incredible! Parhelia is STILL champ as far as number of TMUs

[megadrive0088]

Is it not true that the Martrox Parhelia has the largest number of TMUs in a consumer graphics chip still, even after R300 and NV30?
Parhelia has 16 TMUs in total, 4 per pipeline. A 4:4 configuration. right?

I'm quite surprised that NV30 didn't have 2 TMUs per pipeline as almost everyone assumed. it was even reported by major websites like Anandtech that NV30 would have 2 per pipe. hmmmm.


how likely is it that R350 and NV35 will move to a 8:2 config? or perhaps a large number of TMUs do not matter as much as they used too...

 

[megadrive0088]

that said, I'd still like to see a Parhelia 2 with plenty of overkill. that is, a full DX9 part, perhaps VS/PS 3.0 - a 256-bit bus again, or perhaps (in the spirit of overkill) a true 512-bit memory bus! And this time, 8 pipes, like R300 and NV30, but 4 TMUs per pipe like the current Parhelia.

It would be able to stand upto NV35, surpass it in many areas, though falling behind in others (like Parhelia vs NV25)

 

[T2k]

Re: incredible! Parhelia is STILL champ as far as number of

Is it not true that the Martrox Parhelia has the largest number of TMUs in a consumer graphics chip still, even after R300 and NV30?
Parhelia has 16 TMUs in total, 4 per pipeline. A 4:4 configuration. right?

Yes, it is. But doesn't worth anything w/ 220MHz.

I'm quite surprised that NV30 didn't have 2 TMUs per pipeline as almost everyone assumed. it was even reported by major websites like Anandtech that NV30 would have 2 per pipe. hmmmm.

What for?
For 128bit?

how likely is it that R350 and NV35 will move to a 8:2 config? or perhaps a large number of TMUs do not matter as much as they used too...

I don't think it's gonna be the hard question...

 

[Nagorak]

Having multiple TMUs is more or less a waste now. Back when the Parhelia was supposed to be released it might have been a good feature, but no longer. I doubt any consumer graphics cards will come out with more than 2 TMUs in the future, and I rather suspect 1 TMU will be the norm. It's easy enough just to loop back for multiple texture passes, and you get a much bigger performance increase from pipes than TMUs. I don't know, we'll see I guess, but the fact that both NV30 and R300 were 1 TMU cards seems telling.

 

[WaltC]

Having multiple TMUs is more or less a waste now. Back when the Parhelia was supposed to be released it might have been a good feature, but no longer. I doubt any consumer graphics cards will come out with more than 2 TMUs in the future, and I rather suspect 1 TMU will be the norm. It's easy enough just to loop back for multiple texture passes, and you get a much bigger performance increase from pipes than TMUs. I don't know, we'll see I guess, but the fact that both NV30 and R300 were 1 TMU cards seems telling.

I would agree completely here. Also, shaders will at some point start taking over from TMUs as well--overall I think the 1/1 ratio has got to be much more efficient. As far as Matrox goes--this could be why their part doesn't clock any higher than it does (which would also explain ATI and nVidia's other motivations for cutting out TMUs.)

 

[Sage]

when the emphasis on new games is programmable shaders, we dont need to use much multitexturing to get the same, and even better, effects. theres really no need to worry about games pre UT2003, as any R300 / NV30 (with the exception of R9500 non-pro... yuck) should be quite capable of performing far above desired speeds, which means that we will need better and faster shaders. ahh one-handed computing... erm sorry, off topic. yes, back to shaders. more TMU's means more transistors, and we are getting to the point where the number of transistors is really becomming a problem. 130billion or more? heheh yeah, i dont think youd be adding one transistor you didnt need onto that sucker.