R350/NV35 Z Fillrate with FSAA

[Geeforcer]

I have never taken a stance on Revenge.

Yeah right

Please find any quote of me saying that I believe it or that I don't believe it.

Please.

Humour me.

Can I play?

http://www.beyond3d.com/forum/viewt...rder=asc&highlight=Tagrineth&start=20

 

[Ailuros]

That link was another entertaining nonsense memorabilia. I'll just stick to Wavey's rather sparse comment on that page:

Considering that with 4X MSAA enabled an entire Rampage chip would have effectively been one GF4 pipeline, but with twice the texture rate, I'd doubt those numbers!

Anyway I did a few quick tests with and without AA modes and I realized how damn hard it is to find a recent fillrate limited timedemo heh.

R300/2400+XP/PC 2700DDR

Elephant Atrium in SS:SE, 1280*960*32 (noTC/16x tri-AF enabled):

noAA/16xAF ---> 2xAA/16xAF = -11.2%
noAA/16xAF ---> 4xAA/16xAF = -22.5%

***edit: I expanded the experiment a bit more, this time with different amounts of texture layers. All following tests with 4xAA/16xAF:

quad to dual texturing = -6.8%
dual to single texturing = -9.5%
quad to single texturing = -15.6%

And that is what I'd call essentially free Multisampling in that occassion. Add to that how many AA samples said card is capable of, add colour compression and three stages hierarchical Z, early Z, fast Z-clear and what not and the picture is pretty clear.

 

[Mummy]

Please find any quote of me saying that I believe it or that I don't believe it.

I was going to link that thread Geeforcer provided, also, can i post our little PM message exchange? it would be pretty funny, eh, software gates ? lololol

You sure didnt say "i believe it", so you are semantically right.. but u were defending and promoting the Revenge hoax like a cheerleader...

O...K... uh... I'm not even saying Rampage could've stood up to NV25, really. It would've killed NV20 and R200

I dont know if the word kill is right, it was probably better though, sure, i agree, we all agree, but that's HISTORY, its OLD CRAP, cant we get rid of that Rampage vs NV20 crap ONCE FOR ALL ? or u have to prompt that shit again and again and again ? (i think that 12 page thread Geeforcer posted was enough on the subject?)

but not much past that. IF ANYTHING it would've nipped at Ti4200's heels in the most expensive configuration... and probably would've meant no GF4MX (which would've been nice)... but beyond that, nah.

Yep, would, probably, maybe, might be, could be, cant we get rid of those mental masturbations? thanks.

One person compares a modern feature with a feature that would've been in a product two years ago, and immediately several rabid Anti-3dfx Cannonsâ„¢ start toasting Rampage

Well, i didnt toast rampage myself, and Ailuros neither, we are just telling you to wake up from your wet dreams.

Do u know why i hate 3Dfx so much nowdays? because of you 3dfx fanboys and fangirls.

I was the first to use a 3D accel card in a pretty good emulator (Playstation one), 3Dfx contacted us back in time giving support and free cards (at least, they wanted to, then the project went to crap so we didnt do anything).

I loved my 3Dfx and i was pretty happy to see that a company was following private programmers in the quest for the emulation (now that doesnt mean 3Dfx was a good-boy company, they werent...) but 3Dfx then started sucking, and in those last 2/3 years in every fucking forum i had to read things like "We'll implement pixel shaders on a Voodoo5", "Rampage would have been faster than NV30", "Yes yes that feature is cool, but rampage, but sage... Gigapixel..... <insert your fav mental masturbation here></insert your fav mental masturbation here>",
from rabid 3Dfx fanboys, so i cant stand that company anymore...

I started mentioning it because I thought it was neat that suddenly this new, famous, brilliant product by ATi has one design feature that parallels something that 3dfx were poised to do

The fact that 3dfx had a feature years ago before others doesnt mean that current products are shit, there are many features that a chip may lack, old or actual ones, doesnt mean that the chip sucks.. there are many ways of doing things.

Rampage lacked many features, had some that others didnt have, so what ? of course only 3Dfx ones are best and ones from NV/ATi suck eh?

Did u ever try to use Pixel shaders 1.0 ? do you know how much they suck even compared to 1.1, for example ?

 

[elroy]

The initial Rampage design was meant to follow Voodoo2.

It was called "R4" because it was the fourth total redesign of the Rampage idea.

IIRC, the first iteration of Rampage was supposed to follow the original Voodoo (i.e. even the Voodoo2 was a stop gap product!). I think that really explains why 3dfx went down the toilet.

 

[bdmosky]

I was the first to use a 3D accel card in a pretty good emulator (Playstation one), 3Dfx contacted us back in time giving support and free cards (at least, they wanted to, then the project went to crap so we didnt do anything).

Which emulator?

 

[Tagrineth]

Sure! The Voodoo5 5500 has 256-bit SDR.

It's called SLI, sweetie.

*sigh* why am I even wasting my time?

I don't see your point here.

It effectively is 256-bit. Or do you not thing 128 + 128 = 256? Let's rewrite the fundamentals of arithmetic now...

Yep and it would have effectively turned into a second edition of the Rampage delay horror series. Of course can you add whatever your heart desires at short notice. Even with an optimal 18 months design to production cycle, the chalkboard design takes up in that case about 3 months. Any change after that means only one thing: delay.

True. But Mojo wasn't even on the chalkboard yet, really. It was an idea in Sellers's head.

No it wasn´t for free. Of course will you come up with some fancy definition of the term "free", but I´ve seen it all so far, heck why not that one too. On a TBDR Multisampling comes essentially fillrate and bandwidth free. Supersampling was bandwidth free on KYRO.

What I meant was something along the lines of, once you get over the hit from multitexturing, the added loss is barely noticeable. Hence "Effectively free with XXX in use".

For the factual relevance here are some old numbers released from 3dfx themselves: MSAA on a 100MHz GP1; approximately 26 fps in 1024*768

http://www.3dconcept.ch/cgi-bin/showarchiv.cgi?show=1390

How is a 100MHz GP-1 relevant to discussion of Rampage?

Oh please give me a break with that old BS. I´m tired of those daydreaming trips. Dave said himself that it was an internal joke and it´s been said over and over again what it was and for what it was intended.

Can we finally put an end to that nonsense?

Whether or not it was a joke, and sure, it probably was, it COULD WORK in CONTROLLED conditions.

Can I play?

http://www.beyond3d.com/forum/viewt...=0&postorder=asc&highlight=Tagrineth&start=20

Cute.

That's me discussing possible specs and saying that I thought the ppt was referring to that spec in a certain way, not saying "It's TRUE! TRUE I SAY!"

I've discussed the possibilities of a theoretical Revenge architecture, I've discussed what the specs could mean, but I've never said any of it was real and never intended to do so. 8)

 

[Evildeus]

Sure! The Voodoo5 5500 has 256-bit SDR.

It's called SLI, sweetie.

*sigh* why am I even wasting my time?

I don't see your point here.

It effectively is 256-bit. Or do you not thing 128 + 128 = 256? Let's rewrite the fundamentals of arithmetic now...

Hmmm, sure 2*128=256 but 2*128 bus doesn't make a 256memory bus, it's just a 2*128 memory bus... So it's not effectively 256 bit....

 

[Tridam]

Sure! The Voodoo5 5500 has 256-bit SDR.

It's called SLI, sweetie.

*sigh* why am I even wasting my time?

I don't see your point here.

It effectively is 256-bit. Or do you not thing 128 + 128 = 256? Let's rewrite the fundamentals of arithmetic now...

Hmmm, sure 2*128=256 but 2*128 bus doesn't make a 256memory bus, it's just a 2*128 memory bus... So it's not effectively 256 bit....

What's the difference ?

 

[KimB]

Hmmm, sure 2*128=256 but 2*128 bus doesn't make a 256memory bus, it's just a 2*128 memory bus... So it's not effectively 256 bit....

What's the difference ?

The V5 5500 had to duplicate textures across the two blocks of memory. This meant both that there was less memory available, and that texture caches were more hard-pressed to optimize external memory accesses (i.e. the same data would occasionally be sent to both chips, when with a one-chip architecture, that data would have only been sent once).

In other words, it's not effectively 256-bit because of the inefficiencies of attempting to manage processing among two separate processors.

 

[Tridam]

Hmmm, sure 2*128=256 but 2*128 bus doesn't make a 256memory bus, it's just a 2*128 memory bus... So it's not effectively 256 bit....

What's the difference ?

The V5 5500 had to duplicate textures across the two blocks of memory. This meant both that there was less memory available, and that texture caches were more hard-pressed to optimize external memory accesses (i.e. the same data would occasionally be sent to both chips, when with a one-chip architecture, that data would have only been sent once).

In other words, it's not effectively 256-bit because of the inefficiencies of attempting to manage processing among two separate processors.

I know all that I made a rather long analysis about this point in my old v5 review. But you can't say that it's not à 256-bit bus. The efficiency or the memory size doesn't matter.

But Evildeus doesn't mention this point. He just says that 256-bit is different from 2X 128-bit. It's not the case.

256 bit = 128 bit + 128 bit = 64 + 64 + 64 + 64...

The bus width is simple to talk about. The efficiency is not. Talking about V5 bus inefficiency or talking about modern memory sub-system of the last GPU of ATI/NVIDIA is talking about the same thing.

V5 5500 256 bit
Parhelia 256 bit
Radeon 9800 256 bit
…

They are all very different. But they are all 256 bits.

Will you make such a difference between K8 940 and K8 939 ???

 

[KimB]

There is one other difference, engineering. Simply put, a 256-bit memory bus attached to a single chip requires a significantly higher pin-density than two 128-bit memory busses attached to different chips. It's a rather different engineering problem altogether.

 

[Tridam]

There is one other difference, engineering. Simply put, a 256-bit memory bus attached to a single chip requires a significantly higher pin-density than two 128-bit memory busses attached to different chips. It's a rather different engineering problem altogether.

I agree. At V5 time it wasn't possible (->expensive) to use a 256-bit bus with a one chip board. But that doesn't change the fact that it is a 256-bit bus.

 

[Ailuros]

Ok let´s waste more of the boards bandwidth:

I don't see your point here.

It effectively is 256-bit. Or do you not thing 128 + 128 = 256? Let's rewrite the fundamentals of arithmetic now...

Only Parhelia and P10 have true 256bit busses. On the other hand do you need an extra analysis for the differences in NV35/R3xx and their memory controllers?

True. But Mojo wasn't even on the chalkboard yet, really. It was an idea in Sellers's head.

Sure who would even dare to doubt you guys in the first place? Did you get an idea in the meantime what Fearless stood for?

What I meant was something along the lines of, once you get over the hit from multitexturing, the added loss is barely noticeable. Hence "Effectively free with XXX in use".

Fillrate free under conditionals, yet the performance penalty would have still been there. Want me to crank out old 3dfx performance claims about the VSA-100´s FSAA? By the minute you can define the difference between marketing hype and reality we might even start to get at least some common ground here.

How is a 100MHz GP-1 relevant to discussion of Rampage?

GP-1 was a low speced TBDR. Again TBDRs handle Multisampling with no fillrate and bandwidth costs and those graphs actually show that. Would you guess in that case a minimal performance penalty? No. Now give me a fair explanation why Spectre would have been more effienct with MSAA on then a TBDR. Yes that thing has only a 100MHz clockspeed, but noAA performance should be still significantly higher than 26fps, don´t you think? I´m talking about pure percentages here.

Spectre would have had a performance penalty with MSAA on non period.

Whether or not it was a joke, and sure, it probably was, it COULD WORK in CONTROLLED conditions.

No serious IHV would use a flawed software HSR implementation officially in it´s drivers. The thing was hidden in the last official driver sets.

Still not tired of this nonsense? Apparently not.

 

[Althornin]

I agree. At V5 time it wasn't possible (->expensive) to use a 256-bit bus with a one chip board. But that doesn't change the fact that it is a 256-bit bus.

I'd be more inclined to argue that it is NOT "a 256 bit bus" because it is NOT a "single" bus!
It is TWO 128bit busses, each on running to a different chip.
Sorry, but two discrete busses do not make up a single wider bus, imo.

 

[Ailuros]

What's the difference ?

She said 256bit DDR for starters....

Althornin,

It doubles bandwidth though in dual chip configurations and that´s probably what they mean. If they should want to indicate that they already 256bit busses back then it´s as nonsensical as all other claims.

Avalanche (if my memory doesn´t betray me) had 1024bit internal and 128bit external (per chip). You can´t say there either that it has a 1024bit or a 256bit bus either. You can use the numbers actually just to calculate the memory bandwidth.

 

[Althornin]

Althornin,

It doubles bandwidth though in dual chip configurations and that´s probably what they mean. If they should want to indicate that they already 256bit busses back then it´s as nonsensical as all other claims.

That was exactly my point....
Im not an idiot.

He just says that 256-bit is different from 2X 128-bit. It's not the case.

I don't see your point here.

It effectively is 256-bit. Or do you not thing 128 + 128 = 256? Let's rewrite the fundamentals of arithmetic now...

You see how i was responding to people who are classify it as if its a single bus?

 

[Evildeus]

What's the difference ?

It doubles bandwidth though in dual chip configurations and that´s probably what they mean. If they should want to indicate that they already 256bit busses back then it´s as nonsensical as all other claims.

That's what i was thinking and why i said it's a 2*128 memory bus, and not a 256 bus, even if it's is equivalent in the bits transmitted...

 

[Reverend]

Whether Tagrineth is a 3dfx or Rampage cheerleader or not shouldn't be stressed. She has a lot of info regarding Rampage (and stuff after Rampage). Either you appreciate the info provided in her posts or you ignore them.

I don't know if Tagrineth is a "3dfx and/or Rampage fan" or not and I don't particularly care. You shouldn't either, because both are dead. If I wanted to, I can provide many Rampage/3dfx information continuously or when the chance is there but I know better.

 

[Ailuros]

I have nothing against presenting information as long as it doesn't get blown out of proportion.

Finally I stopped believing in miracle-performers a long time ago and especially in vaporware and paper specs. On paper most architectures look impressive, don't they?

You shouldn't either, because both are dead. If I wanted to, I can provide many Rampage/3dfx information continuously or when the chance is there but I know better.

3dfx was very well aware before it was over that Spectre would have been a fair competitor to NV20 from all perspectives, while they expected to gain far more ground with Fear. Please tell me that my memory has abandonded me as much that I recall things in the reverse order. And we don't need to get over the same things over and over again, that's true, for three whole years now. There were no miracles in Spectre from my standpoint, being a former 3dfx-fan myself doesn't change that fact one bit. There's a huge difference between having a preference and wearing blindfolds.

 

[Xmas]

Finally I stopped believing in miracle-performers a long time ago and especially in vaporware and paper specs. On paper most architectures look impressive, don't they?

I don't even think Rampage looks impressive on paper (or did back then)