In the test 15 years ago: Intel's Core 2 Extreme pushed multi-GPU setups to the limit


In a test 15 years ago, ComputerBase examined how much multi-GPU systems with up to 4 GPUs scaled with the processor clock (test). Intel's Core 2 Extreme X6800 made the graphics card better than an Athlon 64 FX-60.

Table of contents

  1. 1 SLI and CrossFire in the test
  2. Conroe brought the GPUs to the limit
  3. OpenGL fully exploited Quad-SLI
  4. Conclusion

< h2 class = "text-width text-h2" id = "section_sli_und_crossfire_im_test"> SLI and CrossFire in the test

After ComputerBase tested a Quad-SLI system for the first time on August 9, 2006 with two GeForce 7950 GX2s (in the test 15 years ago three weeks ago), the question arose to what extent the performance in this article was still different from the Athlon 64 FX-60 used had been limited. The Intel Core 2 Extreme X6800 (in the test 15 years ago) showed very large performance reserves in benchmarks with a single GPU and was naturally the even faster choice for another test.

In addition to the two GeForce 7950 GX, an SLI combination of two GeForce 7900 GTX and a CrossFire system with a Radeon X1900 XTX and a Radeon X1900 CrossFire Edition had to assert itself in the test. The processors used were the AMD Athlon 64 FX-60 with 2 × 2.6 GHz and the Intel Core 2 Extreme X6800 with 2 × 2.93 GHz. In addition, a Core 2 Extreme X6900 was simulated by increasing the clock rate of the X6800 to 3.2 GHz.

Quad-SLI and a 30-inch monitor

Conroe brought the GPUs to the limit

The Quad-SLI team was able to gain an average of 13 percent in 1,280 × 1,024 pixels when switching from an Athlon 64 FX-60 to a Core 2 Extreme X6800. With 11 percent more performance, the system with two GeForce 7900 GTX looked similar. The jump with the CrossFire team, which only grew by 8 percent, was less big. The Intel mainboard used represented a potential limitation here, as the GPUs each had 8 instead of 16 PCIe lanes. The increase in the clock rate of the X6800 to 3.2 GHz resulted in marginal increases in performance of 1 to 2 percent in all three systems.

«Previous rating – GPU scaling 1280x1024Rating – GPU scaling 1600×1200 Next»

A similar picture emerged in 1,600 × 1,200 pixels, although the distances between the Athlon and the Core 2 Extreme were smaller. Due to the higher processing load, the GPUs were better utilized and the influence of the processor was correspondingly smaller.

OpenGL played Quad-SLI to the full

As in the original test of the GeForce 7950 GX2 SLI team, the Quad SLI system could not prevail on average against two graphics cards in every resolution. This was due to the fact that in DirectX 9 titles, due to a Microsoft limitation, only three frames could be calculated at a time. Nvidia therefore had to switch to a special rendering mode called “AFR of SFR”, in which two graphics cards worked in parallel on one image instead of each graphics card calculating on its own image. The low efficiency of “AFR of SFR” caused poor performance under DirectX 9. There was no such restriction in OpenGL and accordingly four GPUs scaled significantly better here.

«Previous Doom 3 – GPU scalingQuake 4 – GPU- ScalingRiddick – GPU scaling Next »

The scaling with the processor performance was therefore significantly better in OpenGL titles. In Doom 3, for example, switching from an Athlon 64 FX-60 to a Core 2 Extreme X6800 brought between 22 and 35 percent when using two GeForce 7950 GX2s. It looked similar in Quake 4 with an increase of 19 to 24 percent. Only in Riddick did the processor change practically no difference despite OpenGL.


The results allowed several interpretations. It was important to keep the general weaknesses of multi-GPU systems and their purpose in mind. Due to the poor scaling in DirectX titles, a Quad SLI system could not be recommended to anyone. The results showed that a faster CPU could give both dual and quad GPU systems a performance boost in lower resolutions – but the intended use was high resolutions. There, the advantage of a faster processor turned out to be significantly less. It was thus possible to establish that investing in a higher-performance CPU was not wasted money, but a model of the caliber of an Athlon 64 FX-60 was sufficient to use several graphics cards sensibly.

In the “Tested 15 years ago” category, the editorial team has been looking into the test archive every Saturday since July 2017. The last 20 articles that appeared in this series are listed below:

  • ATi's Radeon X1950 XTX was a top model for 399 euros
  • Thermalright's SI-128 was good, but not perfect
  • Microsoft WHQL torpedoed Nvidia's Quad-SLI
  • An overclocked GeForce 7600 GS with 512 MB of memory
  • Intel's Core 2 Duo E6600 and E6700 cleaned up with AMD
  • The absolute price-performance tip GeForce 7600 GST
  • The Core 2 Extreme X6800 left everyone behind
  • The Zalman VF-900 Cu dominated them all
  • Three GeForce 7950 GX2 from loud to fast
  • GeForce custom designs in extra fast and silent
  • The ” Gamer-Cube “, which only ventilated via the power supply
  • Nvidia's double decker alias GeForce 7950 GX2
  • Intel's Core 2 Duo E6700 shone in the preliminary test
  • Nvidia's nForce 500 for AMD's AM2 socket with DDR2 RAM
  • The best Radeon X1900 XTX came with a Zalman cooler
  • PCIe x8 versus PCIe x16 for multi-GPU systems
  • ATis X1900 GT bit its teeth on Nvidia's 7900 GT
  • PhysX accelerator for 299 euros from Ageia
  • The GeForce 7600 GS was a stunner for 125 euros
  • HTPC cases from Lian Li and Silverstone

Even more content of this type and many more reports and anecdotes can be found in the Retro corner in the forum of ComputerBase.