Why? Do you have any specific 4GB sticks to recommend?

- I personally ordered a 2x4GB kit from Kingston (DDr-3 1600)- this one seems good; RipjawsX Series for Sandy Bridge - 8 Go (2x4Go) DDR3-1333 (PC3 10600 / 10666) - 9-9-9-24-2N - 1.5V - F3-10666CL9D-8GBXL

Florian,Thanks for the information I would not use anything slower than 1600MHz with CAS Latency 6 or 7 on Core i7 or Sandy Bridge system. I wonder if there are any low latency 4GB sticks available on the market.

Hi all,I looked at those 2x4GB kits you mentioned, but I don't like the higher latencies. I'd much rather stick to CL6 (7 at 1.5V) than CL9. The lowest latency 2x4GB kits I found was corsair vengeance at CL8. Thanks for the recommendations anyways.Edit: Just noticed that those sticks are only 1333MHz. I don't want to lose performance through ram. I want to stick with 1600MHz.

Edited January 16, 201115 yr by Alex Sarfati

Mushkin Redline 1600 CAS 6 DIMMs are 6-8-6-24 at 1.65v, and mine run just fine at 7-7-7-24 and 1.5vRegardsBob ScottColonel, USAF (ret)ATP IMEL Gulfstream II-III-IV-VColorado Springs, CO

Hi Bob,For my own edification, if the i7-2600 is rated for RAM 21GB/s (1333 MHz) can an advantage be realized in running 26MB/s (1600MHz)?I ask you this out of shear respect for your technical knowledge and my further understanding.

Hi Gary,Are you talking FSX based or general? I suppose in FSX no real performance increase would be seen, considering the simulators thirst for CPU clock speed. In other programs, multitasking etc it probably would be noticed more significantly.

Hi Bob,For my own edification, if the i7-2600 is rated for RAM 21GB/s (1333 MHz) can an advantage be realized in running 26MB/s (1600MHz)?I ask you this out of shear respect for your technical knowledge and my further understanding.

There's no difference on the desktop outside of synthetic benchmarks. Real-world applications will not run appreciably faster just by having faster RAM. Particularly with Sandy Bridge since you cannot increase the base clock. Here's a good review on the subject.

There's no difference on the desktop outside of synthetic benchmarks. Real-world applications will not run appreciably faster just by having faster RAM. Particularly with Sandy Bridge since you cannot increase the base clock. Here's a good review on the subject.

Hi Max,Thankyou for that information! I thought that myself, however I thought that performance differences may also be noticable in RAM intensive programs? Please correct me as required.

Hi Max,Thankyou for that information! I thought that myself, however I thought that performance differences may also be noticable in RAM intensive programs? Please correct me as required.

Good review Techguy (thxs), I guess I am just trying to understand the concept of increased frequency in comparison to design specifications. To further simplify my question, does OC’ing the multiplier also OC the IMC?I have detracted from the review that memory speed in excess of 1333MHz is not benefit when looking at Cinebench and other real-world tests (nor does 4GB vs. 16GB make a difference).Even simpler I was wondering if the maximum design spec of the 2600 was finite at 21/GBs hence any additional bandwidth would just bottleneck the IMC (although no indication of bottlenecking is evident in the test results, not sure if those tests are definative).

There's no difference on the desktop outside of synthetic benchmarks. Real-world applications will not run appreciably faster just by having faster RAM. Particularly with Sandy Bridge since you cannot increase the base clock. Here's a good review on the subject.

Thanks for the link to that review :Applause: Very interesting :Thinking:

Good review Techguy (thxs), I guess I am just trying to understand the concept of increased frequency in comparison to design specifications. To further simplify my question, does OC’ing the multiplier also OC the IMC?I have detracted from the review that memory speed in excess of 1333MHz is not benefit when looking at Cinebench and other real-world tests (nor does 4GB vs. 16GB make a difference).Even simpler I was wondering if the maximum design spec of the 2600 was finite at 21/GBs hence any additional bandwidth would just bottleneck the IMC (although no indication of bottlenecking is evident in the test results, not sure if those tests are definative).

No, the memory bandwidth isn't capped at design speed, any more than the CPU clock speed and throughput is limited...the design specs are used to assure that virtually every CPU will perform with an extremely high degree of reliability at the design speed...or in other words that the odds that any given CPU (produced within design tolerances) will not be able to hold its own if run at spec speeds are statistically nil. The i7 Nehalem CPUs are also designed for 1333 MHz RAM, and they are also capable of running with much higher memory bandwidth at higher speeds (which is why the CPUs have a memory multiplier at all).What I didn't like in this article, as in many like it before, is that if you read the fine print, they perform the tests at low CPU clocks...around 3 GHz was what this article's fine print quoted. Memory bandwidth limitations most often show up when you get the CPU clocked up fast enough that the memory subsystem can no longer keep up with the CPU...I found that at speeds around 4 GHz and above on my i7, fast RAM with tight timings made much more of a difference than it did down at 3.33 GHz stock speeds. And most of the tests also only test memory bandwidth effects on the CPU...I don't see anything about DMA bandwidth to the GPU(s).I think it's reasonable not to pay through the nose for exotic RAM, but I still think 1600 MHz at CAS 6 is the sweet spot for performance before costs go parabolic and any marginal improvement in performance is no longer worth it.RegardsBob ScottColonel, USAF (ret)ATP IMEL Gulfstream II-III-IV-VColorado Springs, CO

Bob, I beg to differ. I actually have experience and MaxxMEM2 results running 1600 cas 6 memory vs 1866 cas 9 memory on my Sandy Bridge system at 4.6GHz. Check out my experience here on avsim.

8GB (4x2GB) Mushkin Redline 1600MHz CL6 1.65V <----Will turn down to CL7 1.5V

I don't think this^ will be happening - at least not if you have the same experience I had. Check out the link I posted above for Bob. My biggest advice to you is DO NOT PURCHASE 4X2GB. Find some good 2x4GB instead. That is solid advice. If you go through with it, let me know if you are able to run them at CL7. It might be interesting to see if your Asus mobo will do it.

Bob, I beg to differ. I actually have experience and MaxxMEM2 results running 1600 cas 6 memory vs 1866 cas 9 memory on my Sandy Bridge system at 4.6GHz. Check out my experience here on avsim. I don't think this^ will be happening - at least not if you have the same experience I had. Check out the link I posted above for Bob. My biggest advice to you is DO NOT PURCHASE 4X2GB. Find some good 2x4GB instead. That is solid advice. If you go through with it, let me know if you are able to run them at CL7. It might be interesting to see if your Asus mobo will do it.

The results you posted are interesting...and I really don't know what to make of the comparative results posted by others--systems with fast memory and low scores, and systems with slow memory and high scores. Also some rumbling on forums like HWBOT that there are problems with Maxxmem not recognizing CPUs, multipliers etc. But something does not compute.Just for grins, I dropped the timings down to 8-9-8-24-2T on my 1600 MHz Mushkin Redlines and pushed the freq up to 1866 MHz. The Maxxmem scores went up almost 9%. Not only can I not explain that, I'm not sure I believe it.Something interesting to ponder tomorrow.CheersBob ScottColonel, USAF (ret)ATP IMEL Gulfstream II-III-IV-VColorado Springs, CO

Hi all,I've already ordered my system I'm affraid. Thanks to my nearest newEgg branch, the parts were dispatched the day after (today) i ordered them. Now I'm only relying on the Los Angeles postal service.

The results you posted are interesting...and I really don't know what to make of the comparative results posted by others--systems with fast memory and low scores, and systems with slow memory and high scores. Also some rumbling on forums like HWBOT that there are problems with Maxxmem not recognizing CPUs, multipliers etc. But something does not compute.Just for grins, I dropped the timings down to 8-9-8-24-2T on my 1600 MHz Mushkin Redlines and pushed the freq up to 1866 MHz. The Maxxmem scores went up almost 9%. Not only can I not explain that, I'm not sure I believe it.Something interesting to ponder tomorrow.CheersBob ScottColonel, USAF (ret)ATP IMEL Gulfstream II-III-IV-VColorado Springs, CO

My present system has 6g of 998805 triple channel Mushkin 1600 (6-8-6-24) Redline Ram. As I intend to buy Sandy Bridge when the dust settles I ordered a set of the 996805 dual channel version so that I will have 8g (4x2g) in the SB. I agree (and have gambled) that these sticks will most likely work at the same frequency with lower volts with looser timings. Well, that is what I hope will happen. Otherwise I will have 5 sticks of ram for sale.Kind regards,

The results you posted are interesting...and I really don't know what to make of the comparative results posted by others--systems with fast memory and low scores, and systems with slow memory and high scores. Also some rumbling on forums like HWBOT that there are problems with Maxxmem not recognizing CPUs, multipliers etc. But something does not compute.Just for grins, I dropped the timings down to 8-9-8-24-2T on my 1600 MHz Mushkin Redlines and pushed the freq up to 1866 MHz. The Maxxmem scores went up almost 9%. Not only can I not explain that, I'm not sure I believe it.Something interesting to ponder tomorrow.

Check this^ out in your pondering. Extrapolating from the results here, it's a pretty reasonable assumption that 1600 (6-8-6-24) memory should outperform 1866 (9-10-9-27). I'm thinking it's because I could only get 4GB of it to actually run at 1600 6-8-6-24 while it was competing with 8GB at 1866 (9-10-9-27). 8GB of slower memory probably has more bandwidth than 4GB of fast memory. Regardless, since you can't get 8GB to run at 1600 (6-8-6-24), it's a moot point. You will be forced to run with it at 7-8-7-24 at best. I couldn't even get that far. Considering what you should be able to find in stock with 2x4GB modules, I think my final recommendation is either:2x4GB DDR3 1866 CL82x4GB DDR3 2133 CL9I don't know what's available in either of these configurations, but they seem reasonable.So far, I'm convinced either my Gigabyte BIOS becomes unstable when overclocking RAM or the Sandy Bridge integrated memory controller (IMC) is not so great at handling overclocked ram. My bets are on the BIOS, though. Considering the difficulty myself and others are having with Sandy Bridge memory overclocking, my advice is to get memory that achieves your desired specs without the need to tweak it. Don't assume you will be able to clock 2000MHz ram up to 2133, for example. Of course that's generally more expensive. Follow this advice and your Sandy Bridge experience will go quite smooth actually.