,5 million pps (We didn't saw third generation engines on DC and also the second generation engines was only a handful and limited in scope and resources since it was rumored from Q1 01 that SEGA would go the 3rd party route) the redesigned DC would be capable for a 4 million pps first generation game.If we accept that a third generation engine maybe doubles the performance in relation with a first generation engine (the cases are many) then we could have seen a 8 million pps game in a redesigned DC maybe as early as Q4 03-Q1 04.For NTSC signal with 300.000 pixels 8 million pps is quiet sufficient becauce we have 135.000/60FPS or 270.000/30FPS
so for a RPG or Adventure game at 30FPS we would have all the time many polygons that they would be smaller than pixel size (not efficient) and for a racing game for example at 60FPS we would have the limit (for efficient usage of polygons) of polygons number if you count that 1/3 of the screen is the SKY.This is for DC level of quality graphics, if we though more light sources and better effect applied to textures (than DC level) that number is maybe optimistic.
c)Original DC 16Mb 100Mhz 64bit SDR main ram 8Mb 100Mhz 64bit SDR graphics ram -- redesigned DC 24Mb 133Mhz 64bit SDR main ram 12Mb 133Mhz 64bit SDR graphics ram.In Q1 99 that would be quit possible with 12-24$ increase in cost but not forget the 299$ launch price.We added 8Mb in main ram for better game logic, physics and polygon transformation and we added 4Mb graphics ram (3Mb for the 4,5 more million pps and 1Mb for texture memory)
d)I would not change the yamaha sound chip although it could be possible to found something better for a little more cash but it would not change anything substanially important about sound (at that cost) or about how the public perceive the Dreamcast power
e)I would not have included a 33k/56K modem (33K in japan) becauce the console gaming industry was not ready for online games at that time, the 56K technology was old news (33k it was o.k. for 1996 level of design) and becauce SEGA devoted a LOT of resources in a feature that it was only going to pay off after several years.We would be made available in Q3 01-Q3 02 a broadband modem seperetaly ala SONY maybe and I am saying maybe becauce like Nintendo I believe that although in Q3 01 the console gaming industry was ready for online games the console gaming public wasn't.
With these changes I think SEGA would have made it and I think it would achieve at the end of Q4 01 (at the same time that the technologically superior consoles Xbox and Gamecube would have lauched) 25 million installed base enough to secure a place in the resources of the industry and a better future for SEGA.
I have really many many more to say (not only technology staff) about SEGA and Dreamcast but I guess it is time to let this tired subject and focus on the future.

 

[Megadrive1988]

my ultimate 'dream' Dreamcast - or Katana - or perhaps another name, would have been something like this:

*a custom PowerPC 604 @ 400 MHz w/ custom extentions or custom vector unit to boost floating point performance by at least twice (maybe 3-4x) that of a normal standard PPC 604. it would produce 2-3 GFLOPs peak, and capable of sustained 1 GFLOP performance.
(slightly more than SH4's 900 MFLOPs sustained / 1.4 GFLOPs peak)

*128-Bit memory bus
- DC had 64-Bit bus

*48 MB SDRAM for main memory for CPU (GPU has access to this too)
-that's 3x the amount of main memory that DC had.

*40 MB SDRAM for video (exclusively for the GPU)
-that's 5x the amount of video memory DC had.

*8 MB SDRAM or other cheaper DRAM for audio
-that's 4x the amount of audio memory DC had.

*96 MB total memory.
-just about 3.7x the amount of total memory DC had.



the primary MEDIA is a custom rom disk, that resembles a ZIP drive / ZIP disc, but with 500 MB of capacity (yes, less than GD-ROM or even CD-ROM) what it lacks in capacity it more than makes up for in speed. providing a wonderful streaming capability to either a harddrive or RAM memory. this would be probably somewhat comparible to Model 3's ROM media in speed.

the secondary media is a standard 24x or 32x CD-ROM drive. both for backwards compatiblity with the 3 other Sega CD formats (SegaCD, Saturn, Saturn2) as well as more standard media for other purposes like web browsers and other software/files, music CD playback, etc. etc.



*a Lockheed Martin 'Real3D-500' GPU - this is a custom GPU with 3-4 times the performance of the combined TWO Real3D/Pro-1000 GPUs in Model 3.
(in Model 3 there is 1 pixel pipe in each GPU, 1 geometry engine in each GPU, 2 pipes total, 2 geometry engines/processors total)

this 'Real3D-500' console-only GPU has 4 pixel pipelines with 2 TMUs each.
(8 TMUs total like Xbox) in front of *each* set of pixel pipeline+2 tmus, is a geometry engine, so 4 geometry engines total in the GPU. the GPU runs at 133.33 MHz (1/3 the clock of the CPU)... giving it a pixel fillrate of about 532 Mpixels/sec and a texel fillrate of about 1064 Mtexels/sec. overall, the raw pixel pushing power of my dream Real3D-500 GPU is inbetween Nvidia's NV10 (GeForce1) and NV15 (GeForce2 GTS) but with more geometry pushing power, higher quality features that are all meant to be used at once, without any impact on framerate or geometry rates whatsoever.

since Model 3 (two Real3D/Pro-1000 GPUs) had a raw transform / vertex/ polygon rate of 60 million polygons/sec (yeah SIXTY million)...and 1.5 million texture mapped square/rectangle polygons/sec w/ ALL features on including AA, in game, sustained, my Real3D based Dreamcast / Katana would have around 4.5 to 6 million such polygons a second. keep in mind that's alot better than the real Dreamcast since it includes trilinear filtering, anti-aliasing, better mip-mapping. and everything else, at 60fps. something the real DC could never do in games.


* my hypothetical Lockheed Martin Real3D-500 GPU is 90% designed by Lockheed Martin Real3D, with contributions from Videologic the companies behind PowerVR and also TriTech the company behind Pyramid3D. about 5% each. kinda like other companies contributed to GameCube's ArtX designed Flipper, like S3. from Videologic, the Real3D-500 GPU gains a Hidden Surface Removal unit, like like part of the Image Synthesis Processor from PowerVR. then from TriTech/Pyramid3D, the Real3D-500 GPU gains bump-mapping and some early pixel shader abilities. the rest though is Lockheed Martin's very best.

*my Real3D based Katana / Dreamcast is backwards compatible with:
Master System cartridge, Master System card, including SMS SegaScope
3D glasses, GameGear cartridge, Genesis cartridge, Sega CD, Saturn, Saturn2 (Real3D+PowerPC upgrade like 3DO M2 but better)

...the Genesis 32X never comes out at all....

*a built in 56K modem from Rockwell (another military-industrial defence contractor company) plus standard 10/100 ethernet like Xbox. no need to buy an expensive and rare BBA.


launches in summer 1999 in Japan, rather than Nov 1998, which allows for superior, stronger hardware, and September 1999 in the U.S., like the real DC did.

equivalent of $350 in Japan. $299 in the U.S. and 220 UK pounds so that UK doesnt get screwed as usual with same unit pricing or whatever you'd call it.

 

[jvd]

that would be crazy expensive .

 

[Lazy8s]

The PS2 unit cost to Sony at launch was estimated at about $485 US equivalent. A lot could be delivered by any technology provider for that kind of budget.

 

[Guden Oden]

This thread is nothing but one big segafan crackpipe circle-jerk wet-dream, jesus some of the posts in here makes me head spin and I'm standing clear on the other side of the internet and I still get dizzy!

Btw, Lazy, do you have some linkage to that figure of yours? I've never seen anyone claim anything that high. I once had a pdf where a PS2 (first generation with the die-cast pin/fin heatsink) was taken apart and cost analyzed, but now I can't seem to be able to google it up again. Typical.

Anyway, I don't remember them coming up with a figure that high.

 

[cthellis42]

I can see that price for, say, demo units and initial cost premiums while starting up the lines and dealing with manufacturing problems, but I don't think it has any real bearing on the true cost of things when it got rolling. The financials don't line up, and I seriously don't think they could have slashed $185 in costs in ~18 months to get the hardware profitable.

 

[jvd]

This thread is nothing but one big segafan crackpipe circle-jerk wet-dream, jesus some of the posts in here makes me head spin and I'm standing clear on the other side of the internet and I still get dizzy!

eh the forum is normaly a sonyfan crackpipe circle-jerk wet-dream.


So its good to get a little something else going on