Thanks for the details, CMKRNL!
I'd still like to note that the gap between a 300Mhz NV31 and a 400Mhz NV30 is *huge*, and it would keep a nice gap for ATI to fill.
400*8 = 3200MegaPixels/s
300*4 = 1200MegaPixels/s
A 350Mhz NV30 would make more sense. It's still a big gap, but it's already much more reasonable. And it should still be able to beat a Radeon 9700 Non-Pro.
350*8 = 2800MegaPixels/s
Now, for the NV34 memory bus... I don't think it makes sense for it to have a 64 bit memory interface. The NV34 is going to be the IGP of the next nForce. The GF4 MX was 64x2 instead of 32x4 for two reasons IIRC: to save money and to make it work more easily on a nForce.
So it makes a lot of sense for the NV31 to be 32x4 ( like the GF4 Ti and the GFFX ) and the NV34 to be 64x2.
But then, how will nVidia cut cost sufficently? To cut memory cost, no 128MB version and they'll use 64x2 as explained above. To cut chip cost, cutting functionality. To cut board cost, less efficient cooler and less features.
So, ideally, nVidia desktop line-up for H1 2003, if ATI products doesn't force them to cut prices, would essantially be:
*speculation*
$499: NV30: 500Mhz core: 4000MP/s, 256MB 500Mhz DDR2 w/ 32x4 bit memory bus: 16GB/s
$399: NV30: 500Mhz core: 4000MP/s, 128MB 500Mhz DDR2 w/ 32x4 bit memory bus: 16GB/s
$299: NV30: 350Mhz core: 2800MP/s, 128MB 400Mhz DDR2 w/ 32x4 bit memory bus: 12.8GB/s
$199: NV31: 325Mhz core: 1400MP/s, 128MB 300Mhz DDR1 w/ 32x4 bit memory bus: 9.6GB/s
$169: NV31: 275Mhz core: 1100MP/s, 64MB 275Mhz DDR1 w/ 32x4 bit memory bus: 8.8GB/s
$129: NV34: 250Mhz core: 1000MP/s, 64MB 200Mhz DDR1 w/ 64x2 bit memory bus: 6.4GB/s
$99: NV34: 200Mhz core: 800MP/s, 64MB 160Mhz DDR1 w/ 64x2 bit memory bus: 5.3GB/s
$69: GF4 MX 440-SE w/ AGP 8X ( NV18 ) : 200Mhz core: 500MP/s & 1000MT/s, 64MB 160Mhz DDR1 w/ 64x2 bit memory bus: 5.3GB/s
That's highly speculative, but it would be a very nice line-up. The only problem which would remain is the Radeon 9500 Pro, which got 2200MP/s at $219 - it beats the 1400MP/s NV31 at $199 very easily. So that line-up would still have several problems...
However, you also mention the NV33, but give no additional details. Could that be what fills in the gap between the GFFX at 350 or 400Mhz and the NV31 at 300Mhz? And does the NV32 exist, or is it a number which will never be used?
Also, CMKRNL says:
For reasons I can't get into, I don't think NV31 will be running significantly higher than 300Mhz on the current process.
Hmm, so you're saying NV31 isn't on the same 0.13 as the GFFX? AFAIK, the GFFX uses TSMC FSG 0.13 - I'm unaware of any slower TSMC 0.13 method. Is nVidia doing as ATI, and going UMC for low-end parts? Would make sense, but nVidia got a so good relationship with TSMC that this would be surprising.
Trident XP4, also on UMC 0.13, is clocked at 300Mhz. So it sounds like UMC 0.13 clock rate is around that. It remains possible, but unlikely.
Anyway, is the NV34 on 0.13 too? Or is it 0.15?
Not really asking for answers on my questions. But any answer would be nice
If we don't get any, speculation is always fun
Uttar