Rumor has it that 3dfx has a new trick up their sleeve, and it has something to do with this:

[Hardware]

If you own a business you have to made some decisions!
When you did wrong 19 of 20 times thats your fault not somebody else!
stb?
0,25?
sdram?
vsa100?
v6000?
t&l?
t-buffer?
etc

 

[lupin]

<< 3) By use of the exclamation point, I somehow think you have a shift key. Make use of it. >>



Seph, what are you? Detective?

Very sharp indeed.

 

[BigToque]

Suppose the V5 will have support for HSR...

What will that do for the performance of the card? much higher framerate? roughly similar framerate? lower framerate?

 

[Wingznut]

Assuming the cpu can keep up, a significant higher framerate.

 

[Hardware]

If there IS a possible way to do a software HSR you can bet nvidia is first (they have a far better driver team)

 

[Doomguy]

IBMer: You have no idea what you're talking about. NVidia's FSAA does not use their T&amp;L unit at all. Both ATI and NVidia use OGSS, while 3dfx uses RGSS. NVidia's FSAA also isnt slower than 3dfx's. They both have very similar performance drops in 2x and 4x modes.

NVidia's FSAA is just as much hardware as 3dfx's method.

Ben and jpprod: Help me fight of the 3dfx guys!

 

[BigToque]

<< If there IS a possible way to do a software HSR you can bet nvidia is first (they have a far better driver team) >>



Why would I count on nVidia being the first with drivers that enable HSR?

How do you know they have a better 'driver team'? Do you know them? Do they call you on the telephone? Just because they choose to put out new drivers each week, doesn't mean that they have a better team, it just means they release what they have at the end of each week. I'm sure 3dfx could do that as well if they wanted to, but dont cause they choose to make sure the drivers are good and stable before they are released.

 

[Wingznut]

Doomguy, are you saying that nVidia's FSAA is every bit as good as 3dfx's?

 

[Mr.IncrediblyBored]

Yes its possible for it to be done in software. The added CPU overhead would be immense.

If this is true, you can expect a large performance increase in high resolution and high color depth. This is where you can afford the CPU cycles to perform the complex algorithms needed.

 

[NOX]

<< <<Then you'll be left with less competition to drive down prices and drive up innovation.>> >>

Why do people always say that? If they don?t do it someone else will, some no name company can easily bring fort the next big thing to the graphics industry and crab a lot of attention. If they don?t innovate, someone else will.

Hardware, this is not some 3dfx vs. Nvidia thread, so relax!

 

[DaveB3D]

Doomguy doesn't know what he is talking about. NVIDIA uses their T&amp;L unit for vertex multiplication... basically they multiple their cords by whatever values to calculate the new x,y value.

 

[Deeko]

<< stb?
0,25?
sdram?
vsa100?
v6000?
t&amp;l?
t-buffer?
etc
>>



Hardware, what the he** are you talking about?

 

[Fozzie]

Define T&amp;L unit though Dave. Yes nVidia uses their hardware to scale and modify the geometry. But I don't think either you or I know exactly what part of the chip is doing what. It may be the Triangle setup engine, if thats even a seperate unit on the core. My point though is that even if you use the T&amp;L unit or another part to do the vertex modifications, filtering and downscalling its all upto definitions as to whether that is hardware or software. Its certainly not being done on the CPU anyway you look at it.

You would certainly know better then anyone here now. How programable are the units of the VSA-100 that &quot;do&quot; the FSAA? From my limited understanding of how the T-buffer setup works it does not seem to be hardcoded.

As for the topic discussion. My current take on it is that it is a call in place for future cores. Just like nVidia has in the 7.X drivers for NV2x functionality.

 

[DaveB3D]

Fozzie,

Kristof and I wrote what I consider the difinitive paper on the subject. check http://www.beyond3d.com/articles/NVIDIA_FSAA

Yes, it is the T&amp;L unit.

I'm not sure what you mean by hardcodered. I mean it is entirely done on the V5 itself. The only thing that the software does is say &quot;hey, use this level of AA and do it like this&quot;. Driver says that and the hardware does it.

 

[hans007]

the v5 is already close to as fast as the gf2 with better image quality, so they done NEED to be faster, but it sure would be good

 

[Soccerman]

LOL! what a RIOT reading this thread! LOL! poor nVidia zealots, won't give us a moment of peace!

ahh, guys lets not open up this wound again with Software vs Hardware FSAA.. it was settled a while ago.. 3dfx's FSAA was declared to be the best in terms of compatability, the best in terms of performance hit (lowest performance hit in other words) however the quality is very debatable, and performance overall, both do very well.

k?

now to the meat of this discussion: HSR in drivers.

first of all, I'd like to say that this makes me wonder if the tables will be turned around yet AGAIN when it comes to the situation with features..

we've known that nVidia will incorperate some form of bandwidth reduction, the most obvious of which would be HSR. we know that it should be implimented in hardware, thus in this situation probably being more efficient (I don't know how 3dfx is doing it so don't ask), however IF 3dfx releases these drivers any time soon, you can expect that it will be before the NV20 is released, therefor 3dfx will have beaten nVidia to the punch.

the other question has to do with RAMPAGE. could it be that V5 is being sort of the testing platform for driver HSR implimentation for the upcoming Rampage which is suspected not to include many, if any at all, memory bandwidth saving techniques?

you have to wonder, will there be compatability problems? what about speed increases?

finally, misconceptions have already turned up!

btw, lets ignore hardware for the time being. we'll have to see benches for sure, but I suspect the performance hit (ie percentage wise, like when comparing FSAA on nVidia and 3dfx cards) will be larger then done in hardware (no pun intended).

Dulanic

&quot;HSR can be done on a fast CPU and you will see a gain. However I would guess anything under 700Mhz wont see any improvement with a software HSR. If the CPU has enough spare cycles to do the HSR, then that leave ALOT more bandwith for the video card to work with. You probably wouldnt see a huge improvement until hardware HSR.&quot;

I do not know anything about 3dfx's exact method of doing HSR, so let us keep from making this assumption, however it is entirely possible, because the CPU has to draw the T&amp;L, then check all Z-buffer values and FINALLY send it to the Video card after eliminating extraneous info.

at least, that's ONE way (the most probable one) of doing it. it's entirely possible that it is done another way, we will see.

ArkAoss are you talking about Kyro, and PowerVR, and Gigapixel?

Doomguy

&quot;NVidia's FSAA also isnt slower than 3dfx's. They both have very similar performance drops in 2x and 4x modes.&quot;

not true, nVidia's method recieves a higher performance hit (percentage wise) then 3dfx's method, simple because their cards weren't designed for it at all. their method is less efficient.

Knightbreed
&quot;If this is true, you can expect a large performance increase in high resolution and high color depth. This is where you can afford the CPU cycles to perform the complex algorithms needed.&quot;
well I wouldn't go so far as to say a LARGE increase, but a significant one (only at higher res, where HSR eliminates Memory bandwidth usage).

as for 3dfx making all those 'wrong' decisions. they might be wrong in YOUR eyes, but I'm beginning to see some of the reasoning behind it. of course, all 3dfx's advances have been behind the scenes, and haven't been very tangable (spell?) until soon.

 

[NOX]

<<the v5 is already close to as fast as the gf2 with better image quality, so they done NEED to be faster, but it sure would be good>>

Uh?better image quality? That?s a pretty broad statement there, what games are you talking about. I?d agree that Q3 may be one of them, but according to me eyes, my V5 wont come close to the image quality my GF2 will produce when playing Half-Life (CS/OpFor), Homeworld etc. UT wouldn?t be a fair comparison, however, the V5 does look much better when playing Motocross Madness 2, so lets be a little more specific.

And no, the V5 doesn?t have to be faster, take for example the G400MAX. Not a fast card but hard to beat the image quality, in fact I had a screen shot of the G400MAX and V5 side by side comparison in UT both cards using D3D. They were identical, the images looked exactly the same only that the V5 had FSAA which was noticeable and mad the V5 look that much better.

 

[DaveB3D]

V5 looks awesome in Homeworld. 1024x768, 4x FSAA and -2.0 LOD bias. Awesome... simple awesome.

 

[Soccerman]

&quot;I?d agree that Q3 may be one of them, but according to me eyes, my V5 wont come close to the image quality my GF2 will produce when playing Half-Life (CS/OpFor), Homeworld etc.&quot;

well seeing as most people measure a card now adays by their Quake 3 performance, we might as well give them SOMETHING to work with no?

btw, have you compared a V5 to a GTS in any of those games? I have a V3, and it does VERY well in homeworld (my limit is the CPU, a K6-2 400). in fact homeworld is a VERY CPU limited game, compared to many others... you can run a Voodoo 3 card easily at 1024X768 on it..

FSAA of course in Homeworld (one of my favourite games!! it's AWESOME!) would take away all those jaggies I get..

 

[Fozzie]

Dave, I've read the White paper and your article on nVidia AA. What I am refering to specifically is are the execution units that do the blending &amp; and the T-buffer modes and whatnot on the V5. Saying &quot;the driver tells the hardware and it does it&quot; is more then a little simplistic no? What I am asking is HOW the hardware does it. When you come down to it you could argue that nVidia AA is hardware because it is virtually all done on the GPU. OTOH it is done in software because it is code that is being run on more general purpose execution units.

What I am saying is that with the last definition the V5 is likely doing it in &quot;software&quot; also if it does not have hardcoded specific units that have only certain modes of operation for 2x and 4x AA only and can not be reprogrammed. Regardless of who is more efficient or compatible. That is what I was attempting to ask. I know enough to know when I don't know enough I guess.

 

[DaveB3D]

Well I'm REALLY tired right now, so I'm not going to go into it right now. However, check the 3dfx T-buffer white paper (check 3dfx.com or 3dfxgamers.com for it). I think that might cover your answer. If not, I'll try and answer it as best I can (ie. as long as stuff isn't confidential ) tomorrow.

 

[NOX]

Yes, at 1024x768 4xFSAA my V5 looks great in Homeworld, but at the same resolution, my GF2 also looks awesome, and the colors seem to be more vibrant then my V5.

<< <<well seeing as most people measure a card now adays by their Quake 3 performance, we might as well give them SOMETHING to work with no?>> >>

I could care less what most people do, I have my own character as to what I want my card to do for me with the games I play.

<< btw, have you compared a V5 to a GTS in any of those games? >>

I?m glad you ask that, because yes I have. I?ve played many games with both cards, and none of them have let me down. The only concern I?ve ever had with my V5 is while playing HL/CS/OpFor/TFC the image seems to be blurred a bit, unlike my GF2, which holds the colors in nice and tight, much sharper look, and everything look cleaner. Thought FSAA is hard to beat when playing those games, and we can all agree that the V5 is champ.

BTW: Both cards are being used under Win2k.

 

[RoboTECH]

NOX, have you farted with the lodbias on the 5500 at all?

your complaints seem to be directed toward low (high, actually!) lodbias settings and/or using the old miniGL

 

[Fozzie]

I understand about the tired thing Dave, if I wasn't so short on sleep I might actually be making sense.

In short my question is if you think that it is accurate to classify the Geforce's AA as &quot;software AA&quot; based on the technicality that parts are being run on non-specific hardware units? And if so do you feel that the V5's AA could still be called &quot;hardware AA&quot; using the same guidelines?

Also do you feel it is/not misleading to the general public to refer to the Geforce's FSAA with the generality as software AA? Given the negative image &amp; generality of it being inherently inferior as a result? Even if it has little or no effect on usage/performance(hypothetically speaking, I am not arguing here that it is or isn't a serious issue on the Geforce series).

What hardware/software does the geometry rotation on the V5?

Thanks,

 

[NOX]

<< your complaints seem to be directed toward low (high, actually!) lodbias settings and/or using the old miniGL >>

Who said I was complaining? I said it was a concern, I?m not complaining about it (don?t put words in my mouth).

I?ve tired many settings, in fact I?ve even tried a few from B3D?s forums (which I am a member of btw, and 3dfx gamers, which I?m also a member of). A few of the settings have improved my performance as well as image. But I still find that my GTS as a sharper image, vibrant colors, and detail on the walls are crisp (talking about HL/CS etc.). And no, I?m not using any miniGL stuff, default only.