Model 3 Step 1/1.5 and Step 2 Video Board Differences

[MetalliC]

This is pretty fascinating. I've not heard of Real3D developing anything after the Pro-1000.

there was at least relatively low cost projects like Real3D Lightning/100, which later turned into Intel 740

I believe work did start on a successor architecture, though, but no one has ever mentioned it being connected to Sega or seeing the light of day in any form. Who else might have developed this chip?

well, Hikaru's GPU clearly looks like Model 3's GPU / Pro-1000 successor and doesn't looks any similar to other GPUs of that era (I've been checking).
why you expecting it was designed by anyone else but Real3D and Sega ?

and in general, why it's so hard for you to believe Hikaru is built using Real3D's technology?

Could it just have been some custom chip by a Japanese vendor?

Hikaru GPU's chips is gate arrays from Fujitsu and NEC, and I'd bet Model 3 GPU too.
At the time, it was common practice for Japanese companies to require that device components be domestically produced.

PS: btw, Hikaru GPU ASICs clocked at 41.6MHz which looks like a slight improvement compared to Model 3 Step 2 33.3MHz

[MetalliC]

I believe work did start on a successor architecture, though, but no one has ever mentioned it being connected to Sega or seeing the light of day in any form.

check this Real3D interview from 1998 or so
https://forums.guru3d.com/threads/an-ol ... ny.280312/
or archived source
https://web.archive.org/web/20000915221 ... r3dint.htm

this part

HGN: On the subject of Sega, how did you work with them on the Model X boards? Did they come in with ideas or specs that they wanted or did you just show them your product? Also what can we expect from Model 4?

R3D: Our relationship with Sega is very much a joint effort where both companies discuss ideas, market requirements, specifications -- basically everything that will go into producing the world's best arcade games. Since the graphics chips and boards we do for Sega are custom for Sega, we really couldn't just show them our "product", because it's not a product until the specs and requirements are set. When we first started with Sega, what we showed them was the sophisticated visual systems we had done for the military. Sega believed (and we believed) we could adapt that technology to the arcade and if this could be done, the level and sophistication of arcade graphics would take a quantum leap forward. Well, as you know, we were able to adapt the technology, which resulted in Model 2 and later Model 3. And Sega continues to enjoy a market leader position in the arcade space.And you ask about Model 4. I know you and your readers are probably anxious to know what's on the drawing board, but I really can't comment on Model 4 or future arcade development efforts with Sega. Hope you understand.

he says he can't comment, but I least he did not denied "future arcade development efforts with Sega" related to "Model 4" development.

[gm_matthew]

Hikaru GPU's chips is gate arrays from Fujitsu and NEC, and I'd bet Model 3 GPU too.
At the time, it was common practice for Japanese companies to require that device components be domestically produced.

PS: btw, Hikaru GPU ASICs clocked at 41.6MHz which looks like a slight improvement compared to Model 3 Step 2 33.3MHz

There's actually pictures of two of the Real3D ASICs decapped here and here and while I'm no expert on the matter, I can tell they're not just gate arrays.

If someone were to decap one or more of the Hikaru graphics chips it might answer some questions.

I presume that Hikaru must have some kind of clock multiplier; the RAM is rated for 125 MHz which is three times 41.6666 so perhaps what's what I presume the Hikaru graphics chipset uses. We're pretty sure that the Model 3 Step 2.x video board doubles the 33.333 MHz clock to run the Real3D ASICs at 66.666 MHz.

[MetalliC]

There's actually pictures of two of the Real3D ASICs decapped here and here and while I'm no expert on the matter, I can tell they're not just gate arrays.

looks sexy! glad to see I was wrong

I presume that Hikaru must have some kind of clock multiplier; the RAM is rated for 125 MHz which is three times 41.6666 so perhaps what's what I presume the Hikaru graphics chipset uses.

okay, if you want more details:
yes, Hikaru have x2 xtal multiplier, it supplies 83.(3)MHz clock to: "Antarctic" command processor ASIC, its SDRAM and cache RAM, and to "Europe"'s SDRAM.
but, everything else - Africa, x2 Atlantis, America, x2 Australia, Europe (i.e. ASICs which does actual gfx processing/rendering work) really clocked at just 41.(6)MHz

We're pretty sure that the Model 3 Step 2.x video board doubles the 33.333 MHz clock to run the Real3D ASICs at 66.666 MHz.

I'm pretty sure in Model 3 Step 2 video board 33.(3)MHz OSC1 output goes to IC58 FCT3807 clock driver and then directly to all the ASICs (Mars-s, Earth-s, Venus, Mercury, Jupiter).
sorry, there is no PLL/multiplier, it really works at 33.3

PS: it is especially funny to see so called "tech specs" at Sega fanboy sites like segaretro, which claims Hikaru GPU clocked at 250MHz

[Bart]

and in general, why it's so hard for you to believe Hikaru is built using Real3D's technology?

It doesn't fit the timeline. I've spoken to a few people who used to work at Real3D, some overlapping the Model 3 era, and no one has ever mentioned a post-Model 3 project with Sega. Real3D was pretty much finished by 1999 and in the year or two prior, they were trying to move into consumer PC graphics with that Intel partnership.

Why would either company keep the partnership a secret? Maybe it really is a Sega custom design. Didn't Konami and a few others bang out custom ASICs a couple years earlier?

Sega by this time had been working with both PowerVR and 3dfx. It's conceivable they also had yet another partner or may even have tried to develop custom IP.

[MetalliC]

It doesn't fit the timeline. I've spoken to a few people who used to work at Real3D, some overlapping the Model 3 era, and no one has ever mentioned a post-Model 3 project with Sega. Real3D was pretty much finished by 1999 and in the year or two prior, they were trying to move into consumer PC graphics with that Intel partnership.

I see your point.
but, as was noted above, most of Hikaru GPU ASICs have relatively low part numbers 608x, which fits somewhere in late 1997 - early 1998 period.
and all but 1 of them works at relatively low clock 41.6MHz.
all of that makes me think it was actually designed in 1997 or a bit later, which is fits in timeline - about mid summer 1997 was finished work on Model 3 Step 2 video board, so Real3D and Sega may have some time and resources for work on (a bit) improved GPU.

Why would either company keep the partnership a secret?

because it is pretty common practice. it's business.
because it is / will be your product with your brand name, and you don't want to promote some partner company (for free).
or, it may be vice versa - you may want to use partner company name to promote your product, and most likely it will not be free for you

anyway, in very many cases we wasn't aware of who actually designed one or another chipset for some gaming device, until it was decapped and examined.

Maybe it really is a Sega custom design. Didn't Konami and a few others bang out custom ASICs a couple years earlier?

I don't believe it may be fully custom desing, there is no point to create from scratch some unique and totally new 3D GPU, it will cost too many resources and will take too many time, especially for a company who not fully focused on this (3D HW development) area.

but using some licensed IP - yes, I think it was possible.

[gm_matthew]

okay, if you want more details:
yes, Hikaru have x2 xtal multiplier, it supplies 83.(3)MHz clock to: "Antarctic" command processor ASIC, its SDRAM and cache RAM, and to "Europe"'s SDRAM.
but, everything else - Africa, x2 Atlantis, America, x2 Australia, Europe (i.e. ASICs which does actual gfx processing/rendering work) really clocked at just 41.(6)MHz

I'm pretty sure in Model 3 Step 2 video board 33.(3)MHz OSC1 output goes to IC58 FCT3807 clock driver and then directly to all the ASICs (Mars-s, Earth-s, Venus, Mercury, Jupiter).
sorry, there is no PLL/multiplier, it really works at 33.3

PS: it is especially funny to see so called "tech specs" at Sega fanboy sites like segaretro, which claims Hikaru GPU clocked at 250MHz

I'm skeptical that most of the Hikaru video ASICs actually run at only 41.6 MHz. For a start, at that speed why would they even need heatsinks? Also, if "Europe"'s SDRAM is clocked at 83.3 MHz then surely "Europe" itself would have to be running that fast in order to use it. Perhaps the ASICs have internal clock multipliers?

At least some of the Real3D ASICs must be using clock multipliers because otherwise Model 3 Step 1.5 would be limited to a maximum fillrate of 33 megapixels per second or 573,672 pixels per frame, and I know from testing that Scud Race definitely outputs more pixels than that.

All of the hardware stats that you see on Sega Retro were written by one guy (Over9000) who clearly has no idea what he's talking about

[MetalliC]

I'm skeptical that most of the Hikaru video ASICs actually run at only 41.6 MHz. For a start, at that speed why would they even need heatsinks? Also, if "Europe"'s SDRAM is clocked at 83.3 MHz then surely "Europe" itself would have to be running that fast in order to use it. Perhaps the ASICs have internal clock multipliers?

it sounds weird and unusual, but it really as was said above.
for example Europe chip have 4 clock inputs: main clock 41.6MHz, PCI clock 33MHz, SDRAM clock 83.3MHz, pixel clock ~52/32MHz depending on video mode.
Antarctic is similar have 4 clocks: main clock, SDRAM clock, cache RAM clock (all is same 83.3MHz), and PCI clock 66MHz
Africa, x2 Atlantis, America, x2 Australia is homogeneous and use only 41.6MHz clock, which makes me think at some point they was designed as one device, while Antarctic and Europe was glued to them later.

why they need heatsinks? - perhaps because they are a bit outdated Fujitsu gate arrays, and even at 41MHz clock they produce too much of heat?
also, as for 1997 it was quite good enough GPU clock.
only Antarctic is NEC's CMOS gate array, which is probably newer and faster, so able to work at 83MHz.

At least some of the Real3D ASICs must be using clock multipliers because otherwise Model 3 Step 1.5 would be limited to a maximum fillrate of 33 megapixels per second or 573,672 pixels per frame, and I know from testing that Scud Race definitely outputs more pixels than that.

... or they may use some hidden surface removal algorithms, to not produce unneeded fill rate load.

btw, 573K pix/frame is very good for the time. for comparison Dreamcast's PowerVR2 had real fill rate up to 1.6Mpix/frame @60Hz, but it was able to process scenes with many times more complex opaque objects (up to 53Mpix/frame), thanks to HSR.

[gm_matthew]

btw, 573K pix/frame is very good for the time. for comparison Dreamcast's PowerVR2 had real fill rate up to 1.6Mpix/frame @60Hz, but it was able to process scenes with many times more complex opaque objects (up to 53Mpix/frame), thanks to HSR.

Model 3 performs hidden surface removal using the z-buffer; here's a link to the Pro-1000 product description, look for "hidden surface removal" on page 21. The way it works is that Earth sends the 1/z value of each pixel to be rendered to the depth buffer 3D-RAM and if there are no other pixels in front of it, the 3D-RAM updates the z-buffer and sends a signal back to Earth to render the pixel. If there is another pixel in front of the one to be rendered, nothing happens.

However, this doesn't improve performance because each pixel has to be checked one by one, and it is almost certainly not performing any kind of occlusion culling. Here is a screenshot of Sega Bass Fishing running on real hardware where the bass gets close to the screen, causing the hardware to become fillrate-limited and run out of time before the floor of the lake can finish rendering:



I tested this in Supermodel by performing an occlusion query and disabling the z-buffer (otherwise it doesn't count pixels that fail the depth test), and at this point it peaks at just over 900,000 pixels per frame.

PowerVR was able to achieve good performance because it performed the rasterization process on up to 32 pixels at once, and then only the visible pixels would be rendered. Performing this kind of tile-based hidden surface removal doesn't help if the hardware can only rasterize one pixel per cycle. (Actually it probably would help early GeForce cards which were limited by memory bandwidth, especially in 32-bit color.)

I can't rule out the possibility that Hikaru has some kind of occlusion culling to help performance, but I am very confident that Pro-1000/Model 3 doesn't have it.

EDIT: Forgot to mention that the Real3D Pro-1000 has a fillrate of 50 megapixels/s per pixel processor (page 12 of the Pro-1000 product description linked above); we’ve always assumed that Model 3 Step 1.0 matches this, and that Step 1.5 and Step 2.x are clocked higher.

Hey Bart, is it possible for you to identify the frequency of the clock crystal on a step 1.0 video board? I can only find pictures of step 1.5 and step 2.x video boards.

[gm_matthew]

Okay, I believe I have conclusive evidence of Scud Race outputting more than 573,000 pixels in a frame which means the fillrate of Model 3 Step 1.5 is higher than 33 megapixels per second.

Here is a frame during the attract mode that really stresses the fillrate:



In total there are 785,006 pixels rendered in this frame, however this does not necessarily mean that all of these pixels are actually rendered on real hardware; perhaps it stops rendering but any remaining polygons are not visible.

Here's the frame after 670,775 pixels have been drawn:



There are significant gaps in the house that do not appear on real hardware, so it must be rendering more than 670k pixels on this particular frame. This would be impossible with a fillrate of just 33 megapixels/s without occlusion culling (which I am confident that Model 3 does not have). This doesn't even take into account the time needed to clear the frame buffer, or any overhead from rendering small polygons; no 3D renderer is 100% efficient in terms of fillrate (except perhaps PowerVR which uses a very different implementation).

I doubt that Sega would have accepted a fillrate of 33 megapixels/s for Model 3 given that the Model 2 manages 50 megapixels/s (which I have confirmed from similar testing in MAME). Heck, the Model 1 manages 36 megapixels/s and has to run games at only 30 fps.

The point is that just because an ASIC is receiving a particular input frequency doesn't necessarily mean that it has to be running at that frequency. For example, the PowerVR chip used by Dreamcast receives 33.333 MHz but runs at 100 MHz.

It is possible that we've been wrong all along and Step 1.5 is no faster than Step 1.0, with both running at 50 MHz (1.5x the 33 MHz input frequency). Or perhaps they both run at 66 MHz and lose quite a few cycles processing each polygon, lowering the effective fillrate. The only way to find out for sure would be to write a custom program to test the fillrate on a real Model 3 board.

Okay, that's enough of me jabbering about fillrates for one day