CPU Cache makes a difference

[EvilAlchemist]

I have not seen a post on this in a while and wanted to share some of my observations.

As a few of you know, I upgraded my FileServer CPU from a E6420 to an X3350.
(The E6420 found a nice home in my HTPC)

Now that I have a quad core, I am running:

• ATI Client
• Uni-Processor Client
• VM Linux (2 Cores)

I overclocked my E6420 to the same speed as my X3350 (2.72Ghz) and
E6420 = 4MB L2 Cache X3350 = 12M L2 Cache

I stated to notice that my Uni-Processor Client on the X3350 was running 10-12% faster on the same work unit.

Was there some other change in the 64nm to the 45nm switch that is causing this or is this just the cache difference?

I would like to hear other thoughts on this issue......

 

[sirmonkey1985]

yeah clock for clock the 45nm quad cores should be faster then the 65nm dual cores in every aspect.. even if you were only using 2 cores on the 45nm quad cores.. the cache does make a difference though and you can see it even in AMD cards.. theres a huge difference between my sempron 3000+ at 2.7ghz and my sempron 3100+ at 2.7ghz.. same processor only difference is the 3100 has 512k cache and the sempron 3000+ had 128.. yet took off almost 45 minutes on the same WCG projects and even in F@H took off just short of an hours worth of folding time.. so im sure theres a difference even between intel processors..

 

[capreppy]

Very interesting read. I had heard clock for clock the 45s were faster than the 65s. SSE4 can make a difference depending on the task at hand, in addition, to the higher L2 cache. I thought I saw/read that SSE4 can make a difference in F@H

 

[metallicafan]

Yeah there was a few new instructions and optimizations with the Penryn core. Here is a quick overview: http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3137

Clock for clock the Penryn should be a little faster in most workflows than the Conroe architecture.

 

[Zero82z]

There's typically a 5-15% performance increase going from a Conroe/Kentsfield to a Wolfdale/Yorkfield. This is due partially to the more efficient manufacturing process, a few tweaks made in the design of the cores, and in some cases, SSE4.1. Cache actually does not have that much of an effect in most tasks, especially considering that your new chip only has 1MB more per core compared to the E6420. I would wager that your performance increases are largely due to architectural improvements.

Also, it may have something to do with the fact that you're using a quad now. Although you are talking about the uniprocessor client, the fact is that now your OS has more CPU resources to balance things around in, and any background processes or other tasks that might be performed on it would incur less of a penalty due to the additional cores.