Don't get fooled by .... nvidia attacking Ati

[Thowllly]

I didn't see this slide before now...

WHAT THE FUCK!!

ok, I don't care if Nvidia and Ati wants to sling mud and lie about each others products, but now Ati has gone too far!

If Ati wants to lie about the nv43 being a 4 pipeline card thats fine by me, I really don't care, it is after all not that big a lie.

But to make it look like Beyond3d supports (or even started) that lie is beyond shitty. B3D is a good and credible site, the B3D review told the truth about the nv43, that despite it looking like a 4 pipeline part, it is really a 8 pipeline part. Now Ati is abusing B3Ds good reputation for it's own gain. That's despicable. I would like to beat the shit out of that fucking asshole that made that slide.

As I said, I don't really care what Nvidia and Ati say about each other, but to drag B3D down into the mud just to score a point against Nvidia really makes my blood boil. Dave Baumann is a saint, if I were him I would probably have gotten an anger induced heart attack by now.

Phew, I feel much better now

 

[PaulS]

Oh please Paul !

Nvidia slides had tons of crap it is stunning that you need to complain about the slides from ATI and had not a word about the Nvidia's ones.

Don't try and portray me as a fan-boy now - I was merely focussing on the ATi slides in that post. The NVIDIA slides were full of PR crap too, but I'm not going to give you a full history of each side everytime I make a post. I was just sticking to that particular issue (one that other people are now picking up on too).

 

[Mendel]

Ati has a right to that reply... but come on, misquoting b3d like that

 

[dan2097]

My intepretation of Dave's review was that basically the 6600GT is 4 pipeline card (with 2x tmus). However the cards pixel shader performance is equivalent to a geforce 6 series card with 8 pipelines.

Basically ATI can say they have double the single texture fill rate which is pretty unimportant overall even now, but the 4x2 architecture of the 6600GT hasnt effected its pixel shader power which is most important.

Dave is that a correct interpretation?

Then again ATI didnt actually say explicitlythey have double the pixel shading performance, they said they had 2xpixel pipelines power whatever thats supposed to mean.

That presentation doesnt really lie IMO (unlike Nvidias did in quite a few places IMO), but it gives half truths or out of context quotes, atleast its choice of game benchmarks is a bit more realistic not that they havnt cherry picked (i.e. the doom 3 graph showing X800XT and 6800ultra on a par with each other) to an extent. That 2xpixel pipelines power does step towards outright lieing though..

 

[Dave Baumann]

It is a 4 pixel pipeline card as it has a maximum peak pixel writing capability of 4 pixels.

Question - if you have a future part with a unifeid shader architecture and 8 ROP's, who many pixel pipelines does it have?

Another question - what do people actually think pixel pipelines are?

 

[Mendel]

However the 3 texture performance differs from a 4x2 design, which Ati are hinting the 6600 would be. Ati just took Dave's review and ripped all objectivity out of it, that's how I'm viewing it. But of course if mr Baumann himself does not mind... I don't know, I'm a bit disappointed!

card fill single multi one texture two three\
6600 GT 1998.2 3800.0 1998.2 1905.1 1336.3
5900 XT 1572.9 3077.8 1560.3 1502.4 739.9


5900 XT three texture fill rate / two texture fill rate = 0.49
6600 GT three texture fill rate / two texture fill rate = 0.70

That's what matters in real world situations, not single texture fill rate.
That's what makes 6600 GT faster than 5900 XT even in no pixel shader situations.

 

[dan2097]

Another question - what do people actually think pixel pipelines are?

Things that ouput pixels to the screen, where the peak number of pixels they can display on the screen per second is directly proportional to number of pixel pipelines * clock speed of said pipelines. By that I mean fps * screen resolution.

Im sure theres something wrong with my thinking about them as by my way of thinking even a radeon ddr can do up to 190fps at 1600x1200 theoretically so newer cards shouldnt need such high fill rates. Are pixels processed multiple times or something, or maybe Im just completly wrong

 

[ChrisRay]

It is a 4 pixel pipeline card as it has a maximum peak pixel writing capability of 4 pixels.

Question - if you have a future part with a unifeid shader architecture and 8 ROP's, who many pixel pipelines does it have?

Another question - what do people actually think pixel pipelines are?

I usually think of them in the traditional sense as the way you described, Each Pipe adds 1 pixel per clock,

With the 8 Rops on the 6600GT, (now I'm no expert on the 6600GT) it still seems capable of performing shader ops as if it were a more traditional 8x1 NV4x card.

 

[Dave Baumann]

However the 3 texture performance differs from a 4x2 design, which Ati are hinting the 6600 would be. Ati just took Dave's review and ripped all objectivity out of it, that's how I'm viewing it. But of course if mr Baumann himself does not mind... I don't know, I'm a bit disappointed!

As a point of techicality you have to separate the notion of the "pixel pipeline" from a "fragement pipeline", basically because things are going to get much more confusing from here.

Think back to Voodoo - it had one chip which what the pixel pipeline and another that was the texture pipeline; essentially the pixel renderer chip is what we would class as the ROP now (although modern ROP's are doing much more) and you could consider that the fragement pipeline has been instered between the pixel pipeline and the texture pipeline (or added to the texture pipeline). There is, pretty much a disconnect between the "4" and the "x2" as there always has been (think TNT2 or VSA-100 - these had always been either 2x1 or 1x2 dependant on what was asked).

Yes, that using the quotes in that fashion is misrepresenting what most people would associate the performance to be under these conditions to be - but the eduation (from our part) needs to be on directions that are happening in the future and how these notions are not as important. At least, however, the the pipeline diagram is still in there.

 

[Mendel]

double post

 

[Mendel]

Another question - what do people actually think pixel pipelines are?

Things that ouput pixels to the screen, where the peak number of pixels they can display on the screen per second is directly proportional to number of pixel pipelines * clock speed of said pipelines. By that I mean fps * screen resolution.

Im sure theres something wrong with my thinking about them as by my way of thinking even a radeon ddr can do up to 190fps at 1600x1200 theoretically so newer cards shouldnt need such high fill rates. Are pixels processed multiple times or something, or maybe Im just completly wrong

two times for two textures... three times for three textures.
Normally a two tmu chip could do two textures in double the speed as compared to three textures and therefore drop substantially when texturelayers were added.

However, this is not the case with 6600... It would appear to act like 4x2 design when doing single or dual texturing, but then again act like 8x1 design when doing triple texturing.

 

[Dave Baumann]

With the 8 Rops on the 6600GT, (now I'm no expert on the 6600GT) it still seems capable of performing shader ops as if it were a more traditional 8x1 NV4x card.

There are 4 colour ROP's on NV43, so only 4 colour values ("pixels") can be written per clock, there are however 8 Z ROP's capable of writing 8 Z/Stencil values per cycle (by contrast NV40 has 16 colour ROPs and 32 Z ROPs).

Edit: Actually, what I should say is that the 4 colour ROP's are capable of either color or Z writes so when only Z writes a being done the 8 Z ROP's comprise of 4 dedicated Z ROP's and the 4 colour ROP's.

 

[jb]

This whole fiasco is just plain sad........ The saddest part is that ATI has taken the high road for 3 years and gotten pretty beat up by nVidia's PR. It's the old addage about if you repeat a lie often enough, people will believe it. Evidently, ATI has decided if you can't compete with nV taking the high road, emulate them! It looks like ATI has decided that to compete with nV, it must behave like nV..... Optimising, shader replacements (Cat AI?) and now this.....

Very sad and disappointing.....

But what else could they have done? I mean you can take the high road for only so long...then its human nature to fall off it. Fight fire with fire as the saying goes or nice guys finnish last if thats a better one.

Not saying it was right of them...but I figured soon or later it was bound to happen...

 

[Mendel]

I'm a bit unsure, btw, about how the weirdness of 6600 "pixel pipelines" affects the penalty of going from bilinear to trilinear, if at all... Anyone benched this?

 

[martrox]

But what else could they have done? I mean you can take the high road for only so long...then its human nature to fall off it. Fight fire with fire as the saying goes or nice guys finnish last if thats a better one.

Not saying it was right of them...but I figured soon or later it was bound to happen...

I agree with you here..... but it's still very sad......

 

[Dave Baumann]

I'm a bit unsure, btw, about how the weirdness of 6600 "pixel pipelines" affects the penalty of going from bilinear to trilinear, if at all... Anyone benched this?

Technically, it shouldn't have that much of a penalty in terms of sampling - with Bilinear it will be able to texture sample up to 8 pixel's worth but only output 4 in a cycle; with Trilinear you'll be sampling 8 pixels over two cycles and be able to output 8 in two cycles.

There might be some performance difference required to handle the different bandwidth requirements though.

 

[Bouncing Zabaglione Bros.]

I agree with you here..... but it's still very sad......

Just goes to show that Nvidia has succeeded in dragging ATI down to their own level (to a small extent). ATI had no choice but to fight back with their own document after all the lies that Nvidia circulated recently. If you compare the two documents, ATI's looks like it was written by an angel.

There comes a point where you have to stand up and fight for yourself even if it means your hands get dirty, because otherwise the only viewpoint out there is the one fostered by the liar and the cheat.

 

[jvd]

It's also rather funny that they say that the 6600 GT is a one trick pony and then compare Doom3 scores between the 6800 Ultra and the X800 XT which have a rather big clock speed difference (pony just left the show slide).

Why ?

The 6800ultra costs 499$ , the x800xt costs 450$ the x800xt pe costs 500$

Who cares what the clock speeds are , if i'm spending 500$ it doesn't matter to me which is clocked higher , it matters what gives the best performance.

Same with processers . I don't care if the p4 is 4ghz vs 2.8ghz of the athlon 64. If the p4's performance is faster than the athlon 64 and cost the same price that is all that matters to me .

You compare price points to each other .

 

[Tweaker]

But to make it look like Beyond3d supports (or even started) that lie is beyond shitty. B3D is a good and credible site, the B3D review told the truth about the nv43, that despite it looking like a 4 pipeline part, it is really a 8 pipeline part. Now Ati is abusing B3Ds good reputation for it's own gain. That's despicable. I would like to beat the shit out of that fucking asshole that made that slide.

Looks like you weren't so furious about this document. Discused

here

 

[Dave Baumann]

It's also rather funny that they say that the 6600 GT is a one trick pony and then compare Doom3 scores between the 6800 Ultra and the X800 XT which have a rather big clock speed difference (pony just left the show slide).

Why ?

Because the document is primarily talking about X700 and the XT has a speed deficit to the 6600 GT (although opposite for the other boards).