R520: 32 PS pipelines? Any truth to this?

geo, the problem with this theory is that you haven't established a motive. Wouldn't the "real" issues with r520 have been enough - all the 24/32 pipe rumours did was make people hold out for the killer r520 once the problems were resolved. A lot less people would've waited for a 16 pipe card (simply because of the pipes=performance craze that has swept the land).

 

I'll give you a possible motive: ridicule the messenger

 

The 24/32 rumors go back to well before problems with R520 were apparent. And who was waiting for what? Folks were waiting for next gen cards. Now they can buy a 7800GTX, or continue to wait for R520. . .which since shortly before GTX release has been morphing into the crippled 32-piper that is lucky to make 16-pipes, a full 8 pipes fewer than 7800GTX.

Look at nagus second post above, re ATI's reputation. Classic preview of the reaction from the great mass who have been lead to expect a 32-pipe card as they come to realize it is "only" 16.

Gamer 1: 16 pipes? Are you kidding me? NV did 24 pipes on 110nm and ATI can only do 16 on 90nm?
Gamer 2: Well, I hear it will be competitive to 7800GTX even at 16 pipes. . .
Gamer 1: But its 90nm, man! Shoulda blown GTX out of the water! What the hell were they thinking?
Gamer 2: Yeah, well, I heard it was supposed to be 32 pipes but they couldn't make it work.
Gamer 1: Obviously! Let's go buy GF7! I don't want no crippled second choice. ATI's sucked ever since they got the Xbox deal!

Looks like motive to me. <shrugs>

But now I'm getting drawn into a little too much NV bashing appearance for my own comfort. So I'll let it go at that.

 

Yeah but doesn't all of that fall down if r520 is simply faster? I don't buy that the "should've been 32 but is only 16" rumours are going to affect people's perception of r520 one iota once performance numbers start showing up. If it's faster nobody will care - if it's slower then they will point to these rumours but it will be a moot point in any case since it's slower no matter what.

I really can't see the benefit to Nvidia here. Nobody is going to buy a 24-pipe card that's slower than a 16-pipe card because it has more pipes. IMO Nvidia would have been served much better by pushing "See they only have 16 pipes and are scrambling with respins to force higher clocks to compete with our great GTX". The 24/32 pipe rumours have elevated the r520 to the level of the coming of the saviour - people are holding out in anticipation for the GTX killer.

 

Not granular enuf --how much faster; how much slower? Most disinformation campaigns are aimed at moving the debate on the margins, not hitting a huge homerun. Politicians call it the "expectations game", and for an example Al Gore would have to have been a cross between William Jennings Bryan and FDR in the 2000 debate for independants (roughly 25%) to not have been "disappointed" in his performance by the time the Bush team got done with him. Right now it seems to me that ATI has been setup for R520 to be a failure if it doesn't hand GTX an R300 vs NV30 *ss-whuppin.

If (if, If, IF) there is design in that result, "cui bono" seems obvious.

 

It probably isn't faster right now. If it was faster ATi would have leaked some benchmarks or some information out there. Even if yields are bad, leaking information doesn't hurt, if anything it would help thier situation and might stop potential gtx buyers to hold off.

I think they have a r520 that can out pace the gtx at stock but not a gtx at 550+ mhz and there is every possiblity nV can and will make an ultra if the r520 ends up being faster.

At this point ATi has to make alot faster then the gtx to make up for lost ground. And to cover themselves for an ultra.

 

I don't see any way that's possible.

I mean, if ATI has really tweaked this design for excellent branching performance, maybe they'll have a chance of pulling an R300 vs. NV30 in some very specific synthetic benchmarks, but other than that, well, it just isn't happening.

Regardless, ATI doesn't need to tremendously outperform the GeForce 7800 GTX. What they need to do to make money is to put out a product that has good power consumption characteristics, good clock speeds, a full featureset, and scales well to low-mid products.

 

ATI will not be able to sustain the ~30% fillrate advantage the R4xx ultra high end GPUs have against 6800Ultras. In order to achieve, that they'd have to clock the R520 at nearly 840MHz and even if it would be possible I'd love to know where they'd get the bandwidth from to feed >13GTexels/s right now.

Through sterile calculations it seems to be though a geometry throughput monster (>1200MVertices/s) and if they optimized their ROPs well enough, the higher pixel fill-rate (>10GPixels/s) might also shine through.

I still think that in between 620-650MHz the R520 has very good chances to compete just fine against a G70 in today's applications. Those that expect to see a repeated performance increase as between R3xx and R4xx are rather daydreaming.

 

Me neither. Nor the reverse. Hence the utility of such a hypothetical campaign as above.

 

Why a 512 bit memory bus of course.

 

It has occurred to me to wonder if building in redundant quads could be a back door way to *also* have a die big enuf. . .

Edit: But probably not. 'Cause stuff like a 512-bit bus tend to be a one way ratchet at the high-end. . .you'd look pretty silly if your 90nm G70 had less bandwidth. . .or the gen after that.

 

Well, it is possible that the R520 will be a broken architecture, though unlikely. But we know that the G70 is not, and thus it just won't be outclassed by a similar amount to the NV30.

 

R520 being broken or not doesn't really matter at all. It's too late to make any impact even if it should end up being faster than G70. NV will prolly have their next chip out there until R520 arrives.

I can't find the post now, but I said it some 6 months ago that nV will rip ATI a new one this time around (without any insider infos). And that's what they did already by releasing the G70 a few months ahead of ATI's product. I bet they knew what's going on at ATI and thus went the low-risk route, holding their faster chip back a few months (32 pipes? 90 nm? Double core? Whatever... ).

I love being right

 

The next generation, from both companies, should be arriving late next year.

But nVidia may put out a higher-performing 90nm version of the 7800 GTX at a similar time to ATI's release of the R520, though I rather doubt it. I'm willing to bet that the 7800 GTX will stay where it is in nVidia's lineup, though the price may be adjusted to meet ATI's offering, and we'll just see nVidia start to replace their current NV4x lineup with lower-end versions of the G70.

 

The price will be adjusted, if R520's launch MSRP is lower than what G70 currently costs.

Agreed on the rest; late 2005 (or even early 2006) sounds like R580/Gxx@90nm time to me also.

 

Theoretically there are many sollutions for many bottlenecks; it just shouldn't cost an arm and leg for each specific timespan. If IHVs will move to wider buses, then maybe during the WGF2.0-era.

 

what in the heck....

Page 48 and people are still talking about 16 "pipes" 24 "pipes" and 32 "pipes".....

I am like in the twilightzone or something.

GTX is not even 24 "pipelines".. it's 16/24.

You guys have got to get off this pipeline thing. The R500/c1 has 3 "pipelines" if you want to get technical.. but it is more powerful than the 520 or the G70.


16 (1-3-1)... at 600+mhz

has everyone forgotten this already?

 

oh you guys are forgetting features as well...

and Chalnoth.. It would be logical that R580 will launch before the end of the year to counter any souped up GTX...

I think that when this is all said and done everyone is going to be plesantly suprised by what Ati releases.

 

Just.. waiting wishing...

 

Where's that ROFLMAO smiley?

Jawed