The best DDR3 for Sandy Bridge, round 2

[Ken g6]

Ever since AnandTech published "Sandy Bridge Memory Scaling: Choosing the Best DDR3", I've been recommending 1333MHz RAM over 1600, especially for gaming. Just to recap, the article showed a 1-2% performance benefit at best, and no benefit for gaming.

One thing that article didn't cover was Photoshop, so I went looking for such a benchmark today. I found a Bit-Tech.net article that, while it doesn't have Photoshop, shows only a 1% improvement in the GIMP. So that settled that.

But then I looked at page 6. It shows a full 5% performance improvement in Civ V! Have we been leading people astray from significant performance improvements in some games?

 

[Capt Caveman]

RAM is so cheap, why not just get DDR3 1600 1.5v? More flexibility and headroom for little money. Heck, I have 16gb of it in my current rig.

 

[StrangerGuy]

DDR3 1600 is only like $1-2 more than 1333 and latter is more or less guaranteed to hit 1600 speeds anyway.

 

[mrjoltcola]

I guess if your question included "... for gaming" then I'd say the difference is inconsequential, but for me, I also compile projects many times a day, and I want to look at the return on investment, and hit the sweet spot of the curve, while still getting a rig that will last me 5 years between builds. 16GB of 1600Mhz cost me $106. If you get a deal like Microcenter is running, a 2600K + motherboard + 8GB RAM cost 450 bucks, 2500k cost < 400, so either of these prices puts a workstation class machine on your desk at prices unheard of 5 years ago. My last rig was a Core 2 Extreme X6800 (2.93Ghz) with 1066Mhz and it ran me about 2k. New rig was 1600 and incuded an expensive HyperX SSD. So its all relative to what each person thinks is expensive.

My simple benchmark (using SuperPi) with an overclocked 2600K, results in more performance advantage for higher levels of parallelism (or maybe just busier bus?).

Example, running SuperPi at 4.4Ghz I get:
4 threads / 1333Mhz / 1million digits = 12.47 sec avg
4 threads / 1600Mhz / 1million digits = 12.13 sec avg
Approx 3%

But running 8 threads showed:
8 threads / 1333Mhz / 1million digits = 16.0 sec avg
8 threads / 1600Mhz / 1million digits = 14.81 sec avg
Approx 7.5%

My guess here is the extra memory speed helped more when all cores (logical and physical) were pegged and keeping the memory bus busier concurrently, but this is just a guess.

It is enough for me to care to spend a couple of extra bucks on RAM when I just spent +80 bucks more on the 2600K over the 2500K. I guess, in my book, there are better things to save $$ on than RAM speed.

Cheers.

 

[Patrick Wolf]

But then I looked at page 6. It shows a full 5&#37; performance improvement in Civ V! Have we been leading people astray from significant performance improvements in some games?

None of these memory articles show a wide range of programs and games. Like CPU's, not all games benefit from a faster one. Based on the bit-tech and xbitlabs articles I concluded that 1600 was optimal and have always been recommending it since the "premium" for it is about $10.

Cheapest 1333 8GB 1.5v kit at newegg is $40.
http://www.newegg.com/Product/Produc...82E16820161468

For $10 more you get this (cheapest kit at time of posting)
http://www.newegg.com/Product/Product.aspx?Item=N82E16820231416