(Sorry to post test results in stages like this but it's unavoidable with the access I have to the system and running this site and answer mail, etc. Later when all tests are complete I'll reorganize this page better, perhaps into multiple pages by topic as I usually do in other full reviews. On 6/4/2001 at 2AM I added graphs of additional tests comparing my G4/500 to the G4/733 and Radeon vs GeForce3.) On 7/21/2001 I added more general comments in the games listing (The Sims, Alice, TR Chronicles and Madden 2000).
GeForce3 w/1045 ROM - No Widescreen FW900 Modes Supported: [added 9/14/2001] I've just tested a new Sony GDM-FW900 with an OEM GeForce3 w/rev 1045 firmware and the latest drivers included with OS 9.2.1. I'm sad to say there are nothing but 4:3 resolutions listed. The Radeon offered more than twice as many total modes, including some widescreen modes. However neither card was very sharp in my opinion at 1600 pixel and above resolutions that were tested. I hope Apple will release a firmware update for the GeForce3 to address this. (There are reports of later than 1045 ROM versions in some recent GeForce3 cards, but it's unknown if the later firmware updates support any of the Sony's widesceen modes.)
(5/31/2001 update) I'm running the GeForce3 card now in my 2xAGP Gigabit G4/500 dual processor system (no other cards installed currently). See Thursday's news for comments on Apple's power spec for the 2xAGP system, which is under the requirement for the card, but in limited use so far it seems ok. I can't guarantee it will be over time or in anyone else's 2xAGP G4 Tower system (use it at your own risk of course since Apple states it requires a 4xAGP Mac and this would likely void any warranty on the card or the system). I don't have a pre-Gigabit G4/AGP system to test.
I've updated the Quake3 scores below from the G4/733 to include results of the GeForce3 in my Dual G4/500 (2xAGP) system. (On 6/3/2001 results were also added for CineBench, RaveBench, Photoshop scroll tests and MacBench Publishing Graphics tests.)
(Original comments/performance tests follow)
(5/30/2001) I've finally gotten my hands on a 2001 G4 with GeForce3 card to test. Included in this report are comments on what I've seen so far and results of some initial tests of performance and compatibility.
The card (and 2001 G4 AGP slot) isn't an AGP Pro form factor.
OS 9.1's Apple System profiler reported the PC133 RAM as "CL2" but in fact the 256MB dimm is marked as "PC133 333" (and I verified the Samsung chips are CL3). The RAM dimm brand/timing may vary by build however and isn't a major difference (CL2 vs CL3), but I'd prefer to have the lower latency RAM.
The first thing I did was disable the ATI graphics card extensions (including OpenGL ATI Renderer) as well as trim some other extensions I do not use (Voice recognition, Multiple Users, Web sharing, USB extensions for devices I don't have, Palm related extensions, FAX, etc.). I then ran Software Update and installed all updates except for CarbonLib 1.3.1, as well as updating to Quicktime 5.01.
The system didn't arrive until about 2 PM yesterday, and it took some time to install applications on the hard drive. However I did have time to see how the system performed with Lightwave 3D 6.5B, Photoshop 5.5, Cleaner 5 (Mpeg), iMovie2 (QT export test), MacBench (Graphics/primitives, Disk and CD tests), CineBench, Throughput, RaveBench, Altivec Fractal Demo, G4Timedemo, Quake3 v117, GLQuake1, and Star Wars/Pod Racer. 2D scores from MacBench's scripted tests were impressive - the highest I've seen. I've not yet run tests of scrolling in apps, etc. however.
For pro-3D apps users, I'm sad to say that the Geforce3 (tested with the 2.1 and also the 2.1.1 drivers) has the same problems in Lightwave 3D as the MX card did with the 1.1.1, 2.0 and 2.1 drivers. (Missing features in scenes and generated previews - shown in this article.) I don't know if the modeler crashes are still an issue with the 2.x drivers however, as I've not tested for that. (Disabling the Nvidia OpenGL extension was a previously reported fix for the image problems and according to one user, also seemed to prevent modeler crashes.)
In addition to the image problems in Lightwave 3D, there seemed to be some transparency issues in games as well - noticed in GLQuake and Driver for instance. (See farther down for a compatibility list of other games run.) RaveBench's transparency test seemed ok however, which actually surprised me. (For info on RaveBench's tests, see the RaveBench illustrated guide I posted a few years back.)
Performance was impressive in GLQuake (appx 289 FPS at 1024x768). To test how the card performed at very high resolutions, Quake3 v117 (Demo1) at 1600x1200/32 Bit color with all options on, HQ settings, etc. ran 63.5 FPS. (The results at 640x480 with the same settings was only appx. 10 FPS faster.) Tests were run with sound enabled and starting with a clean .cfg file. (I'll run tests with Locki's config also just to see what the max rates are.)
More info and test results/comparisons will follow later (as well as graphs of results) but here's a few examples:
Photoshop 5.5 Tests
(21 Filter PSBench test - 10MB image size - lower times are faster)
- G4/733 w/GeForce3: 76.1 seconds
- G4/733 w/Radeon: 76.4 seconds
- G4/500 Dual w/Radeon: 76.7 seconds
- G4/500 w/Rage128 Pro: 101.2 seconds
- PIII 1GHz w/GeForce2 DDR: 85.8 seconds
The G4/733 with Radeon tests were done by a reader, noted in a previous article here. (I didn't expect the graphics card to make any real difference in these filter tests, but just posting that score as a FYI.) G4/500 single and PIII 1GHz scores from this prev. article. (For image scrolling tests in Photoshop 5.5 between the Radeon and GeForce3, see the graphs farther down this article.)
Cleaner 5.02 Tests
Cleaner 5.02 is the latest version of the popular Media Cleaner Pro utility by Terran (now owned by Media100). This test used a 320x240 Sorensen QT movie converted to MPEG. (Using wizard settings of CDROM, 4x, High-End, MPEG format. Data rate avg 2.0Mbits/sec.) Timings were from Cleaner so there's no human stopwatch error.
- G4/733 w/GeForce3: 5 Min. 36 seconds
- G4/500 Dual w/Radeon: 7 Min. 19 seconds
iMovie 2 Tests
(Export tutorial movie w/no transitions to CDROM Medium settings - lower times are faster.)
- G4/733 w/GeForce3: 1 Min. 32 seconds
- G4/500 Dual w/Radeon: 2 Min. 15 seconds
The G4/500 DP tests were run with QT 5 preview 3, but will be retested with QT 5.01.
Altivec Fractal Demo Carbonized version
- G4/733 w/GeForce3: 2,650.5 MegaFlops (appx. 2.65 GigaFlops)
- G4/500 Dual w/Radeon: 3,283.6 MegaFlops (appx. 3.28 GigaFlops)
The latest carbonized version of this Fractal benchmark (available here) uses both CPUs in a dual G4 system (plus other optimizations I suspect), since results were a shade more than twice as high as the previous version reported on my dual G4/500.
(Set to "Amazing quality", millions colors)
- G4/733 w/GeForce3: 84.8 FPS
- G4/500 Dual w/GeForce3: 64.1 FPS
- G4/500 Dual w/Radeon: 64.3 FPS
The G4/500 has 2X AGP and 100MHz bus speed, the G4/733 has 4xAGP and 133MHz bus (as well as the faster GeForce3 card of course). (G4Timedemo is available here.)
I've not yet tested the Radeon with the latest ATI extensions and OpenGL 1.2.2 included on the G4/733, so for now the game scores below are just from the G4/733 with GeForce3.
Notes on Radeon vs GeForce3 Hardware: As you may already know, the Radeon Mac card has a core chip clock and memory clock speed of 166Mhz (166/166). It uses DDR video ram so the effective memory clock is twice that rate (333Mhz). The GeForce3 Mac version is assumed to have the same clock speeds as the retail PC versions - 200Mhz core clock, 230MHz memory clock. It also uses DDR memory, so the effective clock rate for memory is 460MHz. The Radeon has 32MB of video ram, the GeForce3 64MB. For more details on the GeForce3 or Radeon card's chips/features - see the articles here on the Video topics page.
Display Modes - GeForce3 vs Radeon: I wish I still had my OEM Radeon card (shown in this comparison article), but several months ago I traded it for an Mac OEM GeForce2MX card (it was the only way I could get an MX card to test at that time), so this comparison uses the Retail Radeon Mac Edition ( reviewed here previously). There may be more graphic modes supported in the retail Radeon than the OEM one perhaps, I'm not sure. Also note the list below was taken while connected to a Hitachi 19" (VGA connector) monitor. Higher refresh rates may be present (at higher res) with other monitors. However both listings are from the same system, same monitor with only the drivers and graphics card being the variables.
[Update: I hope this is some sort of optical illusion, but with the GeForce3 connected to a Sony F400 19" flat screen monitor there appears to be a pulsing effect seen in brightness/flicker at times (seen at both 75 and 85hz settings, the highest rate available with the card at 1024x768).]
I don't have the full company claimed resolution listings for the GeForce3, but ATI's full range of resolutions on the Radeon are listed on the hardware/specs page of my fall 2000 review. Higher refresh rates are listed there from the original spec sheets than are seen when connected to my 2+ year old Hitachi display (or for that matter on the retail box).
SuperDrive's Audio Extraction Rates: The Pioneer Superdrive is a 2x DVD writer, 8x CDR, 4x CDRW and 32x reader. To test audio extraction rates I used Toast's audio extraction utility (AIFF, 16bit, 44Khz stereo settings) to extract the inner and outer track from an audio CD. (overlap was enabled, but did not seem to affect rates even when disabled.)
- Inner Track: 626-738KB/sec
- Outer Track: 1,323-1,394 KB/sec
This drive's revision (firmware version I assume) was reported by ASP as 1.33.
2D Performance - GeForce vs Radeon: For 2D performance tests I ran MacBench5's scripted emulation of desktop publishing apps (the "Publishing Graphics" test) at 1280x1024/Millions colors with both the Radeon and the GeForce3 in the G4/733 system. The GeForce3 with its faster clock speeds won this test, as well as the Photoshop scrolling test.
Now the same tests with both cards in the Dual G4/500 (2xAGP/100Mhz bus) system:
I also ran MacBench's complete graphics primitives tests. I'll have to post those scores on a separate page as the list is very long. The GeForce3 was faster in most of those functions, except for the DrawChar, DrawText and DrawString functions - where both cards were within 1% of each other.
Photoshop Scrolling Tests: Time to scroll an 18MB/300dpi image file vertically and horizontally at 400% zoom (1280x1024/Millions color display mode). Cards were tested in both the G4/733 and the dual G4/500. Interesting was the dual G4/500 system was faster at this test.
The tests above used a stopwatch, so there's some margin for error. Interesting was that the both graphics cards were slightly faster in this test when in the G4/500 Dual Processor (2xAGP) system than in the G4/733. I know Photoshop has support for dual processors, but was still surprised at the results.
ThroughPut 1.5 Tests: Rene Trost's ThroughPut 1.5 benchmark was also ran. Here's how Rene describes this 'pure' benchmark:
ThroughPut is a little application, written in PPC assembler, that tests how much data your Mac can
push through the PCI or AGP port to feed your graphics card with data."
See his web page for more details, but basically his benchmark tests the amount of data the system can deliver to the graphics card using several modes - CPU (32bit stream), FPU (64bit stream), Altivec (128bit stream) and CopyBits (256bit busmaster). Altivec requires a G4 CPU of course. As you can see from the graph below - the 2001 G4 system produced much higher scores in general. Note the ATI card however produced higher busmaster scores (indicating better drivers).
The 2001 G4s have a 4xAGP slot, 133MHz system/memory bus vs 100Mhz bus and 2x AGP slot of the Gigabit G4, as well as improvements in the motherboard chipset which is said to improve PCI card performance.
CineBench2000 Tests: I used Maxon's Cinebench 2000 benchmark (available here) to test both the Radeon and GeForce3 cards in the G4/733 and the dual G4/500. Cinebench reports a score for Software shading (no opengl hardware acceleration), OpenGL Shading (hardware accelerated), a single CPU raytracing score and a Dual CPU raytracing score. (The G4/733 has only one CPU so there are no results for the dual CPU test.)
As you can see from the scores - the graphics card made literally no difference in this benchmark. However the dual CPU system shows a significant benefit from software that takes advantage of multiple processors.
RaveBench 1.1 Tests: Although there were transparency issues with the GeForce3 (and GeForce2MX card) in some games like Driver, RaveBench's transparency test looked OK. I ran tests of the Radeon card (driver v1.1.1) and the GeForce3 card (2.1 drivers) in Ravebench. The graph below shows the results of each of the tests. (Higher scores are faster.)
For info on RaveBench's tests (including screenshots of each), see the RaveBench illustrated guide I posted a few years back.
Memory Bandwith Benchmarks: Newer Tech's GaugePro 1.1 (last version they released) is not fully compatible with the 7450 CPU, it reports an unknown CPU type and a CPU speed of 399MHz instead of the actual 733MHz. Memory bandwidth results with it on the G4/733 were only 145MB/sec - much lower than even G4/450 cpu upgrades in a B&W G3 (over 200MB/sec typically for previous G4 systems using GaugePro's memory test in OS 9). For more on the low memory bandwidth benchmark results with G4 CPUs after the 7410, see these articles:
- G4/733 w/GeForce3: 289.3 FPS
Quake3 Radeon vs GeForce3 in G4/733: Q3 v117 used, 140MB allocated. Demo1 test, all game options on (marks on walls, HQ sky, etc.), 32-bit mode, High Geometric detail, texture quality slider one notch down from max, bilinear filtering, clean .cfg file. cg_gibs="0" in the autoexec.cfg file)
Quake3 Comparisons - G4/733 vs G4/500 Dual CPU: Same settings as above, but this table also includes results of Radeon vs GeForce3 tests in my 2xAGP G4/500 Dual CPU system. NOTE: All GeForce3 tests used OpenGL 1.2.2/Nvidia drivers 2.1. The G4/500 DP system used ATI Radeon 1.1.1 drivers, OpenGL 1.2.1 and OpenGL ATI Renderer 1.2.3. The Radeon tests in the G4/733 used the ATI extensions installed at the factory (later versions), OpenGL 1.2.2 and OpenGL ATI Renderer 1.2.4. (I re-ran the Radeon/DP 500 tests this weekend, as since my last run that system had a firmware update and many system/OS software updates. Tests this weekend showed much lower 640x480 scores than I had previously with the Radeon in that system.)
Quake3 v117 Performance
||55.0 FPS ||49.4 FPS
||55.3 FPS ||45.1 FPS
||53.7 FPS ||28.8 FPS
||54.0 FPS ||36.2 FPS
The fractionally higher score (0.3 FPS) at 1280x1024 that at 1024x768 with the GF3 is just due to normal run/run variations. It is odd the Radeon had a 0.8 FPS higher score at 1600x1200/32 in the DP 500 than the G4/733, but that's still less than 1 FPS variance and perhaps the driver/OpenGL differences were the reason. (The Radeon card at 1600x1200/32bit is fill rate limited as you can see from the drops in scores as resolutions rise.)
In quick tests, the GeForce3 seemed to have no image problems with 4x4EVO, Descent 3 OpenGL, Deus Ex (1.01), Star Wars Pod Racer and RainBow Six (although they were not extensively play tested). Descent 3 had some problems when I tried to run at 1024x768 mode however. (The readme noted a problem with DrawSprocket 1.7.4 when running the game at higher than 640x480 mode, but the G4/733 system had v1.7.5 which has the same issue according to a reader's Descent3 tips in the forums here. The readme says putting DrawSprocket v1.7.2 in the Descent3 folder should fix that.) There were some occasional "pauses" in gameplay with several games (perhaps due to music playing from the game CD? the DVR drive doesn't seem to be that fast a CD reader). I don't have a copy of "The Sims" to test yet. (A reader had asked if the GeForce3 ran this game without any image problems.)
So far no functional problems have been seen in limited tests of the GF3 in the Gigabit G4, but remember this is not an system Apple recommends for the GeForce3. (Should any problems arise down the road...)
Beyond the Numbers:
Performance is not the only thing that matters if there are compatibility or image problems with an application or game you use. As noted above, there's still issues with Lightwave 3D and the Nvidia OpenGL drivers (at least with 2.1.1 and earlier tested here). I was contacted by Apple on the driver issues and believe they are now working on these problems, so hopefully a future update will resolve them. In the meantime I'm including comments on game tests and 2D image quality with the GeForce3 card based on tests to date.
GeForce3 Game Compatibility Reports:
- Alice: This was one of several games I bought on sale at Macworld NY in July. Although once during a transition from a cutscene I saw some stray polygons, in general the gameplay I've seen so far looked good - very smooth and with good image quality. I have not tested the game under OS X (yet). (The OS X nvidia drivers are basically MX versions, not optimized for the GeForce3 as of v10.0.4. However OS X 10.1 announced at Macworld NY (due in September), is to have optimized GeForce3 drivers.
- Tomb Raider Chronicles & Editor: Another game I bought at MWNY ($35).
It looked and played fine in limited tests and the Level Editor worked, although the game doesn't seem to use Volumetric Lighting like the PC versions. I have not tested the game under OS X (yet).
- The Sims: To add to the original reader report,
in July at MWNY I bought a copy of The Sims and tested it on a
G4/733 w/GeForce3 (running 2.1.1 drivers). The game looked fine
and played well, at least from my experience completing the Sims' tutorial
which showed no image problems and smooth gameplay.
(Earlier Reader report)
Saw your mention on the front page that a reader asked if the GeForce 3 runs
The Sims correctly. I got a new G4/733 with the GF3 card last Thursday, and
I've run The Sims for several hours now with that configuration with out any
problems/artifacts at all.
The one issue I've noticed - my previous machine was an upgraded 8500
(300Mhz G3) with a Voodoo 5. While the new machine runs MUCH faster, I miss
the anti-aliasing of the V5. Sims is one of the programs where I really
notice the jaggies now, even at 1024x768.
As I mentioned in the Voodoo5 card review here last year (see video articles page), the FSAA of the V5 does look great in some games like Star Wars Pod Racer, where no other card looks as good as the Voodoo5 PCI with Anti-Aliasing enabled. (Granted it was too slow for some FPS games, but for sports games, driving games, Sims, etc. the performance should be adequate). FSAA is an option in the PC Nvidia drivers, but it's not present in the Mac drivers to date. (Same story with the ATI Radeon.) BTW: Some reports I've read from ex-Voodoo5 owners that bought a PC GeForce3 (where you can enable FSAA) noted the Voodoo5's FSAA had noticeably better image quality. It's a shame there will likely be no future 3dfx driver updates since they were bought out by Nvidia. For more info on the Voodoo5 PCI, see the previus review here from last year.
- 4x4 EVO: No apparent image quality issues in limited play (looked good to me at 1024x768 mode), but there was some occasional pausing in the game. Perhaps related to the fact the DVR (Superdrive) does not seem very fast as a CD reader and music was enabled.
- Descent 3 v1.4.1 OpenGL: Problem with Drawsprocket version installed on G4/733 if resolutions were increased from the default 640x480. (Noted in the readme to put v1.7.2 in the Descent 3 folder but I don't have any version that old that I can find - and searching Apple's support pages doesn't show any links to download the older versions.) Otherwise it seemed ok, again with some occasional pausing. FPS meter in the game engine seems to max out at 62.5FPS, but minimum FPS seen during first level play was about 47 FPS.
- RainBow Six: Seems ok in limited play. Another game that they recommend you run at 640x480 resolution.
- Deus Ex (v1.01): No problems seem in limited play. (See tips in Unreal Tournament listing below regarding ini file mods.)
- Driver: Transparency not correct (water, etc.).
- Shogo Demo: Perhaps the 2.1 drivers solved the previous issues noted with MX cards/1.1.1 drivers. Graphics nothing to write home about, but other than some occasional sparkling it seemed ok.
- WaterRace 1.01 (Rave): The 'wake' from the boat (white foam/water spray) does not look as smooth as the Radeon card.
- Madden 2000: I bought this game for $10 at MWNY and it looked fine in playing a demo game on the GeForce 3. The mouse cursor however was much too fast - very hard to control in the main menu screens. However the game characters, etc. looked fine and I saw no problems playing a 5min/QTR game.
- MacMAME: Ed Lai also commented on MacMAME -
"I tried MacMAME again and checked out the options for the RAVE plug-in. Guess what... It doesn't automatically switch the screen to the best resolution like the ATI plug-in does. That means that I was running MacMAME in 1600x1200 resolution! When I switch my screen resolution to 1024x768 and then ran MacMAME w/RAVE plug-in, everything was fast and smooth again."
- Unreal Tournament: To allow OpenGL to be an option (other than RAVE) you need to modify the UnrealTournament.ini file to include the settings noted in Nvidia/OpenGL/UT tips from the FAQ gaming section. (Note: this tip is also said to work for Rune.) Another tip to eliminate the "black boxes" around some projectiles is to set 32Bit mode with "Use32bit=1" in the OpenGL section of the ini file. (Note these tips are said to also work for Rune and DeusEx.) Other mods some make to the ini file (previous Radeon tips for DeusEx/UT) are setting "TextureComposite=True" and enabling 32bit textures (which are disabled by default).
DVD: The version 2.6 DVD player (which has a Nvidia and ATI specific extension) and/or driver version seems to have cleared up some past issues with DVD quality on the Nvidia cards, at least based on some quick checks with the GeForce3 in the G4/733.
2D Image Quality: I use a Powerbook 16-20 hours a day, so I'm used to viewing LCDs. Using my Sony F400, the Geforce3 at 1600x1200 seemed to get less sharp at the max refresh (85Hz) listed. At first I saw some "ghosting" of some text at 75Hz, but switching from a lower refresh rate back up seemed to have cured that. However it's still not razor sharp at that resolution. I'll be comparing the Radeon card at that resolution on the same monitor as soon as I can.
Note that there has been some reports of PC GeForce3 cards like the VisonTek having poor quality 2D from some owners. I did not see the "bleed" noted in some of those reports. I'll dig up and post links to PC forums on this issue (thought to be on-card filtering related). Not all PC GeForce3 owners noted this issue and it maybe isolated to specific card batches/brands.
Although a LCD monitor can have "tearing" in games (if Vsync is disabled) if the framerate exceeds the refresh rate, and they're not as good a gaming display as a CRT - they at least do not have the issues of blurry high resolution modes. (At least at the max native resolution they support - however lower resolutions if scaled to full screen on LCD displays have text far less sharp than the native (max) resolution).
I'll be posting the results of other game tests later. In the meantime, here are some Unreal Tournament and Quake3 scores from an identical G4/733 w/GeForce3 sent by Michael Nelson.
Unreal Tournament v436/Wicked 400 Benchmark
800x600 /medium/medium/32bit/decals/dynamic lights
18.59 min/42.09 max/27.92 avg
800x600 /medium/medium/16bit/decals/dynamic lights
19.45 min/41.27 max/28.54 avg
1024x768 medium/medium/32bit/decals/dynamic lights
18.04 min/40.29 max/27.56 avg
1024x768 medium/medium/16bit/decals/dynamic lights
17.51 min/40.41 max/27.56 avg
Quake III v1.17 Demo001
800x600 32bit/vertex/high quality/max texture/marks on walls off/ejecting brass off/32bit texture/bilinear:
800x600 16bit/vertex/high quality/max texture/marks on walls off/ejecting
brass off/16bit texture/bilinear:
1024x768 32bit/vertex/high quality/max texture/marks on walls off/ejecting
brass off/32bit texture/bilinear:
1024x768 16bit/vertex/high quality/max texture/marks on walls off/ejecting
brass off/16bit texture/bilinear:
1280x1024 32bit/vertex/high quality/max texture/marks on walls off/ejecting
brass off/32bit texture/bilinear:
1280x1024 16bit/vertex/high quality/max texture/marks on walls off/ejecting
brass off/16bit texture/bilinear:
1600x1200 32bit/vertex/high quality/max texture/marks on walls off/ejecting
brass off/32bit texture/bilinear:
1600x1200 16bit/vertex/high quality/max texture/marks on walls off/ejecting
brass off/16bit texture/bilinear:
1600x1200 32bit/lightmap/high quality/max texture/marks on walls on/ejecting
brass on/32bit texture/trilinear:
Michael later sent other tests at different resolutions/settings:
Quake III v1.17 Demo 1
640x480: Simple Items on/Marks on Walls off/Ejecting Brass off/High Quality
Sky off/16bit colour/Low Geometric Detail/Min texture quality/16bit
Textures/Bilinear Filtering: 94.5FPS
Maximum Quality Setting:
1600x1200: Simple Items off/Marks on walls on/Ejecting Brass on/High Quality
Sky on/32bit colour/High Geometric Detail/Max texture quality/32bit
textures/Trilinear Filtering: 54fps
Wicked 400 Benchmark
640x480: 16bit/low/low/decals off/dynamic lights off:
21.22 min/44.8 max/32.23 avg
Maximum Quality Settings:
1024x768: 32bit/high/high/decals on/dynamic lights on:
15.04 min/37.66 max/24.21 avg
I also did some testing in Deus Ex in the same manner (Fastest, Medium, and
Deus Ex Intro
640x480: 16bit/Low World Texture/Low Model Texture/Detail Texture off/Decals
800x600: 32bit /High World Texture/High Model Texture/Detail Texture
Maximum Quality Settings:
1024x768: 32bit/High World Texture/High Model Texture/Detail Texture