|
|
|
Is
the PCI version of the Radeon a lot slower than the
AGP version? Post
Date: February 23rd, 2001 Many
of you don't have an AGP port so you'll need a PCI
version of the Radeon if you want more 2D or 3D
speed. You ask me, "How much speed do I give up
using the PCI version? Is AGP really that much
faster?" The answer is, "It depends." It depends on
what application you run. It can vary from "NO
DIFFERENCE" to as much as "33% DIFFERENCE." Here's
some 3D game data to illustrate the
point. First,
Quake 3 Arena 1.17 running Demo001... Here
is Quake 3 Arena Beta 1.27g running
Demo127... And
now Unreal Tournament 436 running the "Wicked 400"
demo file... PERFORMANCE
ANALYSIS With
just this limited testing, you can see that in
certain cases the AGP version of the Radeon smokes
the PCI version. Yet in others they are even in
speed. One
way to account for the variation is the fact that
Unreal was using RAVE or DirectX acceleration and
Quake 3 Arena was using OpenGL. Keep
watching this page as I will be adding the results
from testing the PCI version of the Radeon on a
Blue'n'White Yosemite with 66Mhz PCI slot running a
G4/500 upgrade and a 9600 with "older version"
33MHz PCI slot, slow bus, and a G4/500
upgrade. As
you can see, I threw in some Pentium 3 test
results. Interesting that it shows its legendary
game superiority running in DirectX mode with
Unreal Tournament but becomes "mortal" when running
OpenGL mode in Quake 3 1.17. Note
that the AGP Radeon equals or beats the GeForce2 MX
in all 3 tests. RELATED
TESTS Michael
Breeden posted his comparison of the
Radeon
and GeForce2 MX using a G4/500. TEST
NOTES Testing
of the Macintosh boards was done on a Power Mac
G4/533 and Power Mac G4/500 under Mac OS 9.1 with
512M of RAM. The
Pentium 3/866 had 512M of RAM and ran Windows
ME. Thanks
to ATI
Technologies,
I was able to test both the AGP and PCI version of
the RADEON along side the nVIDIA GeForce2 MX.
(I
would have included the Voodoo5 5500 but with 3Dfx
being bought out by nVIDIA, I didn't want to
torture you with what might have been.) QUAKE
3 DEMO1 -- Id's Quake
3 Arena (1.17)
was set at 1024X768, 32 bit, with Maximum Geometric
Detail, Maximum Texture Detail, Lighting Map,
Trilinear Texture Filter. TEST METHOD: When the
main screen appears, I press "~" and enter
"timedemo 1" (return) and "~" once more. Then I
click on DEMOS and run Demo001. Once it finishes
and returns to the main screen, I press "~" once
more to get the frames per second
readout. QUAKE
3 DEMO127 -- Quake
3 Arena Beta 1.27g
was set at 1024X768, 32 bit, with Maximum Geometric
Detail, Maximum Texture Detail, Lighting Map,
Trilinear Texture Filter. TEST METHOD: When the
main screen appears, I press "~" and enter
"timedemo 1" and then "demo demo127" (return). Once
it finishes and returns to the main screen, I press
"~" once more to get the frames per second
readout. UNREAL
WICKED 400 -- Unreal
Tournament (436)
was used with the Wicked400.dem document.
Preferences were set to 1024x768, 32bit, High World
Texture Detail, High Skin Detail, Minimum Frame
Rate=0. TEST METHOD: TimeDemo statistics were
turned on. Practice Session was selected. Instead
of pressing the "shoot" key to start, I press the
"~" key. When the console comes up, I enter
"demoplay wicked400" (return or enter). Then press
"~" again (to make console disappear. When the
fight sequence is done and the words Unreal
Tournament appear alone on the screen, I press "~"
again to get the minimum, maximum, and average
frames per second. Graph above is AVERAGE.
(Wicked400.dem can be obtained from
3Dpulpit.com) SPEED
TEST RESULTS
from Bare Feats (by CATEGORY) Some pages
are translated into FRENCH
by MacGeneration. LINKS
to SPEED tests on other web sites HOT
SPEED DEALS DOWNLOADS
that add more SPEED SPEED
UPGRADE
guide RETURN
TO bare feats HOME PAGE © 2001 Rob Art
Morgan, publisher of BARE FEATS
by rob
ART morgan,
mad scientist
Gotta Question? Comment? Want to advertise?
Email rob-art@barefeats.com