We have, including the author of this text, and it has been so ever since the beginnings of GPU-accelerated 3D graphics in the era of Pentium 2 computers, when the main question was whether to take a Voodoo2 or Voodoo3 graphics card (Voodoo3 was only fully functional on Pentium 3 processors). The problem of having a GPU stronger than the accompanying CPU can handle results in overkill, which basically translates to a smaller than expected performance gain. The graphics card is limited by the CPU and its performance can literally be halved, and if you have no money left to replace the processor as well, you wasted it. It is also known that inexperienced users buy a new graphics card, compare its performance to articles and reviews on the net and have performance nowhere near those numbers. This is due to reviewers' habit to use the strongest CPU available to them when testing a new GPU, and even those are usually heavily overclocked. Since these CPUs produce massive amount of “points”, those results simply cannot match with a regular user's ones. This is precisely why we try to avoid testing in such circumstances, simply because the primary characteristic of a test must be validity. This article is dedicated to that exactly – the question of how much a processor affects the GPU.

Testing

In order to have a picture as clear as possible, we had to cover a wide range of possible combinations. Since this is particularly time-consuming (and time is the most scarce of all resources), we decided to pair four nicely differentiated processors and graphics cards. The idea was to test each possible combination of these eight components in two common resolutions. The first one is 1280x1024, inherent to all 17” and 19” monitors, and 1600x1200, which is an exclusive property of major displays. Although the swarm of cheap 20” and 22” monitors with a resolution of 1680x1050 might have made us choose this one, we still decided to go with 1600x1200, since it contains more pixels overall, which makes it more demanding. To sum up, the total number of tests performed was 192! This should, if nothing, provide us an insight within the dependence between processors and graphics cards. Since we had a limited number of tests, we decided to take a look at the more “practical” use of hardware, so we tested everything in actual games, save one synthetic benchmark – 3DMark05, simply because it gives higher numbers as scores, so we can deduce exactly what the difference in performance is.

CPU-choice

The processors we chose were a Core2Quad Q6600 as the strongest one, Core2Duo E8200 as a thoroughbred C2D processor manufactured in 45 nm (which is known as the choice of mid-class gamers due to 6 MB of L2 cache it has, low power consumption and a higher clock than the Q6600), a C2D E2180 processor, with a clock of 2 GHz and 1 MB of L2 cache and finally, Conroe-based Celeron, with a clock of 1.6 GHz and 512 KB of L2 cache.

GPU-choice

We needed to cover quite an amount of “territory” with only four cards, but we decided to concentrate on the mid-range, i.e. expel the weakest and strongest models. The reason for such a decision is obvious – even an IT layman can deduce that a 50€ graphics card will hardly make a difference between a 1.6 GHz and a 4 GHz CPU. It's the same situation vice versa – even a quad-core CPU cannot realise the full potential of a Radeon HD 3870X2, let alone weaker models. Anyway, the bottom of our selection was the Radeon HD 3650 with 256 MB of RAM, installed in many configurations with a popular price tag. Following it is a Radeon HD 3850, which is a serious opponent to GeForce 8600GTS and presents the best possible buy in the ~120€ segment. One step ahead, we have the GeForce 9600GT, Nvidia's newest and the prime representative of GeForce 9 series. We chose GeForce 8800GTS 512 MB as the strongest model. This palette was the optimal one, and by increasing the number of models, we would probably get a more precise, but insignificantly better analysis.

What it looks like

Have a look at the charts and all your questions will be answered. To start with the weakest card tested, the HD 3650. This card is targeting gamers who play in 1024x768, so it was quite a torture to test it here for its limited number of ROPs and a generally weak GPU. Naturally, the processor cannot help much here and scaling is basically non-existent. However, it was noted that Celeron scored lower results with this card than other CPUs. The first of the “more serious” CPUs, Pentium E2180, already provides benefit for the card, but that's about as high as it gets. Although it might seem to you that the strongest processor (the Q6600) provides much better performance, have in mind that the price difference is ~130€, and that is an amount of money that can get you a far better graphics card. The following two cards are quite similar in terms of CPU: GPU ratio, so we can state that the E8200 is definitely the optimal solution for them. One should note, however, that the E2180 is also a more than decent choice, since it provided comparable performance as well. If you are prone to overclocking, you are in for a treat, since the E2180 can be clocked at much higher frequencies than the default one. The 8800GTS doesn't make a huge difference between the E8200 and the Q6600, but the E2180 is too small a fry to keep up with this GPU.

Four cores against two?

If you are not too familiar with the topic, it has probably surprised you that dual-core CPUs often had better results than the quad-core one. This should not stun you, because quad-core optimization is still not that present, which automatically reduces the gains in performance. Very well, but how can the quad-core be slower than a dual-core processor? Well, the E8200 is clocked 400 MHz higher than the Q6600, and the new 45 nm production process give it certain instruction improvements which speed it up a little bit in addition. Advantages of Q6600 simply cannot be seen in applications which don't make use of it. Therefore, if money is a limiting factor for you, invest in a better graphics card rather than a quad-core CPU.