When I first heard of overclocking competitions, I was skeptical. I knew they existed, but I figured they were limited to an outrageous few with access to exotic cooling methods, like phase changers and liquid nitrogen, and custom-hacked hardware, like voltage-modded motherboards and graphics cards, not to mention more money than sense for processors like Intel Extreme Editions. I also figured the most recognition competitors got for their accomplishments was posting scores on benchmarking sites like 3DMark’s Orb or obscure forums that receive an influx of new posts each time a faster piece of hardware comes along.
To my surprise, this arena isn’t puny by any means. Gigabyte has turned friendly competition into a full-fledged international championship, using eleven regional events around the world to wring out the best from the rest when it comes to overclocking. The inaugural Gigabyte Open Overclocking Championship, or GO OC 2008 for short, took place September 25th in Taipei, Taiwan, and we were there to witness the madness.
Over 40 participants were invited to the final competition, but as in any coordinated event this large, a few people, including some favorites from Greece and Germany, were unable to make it. In the end, 23 teams arrived representing 20 countries. Most of the teams had two members, but a few showed up with only one. The teams were pitted against each other in a full day of overclocked benchmarking with different applications, followed by a round of freestyle record attempts.
|Registration:||7:30 AM — 8:00 AM|
|System Setup:||8:00 AM — 11:00 AM|
|Opening Ceremony:||11:00 AM — 11:20 AM|
|Final Presentation and 10 minute break:||11:20 AM — 11:30 AM|
|Round One — Everest (Bandwidth)||11:30 AM — 11:55 AM|
|Round Two — Super PI:||12:05 PM — 12:30 PM|
|Round Three — 3DMark 2001:||12:40 PM — 2:00 PM|
|Round Four — 3DMark 2006:||2:10 PM — 3:30 PM|
|Award Ceremony One:||3:30 PM — 4:00 PM|
|Setup for Freestyle:||4:00 PM — 5:30 PM|
|Freestyle Contest:||5:30 PM — 7:30 PM|
|Award Ceremony Two:||7:30 PM — 8:00 PM|
The venue for the event was Taipei’s Grand Hyatt hotel, which is literally right across the street from the famous Taipei 101 tower—the world’s tallest completed skyscraper. The event was held in the largest conference room the hotel offered.
Given the number of countries involved, I was particularly interested in getting a feel for how popular overclocking and technology in general are around the world. Considering the enthusiasm teams from Asian countries showed at the event, it certainly seems like North America is a little more timid in this regard.
Tension was nowhere to be found before the event as participants mingled in the foyer outside the conference room, reminiscing about the days of the Celeron 533A or talking about the records they were planning on breaking during the competition.
Much to my surprise, almost all the contestants already knew each other, either from previous events similar to this one or online forums like Xtreme Systems. Xtreme Systems owner Fugger was at the event competing on team USA-2 with fellow forum favorite Vapor.
The necessary logistics were handled very smoothly by Gigabyte, which had each workstation decked out with top-of-the-line enthusiast hardware and a few essentials, including paper towels and a thermal mug for holding liquid nitrogen.
The CPUs were hand-picked by judges and fellow enthusiast Hicookie and pre-tested for performance to ensure that no team would have a significant advantage or disadvantage due to the luck of the draw.
Where possible, the system hardware was Gigabyte’s own, including the motherboard, Radeon graphics cards, and even a 1200-watt modular power supply.
The memory, however, was provided by Corsair. Western Digital offered up a speedy VelociRaptor hard drive for each team, as well.
ViewSonic graciously donated a multitude of 22″ 1680×1050 monitors to round out the competition platforms.
Liquid nitrogen, more affectionately called LN2 by most attendees, was the star of the show for providing temperatures as low as 160 degrees Celsius below zero before rapidly boiling at -149°C. The handling of the LN2 was split between members of Gigabyte’s staff, a few qualified technicians, and generally one member at a time from each team.
A stage complete with spotlights and an audio system fit for a small stadium set the mood for the opening show, with cheerleaders shouting “Go.. O.. C.. GO OC” and laser lights scurrying around the expansive conference room.
You can’t get record-setting benchmark scores with basic gear, so contestants brought along plenty of their own goodies.
The first task teams undertook was to prepare their fresh hardware for the harsh conditions of below-freezing temperatures by waterproofing any piece of electrically conductive material with whatever means available.
A huge range of methods were employed, including spray-on aerosols, self-adhesive rubber foams, Vaseline and silicone grease, nail polish, and even putties and poster tack.
Care had to be taken not to disrupt components that have to remain clear of insulation for the system to operate, and some teams came better prepared than others for a quicker dressing of their hardware. Among my favorites was team USA-1’s technique of bringing a pre-fitted foam sheet with cutouts for each large motherboard component.
Putty seemed to take the most time to apply, and while the pre-fitted foam sheet was generally the fastest method used, most teams did alright with one of the grease solutions or a combination of techniques.
The Radeons were given just as much attention, with similar materials used to shield the area around the GPU socket while keeping the core and memory clear for contact with the graphics cooler.
In a move that I found safe but smart, some teams assembled their system as quickly as possible since time was such an issue, checking the hardware for any duds. Gigabyte allowed each team a single swap for each component in their system.
If there was a key component element to competition besides the hardware provided, it was the heatsink tube used to cool the processor. Every team used a tower (normally copper) to hold liquid nitrogen and transfer heat away from the processor to the frigid metal.
To prepare chips for the insane heatsinks, insulation had to be carefully applied to maximize core contact while ensuring that condensation didn’t short the electronics.
You can’t hold a few pounds of copper to a motherboard safely with a standard retention mechanism, so each tower came with its own customized mounting hardware, ranging from six-inch bolts to woodworking bar clamps.
Teams weren’t as picky as you might expect when it came to thermal interface compounds. When asked why they weren’t using a well-known product such as Arctic Silver, most said that at these temperatures, the one-to-two degree advantage provided by more exotic compounds isn’t worth the additional cost considering just how much compound teams go through in an average overclocking session. That makes sense to me.
It’s a brute-force approach to cooling, but one can’t deny the results that can be obtained by almost directly exposing a processor to an extremely cool liquid.
As for cooling the north bridge chip, I was surprised to see that most teams left the stock cooler in place and just directed huge amounts of air at it. When asked about this, most contestants informed me that the cooling effect of liquid nitrogen on the CPU tends to keep the north bridge relatively cool as well. That wasn’t good enough for some teams, though.
While the majority of teams left the stock water block on the motherboard and just directed a fan or two at it, a couple put the stock cooler to use and hooked up a dedicated water cooling system. A few even removed the barbs completely and positioned their own liquid nitrogen-ready towers over the north bridge chip.
Voiding your warranty
In such an extreme environment, there isn’t any questioning potentially dangerous enhancements like voltage modding—it’s a given part of the competition.
Graphics card makers keep power regulation on a tight leash, but if you can create a short in the right place on the circuit board and insert a relatively low impedance resistor in the new circuit path, you can safely deliver higher voltages to the GPU, allowing for higher clock speeds.
Each team went about voltage modding in similar ways; a soldering iron, some variable resistors and wires, and steady hands were pretty much all that was needed. More interesting was how the participants kept reference materials detailing the modifications.
All the teams used the same hack. Once one person in the community finds the proper points to short and valid loads to introduce to the new circuit, a high-resolution image of the surgery is all that is needed by anyone who wants to duplicate the mod.
Part of the process involved checking the properties of the circuit before, during, and after the mod to ensure nothing went wrong. For this task everyone had their trusty digital multimeters ready, with some teams leaving probes permanently attached to the key points to monitor levels while everything was running.
What really stood out about the modding was the fact that this was all taking place under the pressure of an extreme time crunch. Any time lost to a soldering snafu could mean the forfeit of a score in an event if the system could not be brought up to speed in time. It should come as no surprise, then, that the more experienced modders had a real advantage. They were practically professionals at working with the minute details of a voltage mod.
Some teams also employed a “droop” motherboard voltage hack.
This warranty-voiding procedure let teams more finely control the stability of the voltage provided to the CPU. Some teams deemed this mod “only marginally beneficial,” but others declared it “essential” at the extreme clock speeds desired.
The motherboard mod took roughly as much skill as the graphics card one, but was a little trickier due to the fact that the solder points were on opposite sides of the circuit board.
My favorite part about these mods was the attention to detail some people were able to put into them in the short amount of time available. Cables were routed neatly to switches that could turn the mods on and off as demanded by some of the teams.
To ensure stability, other teams ran variable resistors that could be tuned once the machine was on.
A couple teams were even watching current levels through ammeters, allowing them to cut power if too much current flow threatened to damage system components.
Trials and tribulation
When it came time for the official event to begin, the stage lit up again and the announcers detailed how the different stages would play out.
There was also some time for executives from companies who had donated equipment to briefly talk about their work and support for events such as this one.
Of course, the competitors couldn’t be bothered to pay too much attention as they hurried to get equipment prepared for the first round of benchmarking.
As if system assembly and modification weren’t difficult enough, time constraints forced some teams to simplify their approaches in order to post scores in the first round.
By this time, most teams had setups at least ready to start, and the liquid nitrogen was flowing freely. Several members of Gigabyte’s staff worked almost non-stop filling insulated jugs with LN2 directly from the tanks.
Although the primary goal of most contestants was to get the CPU as cold as possible, often times that pursuit caused an entirely different problem.
Due to what’s known as the “cold bug,” most CPUs will simply freeze up and stop working entirely if their temperatures plunge too low.
The cold bug is especially tricky because you don’t know what temperature a processor will bottom out at until you hit it. Most teams that ran into this problem encountered it more than once when trying to pin down whether their systems were freezing because the processor wasn’t cold enough or because it was too cold.
Some teams spent a lot more time prepping their hardware before diving into the software side of things.
For those teams who weren’t as far along, the time was running out to get their systems up and running.
The pre-fitted foam team USA-1 brought wasn’t enough to counteract the time it took to perform all their intensive mods.
It would have been easy to slow down at this point, considering the discouraging feeling that comes from not being sure if you’ll have everything ready on time, but an abundance of Red Bull helped to keep everyone going.
The P45 event — rounds one and two
Gigabyte had intended to enforce a clear-cut line between the time allocated for system setup and preparation for the first round of testing, but that line quickly blurred as teams scrambled to get their systems to boot.
The setup for recording high scores and keeping everything updated was pretty slick. Judges ran back and forth from the main table to the teams who had just posted their highest score, signaled by a raised hand from one of the participants. Once a screenshot was taken, the officials would record the score on a master Excel spreadsheet that was displayed for all in the conference room to see.
In addition to the scores, a countdown timer kept teams on edge by providing a constant reminder of how much time was left until the end of the round.
Problems including unstable operating system settings and driver issues plagued some contestants, who struggled even to get a stable system at conservative clock speeds.
Some folks discovered much sooner than others that the Radeon 4870 drivers didn’t like it if the graphics card’s fan was disconnected. Everything would start up as usual, but a hard blue screen of death awaited most once Windows loaded driver files.
Getting a high bandwidth/read score in Everest was the first competition, and although team USA-1 actually had everything almost ready to go by the time some teams were turning in their scores, they still faced stability issues due to the GPU fan driver bug.
Team Hong Kong won the first round with 11,578 points in the Everest benchmark. Turkey placed second, only eight points behind, with China-1 in third with a score of 11,301, Sweden-1 in fourth with 10,974, and China-2 in fifth with 10,611.
For the second round, contestants had to compute Pi with SuperPI configured with the 1M setting.
At these obscene speeds—most well above five gigahertz—calculating Pi to a million places only took between eight and 10 seconds.
Although Indonesia, Hong Kong, and a few others got great results right away, other teams took time to perfect their setups and started posting scores with only minutes left in the round.
Sweden-2 pushed their CPU to an amazing 5.6GHz, capturing the best time of 8.172 seconds. Singapore scored second with 8.187s, Russia third with 8.188s, Indonesia fourth with 8.250s, and France fifth with 8.266s.
The P45 event — rounds three and four
You might think teams would get a break before starting a completely different set of benchmarks, but the 3DMark 2001 test was up next, and teams only had an hour to prepare for it, just like in previous rounds.
Without a real break yet, it was a wonder more teams weren’t slowing down by the third round, especially considering that most had traveled across the world for the event.
A rotation of live DJs kept spirits up, along with the constant commentary from the MCs and prodding from journalists who wanted some more details on what each team was doing.
Food was provided, but most contestants didn’t have time to socialize while they ate. Most brought refreshments back to their workstations, where it sat next to the same substances used to keep their systems running.
I didn’t hear any reports of someone drinking liquid nitrogen by accident, but plenty of teams did the opposite, using their re-shaped Red Bull cans or water bottles as funnels to help pour coolant.
Team USA-1’s ingenuity started to pay off in the second round, as they got some frosty temperatures and outstanding clock speeds from their Radeons.
Most teams had heavy insulation on the back of their GPUs, discouraging condensation and frost buildup. However, for the teams who simply used some grease or thin clearcoat, this wasn’t all that uncommon of a sight.
3DMark success came down to not only who had the highest overclock, but also who was able to tweak his system effectively.
The top place exchanged hands many times during the hour of testing for rounds three and four, but USA-1, Sweden-2, and Russia pulled ahead by a decent margin in the last five minutes.
In the end, an extremely good CPU overclock and a moderate GPU overclock reigned supreme, as team USA-1 captured first place with 103,303 3DMarks. Sweden-2 wasn’t far behind with 102,445 3DMarks, with Russia trailing in third with a score of 101,639 and Italy in fourth with 101,028 3DMarks.
3DMark 2006 benefited from largely the same hardware tweaks as 3DMark 2001, but knowledge of specific driver and registry settings helped some teams get ahead. Higher clock speeds are only one part of the equation in 3DMark; one can improve scores by tweaking registry settings to bring the level of detail in the benchmark way down.
Team USA-2 found this out the hard way, and consistently came up short of the scores posted by Sweden-2, despite that team running a lower GPU clock. There was a bit of discussion on the subject, and eventually it was agreed that Sweden’s higher CPU clock speed and reduced level of detail did more for the 3DMark score than USA-2’s higher GPU clock.
In round four, team Sweden-2 took top honors with 29,630 points in 3DMark 2006. USA-2’s 29,463 3DMarks was good enough for second place, while France scored bronze with 28,481 3DMarks, followed by Indonesia in fourth with 28,231.
The freestyle event
After an awards ceremony for the winners of the the P45 event, the participants finally had a decent chunk of time to relax and re-tool their setups for the next stage of the competition.
For the freestyle event, contestants were handed a different task: set as many new world records as possible. There were a few guidelines to follow. The records to be beaten mostly had to come from programs officially tracked by hwbot.org, limiting the playing field to SuperPI, PiFast, wPrime, CPU-Z validation, PCMark 2004, 2005 and Vantage, 3DMark 2001, 2003, 2005, 2006, and Vantage, and AquaMark 3. Otherwise, teams were free to configure their systems as they saw fit, just as long as each used a Gigabyte motherboard with an Intel chipset, an Intel processor, and a graphics card powered by an ATI GPU.
For most teams, this was a chance to really bring out their best gear and combine it with a healthy supply of LN2 to get some mind-blowing numbers.
Along with different processors and graphics cards, teams often used more elaborately modified motherboards, some complete with auxiliary circuit boards laden with switches and resistors.
Overall, there was a lot more personality in the freestyle round, as teams chased records they were most passionate about with whatever they had on hand.
Team Japan attempted to set a new 3DMark record with a wildly overclocked quad-core processor (they were up to 5.6GHz when I checked on them last) and two Radeon HD 4870 X2 cards running in CrossFire. Sadly, even with their ridiculously cold -150°C processor temperature, the team was unable to breach the current 3DMark 2006 record of ~33k points.
Since folks from around the world are constantly battling for the highest 3DMark scores, it’s no surprise some teams went for less popular records.
Team Vietnam combined a ridiculously fast front-side bus with an E6550 and went down the line, setting records in CPU-Z, Super Pi, and wPrime before moving on to try the same with an E8200.
Some teams simply tried setups too ambitious, whether they were going for unheard-of clock speeds with the latest processors or new records in 3D benchmarks. In arguably the most ambitious of all these attempts, team Singapore wowed spectators with a dual Radeon HD 4870 X2 setup that featured liquid nitrogen cooling for each GPU.
As impressive as it was, team Singapore had an awfully difficult time with their ridiculous setup. They eventually got it stable enough for a complete 3DMark run, but their scores weren’t any higher than what Japan was getting with air-cooled GPUs.
Insane setups were the whole point of the contest, really, so who could fault teams for going all out and combining the fastest components available to set new records? Team USA-2 took this idea and ran with it, incorporating a RAID 0 array of two Intel X25-M SSDs into an already blazing fast QX9770 rig equipped with a Radeon HD 4870 X2
Considering the amazing performance of Intel’s new solid-state drive, it was no wonder a system equipped with two of them burned through system benchmarks like a Hummer limo through gasoline. The combination proved powerful enough to set not only a new record for the QX9770, but also a record overall score for PCMark04.
Winners and conclusion
The main event, officially titled the Battle for the Best P45 OC, combined rankings from various rounds. For the first two rounds, six points were given for first place, five for second, and so on, all the way down to two points for fifth place. For the 3DMark rounds, first place netted 10 points, second garnered eight, third six, and fourth, er, four points. With this weighting, team USA-1 collected enough points to tie team Russia with 10 points each.
According to the rules, any ties would go to the team with the higher 3DMark06 score. Russia’s 27,945 3DMarks were just barely enough to give them 2nd place overall ahead Team USA-1’s score of 27,847.
There can only be one winner, though, and that glory was well earned by team Sweden-2, which placed first in round two, second in round three, and first again in round four.
Truly the “Swedes” (remember, SF3D is actually Finnish) boasted the most well-rounded overclocking and benchmarking skills, considering their impressive 26-point total score.
Gigabyte Vice President of Marketing and Sales in the Americas, Tony Liao, gave the champions their check and congratulated them personally.
The first awards ceremony was quite the event on its own, complete with dramatic light shows for each winning team and an enthusiastic head judge exclaiming the achievements of the competitors.
As for the freestyle event, team USA-2 took the win for setting not only four new category records: two each in PCMark04 and PCMark Vantage with first a QX9770 and then an E8600, but also a new overall record score in PCMark04.
Fugger’s decision to bring a couple of scorching-fast Intel SSDs and use them in a complete system benchmark like PCMark was smart, and it was good he did considering that team Vietnam set four new category records themselves with two lesser-known processors. This left the Vietnamese team in a very close second place.
Despite exhaustion, everyone seemed pretty enthusiastic at the end of the event. With any luck, the Gigabyte Open Overclocking Championship will become an annual event that draws more attention next year.