Earlier this month we previewed AMD's Llano architecture in a notebook environment. Now we have the desktop version with a 100 W TDP. How much additional performance can the company procure with a loftier thermal ceiling and higher clocks?
This one’s always easy to predict ahead of time. iTunes is a simple single-threaded app that scales similar to WinZip and Lame. That means the IPC-oriented improvements Intel made to Sandy Bridge (versus Nehalem) translate into big gains, while quad-core computing ends up meaning very little. The 3.4 GHz Phenom II X4 naturally finishes ahead of the 2.9 GHz A8-3850, then.
MainConcept is another story entirely. Well-optimized to exploit threading, the fast Phenom II clenches a first-place finish, and AMD’s 2.9 GHz A8-3850 falls into second place. The dual-core Core i3-2105 cannot keep up, and it takes a last-place spot.
The same holds true in HandBrake, and indeed you’ll find that, on average, most media-oriented applications that rely on processor performance favor cores over clock rate. The Phenom II X4 takes first place again, and A8-3850 comes in second. The Core i3’s two cores with Hyper-Threading take last for the second time in a row.
Ah, but we love mixing things up. Both MainConcept and HandBrake center on software-based encoders. They’re both limited to the performance of your processor. But Intel and AMD uniquely support hardware-accelerated transcoding features, too. Core i3’s HD Graphics 3000 engine includes Quick Sync, while A8-3850 includes UVD 3-based decoding acceleration and encode acceleration through the GPU’s 400 ALUs.
I’ve received emails questioning the use of Quick Sync as cheating in favor of Intel, since it employs a black box of sorts to alter the output quality. That's actually an interesting angle to explore. The fact of the matter is that any time you use hardware acceleration, including AMD’s, to parallelize a transcoding workload, the output file is going to deviate from the source. If you want to explore the implications of utilizing hardware to speed up your transcodes, check out Video Transcoding Examined: AMD, Intel, And Nvidia In-Depth, where we go into depth on this very issue. If quality is your main concern, disable acceleration features altogether and stick with software.
CyberLink’s MediaEspresso 6.5 not only allows us to test software-only, but it also supports Intel’s and AMD’s respective hardware-based capabilities.
Amazingly, Quick Sync is the only technology worth using. MediaEspresso correctly defaults to software mode when we fire up the 890GX and Llano-based configurations for the first time, as the benefit to using hardware-based decode and encode is minimal.
Higher-end discrete cards with more general-purpose compute power might fare better, but these lightweight integrated engines simply don’t have the displacement to drag race against Intel’s fixed-function implementation.
- Meet AMD’s Desktop Llano-Based Lineup
- Dual Graphics: How Does Is Perform?
- Dual Graphics: Not Always Your Best Bet
- Storage Performance
- Making Memory Performance Matter Again
- A Word On Overclocking Llano
- Test Setup And Benchmarks
- Benchmark Results: PCMark 7
- Benchmark Results: 3DMark Vantage
- Benchmark Results: Sandra 2011
- Benchmark Results: Metro 2033 (DirectX 10)
- Benchmark Results: Call of Duty: Modern Warfare 2 (DirectX 9)
- Benchmark Results: World Of Warcraft: Cataclysm (DirectX 9 And 11)
- Benchmark Results: Content Creation
- Benchmark Results: Productivity
- Benchmark Results: Media Encoding
- Power Consumption