magic driver crap

Toggle sidebar Toggle sidebar
cefoskey

cefoskey

Distinguished
Jan 20, 2003
440
0
18,780
After reading the last few threads about Nvidia's magic drivers, Im starting to see a developing trend: while all the nvidia fans keep yammering on and on about these driver speed improvements, whats to say that ATI isnt writing drivers to do the same. Also, if Nvidia isnt merely "optimizing" their code like ATI did (and got chastised for) where is all this free performance coming from? Dont say its shader 3.0 crap because almost no retail games take advantage of it right now.

Plus, what makes everyone think that DX9.0c will only improve the performance of the 6800? Seems to me like ATI is just quietly working in the wings (and selling REAL CARDS WE CAN ACTUALLY BUY)

"Who is General Failure, and why is he reading my drive?"
P4 3.0 HT, Intel D865GBF, 512MB Crucial PC3200 DDR, WD 36GB Raptor 10,000RPM, BBA Radeon 9600PRO, SB Audigy, Hauppage WinTV
 
look sm 3.0, Crytek's 1.2 patch is already being tested. We are going to be recieving the source for the update soon. So guess what all the folks that have the x800 series are going to miss out on some really nice graphics in Far Cry plus a nice performance bosst.

ATi's new drivers, what new drivers they have being using updated versions form thier r300 line.
Even Dave Otten said that.

Thats why there won't be huge performance gains form ATi's new drivers.

The reason Dx9c will only improve on the 6800 is becuase the newer driver set for Nvidia is specifically made for 32 bit precision and Dx9c is specifically made for that.

Dx9b isn't as strict about it in most cases. ATi can't make a drivers set to get those enhanced features becuase it will hurt them.

Btw best buy has the GT?
 
They had the GT for preorder for 4 hours.

"Who is General Failure, and why is he reading my drive?"
P4 3.0 HT, Intel D865GBF, 512MB Crucial PC3200 DDR, WD 36GB Raptor 10,000RPM, BBA Radeon 9600PRO, SB Audigy, Hauppage WinTV
 
BTW entium, if the DX9c code is taking advantage of 32bit precision, wont most of the 24bit cards create rendering artifacts while using those paths? that would be far too damaging to the core of the industry, which is midrange cards (none of which have 32 bit precision). Does the game code or DX9 code dictate 32/24 bit precision

"Who is General Failure, and why is he reading my drive?"
P4 3.0 HT, Intel D865GBF, 512MB Crucial PC3200 DDR, WD 36GB Raptor 10,000RPM, BBA Radeon 9600PRO, SB Audigy, Hauppage WinTV
 
no there is a fall back still, but dx9c is optimized for the 32 bit paths.

neither the game code or the api does, game code is always 32 bit. Drivers are the ones that dictate it.
 
here is an example.

the reason why runtime compiling on the nV cards are better then the x800. It chooses the best and fastest paths for thier cards right?

Typical normal map shader.

4 textures are sent through 4 pipelines

lets say each pipeline has 1 pixel shader unit.

that means we can use 4 pixel shaders at the same time without any significant performance loss. there is no need for branching and loops.

In truth anything above 2 pipelines and 2 pixel or vertex shaders being used at the same hurts performance greatly.

The way SM 3.0 works (very simplified), when loops and branches are added we can now use the pixel shaders similultaniously without getting hurt with extra passes.

Take this into concideration and the way nV's shader compiliers are set up. Thats where the performance comes from. Even though a game that is not SM 3.0 nV will still pick the best paths to use.

(this is speculation but there is no other explination thats possible)
 
while all the nvidia fans keep yammering on and on about these driver speed improvements, whats to say that ATI isnt writing drivers to do the same.
Click to expand...

have you noticed how the 12 pipeline version of the R420 performs the same as the 16 pipeline version? its the same architecture, and im sure its something wrong with the current drivers where the extra 4 pipelines arent being utilized or something. we will probably see drivers soon that will fix it





-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>
<A HREF="http://www.subhi.com/keyboard.jpg" target="_new">This is you, interweb junky</A>
 
extra pipelines only come into effect when there is a bottleneck there. And I don't know any game that stress more then 8 pipelines unless aa and af is turned on this is when bandwidth usage is more then doubled. And the pro falls back on any of the newer games
 
are you sure about this? sinse the R3x0 and R420 are quite similar, where is teh extra performance coming from then? the clock speed increase wouldnt explain it

-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>
<A HREF="http://www.subhi.com/keyboard.jpg" target="_new">This is you, interweb junky</A>
 
i know for a fact that the Q3 engine will use all available pipelines in pairs of 2

-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>
<A HREF="http://www.subhi.com/keyboard.jpg" target="_new">This is you, interweb junky</A>
 
well can't really use the q3 as a good example, cuase its ogl, ati has issues with that, and then. Q3 doesn't stress pipelines at all. Its more of a polygon thing with the Q3 engine and later games.
 
also once framerates go above monitor refresh rates its totally inaccurate, it could be used for relative performance differences but not absolute.

how can you measure something when your scale is too small?
 
i wish davepermen was still here so that he can argue against you


because i dont know crap :)
-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>
<A HREF="http://www.subhi.com/keyboard.jpg" target="_new">This is you, interweb junky</A>
 
well look at this way,

the difference from the Q3 engine and Doom 3?

The engines are very similiar structure wise. Doom 3 adds shader support, and all lights are now dynamic with volumetric shadows on everything.

When using shaders like normal mapping, what is getting stressed is the pipelines (calculations done on the gpu end is incosiquental for the most part because of the way the shader is optimized it should only use the least amount of gpu).

Since GPU's are getting more power more polygons is the always there. But the huge leap in model quality is all at the texture level. There is no way any graphics card today can do 20,000 k polys per character without crushing it (well unless its tesselated after the low poly is sent through the port) because of bandwidth. Animated characters at every scene must be sent through the agp port at every frame because skeletal animation is done at the cpu level because of physics.

So logically what graphics cards companies and ogl commitee and ms have done was improve texture and lighting qualities to fool the eye into thinking there are alot of polys there. Normal mapping and displacment mapping techniques are very old from the 80s manybe even older.

The problem was GPU's weren't fast enough to handle the amount of calculations plus there were no porgrammable shader to do it till a few years back Gf 2 had some programmable shaders but not till Gf3 came out it was fully exposed. So really before this there was no way of doing advanced multi texturing like normal mapping. So there was no need for many pipelines. Thats why even though the Quake 3 engine does support it, its not that good with it.


Now what happens with the artecture, pipelines are really not being stress on the Q3 engine becuase there are no advanced shaders. So the programmable pixel and vertex shaders are doing nothing.
 
There is no way any graphics card today can do 20,000 k polys per character without crushing it (well unless its tesselated after the low poly is sent through the port) because of bandwidth.
Click to expand...
well, one coudl use this to argue that the x800 is better :)


in a few fillrate tests, it had like almost double that of the 6800 in some areas. that, along with 3DC, could mean some impressive visuals.


i realize that displacement mapping is supposed to be better than normal mapping.. but from what ive seen, normal mapping on the x800 is VERY impressive looking. can the same level of detail be attained with displacement mapping? i mean can the same amount of polys be drawn so taht it looks as good? and does displacement mapping increase the number of polys the 6800 can crunch, or does that come from the branching/looping in DX9c?



i get what your saying about Q3. unfortunately, i dont have enough knowledge about the the pipeline structure to totally grasp it, but i do see what your saying. it DOES explain why the x800pro is performing the same as the x800xt, because they have the same number of vertex pipelines (6)
-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>
<A HREF="http://www.subhi.com/keyboard.jpg" target="_new">This is you, interweb junky</A>
 
Oh yeah displacement mapping is much better then normal mapping. It has to use 32 bit textures no matter what there is no compression for it :).

Displacement mapping is not what increases the polycount.

There are two types of displacement mapping, virtual displacement (parrallex mapping, what Unreal 3's tech demo uses) and camera dependent adaptive tesselation with displacment (this is what increases the polygon counts) (lets use CDATDT for ease). The unreal 3 engine can't do CDATDT without reasonable frame rates on any card.

Our engine Shiva 3d, can use tesselation on ATi cards as well (over all our performance is about 8 times faster then Unreal 3). But it can't do it as efficiently as the Gf 6800. Because the Gf 6800 has dx9c and SM 3.0. ATi's cards are forced to use two passes to do this, where nV only uses 1 pass. (each pass all geometry is recalculated so the ATi x800 is forced to do double the cacluations)

Its not that the x800 and the gf 6800 can or can't do it. It is, is it feesiable. :). With the new advances in physics we have to retain higher frame rates :). Our engine can actually push 6 million polys to the screen at 30 fps on a x800 and the same on the gf 6800. On average most of our models are 10-15k each. Characters are going to be right around the 15k mark. But we won't use more then 1.3 million polys for per viable area. Leaveing us at around 70-80 fps which is where we want to be before physics and ai are added in.

When we start getting to that 20k (I think it was actually at around 25k) per animated model limit wierd things start to happen :). For the longest time I though it was something with my code. Our frame rates suddenly go down by almost half on both cards. Pci express should help out here :).

You have to make sure which fillrate test you are looking at. I did the same mistake, I didn't seperate the two types of fillrates texture and vertex. And that really confused me hehe.

Polygon fill rates for both cards are very similiar :)
 
im talking about pixel fillrate. and yea i realize that normal mapping just gives the impression of high polygon models


but can displacement mapping acheive the same level of detail? i mean those movies of the Unreal3 engine are DAMNN nice looking. and its using 3DC 😉 sinse displacement mapping increases the polygon count, wouldnt that decrease performance, compared to using 3dc?

-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>
<A HREF="http://www.subhi.com/keyboard.jpg" target="_new">This is you, interweb junky</A>
 
Why is PCI express going to improve your FPS? From what I understand, no modern card can saturate the AGP bus, and the first major uses are going to be video editing, where the card has to send a lot of info back to the CPU. This was because the AGP bus has lousy upload bandwidth.

"If I owned this place and hell, I'd rent out this place and live in hell" - Toombs
 
Why is PCI express going to improve your FPS? From what I understand, no modern card can saturate the AGP bus,
Click to expand...


because he said that with his engine, the skeletal physics had to be passed back to the CPU to be processed, and models that have 20k polygons are just too much for the agp bus to handle

-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>
<A HREF="http://www.subhi.com/keyboard.jpg" target="_new">This is you, interweb junky</A>
 
Its not going to improve FPS well up to a certian point it will once you have texture flipping but it will allow faster data access from the ram
 
first of all magic drivers aren't crap, that's just jealous talk from ATi. Just because nVidia has a 100 times better driver team than ATi doesn't mean nVidia has to cheat. what you are saying is that, all chipset makers, all sound card makers, all graphics card makers, or anything that uses a driver cheats when teh company releases a new driver that will increase performance, don't limit yourself to just graphics cards, chipset driver could increase performance too. Therefore, your first point is BS. I don't agree with Troll Entium here cause he can't provide any prove or anything even worth reading so far. I don't believe these magic drivers will do what Mr.troll here say it will do, but i do believe what ever Mr.troll said will happen in the future as nVidia continues to increase driver performance. Second, Shader 3.0 isn't crap, that's like nVidiots saying DX9 is not necessary back in the Ti4600 days. no retail games? what's farcry then? DX9.0c may not increase nVidia's performance, but 9.0c will make nVidia run faster than ATi in future games where 9.0c tech is actually applied. That makes your second point BS. SO..... what are you ranting about again?

Also Mr.Troll, no matter how high ur post count is, no one will regard you as someone that knows anything besides made up, NDAed info you been putting out. SO! stop posting multi threads to one question, it just makes you even more annoying.

RIP Block Heater....HELLO P4~~~~~
120% nVidia Fanboy+119% Money Fanboy
GeForce 6800 Ultra--> The Way we thought FX 5800Ultra is meant to be played
THGC's resident Asian and nVboy 😀
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom