3D TestsWe ran our test machines through a series of 3D applications, including a 3D animation package and several 3D games. The desktop G4 was tested with its original ATI Rage 128 graphics card, while the PowerBook was run with its built-in first generation Radeon Mobility graphics chips. In an effort to test the new system as thoroughly as possible, we tested the dual G5 with the three high-end graphics cards users can purchase for their new systems: the OEM version of the 8x AGP ATI Radeon 9600 Pro and the 8x AGP OEM NVIDIA GeForceFX 5200 Ultra, and, thanks to the good folks over at ATI, a retail version of ATIís 4x AGP Radeon 9800 Pro.
Walking into these tests, we were expecting the Radeon 9800 Pro to win almost every contest, hands down. With its superior specs, including 8 pixel pipelines, a 3.0 Gpixel/sec fill rate, 680 MHz memory clocks, 256-bit memory bus, 21.7 GB/s bandwith and 128 MB onboard memory, the 9800 is superior to the other two cards in almost every aspect; except for its 4x AGP rating, as well as its 380 MHz core clock speed Ė slightly slower than the Radeon 9600ís 400 MHz clock. While this adage held true in most tests, the GeforeceFX 5200 Ultra showed surprisingly strong results in some of our Unreal Tournament testing.
For a detailed rundown of the three graphics cards, check out IMGís recent review of the ATI Radeon 9800 Pro.
CineBench 2003The CineBench 2003 3D benchmarking utility is built upon Maxonís Cinema 4D 3D animation and rendering software. The utility runs through a series of highly-complex 3D renders, testing a CPUís ability to render 3D environments, both with and without OpenGL optimization. The test results reflect well a computerís speed by combining both its CPU speed and its graphics cardís power.
CineBench averages results from renders of two different projects to determine a computerís 3D speed. The first scene contains 37,000 polygons and 1,064 objects, and the second scene contains 70,000 polygons, but no textures. Each set of tests provide three results - the Cinema 4D Shading Test renders out the scene with the 3D packageís software shaders, measuring the unitís raw CPU power. The OpenGL Software Test renders the scene with the graphics card handling shading and transformation, while the CPU shader handles the lighting. The OpenGL Hardware Test renders the entire scene in hardware acceleration only, relying only on the graphics card to do the work.
CineBench 2003 8.1 was used on the G4 computers, with the G5-optimized Cinebench 2003 8.1 beta running on the new computer.
OpenGL Hardware Lighting
The OpenGL Hardware test relies solely on the graphics cards to render out the images. Not surprisingly, ATIís Radeon 9800 Pro performed best in both tests, pumping about 11% more polygons per second than the 9600 and 38% more than the GeForceFX 5200 in OS X 10.2.8. Interestingly, OS X 10.3, despite sporting improved OpenGL support, slowed down the ATI cards marginally, and increased the 5200ís output rate by about 10,000 polygons per second. Despite these changes, the Radeon 9800 still outperformed the other two cards with similar margins than in the 10.2.8 tests. And for those keeping tabs, the G5 equipped with the factory-installed Radeon 9600 ran over 43 times faster than the 400 MHz G4 equipped with its Rage 128 card.
OpenGL Software Lighting
Using a mix of software lighting and hardware accelerated shading, the OpenGL Software Lighting test uses both the processor and graphics card to render out its scenes. While the G5ís 6.8 times speed increase over the first generation G4 desktop model is no surprise, what did surprise us was the results between the three competing graphics cards on the G5. As a mix of CPU and GPU work, we were expecting the results to mirror the OpenGL hardware test results, with ATIís cards showing a definite lead; however, our final results show all three cards performing very closely, with only a 2.2% difference between the fastest card Ė the Radeon 9800 Ė and the slowest, the GeForceFX 5200 in OS X 10.2.8. The margins were even smaller in our OS X 10.3 tests, with the Radeon 9800 performing only 0.7% faster than the GeForceFX 5200.
Cinema 4D Shading
In the software-only shading test, the G5 easily surpassed both the G4s, running about 4.8 times faster than the 400 MHz, and 3.6 times as fast as the 667 MHz. Since the Cinema 4D Shading test relies solely on the computerís processor to render out the scenes, itís no surprise that the results are all within a one- to two- percent spread. Although the Radeon 9800 ruled in OS X 10.2.8, the Radeon 9600 actually beat out ATIís flagship graphics card in OS 10.3.
Cinebenchís final tests raytraces a single frame from a ďreal-lifeĒ scene, lit by 35 light sources, 16 of which use shadow maps and cast soft shadows. Raytracing is a software-only rendering algorithm that produces some of the most realistic lighting and natural-looking images. Cinebenchís first tests a single processor, then repeats the render a second time, splitting the work between all available processors.
Being a CPU-only test, the results of the test are all very close between the three cards, with the Radeon 9600 configuration performing slightly faster than the Radeon 9800, which we attribute to the cardís faster 8x AGP bus speed.