After our stock tests with the Corsair PC12800 C8-8-8-24 RAMs it's time to discover if the newly added multipliers can unlock more CPU performance. We exchange the Corsair Dominators for 2 x 2Gb G.SKill Flare RAMs. Since we don't have specific Ivy Bridge RAMs in the lab we opted to go for the Powerchip based ( PSC ) 2000C7 Flare kit. We know that these can do at least 2600mhz on AMD FX, let's hope for similar speeds on the Ivy Bridge IMC. We quickly discovered that 2800mhz would not work as we couldn't get stable into Windows at all. But 2600mhz didn't seem to be a problem.
The RAM dividers we have tested are:
- DDR3 1333MHz C9-9-9-27 1.65Vdimm
- DDR3 1600MHz C9-9-9-27 1.65Vdimm
- DDR3 1866MHz C9-9-9-27 1.65Vdimm
- DDR3 2000MHz C9-12-10-35 1.65Vdimm
- DDR3 2133MHz C9-12-10-35 1.65Vdimm
- DDR3 2200MHz C9-12-10-35 1.65Vdimm
- DDR3 2400MHz C9-12-10-35 1.75Vdimm
- DDR3 2600MHz C9-12-10-35 1.77Vdimm
Our i7-3770K was OC'ed at 4500mhz ( 45 x 100 ) at 1.25Vcore. The CPU OC strains the CPU some more, so if this is stable, then stock rated speeds surely are too. With the CPU OC the duration of the benchmarks was shortened too, so a win-win situation :)
First up is SuperPi 32M, stressing one core and depending massively on bandwith. With Sandy Bridge we got decent scaling up to DDR3 1866MHz. Going from DDR3 1866 to 2133MHz was still showing some improvement, but it was far less noticeable.
Massive scaling from DDR3 1333 to DDR3 1600MHz with over 10 secs gain. And it keeps on scaling with smaller increments up to 2600mhz. A whopping 6 secs gain from 2133 to 2600mhz means this new CPU technology will be a bencher's delight. The AIDA64 bandwith tests show the READ and COPY tests scaling hard with more RAM speed.
The Latency improves nicely with more raw RAM MHz. You want the best out of this Ivy Bridge ? Then pair it with a decent set of rams. The CPU Queen of AIDA is purely CPU based and stays pretty stable, no matter wich RAM divder selected. But the Photoworx test adores more bandwith. Maybe you already noticed that the 2200Mhz divider results seem a bit flawed. Maybe bugged on the Gigabyte board, only further testing will tell.
The Maxon Cinebench final score goes up slightly, but don't expect massive gains there by just using faster RAMS. Similar outcome for the multi-threaded SuperPi test : Y-cruncher. This one stresses all the CPU cores flat out and can be considered as a decent stability test for your IMC, CPU and RAM.
Finally to round off the Memory Divider testing we use the X264HD 4.0 encoding test on the i7 3770K CPU. Nice scaling action, for an encoder an FPS gained is a treasure.
Maybe better to wait on buying one till Intel addresses this issue and switches back to the paste used in SB. If they plan to do that at all?
Other option would be to tear off the heat spreader but I'm not going to risk that on a 300€ CPU