This was the first time this CPU was tested with 2D performance benchmarks, so learning the capability of the CPU was important. Starting off with SuperPI and PIfast, we attempted to push the CPU to its limits. The CPU ran over 6 GHz without issue, but with the memory cold as well, pushing the IMC was also a priority. SuperPI 1M and 32M proved very difficult, but PIFast was happy to run and was the only bench with a satisfactory result.
You’ll notice in the screenshot above that the CPU voltage is at 1.88V, the setting the the BIOS was actually at 1.85V, but the Classified motherboard tends to overvolt a bit with LLC enabled. I point that out because this was a critical difference I noticed during testing. So keep that in mind as you read the rest of the results. This result did not match my previous personal best of 13.5 seconds with a 980X, but it was very close.
SuperPI 32M was the most important personal goal during these sessions, my personal best was previously set at 6:07 with a 980X. But I was unable to break past the 6GHz barrier with this configuration. Voltages were set between 1.75V and 1.9V, and optimal was in the 1.82V to 1.85V range.
With only 25L of LN2, my testing had to be cut short on 2D, so the PIFast result stands alone.
For 3D performance benchmarks, I added two eVGA GTX 285 Classified video cards to the system, due to the limited LN2 I ran the CPU & GPUs on dry ice only. This was the first time these GPUs have been run with extreme cooling, so I had no idea what to expect. However, I ended up being fairly disappointed by their lack of clock increases versus the water cooled testing I’d done with them previously. However, getting them cold did have one very positive effect. The voltage needed while using water cooling was 1.3V vGPU, while running dry ice they were happy to run the same speeds at stock voltage of 1.18V vGPU. What was strange was that with more voltage the system had difficulties.
Unfortunately, this did not allow me to evaluate the GPUs with very high heat loads, as desired. One of the greatest benefits of good thermal compound is its ability to transfer high heat loads very efficiently. With the Ceramique, I was able to keep the pot temps between -59C to -62C during the 3DMark Vantage game tests.
After allowing the system to dry out for the night, it was set up again with the eVGA Frostbite. After getting the system up and running, we were actually shocked how different it felt. This review was approached with a fair bit of skepticism, however, within about 30 minutes it was starting to fade. Loading up the statup settings from the previous day and slowly bringing the system down to temperature was unassuming, but once cold, the system was definitely behaving better (read: more stable). I started with lower voltages, following the same pattern as the day before, and right off the bat I was able to run SuperPI 32M at over 6GHz!
Not only that, but the voltage was set to only 1.75V in the BIOS. Excitement started to swell as we would assume it would have continued to scale with more voltage, as it had the first day of testing. However, it did not. This was actually a bit confusing at first. But k|ngp|n’s words came back to me, “On other brands I have used in the past, the paste can act as an insulator and can be really stiff or difficult to work with.” And I realized I was probably seeing this exact phenomena, the Frostbite allowing the pot to do what it was intended to do, and control the CPU temp. With the Ceramique, it seemed I had to run high voltage to compensate for the lack of adequate temperature control.
However, a bit counter to this theory, with the Ceramique, the CPU was cold bugging a -130C. However with Frostbite, the coldbug didn’t hit until -145C. This most likely shifted due to the different combination of voltage levels and temperature control.
After getting the score above, we pushed a bit to run again with memory at 6-6-6-18 and also at higher clocks, but it turned out (as it so often does) the first successful run was the best. I moved back to PIFast and was able to match my score from the previous day, but not better it. Since the CPU seemed quite happy with these settings, we decided to see if it was capable of running them through 3DMark 05 and 06. Reconfiguring briefly with a strong GTX 580 on water, we were able to nail some decent results.
After changing the system configuration back over to the dual GTX 285 GPUs, we went back to work on 3DMark Vantage as we had the day before. But the Frostbite didn’t work any miracles this time. Pot temps were slightly different, hovering between -58C and -62C when loaded, but hardly enough difference to be conclusive. It would seem that Ceramique is sufficient at these heat loads and temperatures. After several more hours, a decent Vantage score was recorded, but the GPU clock still did not exceed the clocks achieved on water cooling. I’m not planning to post that score until the end of the month, however, because its currently the best score OCForums (my HWBot team) has for the G200 category for this month’s challenge. I’ll update this article later this month, or you can check the competition page when it wraps up.