wasn't anyone surprised at how ATi achieved 300+ mhz with the Radeon 9700?

[Soccerman]

title says it all.. I totally didn't think of that, partly becuase I wasn't sure what they were doing already (both nVidia and ATi), and partly because for some reason I never thought of packaging as a solution to their .15micron problem.. will nVidia do the same, even though they have the .13micron advantage? they don't seem to have done so yet on a video card (though mobile video seems to be another story).

 

[BFG10K]

I wasn't surpised at all, given the Ti4600 is running at 300 MHz core as well. In fact a golden sample Ti4600 is rated to do 310 MHz core, manufacturer guaranteed.

 

[Soccerman]

so you sort of saw the flip chip package coming? the R300 is something like 110 million transistors, which is nothing like the GF4's 60 something million (I think it was 64 mill).. so assuming a similar clockspeed is foolish IMHO.

 

[gururu]

I guess that it involved some snazzy engineering. It seemed to be a matter of do-or-die.

 

[zephyrprime]

I've always been surprised that video card processors are as low clock speed as they are. I guess that they must not be very finely pipelined. This is too bad since I think it might be possible to pipeline a GPU more than a regular processor. (adjusted for the several rendering stages impact on pipelines of course.)

 

[merlocka]

so assuming a similar clockspeed is foolish IMHO

Intel took .18u 42million transistors to 2GHz where AMD's 37million transistor .18u design topped out at about 1.4GHz. There's more to it than clockspeed and number of transistors, so assuming that nVidia squeezed .15 to the limit is also foolish.

FCBGA give you the advantage of better cooling, but it's more expensive. Packaging isn't snazzy engineering, and if ATI decided that it was economical enough to use FCBGA for their GPU, nVidia will probably be looking at the same.

If ATI announces that their part will do my laundry for me, I'll be suprised. If they announce it at 325MHz, I'll yawn and give my yams a scratch.

 

[BFG10K]

so you sort of saw the flip chip package coming?

Not at all, I just didn't think that nVidia had maxed out the current 0.15 um process.

There's no denying that ATi did a nice job though but I don't think I'd class it as a miracle.

 

[jiffylube1024]

Originally posted by: merlocka
so assuming a similar clockspeed is foolish IMHO

Intel took .18u 42million transistors to 2GHz where AMD's 37million transistor .18u design topped out at about 1.4GHz. There's more to it than clockspeed and number of transistors, so assuming that nVidia squeezed .15 to the limit is also foolish.

FCBGA give you the advantage of better cooling, but it's more expensive. Packaging isn't snazzy engineering, and if ATI decided that it was economical enough to use FCBGA for their GPU, nVidia will probably be looking at the same.

If ATI announces that their part will do my laundry for me, I'll be suprised. If they announce it at 325MHz, I'll yawn and give my yams a scratch.

^^^^ Hahaha, I agree 100%.

 

[Killrose]

I was surprised, but does'nt the flip chip pagaging help out here a little bit?

 

[Soccerman]

ya, that's the whole point. the flip chip packaging is one of the major reasons why a near doubling in transistor count while sticking with .15 micron doesn't kill the clockspeed..

as for the P4 @ .18 vs Athlon @ .18 thats a terrible comparison..

first of all, the P4 topped out at I would say 2ghz like u said, but the Athlon is just now topping out at 1.73 I think. I'm not surprised that the higher transistor count package was designed for higher clock speeds, because that's what the pipeline is all about, whereas I don't see anything these days on how many stages the pixel pipeline is on a video accelerator (because as we all know the 8500 and GF4 both theoretically can pump out 4 dual textured pixels PER CLOCK), or anything like that.

I guess that answers the question on why video accelerators are slower mhz wise than CPUs. they have an extremely high IPC (instruction per clock) count..

my point is, ATi doesn't appear to have necessarily sacrificed IPC to achieve the same thing. I mean we all know the GF3 for some reason was a bit quicker clock for clock than the 8500 (most likely a better design, ie, better placement of cache or something), and we all know the reason for the near doubling of transistor counts (adding DX9 compatability).

 

[merlocka]

as for the P4 @ .18 vs Athlon @ .18 thats a terrible comparison..

oops, forgot Palomino is still .18

but double oops, this actually make a better point. The .18u Thunderbird and the .18u Palomino share the similar IPC yet the Palomino (which is faster clock 2 clock vs the Tbird) has scaled to 20-30 percent higher clockspeeds with reduced power consumption per clock and it has (albeit slightly) more transistors.

Regardless of whether you like that analogy or not, CPU's moved to FCBGP a long time ago to improve thermal dissapation (and improve clockspeeds) as well as improve pinout routing. Intel took their 850 chipset to FCBGA, both the new 845 north bridge's use it. Gigabit ethernet MAC's even use it.

So, again, no. I'm not suprised that ATI moved the R300 to FCBGA and I'm not suprised that they can produce a 300MHz+ .15u GPU.