As you can see from tests like the ATI Shader Demos on the 3D tests page, the 9700 Pro has the potential for much higher performance than previous cards if the application code takes advantage of it. Performance of any card depends on several 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
As you can see from the results below, with the games I had available to test, performance can vary widely depending on those factors. (Again I'm hoping to have a copy of UT2003 to test next week, which should better utilize the 9700. If anyone knows of a good, repeatable FPS test for MOHAA and SOFII let me know.) I tried to stress the cards as much as possible with max FSAA settings, high resolutions, etc. - but in many cases the results are more CPU bound than video card.
Quake3 Arena Tests
Yes, it's an "old" game now, but still popular and one of the few games on the Mac that seems to scale very well with graphics cards. (See many of the other examples of game tests that show CPU bound/flat rates with high end cards.) All tests used the latest Quake3 1.32 version, with Quake3's "high quality" settings, all game options on. (r_smp set to 1 to use dual CPUs, and my typical "s_chunksize 4096" vs default 2048 was used. No other config file changes.)
This is one example where the 9700 Pro blows away even the GeForce4 Ti at high resolution - literally 50% faster at 1600x1200 mode. Not all games scale as well as Quake3 however. (Use of custom configs like boli's 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.)
Red Faction 1.2 Tests
Red Faction 1.2 supports FSAA with the ATI Radeon series, so it was used to test (via builtin timedemo) at the max game settings possible - 4x FSAA at 1024x768/32Bit color.
This was one of the more dramatic examples of the 9700's much higher potential compared to the previous 8500/9000 cards. (Not in the rates - the game seems capped at 60FPS, but in the delta compared to the 8500/9000.)
Jedi Knight II Tests
Clearly with high end cards like the 9700 Pro and Geforce4 Ti, this game is CPU bound (note literally the same FPS rate from 1024x768 to 2048x1536 with the 9700 Pro). I can only assume the slightly lower 9700 Pro FPS rate than the GeForce4 Ti is due to nvida's drivers being more efficient and/or the ATI drivers not being fully optimized yet. (The 9700 Pro's framerates even at the low end where the graphics card is not taxed shows this IMHO. Either that or the game code isn't taking advantage of a faster path in the ATI drivers.)
4x FSAA tests were run at 1024x768 (since that's the limit I believe of super-sample AA mode - the 9700 supports Multi-Sampling which allows FSAA at higher resolutions, but that's not enabled yet in the drivers.) The Nvidia Geforce3 and 4ti cards were tested with both 4xFSAA and with the (optional) Quincunx AA mode enabled. (Qx in the graph denotes Quincunx mode enabled.)
The next graph shows non-FSAA tests (but at the highest quality settings). Only the 128MB cards (Geforce4 Ti vs 9700 Pro) were tested at 2048x1526, as the 64MB cards had problems running that mode.
As you can see, the 9700 Pro shows literally no change in FPS rates from 1024x768 all the way to 2048x1536 - where even the GF4ti starts to drop a bit at the max resolution. If I had a dual 1.42GHz that may help feed the 9700 Pro better than the dual 1GHz test system. The Nvidia card's slightly higher rates even at the low end suggest their drivers are a bit more efficient.
Return to Castle Wolfenstein Tests
Again this game is CPU bound with the 9700 Pro (note the same FPS rate from 800x600 to 1600x1200), where with the other cards (including the GeForce4 Ti) performance drops as resolutions rise to 1600x1200. A dual 1.42GHz model (instead of this dual 1GHz) should show higher rates with the 9700 Pro.
Again the 9700 Pro shows literally a flat rate regardless of resolution, where even the Geforce4 Ti begins to show its limit at 1600x1200. I don't think the 9700 Pro is getting enough data to really show what the card is capable of, at least with this system.
Unreal Tournament X PR3 Tests
(I regret not having a copy of Unreal Tournament 2003 to test, a test copy of should arrive next week for testing and I'll add the results here when it does.)
Note: First let me say something that of you already know,
Unreal Tournament is not a good test of video card performance. It's more CPU bound than any game I know. Even at the max settings and 1600x1200 resolution, recorded demo playback rates were much lower than typical rates seen during play. 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. UT X PR3 was set to Maximum texture/skin detail and MinFPS = 0.
Since UT X PR3 supports FSAA with ATI Radeon cards, I ran tests of the 3 ATI cards with 4x FSAA, 1024x768 mode (highest possible with super-sampling I'm told), to put as much stress on the cards as possible.
The 9700 Pro's AVG framerate is more than 50% faster than the previous ATI cards at this FSAA test.
I then ran demo tests without FSAA, with resolution set to 1600x1200 (still Max texture/skin detail and MinFPS = 0).
During 1600x1200 play with the 9700 pro, typical framerates were from 70 to 100FPS. Again a faster mac like the dual 1.42GHz would better feed a high end card like the 9700 Pro, but with the dual 1GHz it was very smooth at 1600x1200 with max details, much more so than the figures above would indicate.