(Ed.: We had tons of fun with the guys from Team IRONMODS during our Overdrive Overclocking Championship. With AMD's Socket AM3 Phenom II processors on the horizon, we asked team member Ton "TiTON" Khowdee to do a pre-launch exhibition of the X3 720 Black Edition--a chip that AMD says should overclock better than the Phenom II X4 940 BE. He obliged, and provided this write-up and the following photo documentation of his efforts.)
This article will be a hardcore demonstration of AMD's latest Phenom II X3 720 Black Edition processor. This is not a traditional review of a processor. We are not focusing on 24x7 stability. Instead, we'll focus on passing benchmarks at the fastest possible speed. We will be putting AMD's latest Black Edition chip through the ultimate torture test. The CPU will be frozen with liquid nitrogen and its voltage cranked all the way up. We are going to dance on the fine line between breakneck speeds and complete CPU destruction. We will go to the dark side and see what performance rewards are awaiting those brave enough to travel down this path.
Here is the platform that will be used for this benching session:
- AMD Phenom II X3 720 Black Edition (Week 49)
- Gigabyte GA-MA790GP-DS4H BIOS:F3
- AMD Radeon HD 4870
- (2x2GB) 4GB Crucial Ballistix Tracer 6400 4-4-4-12 DDR2
- Western Digital 74GB Raptor
- SilverStone Zeus 1200 Watt Power Supply
- Danger Den Torture Rack
- AMD stock CPU fan
- IRONMODS Sumo XL Solid Copper LN2 Pot
Before we can gauge the untapped power of the AMD's Phenom II X3 720 Black Edition processor,we must first set a baseline. We will be using a small set of benchmarks to capture the results. SuperPi 1.5, a long time favorite of overclockers, is a single-threaded application that calculates the number of digits after the decimal in Pi. Similar to SuperPi, WPrime is a relatively new multi-threaded application that takes advantage of all cores on a processor. Futuremark's 3DMark06 will be used to see what type of performance gain can be achieved in graphics benching.
All of the baseline screen shots will have several CPUz windows open showing the system speeds (check out the photo gallery for baseline figures on all of the benchmarks we used).
- SuperPi 1.5 @ 24.609 seconds
- SuprePi 1.5 32m @ 28min 27.703 seconds
- Wprime 32 and 1024 @ 18.797 seconds and 600.31 seconds (10min .31 Seconds)
- 3DMark06 14k, CPU Score @ 3463
Overclocking with LN2 requires some special equipment. The most important hardware used in testing with LN2 is the pot--a container that is designed specifically to cool the processors at sub-zero temperatures. Pots are primarily made from aluminum or copper, and optimized for either dry ice or liquid nitrogen. Copper is the ideal material for sub-zero benching due to its ability to hold temperature very well. The Sumo XL Copper CPU pot was designed and developed by Cpt.Planet, who is also a member of Team IRONMODS.
Now that the stock baseline benchmarks are done, it is time to prepare the motherboard for LN2 testing. For this bench run, we will be using nail polish and foam insulation. This is only one way to insulate a board; there are several other methods that range from using petroleum jelly to applying kneadable erasers.
First, the stock CPU retention bracket is removed from the motherboard. A thin coat of clear nail polish is applied and allowed to dry. This gives the board a thin, waterproof coat of protection. After the nail polish has dried, a layer of foam insulation is applied around the socket. Appropriate notches are cut to accommodate the various caps and MOSFETs. After the CPU is installed, a second layer of foam insulation is applied, which only reveals the top of the processor. The goal of insulation is to make an airtight seal to avoid condensation. Nail polish and foam insulation are also applied to the back side of the motherboard before the CPU pot is mounted.
The pot is set into place using threaded rods and knurl nuts, which keeps the pot in place and the pressure even on the processor's IHS. Additional foam insulation is placed around the pot to prevent condensation and help maintain its temperature.
Once the motherboard is insulated and the CPU pot mounted, it is time to freeze the processor. When the pot is mounted, the temperature is ambient. It will take a couple of full pours of LN2 to bring the pot down to the proper temperature. It takes almost 1 liter of LN2 before the CPU reaches -190 C.
As the LN2 boils off, cold vapors pour out of the pot and down on the the motherboard, bathing the Crucial Ballistix Tracer modules. A key benefit to benching with LN2 is that the surrounding components have a constant sub-zero fog to help keep the chipset and MOSFETs cool.
With the Phenom II, AMD has overcome a huge headache commonly referred to as the “cold bug.” Most processors in the past would have different types of cold bugs. A cold bug could effect stability during overclocking, while severe bugs would completely shut down the computer or prevent it from booting. Not all chips have the bugs at the same level. Some of the original Phenoms experienced bugs as high (or low, depending on how you're looking at it) as 0 C, while other chips went down to -40 C before running into problems with lock-ups.
The Phenom II has been a great joy to overclockers so far, who no longer have to worry about the cold bug. The processors have even been cooled below -200 C using liquid helium without issue. The reason you didn't hear about these exploits before was that they weren't possible.
CPU-Z is a program that reports the current speed of the processor, memory, and other valuable information. CPU-Z is not a benchmarking program, as you probably already know. The validation file can be uploaded to CPU-Z's server and checked for its authenticity. All top benchers use this to prove that they have achieved a valid speed. These type of benches are fondly referred to as suicide screen shots. The only goal of this test is to get the highest speed possible without any regards to the condition of the hardware.
With 1.888 volts, a multiplier of 29x, and the temperature at -197 C, we were able to capture a suicide screen shot at 5.8 GHz with the AMD Phenom II X3 720 Black Edition. This processor gave a remarkable 5.8 GHz validation, which is a 3 GHz boost over the stock 2.8 GHz speed. With the voltage cranked up to 1.94 V, we attempted to join the 6 GHz club. Even with the additional voltage, we did not gain any additional clock speed.
SuperPi 1M is a very short benchmark, and it only runs on a single core. With multi-core processors, not all cores perform equally well. Many overclockers pick the strongest core to run the benchmark by changing the application's affinity in Windows Task Manager. This is typically why SuperPi 1M speeds are very fast compared to other benchmarks, which either put strain on the core for a longer time or require all cores to perform equally well.
At a speed of 5.6 GHz, we were able to run SuperPi 1M in 13.000 Seconds. That is more than 89% faster then stock run at 24.609 seconds.
SuperPi 32M takes much longer to run, so speeds are typically slower then SuperPi 1M. We were able to successfully pass SuperPi 32M at 5.2 GHz. Core 1 on the processor was much stronger then the other cores. If you look at AOD, you can see that Core 1 was running at 5.2 GHz, while Core 0 and 3 were at 4 GHz. The ability for AMD processors to adjust each core speed independently allows for great overclocking flexibility.
At 5.2 GHz, we were able to run this particular bench at 16 minutes and 36 seconds, compared to over 28 minutes at stock. This is an increase of over 71% in performance.
As we move along the benchmarks, each test becomes more and more difficult. WPrime utilizes all 3 cores to compute the number of digits after the decimal in Pi. With all cores running at full speed, the processor is pushed to its limit.
For WPrime 32M, we ran all 3 cores at 5.1 GHz and completed the bench at 10.609 seconds versus the stock results of 18.797 seconds. The results under LN2 are 77% faster.
For WPrime 1024m, we ran all 3 cores at only 4.8 GHz, which is still a 2 GHz overclock. We completed the benchmark in 353.703 seconds, versus 600 seconds at stock. We achieved a 70% increase over baseline results.
Futuremark's 3DMark06 is a graphics benchmark, which has its own score specifically for the processor. During the baseline run, AMD's Phenom II X3 720 Black Edition earned a score of 3463. With the processor running at -190 C at 4.5 GHz, we obtained a CPU score of 5262, which is over 52% faster. The performance increase also gave a nice 3,000 points increase to the overall score.
Now that we have completed all the benchmarks using LN2, it is time to clean up the frozen mess. After a couple of hours of benching, frost and snow have built up on the threads and hold-down. For longer benchmarking sessions, many overclocker wrap the pot with a second layer of insulation to prevent condensation. As long as the pot is cold, there is no problem with frost. In the event the area warms up, the frost will melt, which can cause electrical short circuits.
Flipping the motherboard over, you can see that the hold-down is completely frozen, as well as the motherboard and DIMM sockets. Luckily, we had that coat of nail polish to keep the board protected from the frost.
After removing the pot, it becomes clear that there is no frost or moisture around the CPU and socket. This means the CPU will live to see another day of benching under extreme conditions.
Today we put AMD's latest Phemon II X3 720 Black Edition processor to the test. AMD has proven it definitely has a winner here, at least for the hardcore overclockers out there. The chip was able to easily bench 100% faster then stock without any significant problems. The X3 720 Black Edition will definitely be a favorite processor for those that love and enjoy tuning their systems beyond manufacturer specifications.