Click  for Thunderbolt upgrades!
Click for Thunderbolt upgrades!


A Click on this Banner shows your site support to my Sponsors

Accelerate Your Mac! logo
The Source for Mac Performance News and Reviews

Review: ATI's Retail Radeon AGP Graphics Card
By Mike
Published: 10/16/2000
(Updated 1/31/2001 for Radeon 1.1.1 Driver Update tests)

3D Game Performance
Intro | 2D Performance | 3D Performance | Game Performance | Video/DVD | Software/TV Controls | Hardware Specs

Most   G4/AGP system owners who are considering the Radeon are probably very concerned with 3D Games performance (or 3D apps performance like Lightwave 3D, covered on the previous page of this review). 3D game performance at resolutions above 640x480 at least has been somewhat of an achilles heel of the ATI Rage128 cards in the past. The Radeon however is clearly much improved in fill rate over past ATI graphics chips and delivered much higher performance at 1024x768 than the Rage128 Pro card. In fact, in most tests at 1024x768 mode results were comparable to the higher fill rate Voodoo5. (but remember the Voodoo5 PCI card is running in a 33Mhz slot, not a 66MHz AGP slot).

This page has framerate (performance tests) in several popular games and also a summary/screenshot section below the FPS scores to comment on other games like 4x4 Evolution (Demo/Beta) and Descent 3 and link to pages with screenshots from those games.

As Rage128 owners already know (and still true with the Radeon), 32Bit mode is needed to avoid the very poor image quality in 16Bit mode - the 'buzzing bees' effect I mentioned in my OpenGL update to the iMacDV SE review (see apps/game test review page for image samples) and Mac 3D Card Round Up article last fall is not present with the latest drivers, but there's still a 'mesh' effect seen at 16bit modes. (Most noticable at 640x480 where the pixels are large, but still present in higher resolutions.) The good news is that 32Bit mode with the Radeon shows no noticable drop in performace, so there's no need to run 16bit anyway. From what I've seen so far, 32Bit mode with the Radeon looks very good.

Game Settings: Rather than fill this already table-heavy page further unless otherwise noted, I used the same game settings as shown in all my Video card reviews with these games and also listed on the Entry Page of the Mac Game/Video Card Performance (FPS) database. The information on how to run these tests is noted in the FAQ's Game topic area.


Quake3 Arena Tests

All tests used high geometric detail, texture quality slider one notch down from max, all game options on, but a low audio setting. No config file tweaks were used. (In fact before the tests I deleted the quake3 config file forcing a rebuild of it to remove the chance that any settings had been modified from prior tests.) Quake 3 was allocated 150MB of RAM.

BTW: Quake3 would not run on the Voodoo5 under OpenGL 1.1.5 - no keyboard/mouse function. Not sure why but I didn't have time to troubleshoot it fully yet. The only things I remember changing since the past tests were the Voodoo5 drivers (1.10 udpate), Gamesprockets 1.7.5 (update) and the USB/dongle driver for Lightwave 3D v6. Even with the ATI card and its drivers removed (incl. disabling OpenGL ATI renderer) and no other PCI cards installed the game would not recognize any keyboard or mouse input. I can only guess it's related to OpenGL 1.1.5 perhaps, but If anyone has any suggestions write me (I wrote to 3dfx also about this). Quake3 scores for the Voodoo5 were based on past tests with OpenGL 1.1.3.


First I wanted to graph the Radeon's performance from 640x480 to 1600x1200 at 16/32Bit modes to show how little performance drop 32Bit mode has on the card. At the higher resolutions the fill rate limits cause both the drop off.
(Update - see below for results of tests with the 1.1.1 driver update, which showed an appx. 16% gain at 1280x1024 mode.)

Radeon Quake3 16/32bit curves

I was disappointed I could not get the Voodoo5 to run Quake3 with OpenGL 1.1.5 (no keyboard or mouse input worked). Therefore the performance comparison below is from past tests with OpenGL 1.1.3. (Only the Radeon scores are with OpenGL 1.1.5, which seemed to hurt the Rage128 Pro AGP performance in tests in the G4/500 dual processor system.)

Quake3 32bit perf. in G4/500


Note: I'm not sure if it was the OpenGL 1.1.5/later ATI drivers or some dual processor overhead, but the Rage128 Pro ran Quake3 slower in my dual processor G4/500 system than the single CPU G4/500 using OpenGL 1.1.3 and the previous ATI drivers (including the Multimedia 1.0 update.) See the Radeon OEM vs Rage128 Pro tests page for dual G4/500 performance.

The following graph compares 16Bit mode Quake3 performance.

Quake3 16bit perf. in G4/500

I also ran some Radeon tests with the Locki config file (ugly but fast config file tweaks).

Radeon Quake3 locki config

For reference, the Voodoo5 using OpenGL 1.1.5 w/locki ran at 71.9 FPS 1024x768.

For Radeon Quake3 Screenshots at 32Bit mode/HQ settings click here.


UPDATE: Radeon 1.1.1 Driver Tests: After ATI released the 1.1.1 Radeon driver update (available here) I tested my dual G4/500 w/BTO Radeon AGP card running OS 9.1 in Quake3 before and after the Radeon 1.1.1 Driver Update. (Full extension set other than Multiple Users, which was disabled. Airport and Speech however were enabled.)
I was surprised at the increase seen after the update (appx 16% at 1280x1024). As a reader noted in email, installing the included OpenGL 1.2.1 update (part of the easy install) shows that the ATI OpenGL Renderer extension is a part of OpenGL 1.2.3. (My previous OpenGL 1.2.1 install had extension v1.2.1.)
The table below shows the Quake3 1.17 Demo1 results with all game settings on, 32Bit mode, with High Geometric detail except where noted. (Higher Geometric detail increases the amount of triangles in objects.)

Quake3 v117
HQ settings
G4/500 DP
BTO Radeon AGP
OS 9.1
FPS Before
Update
FPS After
1.1.1 Update
640x480
High Geometric Detail
49.454.6
640x480
Med. Geometric Detail
51.256.4
1024x768
High Geometric Detail
47.952.8
(*54.3)
1280x1024
High Geometric Detail
39.445.6
(*46.5)

(* FPS in parenthesis is using Medium vs High Geometric Detail setting.)


GL Quake 1 Tests

Rather than use Quake 1 RAVE, I decided to compare the Voodoo5 vs the Radeon using GL Quake 0.7 with the pOx code mods. GL Quake was allocated 100MB of RAM. Both cards delivered some impressive framerates. Both the Voodoo5 and Radeon used OpenGL 1.1.5 for this test. No config file tweaks were used.

GL Quake1 perf. in G4/500



Quake 2 v1.03 Tests

Quake 2 tests were run using the same settings I normally do for all video card reviews. (No 8 Bit Textures, no Vsync, High detail textures, low audio/CD audio disabled). All results are in frames-per-second, higher is better.

There are two demos run - the Demo1 is the standard ID demo included with the game. The 2nd tests were run with the more demanding (much more action, players, firing, etc.) Crusher demo that is used to show what the worst case framerates during heated play would be.

The Radeon and Voodoo5 used OpenGL and were run at higher resolutions. (Voodoo5 screens show severe corruption at 1600x1200, but I still listed the results.) Other card results in the G4/500 used OpenGL 1.1.3 from previous tests.

To ensure an unmodified Quake 2 config file - I deleted the existing so that Quake2 would recreate it before starting the tests. Quake 2 was allocated 150MB of RAM.

Quake2 Demo 1 results

Quake2 Crusher results

For Quake2 Voodoo5 2xFSAA and 4xFSAA results, see my original Voodoo5 review game tests page



Unreal Tournament 32Bit Tests
CityIntro & Wicked400 Demo

Note: First let me say something that many readers already know, as do most reviewers (at least I hope they do since it's pretty much common knownledge). Unreal Tournament is not a good test of video card performance. It's more CPU bound than any game I know. There have been discussions on this in the forums many times, and some have said certain video card review sites have stopped using UT as a video card test, instead using it for CPU tests. The Cityintro is literally a movie, not a real test of what happens in the game (there's no firing, player action, etc. during the Cityintro flyby). It's only shown here since the real test of UT (Wicked400 intense action demo) shows very similar results for most video cards in the same system.

Some asked why I didn't list Voodoo5 glide mode (16bit) tests in this review. I tested only 32bit due to time constraints and the fact is was a mode that all cards in this review supported, and looked best running. (Only 3dfx cards can run Glide mode.)

Although it's a 16Bit mode test, I've added Glide/Voodoo5 results in the tables below, both in the G4/500 (33mhz PCI Slot) and a G4/450 2MB cache upgraded B&W G3 running the Voodoo5 in a 66mhz PCI Slot. Compare those to the 32Bit OpenGL scores of the other cards and you begin to get the idea that UT is not a great video card benchmark. Of most importance is the Wicked400 tests, since that's a demo of actual intense gameplay, not a simple movie-like flyby like Cityintro.

UT Benchmarks Used: Both Cityintro (simple flyby) and Wicked400 (actual gameplay) demos were used with Unreal Tournament (full version) with the latest 425a update applied. All tests were run at 32-Bit (Rave mode and OpenGL for the Radeon, OpenGL mode for the Voodoo5), Medium detail/texture and low audio settings. Min desired FPS was set to "0" (zero). UT was allocated 160MB of RAM. All tests used OpenGL 1.1.5. (Voodoo5 scores were lower than expected and had image quality issues as noted below.)

All results are in frames-per-second (FPS), higher is better. Since the UT's timedemo stats reports min, max and average framerates, all are listed below. 2D graphics mode was set to 1024x768, millions colors.

Unreal Tournament
CityIntro Tests
32Bit Mode

(All tests in G4/500 except as noted)
AVG
FPS
MIN
FPS
MAX
FPS
Radeon AGP RAVE
(640x480)
45.22 26.51 91.85
Radeon AGP OpenGL
(640x480)
37.26 17.17 87.44
Voodoo5 OpenGL
(640x480)
33.69 10.93 85.64
Voodoo5 GLIDE
16Bit (640x480)
41.66 24.35 61.14
Voodoo5 GLIDE
16Bit (640x480)
(66MHz PCI slot
G4/450 2MB cache)
41.80 25.20 61.40
1024x768 Resolution Results
Radeon AGP RAVE
(1024x768)
36.74 21.91 77.22
Radeon AGP OpenGL
(1024x768)
23.18 14.27 42.58
Voodoo5 OpenGL
(1024x768)
32.71 10.83 86.10
Voodoo5 GLIDE
16Bit (1024x768)
40.40 23.73 61.18

NOTES:
I'm was surprised at the low scores with the Voodoo5 with OpenGL 1.1.5 (drivers v1.10). I could not run high quality texture/detail settings with the Voodoo5 and even with medium quality settings at 1024x768 I saw huge polygons (black triangles) during the tests. Perhaps my Voodoo5 card is suspect, or there's some other issue with my G4/500 (single cpu) and the Voodoo5. Glide mode (16bit 3d API) is Unreal Tournament's highest performance API (only 3dfx cards can run Glide mode). As I said previously, most video card review sites know (some don't even use it anymore) that Unreal Tournament is not a good test of video card performance. Glide mode is usually a better performer since that was what the engine was original designed for, but it's still not a magic bullet. All UT results here are 32bit mode except for the Glide/Voodoo5 tests as noted and there's not a huge difference between them, especially in the Wicked400 demo that better represents actual intense gameplay.

Wicked400 is a recorded demo with intense action, weapons firing, multiple players, etc. and is used to show perhaps worst case framerates during actual play. (The entry page of the FPS database has a link to the Wicked400 demo ZIP file)

Unreal Tournament
Wicked400 Tests
32Bit Mode

(All tests in G4/500 except as noted)
AVG
FPS
MIN
FPS
MAX
FPS
Radeon AGP RAVE
(640x480)
21.59 13.86 33.89
Radeon AGP OpenGL
(640x480)
19.62 10.86 29.60
Voodoo5 OpenGL
(640x480)
18.74 11.62 30.56
Voodoo5 GLIDE
16Bit (640x480)
20.37 12.39 33.34
Voodoo5 GLIDE16Bit (640x480)
(66MHz PCI slot
G4/450 2MB cache)
20.71 12.80 35.70
1024x768 Resolution Results
Radeon AGP RAVE
(1024x768)
20.51 13.60 29.80
Radeon AGP OpenGL
(1024x768)
18.60 10.77 29.37
Voodoo5 OpenGL
(1024x768)
18.06 10.67 30.11
Voodoo5 GLIDE
16Bit (1024x768)
19.61 12.96 33.20


Unreal Tournament as mentioned before is not a good video card benchmark and delivers low scores with most every video card I've tested (esp. the Wicked400 demo).

Other Game Test Comments and Screenshots:

I ran quick tests with several other games with the Radeon card and wanted to comment on them:

  • Dues Ex: As others have reported in the forums here with the BTO Radeon cards, the current drivers produce something I call 'sparkling' in Deus Ex (v1.01 tested). As you move around a room you'll see white dots/lines appear on the screen (they come and go). I don't have screenshots of this game but wanted to note this issue. I've reported this to ATI also. (I tested at 32bit/high detail settings.)

  • 4x4 Demo (beta): I ran 1024x768 mode, high detail/texture settings. Image quality was excellent, much better than I expected. The sky, ground, etc was very detailed. I grabbed some screenshots during gameplay but the resulting files were illegible (image broken up with horizontal lines, etc.) at 1024x768 mode. Screenshots at 640x480 were OK, but didn't look as good as 1024x768 mode. For 640x480 size captures see the 4x4 Screenshots page.

  • Descent 3: I first tested Descent3 v1.3.x, which had some missing textures if 'mip-mapping' was selected. I updated to the current 1.4.1 version and this problem disappeared. In comparing 640x480 OpenGL performance in Descent 3, both the Voodoo5 and the Radeon pegged the FPS meter at 62.5 FPS often, but the Radeon would dip below 20 FPS in combat where I never saw the Voodoo5 drop below 20 FPS. However attempting to max quality settings with the Voodoo5 resulted in a locked up game. Image quality in OpenGL mode looked very similar between the two cards. I grabbed screenshot files from the Radeon only as the V5 tests were run at the last minute. Screenshots were TGA files and didn't look as good as actual in-game quality in my opinion (some banding showed in the screenshots that I did not notice in the game). See the Descent 3 Screenshots page for samples.

  • RainBow 6: Very limited tests, but the game seemed fine on a quick trial with the Radeon. Some (fairly lame scene) screenshots are here.

  • Quake 3 Screenshots: Radeon samples are here. (Quake3 would not run on my G4/500 after installing OpenGL 1.1.5 - no keyboard/mouse function as noted above.)

I also wanted to test Diabo II on the Radeon (and will in the next day or so, but as I'm writing this it's 5AM and I'm too tired to install it :-).


Related Links:
For over 2,000 (and growing) records of Mac Game/Video card performance with most every system and video card - search my Mac Game/Video Card FPS database. For other graphics card reviews and articles, see the www.xlr8yourmac.com video cards page.


The next page of this review covers the Radeon's performance in Quicktime and DVD.


Index of RetaiI Radeon AGP Review

Intro | 2D Performance | 3D Performance | Game Performance | Video/DVD | Software/TV Controls | Hardware Specs

- or -
Back to www.XLR8YOURMAC.com


Copyright ©, 2000.

No part of this sites content or images are to be reproduced or distributed in any form without written permission. All brand or product names mentioned here are properties of their respective companies.