New AMD Demo Shows Llano Outperforming Sandy Bridge

Much of the talk about AMD products has centered around Bulldozer of late, but Llano is on track for launch this year as well. AMD has released a new video pitting Llano against Intel's Sandy Bridge, with results that (un)surprisingly favor AMD's own solution.

According to Godfrey Cheng, AMD's director of Client Technology, Llano was designed on the following hypothesis:  "People are using more modern workloads like 3D graphics, HD video and Internet surfing in a much more prevalent manner...we all dabble with spreadsheets and word processing...but any modern x86 CPU-based PC can handle these workloads with ease.  But with these modern applications, the capacity to multitask, improve image quality and enhance power efficiency are much more important than raw x86 performance in determining how good a consumer’s experience is with a particular PC."

He writes further: "AMD’s “Llano” and Intel’s “Sandy Bridge” are roughly equal in size and transistor count...But...An analysis of the two components’ die area shows that AMD has invested much more heavily in graphics, parallel compute and video whereas Intel has invested much more of its silicon area in improving classic x86 performance."  That's where this video comes in.

We can't speak to whether or not the two systems were comparably configured and the representativeness of AMD's chosen workload is, as always, open to debate. One thing we did note, however, is an interesting difference in the execution speed of the Microsoft Excel benchmark. (This may be the first time the words "interesting" and "Excel benchmark" have ever been used in the same sentence without the inclusion of a "not" --Ed). We calculated our results by taking multiple screenshots per second and comparing both the displayed graphs and the distribution of color in the data columns linked to those graphs. This last is essential—some of the early graphs are very similar.

When the Excel benchmark 'starts' at 1:07, the Intel 2630QM is running roughly two seconds behind AMD's 3510MX. By the time the additional video workload is introduced at 1:18, that gap has shrunk to just a fraction of a second. This implies that we'd see Intel overtake AMD were the test left to run a few seconds longer and that Intel's investment in traditional x86 areas does pay off in those types of computation tests.

The really interesting bit, however, is that once the CyberLink video starts at 1:19, the Intel Excel window redisplays the same graph it displayed at 1:09 and does nothing thereafter. From 1:19 - 1:25, the Intel Excel window doesn't update, while the AMD system goes on drawing a set of graphs we've never seen before. When we see the Excel graphs again at 1:41, the Intel graph is the same as the AMD graph from 1:12. When AMD adds a 3D rendering component, the Intel side of the Excel graph freezes even longer (10-11 seconds).

Our Llano reference test platform from late last year.

It's entirely possible that these discrepancies were caused by Sandy Bridge-related driver issues, a specific application performance issue, or a bit of visual skullduggery from AMD (the video sequences could have been timed to show the worst side of Sandy Bridge.)  If none of these issues exist, it suggests that Sandy Bridge runs into trouble fairly quickly when handling CPU and GPU workloads simultaneously.

The question of which CPU is faster may come down to home much multithreading 'typical' users really need. We saw good reason to suspect that Sandy Bridge would've caught Llano in Excel when juggling the 3D demo—we'd be very curious to see this test repeated with 720P-1080P flash content standing in for CyberLink PowerDVD. Much of the  video users watch these days, including YouTube, Hulu, and most news sites, is played back using Flash inside a browser environment, not via disc-based media.