How much does RAM speed affect Sandy Bridge performance?

[qubit]

Even before we consider the results of our performance testing, it's interesting to note that Sandy Bridge makes higher-speed memory more and less appealing. On one hand, the unlocked memory multiplier present in K-series CPUs makes setting a higher memory frequency almost as trivial as changing any other BIOS setting. At the same time, the fact that base-clock overclocking is essentially a dead end for Sandy Bridge CPUs means that faster memory isn't required to keep up with higher base clock frequencies. The only reason to buy faster memory for a Sandy Bridge rig is if it's going to improve performance.

So, is it?

It does seem like Intel is dumbing down overclocking and making you pay a premium for the privilege, doesn't it?

Tech Report

 

[micropage7]

yeah, playgroup kids know that nice cpu paired with fast memory would offer best performance
it would be 'useless' if you pair fast cpu with slow memory coz it would hold the max performance that could be bringing out by it

 

[LAN_deRf_HA]

This would have been more informative had they compared it to other platforms. Thankfully we have the maxxmem thread to show off just how much better sandybridge memory performance is http://www.techpowerup.com/forums/showthread.php?t=124143

 

[sneekypeet]

meh, you use crap ram you get crap results. I would have appreciated that test to have been done with ram that can hold realistic timings. 9-9-9 is poop. Why spend all the loot for that kind of setup to kill it off with cas9 1600? (yes I know its a 2133 cas9 kit, but those 9-11-9 timings get me 2400mhz on most of the ram I own)

Maybe I'm looking too deep.

 

[yogurt_21]

meh, you use crap ram you get crap results. I would have appreciated that test to have been done with ram that can hold realistic timings. 9-9-9 is poop. Why spend all the loot for that kind of setup to kill it off with cas9 1600? (yes I know its a 2133 cas9 kit, but those 9-11-9 timings get me 2400mhz on most of the ram I own)

Maybe I'm looking too deep.

they ran 1333 at 7-7-7 just doesn't seem to make alot of real world performance difference. Which is interesting considering x58 runs better on tighter timings rather than higher clocks and p55 runs better on higher clocks.

 

[sneekypeet]

well yeah but I mean the 1600 testing could be much better. I see too many kits at 6-8-6 to justify buying 9-9-9 1600 sticks these days. Just seems pointless to test 1600 at those timings. If run at 6-8-6 it would likely come in on par with his 2133 numbers in my opinion.

 

[cdawall]

these are the same stupid tests that came out when DDR3 was new DDR2 was beating it when you ran a 1066 cas 4/5 kit when you compare to 1333 cas 9

 

[LAN_deRf_HA]

A lot of recent platform reviews use shitty ram, my guess is it's because that's what manufacturers are making these days so that's what they can get sent to them for free.

 

[mastrdrver]

I still can't believe that Intel used the L3 cache for part of the frame buffer for the GPU. What kind of OS uses less then 6MB for just sitting on the desktop?! Its like they picked right back up from 2000 when the dropped the i740. No wonder they need all that L3 cache bandwidth. All of the information being swapped in and out of it must be daunting.

I have no hope that they will ever learn how to make a GPU by themselves. Not going to even get in to drivers. :shadedshu

meh, you use crap ram you get crap results. I would have appreciated that test to have been done with ram that can hold realistic timings. 9-9-9 is poop. Why spend all the loot for that kind of setup to kill it off with cas9 1600? (yes I know its a 2133 cas9 kit, but those 9-11-9 timings get me 2400mhz on most of the ram I own)

Maybe I'm looking too deep.

Why use the expensive stuff with tight timings when the difference only shows up in memory benches? Once back in to the everyday grind of things the benchmark difference disappears in to something undetectable.

 

[sneekypeet]

ummm AFAIK 6-8-6 at 1600 isnt expensive, and it would show in all the benches he did, not just a memory bench

Thats expensive? Computer Hardware,Memory,Desktop Memory,240-Pin DD...

 

[cdawall]

ummm AFAIK 6-8-6 at 1600 isnt expensive, and it would show in all the benches he did, not just a memory bench

Thats expensive? Computer Hardware,Memory,Desktop Memory,240-Pin DD...

most 2000CL9 will do cas 6 and it was even cheaper when i got mine

 

[LagunaX]

Please see my mini review of $79.99 G.Skill F3-12800CL6D-4GBXH (2x2GB) 1.5v.
It is rated XMP DDR3 1600 6-8-6-24 2N 1.5v.
Overclocked to DDR3 2133 7-10-8-27 1T 1.6v.
Reference the AIDA64 bandwidth red/write/copy/latency scores.
I'm not sure how accurate the gflops time and amount are in linX.
Draw your own conclusions and let me know:
http://www.xtremesystems.org/forums/showthread.php?t=266839

 

[mastrdrver]

ummm AFAIK 6-8-6 at 1600 isnt expensive, and it would show in all the benches he did, not just a memory bench

Thats expensive? Computer Hardware,Memory,Desktop Memory,240-Pin DD...

Ok, so its been a while since I looked at dual channel kits.

You are right that it would show extremely marginal differences in the TR piece on anything but memory benches. Seeing as the hard drive is the biggest bottleneck on any system with a mechanical one, any differences are negated by that fact. Until ssds become common place, its like having a 300mph train on a railroad track the size of Dayton. Sure it can go 300mph in a circle but what's the point?

ht4u.net did a really nice review on Sandy Bridge. I wish one of the english sites would have done as thorough review as they did as it is quite impressive. It is where I first ran across the fact that Intel's iGPU can't deal with angel dependent AF even on a Radeon HD 3000 level. :shadedshu