Review: Mac Nvidia 6800 Ultra DDL Graphics Card Published: 10/20/2004 (updated 10/29/2004 for Splinter Cell benchmarks)

Intro/Specs | 2D/Video Performance | 3D Performance | Game Performance

3D Game Performance

As you've seen from some other test pages, the 6800 often hasn't shown a big gains over the older design 9800 (with more mature drivers), although few if any apps currently are really using most of its features, or its most efficient paths. (Sounds familiar doesn't it... par for the course with new graphics cards.) To repeat comments I made in previous reviews, performance of any graphics card depends on factors such as: The system's ability to feed it data The quality/efficiency of the drivers The application/game code (efficiency and support of card features) The limits and features of the card design For game tests high resolutions/high detail settings were used in the hopes that even for highly CPU bound games, that would help show some differences in the cards. I tried to stress them as much as possible but in some cases as we've seen before the results are more CPU bound than video card. I'm hoping drivers and games are updated to improve the results seen currently. (Halo 1.0.5.3 for instance shows a huge drop in performance using the NV shaders.) Splinter Cell 1.0 Benchmarks: I'm late posting this but I've run some Splinter Cell for the Mac benchmarks using a standard timedemo file (attempts to use beyond3d's benchmark file failed to run properly, due to the control prefs setup apparently). There's no updates to the game yet (as of late Oct. 2004) and it does not support FSAA, but high quality settings were used (as you can see in the command line used to run the benchmark in the readme file I made that I've included in the mac splinter cell timedemo download). SHADOWLEVEL High, SHADOWRESOLUTION High, EFFECTSQUALITY VeryHigh, SHADOWMODE Projector Here's a graph of the Average FPS rates of both cards from 640x480 to 1600x1200 mode. (Min/Max rates for each are in parenthesis in the bar for each resolution. I did this rather than graph avg/min/max for each resolution in separate graphs to save space and bandwidth.) Call of Duty MP Timedemo: Although I don't get a lot of time for games, this is one of my favorites. Even with the initial driver release, the 6800 delivers significantly faster performance than the 9800 Pro. Unreal Tournament 2004 (v3323) Tests: Although it's limited to 1600x1200 resolution max currently, I used Santaduck's UT2004 benchmark utility (www.santaduck.com/ut2k4/santa_toolpak_bench_ut2004.sit) to automate running tests with the latest (to date) UT2004. For FSAA/AF tests, I had to manually edit the config files. I've included both Flyby and Botmatch tests even though the latter is heavily CPU bound (people still want to know the results, as flyby tests don't equate much to actual gameplay performance.) Once you enable FSAA, the 9800 takes nearly a 50% performance hit at these settings in the Flyby test, but then it's ATI's previous generation, not an X800. In Botmatch tests, with 4x FSAA (only) enabled the rates aren't significantly lower than w/o FSAA at this resolution. (Since they were already well under either card's limits - but adding 8x AF enabled does lower framerates esp. on the OEM 9800.) For the 4x FSAA + 8x Anisotropic Filtering tests, I tried both MultiSampleHint2 (default - "nicest smoothing") and MultiSampleHint1 ("fastest smoothing") modes for the Nvidia card (from a note on UT2003 FSAA modes in the FAQ's gaming section here) just to see what difference it would make. (The botmatch test is too CPU bound to show much difference, although there was some benefit in the flyby.) Note with 4xFSAA + 8x AF, the OEM 9800's AVG FPS rate takes a significant hit at this resolution (drops from 52 to just under 39FPS). Although Botmatch tests are heavily CPU bound, at these settings the 6800 does show a clear advantage where the older design 9800 is running out of breath. Quake3 Arena Tests: Yes, it's long in the tooth, but one of the few games on the Mac that scales very well with graphics cards AND shows a big boost from dual processors. I included tests at lower res. like 640x480 as sort of a driver efficiency check. The last Quake3 1.32 'G4' version was used with Quake3's "high quality" settings PLUS high geometic detail and max texture quality, with all game options on. Audio not disabled, but music volume set to 0. (r_smp set to 1 to use dual CPUs and my typical "s_chunksize 4096" vs default 2048 was used as well as "com_hunkmegs" set to 256 and "s_mixadhead" set to 0.1.) The 6800 shows a huge advantage at high resolution as the 9800 runs out of breath. Not all games scale as well as Quake3 however. (Use of custom configs like boli's (homepage.mac.com/boli/boli_q3config.cfg) can deliver much much higher rates at the expense of image quality, but I figure if you're going to buy a high end card, then you want all the eye candy ON. And with rates like these - there's no need for reducing image quality.) HALO Tests: I used Halo 1.0.5.3 (latest to date) with maximum detail/image quality settings (shadows/specular/particles, etc.). Only Lens Flare was set to Low and Audio Quality set to Low (variety left at medium setting). I tested at 1280x1024, 1600x1200 and 1920x1080 with no FSAA, 4x and 6x FSAA. The 6800 Ultra card was tested with ATI and NV vertex/pixel shader settings. As you can see from the results, currently using NV shaders has a big impact on performance. Using ATI shaders on the 6800 had much better performance than NV shaders, but some parts of the demo showed image problems at times (like bands in the ocean on the flyover, peppering of the ship's upper interior during some FSAA tests, but not all). Hopefully the NV shader problems will be fixed in a future game and driver update. BTW - although Particles were enabled in the game video setup menu, the timedemo reports always listed them as "Off". (I almost didn't bother graphing many of these results since the rates are so low they're not playable. But several people wanted to see how bad really high res/settings were in Halo even on a Dual G5.) Note: The OEM 9800 Pro had scrambled video at 6x FSAA/1920x1080. Return to Castle Wolfenstein Tests I used the same RTCW Multiplayer demo test used in previous Video card and CPU upgrade reviews, sent by a reader years ago. All tests used High Quality game settings, audio enabled, but music volume set to 0. Jedi Knight II Tests Clearly with high end cards at least, this game is heavily CPU bound (flat FPS rates across resolutions) unless you kick in FSAA. This isn't a great game for benchmarking, but since it also supports Nvidia's Quincunx FSAA I included it. 4x FSAA tests were run at 1600x1200 only. (Qx in the graph denotes Quincunx AA mode.)

Index of Mac 6800 Ultra DDL Review Intro/Photos | 2D/Video Performance | 3D Performance | Game Performance - or - Back to www.XLR8YOURMAC.com

© 2004. All brand or product names mentioned here are properties of their respective companies.