Ivy Bridge Benchmarks Surface

As the release of Ivy Bridge grows closer, there has been much speculation regarding what sort of performance boost we will see over a similar Sandy Bridge processor. Intel gave us an overview of the new 22nm process at last year's Intel Developer Forum, and the information that was released indicated an expected performance boost of approx 10% for Ivy Bridge's CPU core and up to over 100% for the GPU. We have now seen some real benchmarks performed by independent testers.

Intel have adopted a Tick Tock regime for the release of new CPU architectures and new process technologies for 5 years now. Traditionally, on a Tick a new process is introduced which will give a small performance boost, and use significantly less power. A Tock introduces a new architecture, effectively new circuitry, using the now established manufacturing process.

This year, we are at a tick, and Ivy Bridge is largely a die shrink of Sandy Bridge on a smaller process. However, Intel have also improved the integrated graphic core, which could be seen as a Tock alongside the Tick of the main processor core. AnandTech have published the results of tests they have performed with an i7-3770K at stock settings. Additionally, a member on the xtremesystems.org forums has posted a couple of screenshots comparing a 2600K to a 3770K at 4.5 GHz.

For overall system performance, AnandTech carried out several tests with Sysmark 2007 and Sysmark 2012. In Sysmark 2012, the i7-3770K showed an overall performance boost of 7.5%. Sysmark 2007 indicated an overall performance boost of 10%. Across a selection of 8 CPU tasks, Ivy Bridge shows an improvement of between 5% and 15%, averaging out bang in the middle at 10%. Testing the integrated graphics performance across 13 tests, the GPU showed a boost between 20% and 50%, averaging out at 33.8% over Sandy Bridge. A single GPU compute test shows an improvement of around 225%, helped along by the DirectX 11 support introduced with this generation.

The report at xtremesystems ran SuperPi 1M and 32M tests on a 3770K and a 2600K at 4.5GHz. The version of SuperPi is outdated and only show results to the nearest second, so no perceivable difference was recorded between the two processors in the 1M test. In the 32M test, Ivy Bridge outperformed Sandy Bridge by 1.8%. SuperPi is a single threaded benchmark, so the difference between this value and the 10% average for the tests carried out by AnandTech can be explained by the improved IPCs.

The second report also recorded the voltage that Ivy Bridge needed to reach 4.8 GHz, and the core temp when fully loaded. When idling at 4.8 GHz, the voltage required was 0.94V, and under load it only jumped to 1.31V, which is lower than most Sandy Bridge chips at the same clock.

Unfortunately, the temperature recorded by CPUID Hardware Monitor indicated the CPU reached over 100C when running Linpack, and around 90C when running Prime 95. Another report from China recently said the same thing about the temps, and it looks like for people with standard cooling systems (air or water) it may not be able to easily hit the same clock as Sandy Bridge, but what it does do will use less electricity.

So far Ivy Bridge looks like it will be well suited to extreme cooling systems, such a phase change or liquid nitrogen. This would most likely be because Ivy Bridge drawing 80% of the power in a core that is half the size, meaning at any point in the core there is more heat to dissipate. And with an increased GPU size, if you use the integrated graphics you will be generating even more heat.

It is still not known when Ivy Bridge will be available to the retail market, but current reports suggest we may see desktop SKUs next month and mobile SKUs in June.