Splinter Cell

We used the for testing Splinter Cell, and ran the cards through the gamut. For those of you who don't know, Splinter Cell's use of the DX9 back buffer breaks anti aliasing, and although you will still take a performance hit with it enabled you won't see any visual benefit. For reference, Halo has the same issue.

1024x768

1200x1024

1600x1200

Here we can see that the FX5950 Ultra beats the 9700 Pro, although not by much. Splinter Cell is very graphically intensive and every little bit helps here. At 1024x768 without any Anisotropic Filtering to hold it back the FX5950 Ultra puts out some good numbers.

Return to Castle Wolfenstein

Powered by the Quake 3 Engine heavily modified, we used the Checkpoint demo which is loaded with explosions and multiplayer action.

1024x768

1600x1200

The first of our OpenGL tests shows a more positive look for the FX5950 Ultra with a high resolution although the numbers are far from being stellar. At a lower resolution like 1024x768 when compared with the 9700 Pro, a card that is over a year older now, the FX5950 Ultra shows very poor comparison performance indeed. It's not until you get into the realms of 1600x1200 that FX5950 Ultra begins to break away convincingly.

Call of Duty

Call of Duty uses a variation of the Q3 engine, and should prove to be a nice OpenGL reference for video cards (and we can get away from the silly FPS readings in Quake 3 / Jedi Knight and RTCW as well). Since the game doesn't include a Timedemo we have our own in house demo recorded.

1024x768

1600x1200

In some of the previous tests done with the FX cards we have noticed that 8xFSAA hasn't been a possibility, with the card defaulting back to 4xFSAA as the memory buffer fills up too quickly. The MSI GeForce FX5950 Ultra is able to run at these settings with Call of Duty although as you can see it does take a pretty big performance hit from 4x onwards. Again the performance of the FX5950 Ultra only just about beats the 9700 Pro when you apply AA and AF.

Overclocking

Overclocking should net you some extra performance, but beware of the risks. Never jump straight to the highest overclock, and make sure you move up a little at a time.

Of note about the FX5950 Ultra is their are two clock speeds. 2D speeds are 300MHz Core/950MHz Memory, while the 3D speed is bumped up to 475/950. There isn't really much point in overclocking the 2D, so I concentrated on the 3D. The MSI GeForce FX5950 Ultra VTD-256 sports Hynix 2ns memory so with this in mind I jumped directly to 1GHz on the memory without issue. At 1.01GHz I had to repeatedly test the card before I was allowed to apply my chosen settings but despite this it remained stable at 1.01GHz. 1.02GHz was however not stable and so I dropped back down to 1.01GHz.

For the Core and considering the BIG cooling involved on the card, I had high expectations and sure enough I wasn't disappointed. I went directly from 475 to 500 without any signs of problems at all. My temperature probe mounted on the back of the card and under the fan didn't display much of an increase in heat (and neither did the control panel applet) at this point and I wasn't worried in the slightest about its current speed. From the 500 mark I went to 525 with similar results although this time the heat had started to increase slightly (2C). From 525 I went to 550 and upwards all the way to 590 where upon I had my second lock up (first being from the memory). 580 proved to be the highest stable on the core with memory at defaults.

Final stable overclock was 565 Core and 1.01GHz Memory. What this did for the Splinter Cell benchmarks can be see below, and I'll leave it up to you to decide if the risk involved is worth the performance increase.

Splinter Cell OC Results

1024x768

NEXT