Hi Folks,
I have been asked to make this post by Hangtime, Bash and a few others
about performance settings and drivers for TNT IIs. I have had my V770 Ultra for
over a year now and have managed to learn a little bit about it in that time.
All my experience is with the Diamond card and drivers but most of this will apply
to all TNT IIs. I have included some benchmark results from 3D Mark 2000 and WinTune98
for my machine for comparative analysis of Diamond 4.01 and Nvidia Det2(5.22) drivers.
There is alot of stuff I want to try and cover here so please bear with the excessive
length and please excuse any format problems as I am very poor at word processing.
This is a very confusing topic because there are many different factors which can
have a dramatic effect on performance. There are also alot of different opinions and
misinformation and even some mystery about what settings and drivers yield the "best"
results. I am no expert so my intention here is to share what I know and ask a few
questions to start a discussion about it and get some input from others as well.
First, you need to run a clean machine with minimal running applications, defrag
your HD often and optimize swapfile etc. A fresh install of Windows can make a huge
difference but I dont know too many people willing to do that regularly, not me
You can also set the primary function of your PC to "Network Server" instead of
"Desktop PC" in the perfomance tab of My Computer/ Properties to open up system
memory somewhat. It helps to optimize your BIOS settings to get the most out of your
hardware and there are numerous guides available on the web which can do a much
more thorough job than I could do here.
www.tweak3d.net is one of the best and I
highly recommend it for all tweaking info.
Next, I would also strongly advocate using a tweak utility called PowerStrip
available at
www.entechtaiwan.com as a fully functional free demo. The only
drawback of not registering it is that it will randomly dump your preferences
at boot-up and you have to reset them. It is an exceptional program that will
allow you to control all aspects of your video graphics hardware in many cases
much better than than the manufacturers software and all through a single, well
configured interface. It also has a graphics system information and diagnostics
tab that will tell you everything about your hardware and let you configure it for
maximum performance.
Once you have your machine set up properly you are down to the card itself and
the drivers. The V770 Ultra is CPU limited up to about 600mhz which means you will
derive little benefit, if any, from overclocking the chipset if under that threshold.
After that point it becomes fillrate limited and overclocking the core clock and
memory can yield signifcant improvement. The abilty of any given card to tolerate
this will vary with your specific setup and conditions, heat buildup being a critical
factor as well as FSB/AGP settings etc. I have mine running stable at 165/210, well
over the stock settings of 150/183 core clk/memory. The practical limits for the
chipset are about 185/225 but very few will be stable at those speeds. In order to get
my machine stable at 133/89 mhz FSB/AGP, I had to disable AGP 2x and it no longer
likes to run above 165 core clk. I did not see the hit in performance I had been
anticipating from that setting and everything runs stable but results will vary
from system to system and specific cards. Others have theirs stable without needing to
disable AGP 2x although I havent seen reports about overclocking the card under
those conditions. Use Powerstrip and see what you can get out of your machine.
The Boost Utility in the Diamond In/Control Tools will do this also but with much
less "control" over the increments than PowerStrip. In my experience, the Diamond
drivers have shown to be more stable than Nvidia reference drivers when overclocking,
which brings us to the next subject, drivers.
Here is where it gets really confusing and mysterious. Drivers probably have
the single greatest impact on performance and image quality of a given harware
configuration and there is much debate about them. As to the question of which
drivers yield the best performance(highest FPS) in both AH and WB, the answer has
ALWAYS been the same, the 4.01 initial release from Diamond. This has been extensively
tested by myself and others and has not changed through 4 releases of official
drivers from Diamond and inumerable Nvidia reference builds, beta and official.
FPS will be on average 20-25% higher with the 4.01s than any reference build.
The 4.02s yield similar results with a slight edge to the 4.01s. The reasons why
are rather mysterious because the 5.22 Nvidia build shows HUGE increases in(see below)
synthetic D3D benchmarks for both single and multitexture fillrate but the same
DECREASED performance in both sims. I have another thread going on AGW H&S Forum
about this that has opened up some interesting discussion about why this is so
and I suggest reading it if you are interested. I e-mailed HT with some questions
but haven't recieved an answer yet as he has been a bit busy lately.
I did manage to ask him about it in the arena and he said he didnt know anything
about specific drivers and their relative performance but perhaps if we continue
that discussion here he may respond more in depth if he has time.
(PLEASE!!! HT)
Another advantage of the Diamond drivers is the the ease of use of the In/Control
Tools over the Nvidia interface which require you to make registry changes to access
all of the cards feature settings. Also, there is a profile feature which remembers
different preferences for each 3D app you run and saves them. This can be convenient
and useful although sometimes it can get confused and needs to be cleared and reset.
Now we get to the driver settings themselves. Basically you are balancing
image quality (eye candy) against performance (FPS). If, like me, you have a fairly
powerful machine, then you want to maximize quality over performance. If, on the
other hand, you are trying to get more from a marginal machine, you will want
to sacrifice quality to get a playable framerate. In my experience, only 2 or three
settings will result in a signifcant change in FPS besides V-synch, which is
a separate topic altogether. They are Anti-aliasing and Auto Mipmapping and Mipmap
Level. Antialiasing reduces the "jaggies" at the edges of different objects and
reducing the supersampling rate or turning it off will yield a few FPS on most rigs.
Mipmapping affects texture quality (shimmering terrain) and also how objects moving
towards or away from you (cons) are rendered. Setting the level to 0 and turning off
Automipmapping will also give a couple FPS on most systems. Diabling Fog Table may
yield 1 FPS or so.
On to v-sync!! More confusion ensues
LEAVE V-SYNC ON!!! Disabling v-sync is only
useful for benchmarking the theoretical capabilties of your hardware. It will result in
an increase in the displayed framerate but you will NOT "see" this as these "extra"
frames are just dumped into the frame buffer because your monitor cant display them fast
enough even though your card can render them. You are also more likely to see tearing
and other graphic anomilies with v-sync disabled and gunnery and bombing will
likely suffer as result as well as screen jumps etc. The key to maximizing FPS
with v-sync enabled is to set your monitor refresh rate as high as possible to find
the right balance that the card will produce and your monitor will display.
This involves some experimenting with different refresh rates to see which yields the
highest stable framerate, stability being the desired effect more just x FPS in any
given load situation. Bloom25 has made an excellent post about this already and I
have little to add except that AH doesnt show this as clearly as WB does in terms
of the FPS displayed and the factors of the refresh rate of the monitor. This was
one of my questions to HT in the e-mail and one I would be most curious to know
the answer to. I asked Pyro online one night and got a "dunno kbman" from him
Okay, I am about done here, finally!!!
I am looking forward to seeing this
discussion develope as others chime in, hopefully HT and Pyro too. There are alot
more questions to be resolved about the driver issues and we might all learn
something valuable here from what flows from that. In the meantime, I hope I have
answered some questions and provided an overview for how to get the most from
your box. Anything I missed just ask and I'll try to help. Once again, sorry
for the booklength post, but there aint no easy way to say it
Good Luck!!
kbman
PS It took me three friggin hours to write this damn thing because of how poorly I type.!! I hope your happy now!!!!
WinTune 98 Test Results: Diamond 4.01
CPU(1) Intel Pentium III@735 MHz
Video BoardDiamond Viper V770 Ultra for Windows 98
Video Mode1024x768@32bits/pixel
RAM128 MB
OSWindows 98 4.10.1998
Area TestedValue
CPU Integer2135.969 MIPS
CPU Floating Point856.0626 MFLOPS
Video(2D)106.9316 MPixels/s
Direct3D128.4961 MPixels/s
OpenGL80.76997 MPixels/s
Memory1467.626 MB/s
Cached Disk136.1452 MB/s
Uncached Disk3.726743 MB/s
WinTune 98 Test Results: Nvidia Det2 (5.22)
CPU(1) Intel Pentium III@735 MHz
Video Board
Video Mode1024x768@32bits/pixel
RAM128 MB
OSWindows 98 4.10.1998
Area TestedValue
CPU Integer2136.904 MIPS
CPU Floating Point856.8002 MFLOPS
Video(2D)88.68829 MPixels/s
Direct3D274.7307 MPixels/s
OpenGL24.55497 MPixels/s
Memory1555.23 MB/s
Cached Disk138.6056 MB/s
Uncached Disk3.767028 MB/s
3DMark 2000: Diamond 4.01
RESULTS
Platform: Internal
3DMark Result: 3200 3D marks
CPU Speed: 263 CPU 3D marks
Game 1 - Helicopter - Low Detail: 64.3 FPS
Game 1 - Helicopter - Medium Detail: 48.3 FPS
Game 1 - Helicopter - High Detail: 23.1 FPS
Game 2 - Adventure - Low Detail: 53.8 FPS
Game 2 - Adventure - Medium Detail: 45.7 FPS
Game 2 - Adventure - High Detail: 31.5 FPS
Fill Rate (Single-Texturing): 137.6 MTexels/s
Fill Rate (Multi-Texturing): 239.6 MTexels/s
High Polygon Count (1 Light): 3840 KTriangles/s
High Polygon Count (4 Lights): 3370 KTriangles/s
High Polygon Count (8 Lights): 2928 KTriangles/s
8MB Texture Rendering Speed: 115.7 FPS
16MB Texture Rendering Speed: 115.4 FPS
32MB Texture Rendering Speed: 108.6 FPS
64MB Texture Rendering Speed: Not enough AGP memory FPS
Bump Mapping (Emboss, 3-pass): 82.8 FPS
Bump Mapping (Emboss, 2-pass): 105.5 FPS
Bump Mapping (Emboss, 1-pass): 156.6 FPS
Bump Mapping (Environment): Not supported FPS
3DMark 2000: Nvidia Det2 (5.22)
RESULTS
Platform: Internal
3DMark Result: 3625 3D marks
CPU Speed: 264 CPU 3D marks
Game 1 - Helicopter - Low Detail: 76.6 FPS
Game 1 - Helicopter - Medium Detail: 54.4 FPS
Game 1 - Helicopter - High Detail: 25.1 FPS
Game 2 - Adventure - Low Detail: 64.5 FPS
Game 2 - Adventure - Medium Detail: 49.8 FPS
Game 2 - Adventure - High Detail: 31.8 FPS
Fill Rate (Single-Texturing): 289.0 MTexels/s
Fill Rate (Multi-Texturing): 265.9 MTexels/s
High Polygon Count (1 Light): 3975 KTriangles/s
High Polygon Count (4 Lights): 3513 KTriangles/s
High Polygon Count (8 Lights): 3015 KTriangles/s
8MB Texture Rendering Speed: 210.1 FPS
16MB Texture Rendering Speed: 205.9 FPS
32MB Texture Rendering Speed: 184.6 FPS
64MB Texture Rendering Speed: Not enough AGP memory FPS
Bump Mapping (Emboss, 3-pass): 113.9 FPS
Bump Mapping (Emboss, 2-pass): 132.1 FPS
Bump Mapping (Emboss, 1-pass): 199.0 FPS
Bump Mapping (Environment): Not supported FPS
