48 shaders vs 32 pipeline?

[Redeemer]

I am kind of new to this, which approach is better? Why is ATI and Nvidia so different in their approach?

 

[Ailuros]

Where's the difference in that particular approach? Both are increasing arithmetic efficiency via increased amounts of ALU units.

We'll most like see on R580 16 SIMD channels with 3 ALUs each and on G7x 32 SIMD channels with 2 ALUs each. While there are fundamental differences and approaches deeper in the pipeline, both inrease ALUs as much as they can.

 

[Redeemer]

What about performance in games.

 

[Ailuros]

What about performance in games.

I can only guess over unreleased products; frankly I doubt though that they'll be any significant difference between them on average.

 

[Richteralan]

I don't think they are that different in terms of performance because the TMU/ROPs stays the same.
Unless there will be some games that have VERY HEAVY use of shaders.

 

[Fodder]

Unless there will be some games that have VERY HEAVY use of shaders.

I think that's the idea ...

 

[ANova]

We'll most like see on R580 16 SIMD channels with 3 ALUs each and on G7x 32 SIMD channels with 2 ALUs each. While there are fundamental differences and approaches deeper in the pipeline, both inrease ALUs as much as they can.

The R580 is 16 SIMD channels with 3 pixel processors each and 1.5 ALUs per processor.

 

[Razor1]

What about performance in games.

For the next year titles its probably going to be very very close, the only real place where we might see differences is with dynamic branching, and that will only show up if nV hasn't improved there performance in this area.

 

[boltneck]

I want a card that has 32 ROPs and 64 SIMD pipelines running at 700mhz.

So what if it generates enough heat to fry a dozen eggs at the same time

Whichever company comes out with one first will get my money. Otherwise I am sticking with the X800 I have now, and the 360 i am going to bet

 

[boltneck]

errr "get"

 

[Reverend]

Unless there will be some games that have VERY HEAVY use of shaders.

Such beasts will not exist unless there is a group of one IHV.

 

[Pete]

I want a card that has 32 ROPs and 64 SIMD pipelines running at 700mhz.

Then you'd better start saving up for SLI 7900s now. Don't forget a PSU, too.

 

[Ailuros]

The R580 is 16 SIMD channels with 3 pixel processors each and 1.5 ALUs per processor.

I'm oversimplifying things in order to understand this whole mess; if I multiply 16 * 4 MADDs times the clockspeed I can reach the claimed 83 GFLOPs for R520.

48*4*0.65*2 vs. 64*4*=/>0.55*2

Does that sound better?

 

[Dave Baumann]

Ail, the ADD's on the second ALU are still a FLOP per cycle.

 

[N00b]

Unless there will be some games that have VERY HEAVY use of shaders.

Which games will that be? UE3 based ones? S.T.A.L.K.E.R? I doubt that we will see so many shader heavy games this year.
Extrapolating from past benchmarks and current rumors (R580 @ 700 48 Shaders 11000+ 3DM05, G71 @ 700 MHz 32PP) I don't see how the R580 will be competitive with the G71 if this year's games are comparable to last years games unless there are any other architectural improvements to R580 that we don't know about yet.

 

[Ailuros]

Ail, the ADD's on the second ALU are still a FLOP per cycle.

I read in some GPGPU experiments about the claimed 83 GFLOPs for R520 and reversed the math in order to reach 16 * 4 MADDs/clk. So it's in the order of 120-something in reality?

 

[Rys]

Yeah, if you count the ADDs, too, you get a higher instruction rate/flop count.

 

[Ailuros]

...and the "mini-ALUs" on G70 are exclusively capable of MULL yes?

 

[Rys]

Nope, those mini ALUs are for PS1.4 modifiers and instructions like RSQ, if memory serves. Only the main ALU pairs can issue a MUL (and NV40 can issue MUL from both, too).

 

[Ailuros]

Nope, those mini ALUs are for PS1.4 modifiers and instructions like RSQ, if memory serves. Only the main ALU pairs can issue a MUL (and NV40 can issue MUL from both, too).

Gosh do I hate brainfarts