The final test was meant to put performance and power consumption in perspective. Step one was to establish a baseline by looking at idle power. The Intel system was sitting at about 12.5 W, while the AMD machine was more power-hungry at 26 W. John was quick to remind us that this was not a final product, and that power should come down significantly with BIOS tweaks and other improvements (though we're honestly not sure how much more AMD can cut power so close to launch).
For this part of the presentation, AMD used a script that added a workload each time the return key was pressed. Workload number one was the Final Fantasy XIV benchmark, workload two an Excel spreadsheet with automated calculations, number three consisted of a 1080p video (Big Buck Bunny), and workload four had SPECviewperf rendering the wireframe model of a car. The final step was to use Windows 7’s Flip 3D to cycle through these programs.
Starting out, the Llano system was running pretty smoothly and drawing about 52 W in Final Fantasy XIV, while the Core i7 was closer to 61 W with frequent dips to 44 W. Why the dips? This happened whenever the integrated graphics engine was no longer keeping up with rendering, basically going idle, then getting back to work. Adding the Excel spreadsheet caused power consumption to climb to about 54 W on the Llano-based machine, pushing Sandy Bridge to between 52 and 68 W (with the pronounced dips, again). However, as soon as Excel was the top window, the frame rate in Final Fantasy tanked even more (because the system was having a hard time prioritizing the tasks, as John explained). Conversely, bringing the game benchmark to the front caused Excel to slow down. The same didn’t apply to Llano.
Playback of the 1080p video using CyberLink’s PowerDVD 10 did nothing to AMD's power consumption, which hung steady at around 54 W. Sandy Bridge, on the other hand, was now up to 70 W and stuttering in the two other workloads. The video remained smooth throughout, though.
Finally, SPECviewperf caused the Llano system to top out at 55.5 W, while Sandy Bridge plateaued at 70 W. However, both the game benchmark and SPECviewperf were basically reduced to slide shows on the Intel platform. AMD’s system wasn’t as fast as it had started out, but was still definitely smoother on all workloads.
Less of an objective yardstick, but definitely interesting as well, was the final twist AMD put on this part of the demo. With all four workloads chugging along, the script then told Windows 7 to use Flip 3D to cycle through application windows after a certain amount of work had been completed. The result here follows the pattern established above: during the time Intel's system managed to “flip” once, Llano did so three times, looking much smoother while doing it, too.
The point to take away here, according to AMD, is that benchmarks focusing on a specific component like the CPU or GPU, will be less relevant with parallel workloads and highly integrated processors like Llano. It won’t matter as much if one system is 20 percent faster in one specific benchmark running on its own because the user will be doing many things in parallel. And when it comes to juggling GPU and CPU workloads, an APU does better. Stepping back, we'd agree, with caveats. But the extent to which AMD's demo might map to the real-world is also fairly unrealistic, too. This is of course something we'll keep in mind for benchmarking moving forward, and if you want to share some of your own multi-tasked workloads with us in the comments section, we'd be glad to put together real-world examples of testing these parts running concurrent metrics.
Asked about the time frame for the introduction of the A-series parts currently known as Llano, John said they would be shipping between May and July, hinting very strongly that we should expect to see a launch sometime around June. Or maybe that was just our interpretation. At any rate, we were shown one more interesting demo from AMD to showcase the strengths of the Fusion architecture and specifically its Llano APUs.