Feature: New Chip May Boost Particle Physics Grid Power
The steady rise of the speed, or clock-frequency, of computer processors has resulted in design challenges for chip-makers in recent years. For example, the rising heat generated by multi-gigahertz processors - said by some to approach the heat produced in a nuclear reactor - leads to difficulty in cooling chips efficiently.
Placing first two, then four, processor cores on the same chip allows several applications to run in parallel with big boosts in overall performance, and lowers cooling bills for computer centers. PCs with multicore chips, such as the Intel quadcore PC launched in Europe last November at CERN, are especially attractive to the grid computing centers affiliated with the high-energy physics community, whose applications require large amount of parallel computing.
“Four completely independent computing jobs can run on the same chip,” says CERN’s Sverre Jarp. “It’s like reserving a table for four in a restaurant, and getting four chefs, one for each person. Multicore PCs are good for high-throughput, parallel computing applications.”
Quad-core PCs are also attractive to scientific communities that face limited budgets, as they are cheaper than buying four individual PCs. The same amount of memory must be purchased in both instances, however. There are also drawbacks to concentrating so much processing power in one place.
“If your quad-core PC fails, all four chefs disappear from your kitchen,” adds Jarp. “A bank might prefer to lower risk by using only single-core PCs, a more expensive option. In the HEP computing community we would do exactly the opposite.”
The HEP community, while it buys large numbers of PCs, isn’t enough to sustain an entire industry, however. Jarp and others are waiting to see if the rest of the computing and scientific community, and the public, will rally around quad-core PCs.
“In the past, HEP took a giant leap forward thanks to commodity computing,” says Jarp. “But will Mom and Dad ask for quad-core? We would like the chip-makers to continue this trend of splitting into more cores, but if the rest of the world doesn’t agree there’s no incentive for them to go farther.”
CERN’s openlab program received an early version of the chips for testing. The program, which has Intel as one of its partners, used its well-defined methods to evaluate the chip’s performance and validate it with HEP benchmarks. The testing included a demonstration of how quad-core processors can speed up the rate of a typical scientific calculation involving a parallelized version of the program ROOT, widely used in the HEP community, by nearly a factor of four over conventional single-core processors. A parallel version of the ROOT program was used to split a data set into four pieces and calculate them in parallel on the quadcore chip.
- Katie Yurkewicz, iSGTW Editor-in-Chief and Francois Grey, CERN