Test setup

All of today's testing has been run on the following:

- AMD Athlon 64 3500+
- 1GB PC3200 DDR-RAM
- Asus A8N-SLI nForce 4 SLI motherboard (Socket 939, PCI Express)
- 74GB Western Digital Raptor
- Pioneer 16x DVD-ROM
- Radeon X1300 PRO 256MB 
- Galaxy GeForce 7300 GT 256MB
- 480W power supply
- Windows XP Professional Service Pack 2
- DirectX 9.0c

The following drivers were used:

- NVIDIA ForceWare 91.31 was used on the GeForce 7600 GS
- CATALYST 6.7 was used on the Radeon X1650 PRO

The drivers 'Quality' mode was used for all testing on the GeForce 7600 GS.  CATALYST A.I. was set to 'Standard' on the Radeon X1650 PRO.

Benchmarks used

Below is a list of all the tests and applications used for benchmarking:

- 3DMark 06 (Version 1.0.2)
Overall score, fillrate, perlin noise and batch size tests, 1280x1024, AA/AF disabled
- ShaderMark 2.1 (Build 129)
All tests, 1024x768, AA/AF disabled
- RightMark 1.0.5.0 (Beta 4)
Pixel and vertex shader tests, 1024x768, AA/AF disabled
- Half-Life 2: Episode One 
Elite Bastards custom timedemo, application AA/AF
- Prey 
Elite Bastards custom timedemo, application AA/AF
- Elder Scrolls IV: Oblivion (Version 1.1)
FRAPS, application AA, control panel AF
- Age of Empires III (Version 1.08)
FRAPS, application AA/AF
- F.E.A.R. (Patch 1.07)
Built-in timedemo, application AA/AF
- Splinter Cell: Chaos Theory (v1.05)
Default timedemo, application AA/AF
- Need For Speed: Most Wanted (v1.3)
FRAPS, application AA/AF
- Call of Duty 2 (v1.03)
Elite Bastards custom timedemo, application AA/AF

Synthetic benchmarks

As per usual, we begin by running through a handful of synthetic benchmarks to check out some of the theoretical capabilities of the board on test.  Our comparison board throughout will be NVIDIA's GeForce 7600 GS (clocked at reference speeds), which is available at broadly the same price point as PowerColor's Radeon X1650 PRO board.

3DMark06

3DMark06 is our first port of call, beginning with a glance at Shader Model 2.0 and 3.0 performance using the benchmark's default settings.

The Radeon X1650 PRO has quite a large core clock speed advantage over its rival, which together with its shader unit configuration lends it quite a large hand in very shader-intensive scenarios such as those seen in 3DMark, despite trailing the GeForce 7600 GS in terms of raw fillrate.  In the benchmark's Shader Model 2.0 testing, the ATI board leads by around 14%.

This performance advantage extends further in 3DMark06's Shader Model 3.0 graphics tests, moving closer to the core clock difference between the two boards and ending up with the Radeon X1650 PRO just under 25% faster.

We can see the massive fillrate advantage courtesy of the GeForce 7600 GS' twelve fragment pipelines using 3DMark's synthetic fillrate test, although of course this doesn't tell the whole story, as it fails to show the pixel shading power of the Radeon X1650 PRO, which we've already seen in action in the first two tests.

3DMark's Perlin Noise test utilises a shader with a very high math to texture ratio - Exactly the kind of scenario that parts such as the Radeon X1650 are designed around.  Having said that, NVIDIA's current generation architectures are no slouches in the ALU department either, although the Radeon X1650 PRO does indeed manage to outperform its rival by around 11% here.

We've talked several times before here on Elite Bastards how ATI's Ultra Threaded Despatch Processor present in all their Radeon X1000 series boards helps with the efficient distribution of batches of pixels to be rendered, and it shows its worth from a synthetic point of view here once again, comprehensively outperforming the NVIDIA part.

Finally, we round off our usage of 3DMark06 to take a look at the kind of performance difference the use of ATI's Fetch4 technology, which can be used with some hardware shadow mapping implementations, makes.  3DMark06 supports Fetch4 as well as NVIDIA's equivalent feature, DST, but this can be disabled by way of a checkbox within the application.  If you want to learn more about Fetch4, you can find an explanation of what it's all about in our Radeon X1900 XTX review.

From our results, you can see a small decrease in performance when Fetch4 isn't put to use, of around 8% on the Shader Model 2.0 tests where it is primarily put to use.

ShaderMark

We now move on to focus specifically on Pixel Shader 3.0 performance, courtesy of the plethora of tests run by ShaderMark 2.1 using a wide variety of pixel shader code.

Its swings and roundabouts between the two boards as far as results go here, with some big wins for the GeForce 7600 GS on some occasions and large advantages for the Radeon X1650 PRO on others.  All in all however, the NVIDIA part comes out on top here, with its more traditional architecture seemingly better to suited to the kind of pixel shading work served up by ShaderMark than the 3:1 ALU to texture ratio on offer in RV535.

RightMark 3D

Continuing to focus on pixel shading performance, our next look is at a number of RightMark's pixel shader tests.

Again, we see both boards show their own strengths and weaknesses across the various shader tests here - The GeForce 7600 GS simply eats up our Pixel Shader 1.4 and procedural marble tests, while the ATI board math-centric shader unit configuration holds an advantage in the Pixel Shader 2.0-based lighting tests.

While both board on test feature five Vertex Shader 3.0 capable units, the Radeon X1650 PRO's higher clock speed give it a substantial advantage with regard to triangle throughput.  This translates directly into the performance gap between the two parts here, with the PowerColor card leading the way across the board.

So, that's the synthetics out of the way, let's move on to some real-world testing!