By Jake Crimmins posted Aug 18th 2010
After a nine hour session of overclocking on liquid nitrogen, well known overclocker Gautam Bhatnagar and I broke the 3DMark 05 dual card world record. With a score of 56,223 3DMarks, the old record was broken by 1290 3DMarks. This score not only broke the previous dual card 3DMark05 record but also took an overall top 10 score in 3DMark05.
Breaking 3DMark records requires many hours of testing and tweaking to get everything just right. For instance while testing our Intel® Core™ i7-980X for this record we found the memory controller on the processor did not like running triple channel at cold temperatures. To get around this issue we used two modules in dual channel mode instead of triple channel. This allowed the CPU to be run as cold as -185°C, which allowed the CPU to scale up to 6.29GHz. This extra frequency allowed for a higher overall score, although the score was only slightly affected by running dual channel.
Gautam pouring LN2
In order to take the record the processor was overclocked to almost 6.3GHz with dual NVIDIA® GeForce® GTX 480s graphics cards on air. The video cards were slightly overclocked with their cores set to 823MHz, and the memory to 4400MHz effectively. Powering the system with the Professional Series® Gold AX1200 PSU provided all the clean and stable power we needed, allowing us to run both of the GTX 480s as well as highly overclock the CPU. The Dominator® GT DDR3 memory (CMGTX2) was clocked to 2154MHz with the timings of 7-7-7-21. Since the benchmark responds very well to tight timings and high memory clocks the specially made and hand-sorted Dominator GT DDR3 memory provided for very efficient scores.
- Intel Core i7 980x
- 2 x Corsair Dominator GT DDR3 Memory — CMGTX2
- 2 x NVIDIA GeForce GTX 480
- Corsair Professional Series Gold AX1200 PSU — CMPSU-1200AX
- Corsair Force Series™ F120 SSD — CSSD-F120GB2-BRKT
- EVGA® X58 Classified
The record breaking system
Overall the nine hour session was a success with very minimal issues. We did run into some condensation on several occasions which led to system instability. However we quickly recovered and were able to run the benchmark with very high system efficiency. The runs were so efficient that at similar clocks we were able to break the previous record by 800 points. After raising the base clock higher to raise the CPU frequency we were able to increase our lead even further.
Don't attempt this at home!
Let me know what you think about our record breaking run in the comments below!