Since we have VIA Nano results...

...so we might as well compare it with something else. Of course, Intel Atom should be the primary competitor, but we don't have its test results yet. What we have is the results of AMD Sempron 140. Well, why not? It's also single-core, of similar cache size and design, is based on the modern 3-way superscalar OoO K10 core with fully pipelined function units. However, AMD Sempron 140 has a much higher clock rate, 2.7GHz, but it's an approximate comparison, a simple extrapolation will do, if we assume that Sempron's performance scales down along with its clock rate. Since the latter isn't very high (compared to what the K10 core is capable of), and that extrapolation is about scaling down (so the 2-channel IMC shouldn't be much better than chipset's), the results should theoretically be close to real. Below is what we got.

Even if we forget about Nano's failure at 3ds max + V-Ray rendering, the virtual 1.6GHz Sempron is 55% faster. We can't compare efficiency per watt, but we can compare efficiency per transistor. Nano has 94M of those, while the K10-based Sempron has 234M. Let's not forget that the latter is actually a dual-core AMD Athlon II X2 with a disabled core (which sometimes can be unlocked). So if our approximate calculations are correct, that chip, if it had one physical core, would count about 130M transistors (including the 2-channel IMC that Nano doesn't have; we consider the complexity of IMC as minor). Meaning that 38% more transistors gives Sempron 55% more performance.

If these results are close to real, the efficiency per transistor only proves that VIA Nano should really be improved. But since there's nothing perfect in our world, we'll return to the performance of VIA Nano as soon as we test Intel Atom in the same conditions.

Afterword

Early in August there was news that FTC and Intel came to an agreement. FTC won't penalize Intel, even having proof of "incorrect" PCMark 2005 results. And Intel promises (yet again) to do the following.

• Inform software developers that its compilers distinguish Intel and non-Intel CPUs and do not always provide the optimal code for the latter. Compensate extra costs to any software developer that would like to recompile its software with a non-Intel compiler (given that Intel libraries which don't have to be changed, according to the agreement, can still be used with non-Intel compilers).
• Won't resort to bribery, threats and other stimuli to force PC makers not to buy competing processors.
• Will support the PCI Express bus for at least 6 years, chipsets and processors with built-in graphics included, so it's possible to connect GPUs from other manufacturers.
• Change license conditions so as to provide more freedom for cooperation and joint technological development to competitors.
• Prolong the license agreement on the release of x86-processors with VIA Technologies for another 5 years, until 2018.

For the win?

We thank Agner Fog for the provided information, motherboard and application.

Write a comment below. No registration needed!