NV30 Memory Architecture?

Most likely the GPU is 512 bits wide internally but is memory bus 128 or 256 bits (is it possible to have both supported in the same silicon)?

If a 128bit bus then maybe is uses something exotic like QDR (200 to 250 MHz) (or very fast DDR (400+)) would be used. I believe it would have a bandwidth of approx 12.8 to 16 Gb/s
With a 256bit bus and standard DDR at 325 to 400 Mhz (Depending upon price/availability/release date) would result in a bandwith in the range of 20.8 to 25.6 Gb/s

The latest drivers (29.42) allude to the NV28, NV30 and NV31.
Just guessing the but the MX would be the NV28, the Intial Release the NV30 and the NV31 would be the performance version.

On a side note, was the GF1 SDR the NV10 and the GF1 DDR the NV11?
I can't remember and even had both of them :(

I guess the real question is, given the current competitor announcements R300,P10,Parhelia) which option makes the most sense?

-Simon
 
NV31 might relate to a slightly altered workstation variant. They've alluded to chip differences with the workstations versions of GeForce 4 but I don't recall hearing a different code name for it - if they've made substancial differences this time then they may chose to give it a different development name as well.
 
All workstation versions of nVidia chips were labelled "GL." I don't know why they'd change this now.

Anyway, an NV31 would almost certainly be a different version of the chip for the consumer-level, though I doubt they'd have a different version at launch. Still, if they do, one possibility would be a low-cost version of the part with only two pipelines or so working on a .15 micron process....
 
Chalnoth said:
All workstation versions of nVidia chips were labelled "GL." I don't know why they'd change this now.

Anyway, an NV31 would almost certainly be a different version of the chip for the consumer-level, though I doubt they'd have a different version at launch. Still, if they do, one possibility would be a low-cost version of the part with only two pipelines or so working on a .15 micron process....

That would fit...The NV11 (GeForce 2 MX) being a cut down NV10 (GeForce).

Stands to reason they would have a cut down NV30.

Though as you say, possibly not at NV30 launch time...
 
One other thing it might be is a "backup plan," in case the .13 micron process doesn't work in time or something.
 
I think that this NV-31 might differ with the memory just like the original Geforce w/ DDR & could therefore just be a change in the DDR-II spec'.
 
I thought DDR-II spec only provided better signaling characteristics and not neccessarily more bandwidth (of course, better signals integrity allows the MHz to go higher but all the reports I've read say they are in the currently in the same MHz range as standard DDR).

If performance is similiar then the other reason to switch would be cost and I would guess that this is not the case even though highspeed DDR is not relatively cheap.
 
DDR-II theoretically double the operational speed. So, it should provide significant performance increase.
 
That would fit...The NV11 (GeForce 2 MX) being a cut down NV10 (GeForce).
The GeForce 2 MX was a cut down GeForce 2 (NV15), it had the NSR onboard just like the GeForce 2 GTS, but with half the pipelines and slower memory
 
Back
Top