bit-tech.net

IDF Spring 2007: Benchmarking Penryn

If you're not yet up to speed with what's happening in the world of Intel CPUs then firstly read Tim's article on the technological breakthroughs of 45nm. Penryn isn't just a die shrink, which usually means cooler processors and higher overclocks, Intel has also included a lot more optimisations as I discussed previously.

Up until now, we've had no quantitative results as to how much better the next generation of CPUs in the Core 2 Duo refresh will be. It’s all well and good telling you about the specifics of the improvements like Radix-16 this or Super Shuffle that, which come with the process revision of 45nm, but how does that translate into real numbers?

Today bit-tech had some time to sit down with Intel and discuss the specifics of Penryn 45nm processors and how it’ll actually benefit the end user.

Whilst we weren’t allowed to use the systems ourselves, the testing done was mostly real world applications rather than canned benchmarks and Intel was upfront with each of the system specifications.

Since the 45nm process is still immature, these CPUs aren't production grade, but we're told that they should offer the exact same performance to the retail processors once they're released. However, we won’t get samples to put through our own testing labs for a couple of months yet because the 45nm manufacturing process needs perfected in order to get yields high.

Penryn Benchmarking 1

Penryn Benchmarking 2
As you can see, there are obvious improvements in the new 45nm Penryn processor. We begged, pleaded and argued that the clock difference between CPUs could overwhelm any noticeable changes. As this was a technology demonstration they should either over/under clock one of the CPUs in order to just show off the difference of the new process. The Intel engineers' argument was that this is a CPU you will buy and will also have its own TDP so overclocking will exceed it, etc. To cut a long story short the guys at Intel wouldn’t budge and that remained their stubborn final position.

They did give us approximate figure of a (liberal) 40% performance increase of a 45nm Penryn CPU over a standard 65nm Core 2 CPU, with just 12% of the performance increment associated to the clock speed increase. Then the rest was split between the larger cache and optimisations at 14% each. As soon as bit-tech gets a CPU in our labs, you can be sure we’ll strip it down and see if these numbers are correct.