What do super computers and overclockers have in common

What do super computers and overclockers have in common?

• By Sebastian Anthony on July 19, 2011 at 11:19 am
• 1 Comment

Share This Article

If you’ve ever overclocked a CPU or GPU you’ll know that adequate cooling is the key to success. Computer enthusiasts might not know exactly why a water cooling kit allows for far greater overclocks — but that’s beyond the point: better cooling equates to faster speeds — that’s all we need to know. Initially, overclockers had to create their own cooling setups with MacGyveresque oversized fans and fish pumps and duct tape, but processors are now so powerful that heat pipes and fancy heatsinks have become commodities that can be found in consumer-grade gear. Believe it or not, though, there’s even a commercial use for watercooling: supercomputers.

Fundamentally, cooler chips run faster. But why exactly? For a start, the resistance in transistors increases with temperature, thus slowing down their switching speed. Metal interconnects suffer the same fate: as copper gets hotter, the resistance increases, reducing the speed and efficiency of the system. The effect of heat is so pronounced that a temperature of 125C can slow down a processor’s frequency by up to 14%. Ideally we want to keep processors at 0C or even lower — but realistically, if we can keep a processor below 40C, very significant frequency increases are possible.

While this boost in frequency is what interests overclockers, supercomputer manufacturers much more interested in the second, highly-desirable effect of increased cooling: lower power consumption. As the individual components in silicon chips get smaller, there are three kinds of power leakage that increase — junction, gate, and subthreshold — but by cooling a processor down, these leakages dry up. For the most part, junction and gate leakage are inconsequential, butsubthreshold leakage is a major problem that exponentially increases as junctions get hotter. Basically, when a transistor is below its threshold switching voltage, current still flows between the source and drain of the transistor. In an ideal world there would be no leakage, but due to their small size and their ever-decreasing supply voltages, some leakage is inevitable.

Now this is where it gets interesting. Because it’s sensitive information, CPU manufacturers very rarely publish the actual leakage figures for their processors — but last year, Fujitusu actually published a paper detailing the exact power and leakage characteristics of its SPARC64 VIIIfx processor. This is the same processor that powers the 8-petaflop Riken K supercomputer, which is currently the most powerful computer in the world (but not if Cray has anything to say about it). The K computer uses 68,544 8-core chips, each with a thermal displacement of 58 Watts. Instead of air cooling, which would result in core temperatures of 85C, water cooling is used to bring the temperature down to 30C. Fujitsu estimates that by simply cooling these chips down, each chip uses 7W less — or, in other words, water cooling results in a power saving of 12%. Multiplying 7W by 68,544 comes to 479,808W — almost half a megawatt of power saved.

The ultimate question is whether it’s worth adding the complexity of water cooling — and once you factor in water pumps, there’s no doubt that the total power consumption might be higher than the air-cooled equivalent. Still, lower temperatures definitely increase reliability (IBM has been water cooling its supercomputers for decades) — and if you can reduce the number of processors by 10% and still achieve the same peak performance, the total cost of the computer can be significantly reduced.

For much more information about the interaction between performance, power efficiency, temperature, and cooling, read Real World Technologies’ story — or read about the K computer

What do super computers and overclockers have in common

What do super computers and overclockers have in common?

• By Sebastian Anthony on July 19, 2011 at 11:19 am
• 1 Comment

Share This Article

If you’ve ever overclocked a CPU or GPU you’ll know that adequate cooling is the key to success. Computer enthusiasts might not know exactly why a water cooling kit allows for far greater overclocks — but that’s beyond the point: better cooling equates to faster speeds — that’s all we need to know. Initially, overclockers had to create their own cooling setups with MacGyveresque oversized fans and fish pumps and duct tape, but processors are now so powerful that heat pipes and fancy heatsinks have become commodities that can be found in consumer-grade gear. Believe it or not, though, there’s even a commercial use for watercooling: supercomputers.

Fundamentally, cooler chips run faster. But why exactly? For a start, the resistance in transistors increases with temperature, thus slowing down their switching speed. Metal interconnects suffer the same fate: as copper gets hotter, the resistance increases, reducing the speed and efficiency of the system. The effect of heat is so pronounced that a temperature of 125C can slow down a processor’s frequency by up to 14%. Ideally we want to keep processors at 0C or even lower — but realistically, if we can keep a processor below 40C, very significant frequency increases are possible.

While this boost in frequency is what interests overclockers, supercomputer manufacturers much more interested in the second, highly-desirable effect of increased cooling: lower power consumption. As the individual components in silicon chips get smaller, there are three kinds of power leakage that increase — junction, gate, and subthreshold — but by cooling a processor down, these leakages dry up. For the most part, junction and gate leakage are inconsequential, butsubthreshold leakage is a major problem that exponentially increases as junctions get hotter. Basically, when a transistor is below its threshold switching voltage, current still flows between the source and drain of the transistor. In an ideal world there would be no leakage, but due to their small size and their ever-decreasing supply voltages, some leakage is inevitable.

Now this is where it gets interesting. Because it’s sensitive information, CPU manufacturers very rarely publish the actual leakage figures for their processors — but last year, Fujitusu actually published a paper detailing the exact power and leakage characteristics of its SPARC64 VIIIfx processor. This is the same processor that powers the 8-petaflop Riken K supercomputer, which is currently the most powerful computer in the world (but not if Cray has anything to say about it). The K computer uses 68,544 8-core chips, each with a thermal displacement of 58 Watts. Instead of air cooling, which would result in core temperatures of 85C, water cooling is used to bring the temperature down to 30C. Fujitsu estimates that by simply cooling these chips down, each chip uses 7W less — or, in other words, water cooling results in a power saving of 12%. Multiplying 7W by 68,544 comes to 479,808W — almost half a megawatt of power saved.

The ultimate question is whether it’s worth adding the complexity of water cooling — and once you factor in water pumps, there’s no doubt that the total power consumption might be higher than the air-cooled equivalent. Still, lower temperatures definitely increase reliability (IBM has been water cooling its supercomputers for decades) — and if you can reduce the number of processors by 10% and still achieve the same peak performance, the total cost of the computer can be significantly reduced.

For much more information about the interaction between performance, power efficiency, temperature, and cooling, read Real World Technologies’ story — or read about the K computer

Hai ! friend`s

coming april `18th university exam start.let we prepare for exam.

Wish`s

Wish You Happy

New Year