Compiling

This benchmark (compiling Ogre 3D sources in Microsoft Visual Studio 2008) is well-optimized for multi-threading, but actually demonstrates quite humble gains.

Java

This is another test nicely optimized for multi-threading. And the boost is already 2.5 times higher than in compiling. Seems sort of random again. But according to stats we've already collected for this benchmark, Sandy Bridge should be most likely faster.

Web browsing

Numbers are rather average for this review.

Audio encoding

Gains range from 12% to 16% (which is humble for this review), although we use a wide range of codecs, proprietory and open-source, all from different developers. Perhaps, the thing is that audio encoding benchmark is one of the most optimized for multi-threading in our test method? Then it means that Sandy Bridge offers considerable gains only in single or few-threaded software.

Video encoding

These results seem to be an indirect proof of the assumption above.

Games

And again it's a total mess without any traceable trends. Still, you can see that Sandy Bridge doesn't prodive any gains in GTA IV. The entire gaming test has demonstrated the lowest performance boost so far.

Conclusions

Well, we didn't hope this test would introduce some incredible revelations. But don't worry, a much more detailed review of the Sandy Bridge microarchitecture is in the works. It will offer much new and interesting information as well as results of low-level architecture tests. As for now, let's formulate a few more or less reasonable hypotheses based on today's test results.

1. The Sandy Bridge architecture has been overhauled. We can't say how radically, but it's much more efficient. It offers 14% boost on average, L3 cache being smaller by a quarter, and other things being equal.
2. Even knowing more details on the new core architecture hardly allows making valid assumptions as to how the performance of specific applications will change.
3. Even though we tested the processors in a multitude of benchmarks, there wasn't a situation when Sandy Bridge was slower than Lynnfield, core clock rates being equal.
4. With much caution we can say that in our tests Sandy Bridge more often demonstrated higher gains (over 20%) in tests poorly optimized for multi-threading.
5. Maybe, just maybe, the previous assumption is related to the smaller L3 cache. So, theoretically, the efficiency of the new core in handling multiple threads may be bottlenecked by such cache size.

The general conclusion is a bit banal: Sandy Bridge is much better than the previous core. Not as much as Core 2 (Conroe) was better than Pentium 4, but it's still a larger step forward in terms of architecture than Nehalem compared with the last incarnations of the aforementioned Core 2.

Write a comment below. No registration needed!