14 DAY TRIAL // According to Gina Longoria, director of the product management and workstation division at AMD, the future's mainstream servers may be built on a combination of GPUs and CPUs, considering that applications can benefit greatly from the thousands of GPU cores. Such a combination would permit such servers to run parallel applications. Some servers already use this principle and, reportedly, AMD will more closely focus on this market starting with 2012.
One example of installment based on a combination of graphics and central processing units is the HyperPower Cluster from Appro. This system is designed in such a way so as to break up and simultaneously execute a myriad of threads and tasks across servers with separate NVIDIA GPU cores and Intel CPU cores. Another example would be the Tsubane GPU-based supercomputer from the Tokyo Institute of Technology. With 30,000 processing cores, this conglomerate can reach speeds of 77.48 teraflops.
AMD hopes to expand the use of such combinations from supercomputers to servers, as GPUs are supposedly better at mathematical and scientific applications than CPUs. Of course, computing using GPUs will remain a niche market over the next two years, so AMD will mostly focus on the multi-core CPUs, according to Longoria. Still, later on, “heterogenous computing” based on graphics processing units might become more widespread. At that point, AMD might choose to focus less on the CPU cores and pay more attention to applications such as video, imaging and face recognition, even making them run more smoothly on servers running GPUs.
“As GPU becomes more relevant, that's a better way of getting performance than [CPU] cores,” Longoria said.
Still, Longoria did add that GPU computing might not always be suited, as such installments only performed brilliantly if applications themselves were written in parallel. Nevertheless, widespread GPU adoption in servers remains a strong possibility. Of course, Advanced Micro Devices is not the only one with this opinion. Dan Olds, principal analyst at Gabriel Consulting Group, also said that GPUs were the means by which server performance could surge the fastest, although he thinks that AMD can, and should, do more for this market.
“I'm glad they are addressing the market, but perhaps they should push ahead and develop the market more,” Olds said. He added that AMD had, so far, been more of a spectator, while NVIDIA aggressively pushed software and hardware for heterogeneous computing and, thus, already had a head start.