Graphics Tech Papers and Mid-Range Chipset Roundup

Tech Papers:

www.bit-tech.net/feature/29
www.bit-tech.net/feature/30

Mid-Range Chipset Roundup:
www.bit-tech.net/review/288

Questions, comments etc

 

temp

just read the review... doesn't the 9600XT run at 500/300?

The RV350 runs at 400/300 (9600pro) and I'm pretty sure the RV360 runs at 500/300 (9600XT)

nice pictars

 

Cheers, goddamn typos! So many figures in notepad!

 

hehe tell me about it, if it's anything like a 3dMarking session it's hell

I oughta scan some images of my workings one day

 

mrhaz there seems to be some areas which are a bit hazy.


You mention the f-buffer but do not mention the very improtant information of card/driver support about current Direct 3D and OpenGL High level shader languages. Microsofts HLSL (High Level Shader Language) and OpenGL GLSlang.

These are very important features which have direct impact on future titles performance and image quality.



Do you know about the NV3x temporary register problem games developers and users have found out?

When running a Pixel Shader program you have to store temp values - at the aproapiate word-lenth requested by the shader program (FP32/FP16) for NV3X.

NV3X has a hardware problem where it can only store 2 FP32 temps OR 4 FP16 temps before there 'quad' pipeline stalls.

FP32 and FP16 should be at the same speed on NV3X, but it isn't bacuase the limited register problem.

It is a hardware - problem that has not fixed but 'worked-around' by reducing the number of temp values stored at one time, by ensuring developers :

1) Use the DirectX '_pp' partal precison 'hint' (FP16 output) when programing DX shaders.

2) In OpenGL, program spersisfic 'Nvidia' codepaths that are programmed to use FP16 - low precision.

Lower Precision levels do decrease image quality.

You speak about instruction lengths but you seem to miss out the large hardware issues the NV3x cards.



Also there is no additional information about Nvidia cards Antialasing (especially higher levels of AA) is a lower quality mix of multi+super sampling, it also is not gamma corrected and non programmable (ATI is programmable) so additional AA versions like centroid sampling AA is not available on Nvidia high end cards.


For more information check out here.

http://www.beyond3d.com/forum/viewt...er=asc&highlight=antialiasing quality&start=0

http://www.3dcenter.de/artikel/ati_nvidia_aa_performance/index_e.php

 

I have just also noticed specification errors in the Nvidia hardware table. No NVIDIA card has 8 Pixel Pipelines.

FX 5800 (NV30) is 4 pipelines with 128 memory and controller.
FX 5900 (NV35) is 4 pipelines with 256 bit memory and controller.

The above also does not take into account the different Pixel Shaders and their implemention into the pipelines which is very a complex situation indeed.

Nvidia say they can do 8 operations a second this is very different to actually having 8 pipelines, it is not compareable.

Have a read of this very detailed thread for some pipeline breakdowns of the NV35 and R300/R350.

http://www.beyond3d.com/forum/viewtopic.php?t=8005&highlight=pipelines

 

Is that the truth or what?

Great articles/features/whatever. My 9600XT should come on thursday

 

may or may

lol

 

can we assume the hottie in the picture is mrshaz?

 

^----- See above.. Soooooo pretty...

 

Ok, on my whistle can we have all the people interested in the tech paper over this side of the room and all the people who want to yak about the lady over here... Peep!

 

Yes, there is a thread in the General form about the lady and her pictures, this is the thread for the Technical part

 

nvidia vs ati

recent benchmarks at bit-tech :

http://www.bit-tech.net/review/288

are those scores for 9800xt and 5950fx right ? what about 9600xt beating 9800xt?

now check out another article with exact same games, it shows ati 9800xt dominates nvidia 5950fx:

http://www.driverheaven.net/reviews/xt59/

or this one with aa/af:

http://www.tomshardware.com/graphic/20031229/vga-charts-13.html#unreal_tournament_2003

 

(The above post was merged with the discussion thread for the article).

Which benchmark specifically are you referring to when you make your comments? The Call of Duty one? From the article....

Gameplay benchmarking makes a direct comparison of exact fps difficult because of the way playing the level slightly differently effects the fps, but for a general comparison it can still be useful because it shows the sort of variation between cards at the specific settings during 'normal use' which for the 9600XT and 9800XT isn't actually all that much with COD @ 1280x1024.

I would prefer to see a 'proper' benchmark myself, but not being familiar with COD I presume that if there was a benchmarking mode or something that enabled you to run through the level in the exact same way, then mrhaz would've used it in preference of the gameplay benchmarking (playing the same map with each card, and taking the fps for each run, which will throw up some anomalies as playing the map exactly the same way is nigh on impossible).

 

mrhaz,

Do you have any comments on my post reflections or about the aticle errors I pointed out erlier?