Yeah, or you could make a super-duper fast T&L chip! Rage 256 Ultra!! :? Sorry for the sarcasm, but ATI's already behind in features. Speed is very important, but they're going to have to start implementing concepts introduced in SM3.0 some time or another, so might as well do it now and learn something from the process. If sacrificing an uber-boost of speed is what it costs, then that's perfectly fine.
Does Dave have ATI SM 3.0 Hardware?
- Thread starter nelg
- Start date
ANova said:
Let me guess, when the DX 9 transition began and the debates heated up between Geforce FX FP16/32 and Radeon 9700 FP24, you were one of the few advocating that everything should be FP16 because this can be faster? If you didn't, Nvidia could certainly have used your logic. ATI's FP24 is just 8 wasted bits, right? Bits are trannies and trannies that could be used elsewhere. Am I reading you right?
Asking which will run faster is very silly. Going by ATI's current track record I would say they would run the same. I say this because ATI has been conservative with transistors and I therefore cannot jusitfy a prediction other than that they would take the saved die space/trannies instead of filling out space.
ANova,
Your making an incorrect comparison between architectures.
I was under the impression that for the Geforce FX the number of 'in flight quads' within the pipeline at one time was a quotient of the size of the register file and the amount of memory a quad takes up.
So FP16 was faster (than FP32) because it was possible to pack two FP16 temporary registers in the same space as single FP32 temporary register -thus reducing register file pressure (which allowed more 'in flight quads' within the pipeline).
It don't think ATI's arch has the same extreme sensitivity to register pressure.
Your making an incorrect comparison between architectures.
I was under the impression that for the Geforce FX the number of 'in flight quads' within the pipeline at one time was a quotient of the size of the register file and the amount of memory a quad takes up.
So FP16 was faster (than FP32) because it was possible to pack two FP16 temporary registers in the same space as single FP32 temporary register -thus reducing register file pressure (which allowed more 'in flight quads' within the pipeline).
It don't think ATI's arch has the same extreme sensitivity to register pressure.
Haven't been visiting this site very often lately hence this late comment from me.
To be honest, I don't quite like what Dave has done wrt his signature (as it was, as well as its edited version).
Of course, he could just be poking fun at some individuals... but he shouldn't have felt the need to...
Anyway, carry on...
nelg said:
To be honest, I don't quite like what Dave has done wrt his signature (as it was, as well as its edited version).
Of course, he could just be poking fun at some individuals... but he shouldn't have felt the need to...
Anyway, carry on...
Unknown Soldier
Veteran
The Rev just sore that he don't have anything similar in his sig??
You could always add a D3 quote/saying.
US
You could always add a D3 quote/saying.
US
I always thought his sig would look better as:Unknown Soldier said:
A little formatting just classes it up and gives it that professional look, ya know?
I don't know about the colors though, I was going brown for his name thinking along the "Tan" lines...and mebbe grey would be better for Beyond3D. :?
I was under the impresion that g.i was a sm3.0 feature that ati has had for the past 3 years now in thier r3x0 cores and in the r42x cores ? So they are implementing concepts .
As for the speed sacrifice i think you will find that ati will manage to increase the speed of the r520 over the r420 and its feature set . After all that is what nvidia did with the nv40 over the nv30
Well if we look at what performance enhancements are possible with the R520 you can see that it might turn out to be quite a leap over R420.
1) The clock speed should be higher. (10%)
2) The inquirer rumor of 2 more processing units per shader.
3) The rumors of 24/32 pipelines. (50%-100% pipeline thoughput)
4) 512MB memory (0%-30%)
5) Faster memory and bandwidth and rumors of an enhance memory interface. (10%)
6) Richard Huddy's slideshow comments about fast 3.0 hardware (10%)
7) Sirerics comments that 32bit can be as fast as 24bit, its all in the implementation.
Of course this is total guesswork.
1) The clock speed should be higher. (10%)
2) The inquirer rumor of 2 more processing units per shader.
3) The rumors of 24/32 pipelines. (50%-100% pipeline thoughput)
4) 512MB memory (0%-30%)
5) Faster memory and bandwidth and rumors of an enhance memory interface. (10%)
6) Richard Huddy's slideshow comments about fast 3.0 hardware (10%)
7) Sirerics comments that 32bit can be as fast as 24bit, its all in the implementation.
Of course this is total guesswork.
I certainly didn't mean to imply that ATI would be sacrificing speed. I was saying that a major increase in speed would be sacrificed. That goes back to my sarcastic statement about a fast T&L chip. You could save all the transistors you want and just work to boost the speed of existing technology, or you could add something new and accept a lesser increase of speed. That's all I was saying.jvd said:
Ostsol said:
I disagree . Remember ati is going from 110nm to 90nm with this . So they can get a drastic speed increase .
They can go to 24 pipe lines and stay at 550 mhz and get 13200 vs the 8800 of today . Couple that with faster memorys and you can get a pretty big leap .
Though it will be the memory that holds them back
How about if they stuck with FP24 and SM2.0? They could probably double the pipes as well as boost the clock speed, giving them quite an extreme card indeed, if a bit behind in features. That was the type of comment that prompted my post in the first place.jvd said:
I agree that a full SM3.0 card will be able to be faster. I don't know how much faster, though certainly not as fast as it could be with an older feature set. It would still be very worthwhile, given that ATI's developer relations would then be adding to NVidia's push for SM3.0 in more games.
Joe DeFuria
Legend
Ostsol said:
Right.
As a gamer, I would personally rather see exactly that card. Stick with FP24 and SM2.0 for now, but make it significantly faster. Because personally, I don't see SM 3.0 being all that much more "capable."
I don't really see the next "feature" inflection point coming until SM 4.0 / DX10 / whatever it's going to be.
If I were an ATI stock-holder though, I'd say a "competitive in both performance and features" is probably what I'd want to see at this point. You don't want to lag behind in features for "too long", otherwise you will lose status as a market and technology leader, which hurts the brand...and thus sales.
So I think what we're all expecting R520 to be..."faster, but not 'hugely faster', plus feature competitive" is the right product for ATI to bring to market. Again...that's actually NOT what I would personally prefer to see.
But I'm sure you're going to be changing your tune as soon as ATi supports SM3.0 and it will be just the bee's nuts then.Joe DeFuria said:
Oh wait, this isn't the Religion, Politics & Socioeconomic Climate forum....is it?
Sorry Joe, reflex.
Joe DeFuria
Legend
hovz said:
If I had to guess, we'd be in the minorty. Marketing wants "checkbox features" for a reason...they tend to work.
We are already seeing games that are SM1.1 and SM3.0 only with no fall back for 2.0. ATI has to move to SM3.0 to stay competitive. Whats the point of having a super fast 2.0 card if future games only support 1.1 and SM3.0.
Consoles are going to dictate the development of future games which is why SM3.0 is more important than SM2.0 or even SM4.0.
Consoles are going to dictate the development of future games which is why SM3.0 is more important than SM2.0 or even SM4.0.
I bloody hate sheeple!
Joe DeFuria
Legend
egore said:
In PC land there are a zillion installed parts supporting SM2.0. That will influence PC shader support.
