The ATI R600 Rumours & Speculation Centrum

I just have few questions?
Why does ATI working higher GPU clock speed for (Like 700-800MHz range for R600)

A. To match Nvidia G80 575MHz GPU performance.
B. To bypass Nvidia G80 575MHz GPU performance.
C. Or R600 delay was made to beat the upcoming G71 (8900GTX)
D. Current clock speed is not enough to fight G80 if R600 was release right now.

 

Most likely because it's been their design goal since the architectural features were set (probably two years ago or so).

 

E. None of the above and because it's how they designed it

 

You should have used this online poll creation website to allow ATI employees to vote anonymously and get a more reliable answer.

 

4 employees voting on a poll with 200 laymen, how is that going to influence it?

 

I'm not going to compare R600 with NV30.

But here is the question for you!
I know this for fact - original goal was for NV30 400MHz GPU: that was set by nvidia design team for their architecture when they were working on. But for some reason they pushed the core for 500MHz because ATI R300 was giving them hard time.

 

So WTF has that got to do with anything?Dude, you`re obsessed with these damned things...really. That comparison is horribly flawed, unless you guesstimate that ATi screwed up horribly(which nV did with the FX, and they were forced to do a steroid kid attempt with their part, the result being the dustbuster, crap availability etc.)

Why did ATi need to clock the darned thing anything?Well, they didn`t quite NEED to, they chose to, just like nV chose to clock the G80 like it did. You choose clocks that fit your architecture, because you want to have everything nice and balanced, you don`t chose clocks by what your competitor has, unless your architecture is incredibly close to his(and that`s highly unlikely).

 

Why ATI needed to clock their R520 X1800XT @625MHz GPU in order to match Nvidia G70 7800GTX 256MB @430MHz GPU ????

 

All I'm trying to say is - if this specifications are accurate with mathematical calculation I came to conclusion that - if you use quad core CPU Intel QX6700 with default setting and also default in 3DMark06, you should end up with score of for single R600 GPU.
800MHz GPU will be 13,500 to 14,500 3dmark points.
1000MHz GPU will be additional 1500+ points more.

If specification are actually lower for R600 then predicted, then I would understand why ATI needed high clock speeds.

 

And why did Intel clock P4s at 3GHz+ while AMD clocked Athlons at only 2GHz+ ? Maybe because P4 architecture was _designed_ for high clock speed, and Athlon was not?... If AMD had the possibility to clock Athlons at 3GHz+ keeping acceptable dissipation and good yields, they would have done so!
Clock speed is not just an adjustment parameter. It is intimately tied with an architecture design.

 

There is some room to play with the clocks, but not much maybe 50mhz + or - which yeah could make a difference. Like we saw with the r420. But I think ATi is won't get into a situation like that again.

 

Not that great but new rumor, the r600 PCB is going to be green?

 

uh, oh, NVidia bought AMD ?!
seriously, if AMD fails to transition to 65nm & K8L , AND ATifails to deliver (not only R600, but failing to get low-mid level DX10 cards at same time as NV), this may become an option

 

That’s the way I see it. I would not say that nvidia implemented their support for different “Capsâ€￾ always in a fast way but at least the offer some support and could learn from the experience.

The difference from this alone may not be large enough to generate a noticeable problem but with the two console GPUs and the somewhat failed R400 projects things have summed up.

 

That would be the end of the world as we know it.

 

I'm sure there are a number of scheduling or bypass paths that cross a significant portion of the core. The fact everyone seems to want to keep the execution logic portion at a relatively small space compared to total die size seems to indicate this becomes a problem at sizes significantly smaller than 400 mm2.

The only exception in recent memory I think was Prescott, which really bloated the non-cache sections. Overclocking tests seem to indicate that it wasn't timing limited 4+ GHz (30+ pipe stages yay!), but other considerations came into play.

It might depend on just how broadly ATI wants the general architecture of R600 to scale. Adding pipe stages for signal propogation might make sense if the chip is a 400mm2 high-end, but would be overly pessimistic for mid-range and lower chips.

On the other hand, having enough stages for good clock speeds at the lower end may hinder the high-end chips. Perhaps that is a factor in the latest rumored clock speeds.

I'd guess that if there are clock domains, those schedulers would be in the slower-clocked domains.

Itanium before hitting silicon was probably over 600 mm2. It was also likely rectangular, so it probably exceeded 25 mm on a side.

 

WHAT?!?!?!

Blasphemy, I'll check this out and get back to you.

 

Why for the love of God is the colour of the PCB of any importance?

 

You obviously never read my title.

 

I doubt the PCB color would change from red to green, red is part of the ATI brand AMD is keeping up