Four cores are good, but eight are better. 40% scalability may not be a perfect result, but it's still very nice.
Here is that perfect result: 85% boost. How? Easy, just take another look at the test method. The process of audio encoding itself is well parallelized (Lame MT, GOGO-no-coda), but this is rarely used in modern encoders. However, we use dBpoweramp which can simply run multiple encodings simultaneously. For this reason, the performance increase should be close to perfect.
From the angle of absolute values, audio encoding has long ceased to be indicative for individual users though, because even a single Core i7-860 can convert a hundred or more music albums per hour. Essentially, it already lets you convert CDs to files faster than they are released.
However, there's an alternative opinion. AllOfMP3.com had that interesting service, online encoding. Their server was used to store lossless music, but a user could download a track in the desired format and bitrate. That website is gone now, but the idea has a right to exist. It's kind of stupid to keep all your music converted into a lossy format like AAC 128/256 Kbps of the iTunes Music Store, for example. Doing this in the 21st century means losing to P2P networks. Paid services are just bound to be more convenient than any illegal ways of obtaining content. Well, online encoding does add more comfort. Of course, it requires adequare performance. A popular service might easily need to provide content to a thousand or more users at once. None of them will be pleased to wait for his or her turn.
We expected a better performance boost. What's the problem then? The problem is that we have had to remove XviD which liked the increase in threads, and the ancient Canopus ProCoder, incapable of loading even a dual-core processor, spoiled the whole picture. What if we remove it and run the tests again?
Now that's another pair of shoes: 1.5x performance boost. You can see that with modern codecs a couple of L5520 catches up with a single Core i7-975EE despite lower clock rates, while X5570 is without a rival among the processors we tested.
We believe no comments are needed. With this customized test method, dual-way machines outperform single-CPU rigs, if identical processors are used, and that's it. Buying them for such prices isn't wise. Instead of assembling a machine based on two lower-end or mid-end Xeons, you can easily buy a mid-end or high-end Core i7. It will be cheaper and simpler, will introduce fewer problems and less noise -- all that at similar performance. Even if you use lots of applications capable of loading more than four CPU cores efficiently, it doesn't mean you need to look for a dual-way workstation. Both competitors are going to release hexacore desktop processors in the near future. Intel's 6-core offering even supports 12 threads.
There are fields of use where multi-way machines will probably be needed forever. But those are too far from traditional standalone desktops. For a long time, mainstream users considered servers a dump for all sorts of files. Or a way to share a printer or fax. However, the client-server model has become an essential part of our lives, having introduced rendering servers, encoding servers and lots of other things. After all, the easiest way to load multiple processors is creating multiple identical tasks, each of which may even be purely serial. This includes terminal servers allowing to replace desktops by thin clients in certain cases. Thus, the capability of installing two (four, six, eight) processors will remain useful for all kinds of application servers forever. At least clients have been thinning while servers have been fattening for at least a couple of decades now. But speaking of desktops, the fame of SMP systems, at the height in mid-nineties, has waned completely.
Write a comment below. No registration needed!