It’s 4 a.m . You’ve already blown off your English midterm paper so that you can beat World of Warcraft, for the eight consecutive time, when all of a sudden the most dreaded thing that gamers fear the most happen — lag. Your computer freezes and the frustration breaks out.
You throw your laptop on the ground, and there goes your unfinished homework, as well as, of course, your $1000 machine.
The aforementioned drastic scene may soon be a thing of the past, due to the research of Huiyang Zhou, associate professor in electrical and computer engineering.
For the last year, Zhou and doctoral student Yi Yang have been working on collaborating the central processing unit and graphical processing unit in computer systems and models.
“Recently, chip manufactures have begun to form hybrid CPU/ GPU chips where both units are manufactured on the same platform but work almost exclusively,” Zhou said.
Zhou’s research takes what experts at Intel and Advanced Micro Devises have designed, central processing units and graphics processing units, and gives the hybrid chips a more practical usage.
“The only benefit of placing the GPU’s and CPU’s on the same chip is lowering the cost and of course creating a smaller form factor which can save space,” Zhou said. “We felt, however, that the proximity of the two chips could provide for numerous unique opportunities that are not yet in the market and go beyond just lowering cost.”
In their current prototype, Zhou and his team, supported by AMD, have the CPU prefetching data for the GPU so that computations are smoother. As a result, the GPU is much more powerful, making your computer faster.
The experiments have already shown that on a set of benchmarks, the proposed pre-execution of the CPU improves overall performance by up to 113 percent and 21.4 percent on average.
“Really intensive programs such as physics simulators and MATLAB are running better because the CPU is being used more effectively than ever before,” Yang said.
“Right now, we are only looking at having the CPU help the GPU perform more complex tasks, but our future goals are to have it the other way around,” said Zhou.
With the GPU as powerful as it is, if it fetched data and information for the CPU, the results would be faster than any processor available.
“The possibilities are truly endless. We feel we are just starting to crack into a new realm, and the future excites me,” Zhou said.
Zhou said his research team filed for an Idea Disclosure Form through N.C . State, but a deal has not been reached with AMD.
“They’re [AMD] ready to pay, but we’re not ready to accept yet,” Zhou said, smiling.
Brandon Walker, junior in computer science, sounded skeptical of the processor’s speed.
“Most of the time, the CPU is what’s slowing you down, so I feel having the CPU help the GPU is more of just a trade-off than anything,” Walker said.
Walker remained firm when he said he wouldn’t give up CPU for GPU , but admitted that Zhou’s research may be groundbreaking if it does relay the appropriate results.
“The benefits are just too great, with a 30 percent boost in performance, coupled with the lower cost and ergonomics design. I am really excited about where this research will go,” Zhou said. “And the best part is, everything has been flawless so far because the idea just makes sense, and thus it just works.”
