G80 & R600 Unified Architectures, what's the LCD?

[Pete]

I'm really not the one to be diving into details/semantics, so this is all just AFAIK, but I'll share what I remember. R580's PS ALUs are vec3+1, thus "vec4." G70's are more flexible but only two-way superscalar: 3+1 or 2+2. As for the "8 vec5s," you're looking at their vertex shader ALUs (each vec4+1, so "vec5") rather than their pixel bits. Wavey made the distinction b/w vec5 and 5D superscalar, but I don't know which is more accurate.

Yeah, I guess over time and driver revisions the comparison #s would get muddled as performance improves, but I'd think the varied nature of the demos and systems would provide a fairly reasonable overall picture. GPU reviews (R600 being the latest example) make clear that the demo affects benchmarked performance as much as the game itself, so getting multiple benchmarks of a game seems preferable to seeing R600 do well in game X in one review and poorly in another (without update drivers as an explanation).

 

[Reverend]

When you're trying to make a decision between two cards, often you don't need to read several 10 page reviews to figure out what's better.

Maybe it depends on a user's specific expectations.

Perhaps you meant "in a hurry to make a decision"?

The specs are often more than enough to figure things out.

Even if history has shown that "good" paper specs usually means good or decent feature-specific performances, history has also shown relying on specs alone (which is what I assume your sentence means) for even that one instance is never advisable IMO.

 

[^eMpTy^]

Maybe it depends on a user's specific expectations.

Perhaps you meant "in a hurry to make a decision"?


Even if history has shown that "good" paper specs usually means good or decent feature-specific performances, history has also shown relying on specs alone (which is what I assume your sentence means) for even that one instance is never advisable IMO.

Well the cases I'm talking about here is comparing specs of the same family to see which is better. Obviously comparing the HD 2900 XT and the 8800 GTS based on paper specs alone would be almost impossible.

But when you're just trying to figure out which of a bunch of cards from the same family is faster (i.e. the X1900 gt, the X1950 Pro, the X1900 XTX or the X1950 XT) the specs are extremely helpful. As well as situations where one card simply outclasses another, but that's not readily apparent to the average user based on the nomenclature (i.e. the X1550 vs the 6500).

I agree completely that benchmarks are the way to go, but in many situations benchmarks are not available, or they're hard to find. And the specs are always just a click away.