I reviewed the article that compared different generations of CPUs at same clock speed (http://www.tomshardware.com/reviews/quad-core-cpu,2499-10.html), so I did the same using the THG’s Desktop CPU Charts 2010. I wanted to compare different CPUs and different architecture at same clock speed, but that was impossible without several motherboards and several CPUs so I used Cinebench 11.5 (single-threaded) benchmark already available at this site. I wanted to see how old my Core 2 Duo 8400 is. Since benchmark results are displayed in points, and I wanted to see how fast the core of each CPU is at same clock speed, I divided the clock speed (since benchmark is using only one core, and that’s the time when maximum turbo kicks in, I used the turbo frequency instead of the normal frequency if a CPU had Turbo) with points (I multiplied the points with 100, example; 0, 93 is just 93 and 1.15 is 115) and I got the frequency that is required to encode 1 point of the result. I could divide the points with frequency, which would give me the point per MHz, but decided against it cause it would result with a lot of decimal points, and the results would be messy. I used all available representatives of Phenom, Athlon, Core 2, Core i3, Core i5 and Core i7 families. Almost all families except Core i3 family had multiple CPUs, I used arithmetic mean to calculate average value per family of processors. I wanted to calculate architecture mean, but since the CPU families and architecture scaled nicely I decided against calculating the CPU architecture mean. I came up with some surprising results. If 2 families have the same architecture, if you want the most efficient family of CPUs (per core & per clock) you should subtract 1 of the name of more expensive family. More features do not mean more performance! Core i3 within the Clarksdale family is faster then Core i5, the same applies to Lynnfield family. I did not test the Gates family, but I can tell you right away, Bill always wins!
After dividing frequency with point and then adding all the numbers and dividing it with number of CPUs I came up with these results. Less is better, cause it means the CPU can score a point with less MHz, and that says it’s architecturally better.
Sandy Bridge Core i5 and Core i7 = 24, 86
Identical results in Core i5 and Core i7 family. No surprise there!
Bloomfield Core i7 = 29, 08
Just 4c CPU and memory controller, no wonder this architecture yields best results (if you don’t count Sandy Bridge, which is built a lot after Bloomfield’s moved in next door, but still, they can’t walk over the bridge 🙂.
Gulftown Core i7 = 29, 535
6c is a little too crowded I guess, but this CPU is for workstations and heavy multitasking, so more cores has a advantage over pure one core efficiency, and this generation of CPUs had a lot of new features, 32 nm saves the day in the case of Gulftown I guess…
Lynnfield Core i5 = 29, 76
Integrated PCIe controller, DMI interface, memory controller… simplicity of Gulftown wins!
Lynnfield Core i7 = 29, 915
Only a little faster uncore clock and hyper-threading yield worse results then Core i5 version of the Lynnfield architecture. Why? Simplicity wins! No hyper-threading means no additional code that slows down the execution… or something like that anyway. This Lynnfield brother has more aggressive Turbo and a few more options. Situation that resembles the situation of Clarksdale CPUs, except in that case, faster brother has no Turbo at all (but still does better then slower brother).
Clarksdale Core i3 = 30, 2
Only one CPU in this family was tested, so the results were clear. This is 32 nm chip, Lynnfield is 45 nm chip, so why did Lynnfield win? Lynnfield has no GPU in it, no AES-NI, and the faster Lynnfield has no hyper-threading. Simplicity wins! Still, this is a 32 nm chip that is almost as fast as Lynnfield CPUs per MHz, it handles excellent in games, should overclock better because it uses 32 nm process, and it costs less then Lynnfield counterpart. Multi-threading apart, perfect gaming solution!
Clarksdale Core i5 = 30, 76
Core i5 Clarksdale version has a lot more features implemented in the chip itself, Turbo is just one of the things that you have to pay the price for, AES-NI is a must have if you encrypt, and so on. 2 Clarksdale Core i5 chips were tested and they both have the same result, even though one has 900 MHz GPU frequency and one has 700 MHz frequency (just like tested Clarksdale Core i3).
Core 2 family = 32, 27
I used both core 2 duo and core 2 quad CPUs, cause their results were practically the same. My opinion is that my Core 2 Duo 8400 is still holding on. Recent article published on this site (http://www.tomshardware.com/reviews/core-i3-gaming,2588-13.html) tested Clarksdale 2c CPU and AMD Radeon 5850 (I still miss ATI in the name of mighty Radeon), the Clarksdale 2c CPU was more then enough for Radeon, and with a bit of an overclocking I bet my Core 2 duo would hold it’s own since it reaches 5 GHz on air cooling on almost stock frequency.
Phenom II = 33, 01
Athlon II = 34, 1
In the article I first mentioned, the one that tested different CPUs, all with 4c and all on 2,8 GHz, the conclusion was that AMD was lagging at least 2 generation behind. Let’s do a little counting. Core 2, Bloomfield, and then came Lynnfield, Clarksdale, Gulftown, and now Sandy Bridge. Core 2 is one, Bloomfield and Lynnfield 2 since they both use 45 nm production process, Clarksdale and Gulftown 3, at 32 nm Intel starts to play hard ball, and now Sandy Bridge. That’s 4! They offer more cores per buck, high frequencies, low price, but the chip itself is going nowhere! They can keep pace with Core 2 Duo, they add a core practically for free, up the frequency, you can maybe unlock a dormant core, sometimes you can even overclock if you hit revision, stepping, changing of the tides, if Lord wishes to help you and so on, but still, Intel overclocks much better, and newest solution from AMD is still a lot slower clock per clock then my Core 2 Duo, and since most of the games don’t utilize more then 2 cores, if all CPUs on the market were produced by the AMD, I would be set for life!
Facebook
Twitter
Instagram