Core change to the G400 chip

[mdhome]

Hello Chaps,
I know this seems like a silly question but what changes would be needed to the core of the G400 chip to bring its 3d speed up in line with the Radeon? Is it a case of just adding a few extra pipelines etc? If so, could these changes be made easily or would they have to start from scratch.


regards Michael

[CHHAS]

The biggest difference is the T&L engine, but putting that aside, a G400 with 4 pipelines and 128 bit DDR ram should be very fast. Hmm ... throw in Dot-3 bumpmapping, HSR and FSAA and I would buy one

[Wombat]

But how much of the core is really left then?

[Cyber-Spanky]

They also need to pump up the clock speed to past 200MHz.

[Ali]

The Radeon T&L engine is as pointless as the Geforce1&2 T&L.

You need DirectX8 T&L nowdays, and wasnt there a rumour a while ago that the G800 needed DX8 to work properly?

Even without T&L, you just need to add a pipeline or two, and even at 150Mhz you would have quite a decent card. 128Bit DDR of course.

Sounds like a G550 doesnt it. ie not a full rebuild, just a modification of the G400 core.

Ali

[mdhome]

So... if we double the number of pipelines and increase its speed up to 200mhz what sort of speed increase could we be lookling at?
Another question also came to mind. How many textures can a G400 process simultaneously? I think a Kyro can do 8.

[CHHAS]

I believe the G400 can do 2 textures per pass (the same as Geforce2). The whitepapers on matrox.com only mentions single-pass multitexturing, so I take that as a 2

The Kyro is a bit different. It's pipelines can only do 1 texture per pass, but due to it's design it can do up to 8 textures (the limit in DirectX8) per geometry pass, how this translates into performance I have no idea

If the G400 had 4 pipelines with 2 textures per pass at 200 MHz, and DDR ram to match, it would have the same fillrate (800 Mtexels, 1600 Mpixels) as a Geforce2 GTS, but without the T&L, so I suppose that performance would be comparable.

And seeing that a lot of people speculate that the current, static T&L engines will be useless soon it might be very interesting ...

[Technoid]

If the "G550" or what it will be called is as fast as my Geforce 1 DDR i will be happy with it!

[Pace]

Although I haven't done very much large scale electronics, I'd say that a lot of the components would be reusable.

T&L for example, was often found as a separate chip on professional workstation graphics cards - I don't think it would be to hard to integrate it.

However, although it's easy to add numbers onto the specifications, that doesn't translate into simply adding transistors easily. The internal logic can change quite a bit at times, simply by going from 3-bit to 4-bit for instance, whereas at other times you could move from 4-bit to 64-bit quite easily

I don't know why you're asking this, but I'm guessing at a timescale reason While the "next generation" chip from everyone's favourite graphics card company could've been completed well before now, they've either been suffering setbacks, or spent too much time on the G450. But of course, the video facilities that will be in the G800 and the 'Plus' could have made chip design difficult, as the core will be getting quite big

Some other considerations are production problems (low yields from manufacture), distribution problems (getting the damn things to the places before they're released for sale) and maybe even the software side isn't ready yet, so they're doing some more coding and testing before any announcements

P.

PS: This info came direct from our inside man Ant