The first tests of Sandy Bridge processors produced a good overall impression. But it didn't answer one interesting question: how did the pure architecture performance change? Although processors we tested had the same clock rates, the Turbo Boost technology — that works differently in Sandy Bridge and Lynnfield — naturally affected the results. So we decided to conduct another testing and try to compare architectures instead of specific products. To do this, we took one Sandy Bridge and one Lynnfield processor, leveled their specifications as much as was possible, and disabled Turbo Boost. Below is what we got.

As you can see, the processors were leveled as much as possible. What we couldn't change was, obviosly, L3 cache (of which Sandy Bridge has less, by the way) and the support for the AVX extension (but benchmarks we used hardly supported it anyway). And if you're wondering, we simply didn't have a Core i5-2600K on our hands when this review was made. That said, let's have a look at test results.

Tests

3D visualization

Note the truly fantastic gain in UGS NX 6. Unfortunately, having two 'black boxes' on hands at once (one is processor, the other is proprietary software) prevents us from making any assumptions that we could support by arguments. Still, logic and reason suggest that such incredible boost is hardly a global breakthrough of Intel's R&D. Most likely, they fixed some annoying drawback of Lynnfield while developing Sandy Bridge. But, once again, that's just our guess.

But scepticism aside, the results are still impressive, average gain being 26%. Simply put, to catch up with a 2.8 GHz Sandy Bridge processor, a Lynnfield CPU would've had to work at about 3.5 GHz.

3D rendering

What attracts attention here is how differently benchmarks react to architectures. The moderate gain in 3ds max (4%), a better one in Maya (13%), and more than impressive boost in Lightwave (22%). We'd like to assume at least something, but it's really hard to do without the knowledge of the internals of processors or proprietory software. So let's just observe and try to make some conclusions based on statistics we'll collect.

Scientific calculations

We've already mentioned UGS NX 6 (this suite is in two groups of tests), so let's focus on other benchmarks. And still there isn't a common trend. Maya, Pro/ENGINEER and Maple are mostly single-core tests, and the gain is moderate. SolidWorks, on the other hand, is too, but the boost is nearly 30%. Mathematica and MATLAB can handle multiple cores, but the gains are so different. This doesn't give an idea of what parts of Sandy Bridge have become faster than those of Lynnfield. And we also don't know which benchmark uses which instructions more often.

Image editing

It would seem that gains grow along with multi-threaded optimization, but then compare the results of ACDSee and Adobe Photoshop. It's just random, or at least seems to be, given what we know about Sandy Bridge (and Lynnfield, actually).

Data compression, decompression

Archivers demonstrate that sometimes Sandy Bridge doesn't offer any advantage at all (i.e. in 7-Zip).

Write a comment below. No registration needed!