Core 2 Duo vs Sandybridge Core i5

[Barfo]

Hey all, does anyone have a good guess on how much faster a Sandybridge Core i5 is than a 45nm Core 2 Duo in single threaded apps, if we assume both have the same clock speed? I'm talking about laptop CPUs btw.

 

[drizek]

Really hard to say. In some apps, tons faster, due to AVX and other extensions. In other apps, maybe double digit percentage points faster. A lot of it comes from HT, so truly single threaded apps don't benefit as much. The other big advantage is the memory controller, but again how useful that is depends on the app.

 

[podspi]

Not a perfect view, but this is helpful:

http://www.anandtech.com/bench/Product/55?vs=289


I usually like to use Cinebench as an easy way to see what singlethread performance looks like (even though maybe I shouldn't). In this case, SB looks ~ 30% faster. Interesting to see they look almost identical in sysmark... Looks to be about 15% faster in Monte Carlo sim, though that is threaded and the i3 gets a boost from HT...

 

[jvroig]

how much faster a Sandybridge Core i5 is than a 45nm Core 2 Duo in single threaded apps, if we assume both have the same clock speed?

I usually like to use Cinebench as an easy way to see what singlethread performance looks like (even though maybe I shouldn't)

I usually have the same concern (looking for the "average" or "most reasonable" clock-for-clock difference), and I also end up looking at Cinebench. Unfortunately, it is rarely satisfying due to being a synthetic test and, more importantly, it actually doesn't give me the clock-for-clock difference I am looking for because max turbo kicks in (single thread).

For other purposes ("stock" settings) this may be ok, since you are comparing apples to apples in a way (out of the box performance on both processors). For my purposes, however, I am interested in the clock-for-clock advantage because of the intention to OC to a reasonable max, turning turbo off in the process. The best I could get was choose an i5 whose max turbo more or less coincides with the clockspeed of the other, non-turbo enabled contender.

It is not perfect, and I am sometimes frustrated at reviews that skip the "clock-for-clock" section. It is understandable, however, since it serves the general population better (presenting "out of the box" performance), and only a few overclockers will have to be inconvenienced for a while.

 

[drizek]

Keep in mind that turbo mode on Core i5 means MUCH better single threaded performance than C2D. Clock for clock comparison is meaningless.

 

[john3850]

Ran SuperPi 1m at the same 4300mhz on a c2d and 920x58 and got a 2 to 3 sec difference.
The i5 has more ipc then my 930 so there will even be a larger difference in superpi scores.
I believe i5 at 4500mhz should get a 8 sec on 1meg pi or a i5 does Pi 30% faster then c2d.