XCOM: Enemy Unknown coming October 9, PC special edition announced
Daily Filter: Tom Clancy's Ghost Recon Future Soldier, Dust 514
Silicon Knights lawsuit has $1 of damages at stake
DmC: Devil May Cry delayed to 2013, coming to PC
Bungie's 'Destiny' exclusive to Xbox, launching in 2013, original Activision contract shows
Megawhat?
by Steve Gibson, Aug 31, 1999 4:58pm PDTIt seems a number of people are confused about the NVidia press release and the numbers they are claiming for megapixels as compared to other video cards on the market. Jack wrote this explanation in clarifying some more an earlier email from Daniel Horowitz which should hopefully clear things up for you guys. Of course if you don't know wtf a megapixel or megatexel is in the first place, you're SOL. Oh yeah, Carnivac has uncovered the origins of the GeForce256. It was originally the G-Force5 but NVidia got their engineers on the ball and juiced it up.
Comments
Yes, I am wrong about the V1, because I apparently got mixed up with all this. Its effective performance is still a little less than 50MPixel/s would imply, because of the overhead needed to multi-pass render. But it is not like the NVIDIA chips. The TNT1/2 can render 2 textures per pixel and the GF256 can render 4 textures per pixel by taking a corresponding hit in pixel rates. The Pixel units can also act as TMUs. Ergo, these NVIDIA chips do not suffer from the overhead of multiple passes, unless the texture/effects load is more than it can handle.
My example about the TNT with multitexture on/off was only to demonstrate the importance of multitexture vs. multipass rendering. Though the Texel-rate would be the same (unlike the V2/V3) the performance would most certainly differ because of the need to do multipass rendering. And that was all I wanted to say, not point at the disadvantages of the 3dfx architecture.
I also think you are right about the S2000. From the overview, it looks like the Texel rate will drop to 350MPixel/s when single-texturing. From a fill-rate standpoint, the Glaze3D still seems to be the most versatile, but in its current configuration it lacks the T&L capacity of the S2000 or GF256. The deciding factor for me will be, \"Who can sell theirs for the cheapest?\"
Okay, what\'s left that we don\'t agree on?
-Ray
-Ray
#53:
You\'re still a little off. The V1 would be at <25MTexel/s because it doesn\'t multitexture. It would need to do two passes, which causes a performance hit.
Your assumption on the TNT2 and V3 comparison is incorrect as well. Multitexturing means that multiple (in this case, two) textures can be applied in a single pass, which is the case for both chipsets. (This differs from multipass rendering because multipass causes a performance hit.) The only difference between the two chipsets is that the TNT2 splits it into two pipelines that can render two pixels simultaneously if multitexturing is not in use, whereas the V2 can only render a single pixel, but has multiple texture units that can apply two textures in a single pass. In order for either card to apply four textures, they\'ll both have to do two passes, again at a performance hit.
The Pixel rate is important because the higher you can get it, the higher resolution you can run at reasonable speeds in. High texel but low pixel (GeForce256) rate limits your resolution. The Savage2000, on the other hand, has a huge Pixel/Texel rate, but is limited in the effects that it can apply without a performance decrease. Because it can only dual-texture, it would require additional passes for more complex effects. Doing a second pass has additional overhead compared to multitexturing, so an S2000 doing two passes would possibly not perform as well as the NV10 when quad-texturing. Compare a TNT with multitexture on vs. off in Q2. Big difference.
Comments please!
-Ray
Hmmmm, makes me wonder what 3dfx is doing? the ol\' ZXABDF buffer or something.... what? exactly. I think 3dfx put too much cash into their pockets and vertical consolidation and not enough into good ol\' R&D.
that stuff gets people\'s attentions
TEST.]B
Matrix-esque midair shotty IHV battles still own me, even though they aren\'t really around nowadays....
BTW, once you get a user name here, you can leave the username and password blank and just hit post and it works. (tho it would probably help if you use the same comp all the time)
Man, that was before summer even started. Geez, it\'s been a while.
Now that the air control is all toned down I may remain the only person to hit all the jump pads with such ease. =[
BTW, what was yer nick? I wouldn\'t remember your IP.
the current subject in the #shugashack channel on irc.enterthegame.com is:
\"Sir, Can I take your daughter to Conneticuit and fuck the shit out of her\"
you know you want to find out what all the talk is about, cum on in!
The advantage over 3dfx\'s architecture at the time was that the Voodoo2 would only have a 90MPixel/s fill-rate regardless of whether it was single or multi-texturing.
Now, as to how it applies to the new chipsets we\'re looking at:
If the 480MPixel/s that the GeForce 256 claims is produced in a similar fashion to the TNT, it could mean that when quad-texturing, it will only have a 120MPixel/s fill-rate. But hey, at least it can quad-texture. When dual-texturing, it will have a 240MPixel/s fill-rate, which is not that much better than the TNT2 now, is it?
The Savage 2000 can\'t quad-texture, it can only dual-texture. So though it can push out two pixels per clock tick, each pixel is only dual-textured. It will still rock for dual-textured games though (at 350MPix/s).
The Glaze3D (yes, I know, vaporware) claims a peak MPixel rate of 600MPix/s when single texturing, but can sustain that fill-rate even when dual-texturing, depending on the filtering method used (assuming bilinear). They gave an honest breakdown of their fill-rate depending on effects:
Glaze3D™ 1200/2400 performance
1200/2400 million texels/s
600/1200 MPIX Gouraud shaded, no texture
600/1200 MPIX single texture, trilinear mip-mapping
600/1200 MPIX dual texture, bilinear filtering
300/600 MPIX dual texture, trilinear mip-mapping
300/600 MPIX quad texture, bilinear filtering
300/600 MPIX environment bump mapping
150/300 MPIX quad texture, trilinear mipmapping
15M tri/s geometry throughput
Their main problem is that the chip does not have onboard T&L, but they do have kickass fill-rate that can be doubled through their SLI-like techniques.
When they do get their own geometry processor online, with the quad-SLI board, it\'ll be over for the competition.
Well, that\'s my analysis of the thing. Please give your comments.
-Ray
and you remember that CFH? I only posted with that slapped on my name for a little while...
I mean, something like the G-Force 256 would sound cool, but I just look at \"GeForce\" and want to pronounce it like the \"zshey force twoo feevty seex.\"
bah
If you have a 15 year old neighbor that is the hottest thing on 2 legs.
Just do her. Get it over with. Do what I did, don\'t be a retard, be smart like me.