Every year, when it's time to update our CPU test method, we play a role of enthusiastic users looking for new upgrades to stay on the frontline of technical progress. Graphics card comes first, as the most progressing part. We also believe that performance boost provided by such an upgrade should be obvious, otherwise it's easier not to bother at all.

And we must say that, judging by performance in games, the yearly progress of graphics cards rarely disappoints us. At least one of the companies usually has time to roll out a new family of GPUs, top-end cards on which demonstrate 1.5x to 2x boost compared with last-year's top-end solutions.

While the progress of CPUs is usually so that, by the time we update our CPU test method, games are bottlenecked by graphics cards even when mid-end processors are involved.

But there's one thing we have to consider, because not only games, but also 3D modelling suites, depend on hardware acceleration of 3D graphics in our universal test method. 3D suites do not benefit from considerable performance boosts. Firstly, because they load CPU much more and there are no specific bottlenecks. Secondly, because graphics card progress in a somewhat different direction these days. They become more and more universal, capable of handling more and more complex shaders, which isn't the key for 3D suites. The most important factor is geometry processing speed, something that rarely means much in games. As a result, drivers optimization for OpenGL come into the picture. This very factor can make even a weaker graphics card a winner, as we confirmed in due time. But that was long ago. Since that time we have accumulated enough indirect evidence that ATI Catalyst OpenGL driver had been revamped. So, it's time to examine the today's situation then.

Tests

To make the review more interesting, we used two configurations of similar price. One had a more expensive graphics card, while the other had a correspondingly more expensive CPU. Let's see which configuration would be the optimal.

Configuration 1:

• CPU: AMD Phenom II X4 965
• Graphics card: HIS Radeon HD 5870
• Motherboard: Gigabyte MA770T-UD3P

Configuration 2:

• CPU: Intel Core i7-860
• Graphics card: Palit GeForce GTX 275
• Motherboard: Gigabyte P55-UD6

Both graphics cards were reference products working at typical clock rates. AMD Phenom II X4 965 CPU also worked at its typical clock rate. Intel Core i7-860 CPU had the Turbo Boost mode disabled, so its clock rate was dynamically adjusted. But this mode is the default for this processor, and most people use it. Other testbed parts included: 4GB of DDR3-1333 RAM, 750GB Seagate Barracuda ES.2 HDD, Vista Ultimate SP1 x64 OS. We used all of our tests utilizing 3D graphics acceleration, except for 3dsmax. The latter just plainly refused to work. You see, we assembled this benchmark manually from the correposnding SPEC benchmarks, the updated version of 3dsmax itself, and related test tasks. It requires some tuning when the testbed is changed. Anyway, we made the diagrams easier to read and omitted certain data. But all the details are traditionally available in this spreadsheet.

3D modelling and CAD/CAM suites

Improving CPU or graphics card had similar effect in Lightwave, but that of the CPU was a bit stronger. But the performance boost wasn't impressive. If your machine is already similar to our basic testbed, there's not much sense in upgrading it like this.

Changing the graphics card added another 39% to performance. Getting a similar effect from the CPU would require something like Core i7-975 EE, which is much more expensive.

The graphics card provided a noticeable, though smaller, performance boost of 13.5%. This was important nevertheless, because you wouldn't get a similar effect by changing the CPU.

In turn, PRO/Engineer preferred the weaker graphics card from NVIDIA. This was actually the only exception. It was probably an echo of old optimization issues of ATI OpenGL drivers. Or it could be a result of some peculiarities of this application which coincided with the NVIDIA GT200 architecture.

This diagram indicates that in this application, the most conservative of all we used, graphics was handled by CPU. But let's see the CPU benchmark before we make any conclusions.

This also proved that changing CPUs had a minor effect (though AMD Phenom II was better). But changing graphics cards resulted in a nice boost. Hmm, some CPU benchmark that is. Anyway, collectively, Intel Core i7 provided its configuration a slight advantage.

This application also seemed to like ATI Radeon HD 5870. Meaning that ATI had apparently solved its OpenGL driver issues.

Write a comment below. No registration needed!