3:8 vertex and pixel shader in X800 and NV40 - coincidence?
- Thread starter tcchiu
- Start date
Evaluation via research and experiments probably. Most likely not only today's software. Both major IHVs have large DevRel teams that deal with what ISVs are working on as in future projects or what developers exactly want.
If X amount of Z is deemed to be sufficient for a specific estimated time-period and it can fit in the transistor budget, then X it will be. I know it sounds a big vague, but there are DevRel or IHV employees active on this board that can correct me or be way more specific about your questions.
If X amount of Z is deemed to be sufficient for a specific estimated time-period and it can fit in the transistor budget, then X it will be. I know it sounds a big vague, but there are DevRel or IHV employees active on this board that can correct me or be way more specific about your questions.
Well it's really hard to get into vertex limited situations on R3xx cards.
Most games are either fillrate limited or CPU (system) limited. (Usually, it depends on the settings.)
So it was logical that the 4:8 ratio would be reduced.
I wonder though if the NV44 will have 1.5 vertex shader pipeline
Most games are either fillrate limited or CPU (system) limited. (Usually, it depends on the settings.)
So it was logical that the 4:8 ratio would be reduced.
I wonder though if the NV44 will have 1.5 vertex shader pipeline
ROFL
I was just thinking of a scenario of not using any MIMDs at all for the value line (which would be cheaper from the HW perspective?), but I doubt it's worth designing a different unit for products of the same generation. It should be easier to just scale it down to a 2-way MIMD or just use the same 3-way as the 6600 has.
Hyp-X said:I wonder though if the NV44 will have 1.5 vertex shader pipeline
My current understanding is that NV44 will be 3 VS, 1 Quad, 64-bit.
trinibwoy said:
Not supporting a 128 bit bus would make the NV44 a tough chip to sell imo. Unless they're going to put some really fast memory on it instead. Anyway, they need some good benchmarks using the 128 bit bus version and then they can sell the 64 bit versions to people that doesn't know better. As a sidenote, try and find reviews of the 64 bit bus versions of the 5200, X300 or 9600. They're not exactly all over the place. I did a quick search for the X300 and the only review i found was using the 128 bit version.
Bjorn said:
Very true indeed, and yet the X300SE has found its way into a truckload of top-tier PCI-E OEM systems. Ironic that the DX7 GF4MX survived far beyond its expected life span as an OEM darling, and the DX9 (sic) FX5200 has sputtered badly after a single cycle with NVDA forced to drastically cut its orders to TSMC and still coping with an inventory glut in the channel. It's a cutthroat, razor-thin margin market that neither IHV can afford to neglect.
Bjorn said:
Its not an X300 specific review, but Anand uses one in their look at Intel 915 Graphics: Graphics Media Accelerator 900.
Hum interesting... apparently we haven't been throwing enough polygons at the R3x0 cards then, and now they take that right from us
I would be interested to see how the 3 vertex-units of the X800 match up to the 4 vertex units in the 9800 series. I don't suppose there are any benchmark figures around yet?
I would be interested to see how the 3 vertex-units of the X800 match up to the 4 vertex units in the 9800 series. I don't suppose there are any benchmark figures around yet?
Similar threads
