Penryns Hang On to Conroes’ Overclocking Torch
The E8400 I got looks to be a great overclocking subject, as was the E6300 which preceded it (details on the E6300 and test rig are in the previous article) and ran at 2.97Ghz instead of the stock 1.87. The E8400’s stock 3.0Ghz clock at 1333FSB was stretched to 3.71Ghz just by pushing the FSB to 1650 without any problems.
Alienware claims they are the only people who can get a 3 Gig chip to 4 Gigs stably, but droves of home builders are doing the same thing, so I was a little disappointed I couldn’t get the chip when I couldn’t get my computer to boot at a 1780 FSB setting. A more realistic 1750 booted, but lacked stability. Seeing no temperature issues (CPU cores were barely above 50 under load) I increased CPU, Northbridge, and Memory Voltage to 1.4, 1.5, and 1.925, respectively. This has gotten the rig through some burn in tests, a couple nights of running Prime95, plus a couple weeks of the heavy use a PC Hardware and Gaming writer puts a system through. CPU-Z screens from all four setups are below. As I become more convinced of the stability over time, I will start to wind down the voltages.
3.94Ghz is a pretty darn good clock speed for the E8400, and furthermore the issue may not be the CPU, but the DDR2-800Mhz memory balking at the FSB. Some people have had better luck hitting 4Ghz or more with DDR2-1066 sticks. Though the Penryns appear to overclock well in general, they don’t always overclock quite that well: people with nicer equipment than mine can find themselves requiring voltage bumps to hold 3.6Ghz. Overall you will probably get something pretty pleasing for your tuning efforts, but don’t sit there and throw electricity at your parts until you smell melting plastic just because your particular chip doesn’t happen to break any records.
The four sets of benchmarks will create some interesting comparisons: the fact that the overclocked E6300 was running at a very similar speed to the stock E8400 should allow us to see how much performance comes from the higher clocks as opposed to other advantages of the Wolfdale.
I use a lot of different benchmarking tools each running several different tests and apply scaling factors so that all the numbers are of a comparable size with traditional Drystone and Whetstone tests. Then I make a weighted average of the numbers, giving most of the weight to the Drystone and Whetstone. The result is a better representation of what a CPU can do because it represents many testing methods, but the scaling means the numbers can still be compared to MIPS and MFloPS. The weighting keeps unruly test results from fudging the more reliable numbers, and the averaging gets pages of data down to a few numbers. Note that I didn’t include frame rate or graphical benchmarks because of their relative reliance on the graphics card as opposed to the CPU.
The chart below shows that the upgrade to the E8400 results in a far from negligible difference. People running their CPUs at stock speed can expect large increases of over 60% in integer performance and over 80% in floating-point calculations. There is also a big increase for overclockers: over 30% MIPS and 50% MFloPS in this particular case. Again, the CPUs you get will overclock differently than those I bought.
The difference between the E6300 running at 2.97Ghz and the E8400 at its stock 3.0Ghz was obviously far smaller, but the newer CPU still represents a noteworthy increase in performance. Despite only having a 1% higher clock speed, the E8400 performed about 8% better overall, while using only 53 watts to the E6300’s 73. The lower Thermal Design Power of 45nm chips may not be useful to everyone, but people looking to trim power bills or for ways to build a small, quiet PC will appreciate it.
L2 Cache and memory access bandwidth were almost identical, in both cases accessing the L2 Cache is more than twice as fast as running all the way to main memory, but the E8400 has a 6MB cache versus the 2MB on the E6300. That may account for a good portion of the performance increase. The 45nm Penryn CPU also shows inter-core latencies only slightly more than a quarter of the older 65nm Conrow (44ns to 161ns).
So you get a big bump in clock speed and other advantages from stepping up (down?) to 45nm, but is it worth ditching your 65nm CPU? The next article concludes the review with a feature and price comparison of the E8400 against other Intel CPUs available for $100 to $300.
CPU-Z info from All Configurations and Benchmark Result Chart