Intel Sandy Bridge CPU In-Depth Look at Overclocking, Memory Timings and More

CPU by leeghoofd @ 2011-02-01

First introduced at the CES, Intel’s new Sandy Bridge CPU architecture is here to flood the mainstream market with over 25 CPUs. Don't panic, most are foreseen for the mobile market and only 9 new models will be introduced for the desktop segment. Coinciding with this new release is also a new socket design. 1155 pins will be the new standard for Intel’s mainstream lineup. Yes you guessed it, Sandy bridge is here to replace socket 1156. Slowly but steadily Clarkdale and Lynnfield will become End Of Life and will be phased out. At the Sandy Bridge Tech conference the representatives of Intel said that the current S1366 i7 lineup (Bloomfield and Gulftown) will remain their high end platform. Time to explore Sandy Bridge...

  • prev
  • next

Ram divider performance

Since the current Sandy Bridge platform is seriously limped in Bclock OCability, let's explore if we can get some extra performance by playing with the available RAM dividers. The tests on the previous pages were all done with 1600mhz ram speed. Only the 1156 i5 760 was tested at 1333Mhz. On my Asus P67 test board I had the choice between : 800/1066/1333/1600/1866/2133 and 2400mhz. (Though I never got the latter to work). With previous Intel chipsets you had to tinker with the Bclock to get the required high ram speeds. Sandy Bridge makes it all, maybe for some, too easy. Just select the required divider, save and reboot. I only tested the 1066/1333/1600/1866 and 2133 divider with the Corsair GTX2 rams. I skipped the 800mhz DDR3 ram as I don't see it useful, as it über cripples the bandwidth. Neither is this ram daily used, not even in laptops...

 

 

As usual we start off with Superpi 1M and Wprime 32. Both being very fast tests and mostly not clearly showing ram efficiency. Normally raw CPU Mhz rules over timings. But we already seem to detect up a small increase in performance as the ram speeds go up. Time to move onto the big brother superpi 32M.

 

 

Now this is more like it. Over 40 secs shaved off the 1066Mhz Superpi 32m time. Wprime is only gaining a few secs, but this benchmark is less influenced by ram speeds or timings. How big are the reads and writes then to accomplish this huge boost ?

 

 

The message seems clear. Copy and Read performance are not brilliant at 1066 and 1333Mhz. (luckily I didn't test 800mhz speeds) Once we move up to the 1600mhz ram divider, the bandwidth department gets a nice boost. Scaling even better at the higher dividers. This socket has got a monstrous bandwidth when bundled with appropriate rams. It even trounces the triple channel performance of it's bigger S1366 brethren.

Let's do some encoding with the X264HD benchmark. Will the extra bandwith provided, give some more headroom ?

 

 

Again observing scaling with more RAM bandwidth. A few FPS more, is for a serious encoder a time saviour. Give your 1155 CPU what it needs : some nice rams running at at least 1600mhz.

 

Moving onto PCMark05, which tells you a bit more about the complete performance of your system.

 

 

Once ram speeds are at and beyond 1600mhz the performance picks up. Bundling such a high performance processor with rams at or below 1066Mhz is a crime. 1866 and 2133Mhz results are very close.

 

 

3DMark01 has always been the love of the S775 platform. Not even the almighty 1366 could take the performance crown. Intel's latest creation however provides some stiff competition. But that's for later. Observing the collected data, more ram speed reflects in better performance. Doesn't take a mad scientist to analyse what's going on here.

 

 

Even though 3DMark06 is less spectacular in scaling then the 01 version, it still manages to get a few points extra. Synthetic tests are nice, but a few games examples maybe will tell you a bit more.

 

 

Far Cry 2's integrated benchmark was run at 1280 x 1024 resolution. Benefiting big time from more ram speed. Rewarding you with a few extra FPS. Though starting off at 150FPS, which is already bloody fast , I don't think you will notice the difference between the 1066Mhz and 2133Mhz setting in games.

 

 

Mafia II seems to be maxed out on this setup. One FPS difference between 1066Mhz and 2133Mhz... the GTX480 runs the flat out...

 

 

I included the Final Fantasy XIV benchmark as it seems to pop up more and more on several forums. Though the outcome of the benchmark is just a number, no FPS. Scaling is there, though what it means for real FPS. absolutely no clue...

 

  • prev
  • next
Comment from Teemto @ 2011/02/02
52x102 here with flares at 2176-6/9/6/24
SuperPi and Pifast stable.
Comment from leeghoofd @ 2011/02/02
So use these clocks too then for 3D Pascal
Comment from Teemto @ 2011/02/02
Yes my master.
Comment from thorgal @ 2011/02/03
So which settings did you use for 5Ghz It's those "just a few settings" that interest me

I always want to learn from a master
Comment from Stefan Mileschin @ 2011/02/04
Quote:
Originally Posted by Teemto View Post
Yes my master.
Is the system stable in 3DMark 2005 CPU test too at those frequencies?
Comment from Teemto @ 2011/02/04
Nope. That's realy the max I could go.
Haven't played around with the other voltages though.
Maybe Albrecht can shed some light if this could improve stability/OC'ability?

 

reply