bit-tech.net

The Best Memory for Sandy Bridge

Comments 26 to 50 of 52

Reply
Kakeshoma 11th January 2011, 22:42 Quote
Very good article! Thx.
caseblue 12th January 2011, 03:00 Quote
I'm happy to see confirmation that 1600 is ideal for basic home use and I'd like to know what kind of success folks have had with buying low latency 1333 and then bumping it to 1600. I've got my eye on G.Skill products mainly and I'm hoping to hear that the 1.5 1333 7-7-7-21 versions will run at 1600 8-8-8-24 or so just fine, even if I have to go to 1.6v. Kindly share your opinions, or better yet experiences, with doing this on P67 or P55 boards. It seems to me like everything is made from the same internals with only the timings, frequencies, and voltages modulated to produce the different models.
aRareKindOfMonster 12th January 2011, 03:37 Quote
Very informative article. I understand that the processor was run at 3000MHz.

Would overclocking a SB processor to the 4GHz or even 4.5GHz mark affect the results?
I have to admit, that I recently purchased a 2x2GB PC14900 CL9 kit, thinking I would have a slight advantage, but your article set me straight. Maybe an OC'ed processor would make me feel better about my recent purchase.

If not, i will have to find a way somebody to sell this kit.
fluxtatic 12th January 2011, 05:35 Quote
Quote:
Originally Posted by wuyanxu
i won't be so sure with that.

i have ran out of memory on my 8GB system, when trying to transcode videos while play MW2, with a VMware of my project in the background. so page file is still needed.

8GB should be standard memory ammount for Sandy bridge. i think Intel should have restricted so that anything less than 8GB memory in its DIMM slots, and it won't boot. that's a sure way to move this forward. i've had 8GB since 2007, waiting for everyone else to catch up.

I would guess this sort of usage isn't common at all. Nice if you've got the horsepower, but not an everyday scenario by any stretch. And force 8GB? I don't know about y'all, but my pockets are a lot deeper than my arms are long. Also, I don't know that I'd be all that happy about being forced to a memory capacity by the proc. I'd rather it be determined by the OS.

I just went up to 6GB (2x2 + 2x1) myself, and have yet to see usage go much over 2GB. As a matter of fact, I handed 2GB over to a RAMdisk (imdisk, which everyone should have, imo) for the cache/temp dirs in Opera and IE. As a matter of fact, if you can stretch to it, get 8 if you think you need 4, 12 if you think you need 8, and throw the excess to a RAMdisk. The speed difference in browsing is surprising, and there's no worries about IE not clearing temp directories like it should. It can defeat the purpose of cache to some extent, but there aren't many sites where it matters if it can pull content from previous cache. This site is the really the only one I tend to leave open in Opera pretty much all the time.

Plus the RAMdisk makes a handy scratch disk for slobs like myself. In writing scripts and such, I have a tendency to collect garbage files that I'm reluctant to delete later (much like my museum of 386 to first-gen-Athlon vintage hardware). I use the RAMDisk as the working dir, and powering off cleans my garbage for me. YMMV, though, especially if you use Photoshop. From what I've seen, it will take all it can and still want more.
Bindibadgi 12th January 2011, 05:56 Quote
Quote:
Originally Posted by aRareKindOfMonster
Very informative article. I understand that the processor was run at 3000MHz.

Would overclocking a SB processor to the 4GHz or even 4.5GHz mark affect the results?
I have to admit, that I recently purchased a 2x2GB PC14900 CL9 kit, thinking I would have a slight advantage, but your article set me straight. Maybe an OC'ed processor would make me feel better about my recent purchase.

If not, i will have to find a way somebody to sell this kit.

You're Welcome :):)

Yes it would - more processor frequency means its execution engine, that initiates the loads and stores on memory bus, runs faster, so you get more performance overall. In fact, a 4.5GHz CPU with 1600 mem would be faster than a 3.3GHz CPU with 2133 mem - it's the CPU that does the work.
aRareKindOfMonster 12th January 2011, 07:58 Quote
Quote:
Originally Posted by Bindibadgi


You're Welcome :):)

Yes it would - more processor frequency means its execution engine, that initiates the loads and stores on memory bus, runs faster, so you get more performance overall. In fact, a 4.5GHz CPU with 1600 mem would be faster than a 3.3GHz CPU with 2133 mem - it's the CPU that does the work.

Thanks for the explanation.

I failed to be clear in my initial question. Allow me to rephrase; would a faster CPU frequency spread the results between memory speeds, giving an edge to faster memory modules, or would it show about the same spread between modules of different speeds? Or, to put it concretely, would the 1600 memory trail farther behind 1866 modules, if an OC'ed CPU would be used?
I have no means of testing it, but your article has led me to believe that the difference would be very minor, insignificant even, especially in games.

I feel disappointed that I came across your article AFTER I purchased the faster 1866 4GB memory kit, when I could have ordered a 1600 8GB kit for a slightly higher price, or just save some money by ordering a 4GB 1600 kit. I believe that I ordered the RAM shortly before or after the article was published. Bummer! :)

Sometimes, you just can't win. :D
perplekks45 12th January 2011, 09:45 Quote
So there's virtually no performance gain when going from 4GB to 16GB? Interesting.
I'd still go for 8+ with Sandy Bridge, just because 4GB is almost nothing anyway.

And what's with the Skimlink-powered ads? :( While not as bad (actually nowhere near as annoying) as IntelliTXT ads I was still quite surprised to see these links.
RonanH 12th January 2011, 10:27 Quote
So for general use with a normally clocked processor 1600 is good. What I'd be more interested in is the effect of using faster memory with an OC'd SB. I know you can use a higher divider to get 1333 memory running at the higher speed but I would imagine that this would have a more detrimental effect.

So my query is does fast memory make a difference with an OC (like most of the people reading this will be running).
himanshudpfoc 12th January 2011, 10:46 Quote
Hi i am new here and i am here to make some new friends.
Bindibadgi 12th January 2011, 11:15 Quote
Quote:
Originally Posted by aRareKindOfMonster
Thanks for the explanation.

I failed to be clear in my initial question. Allow me to rephrase; would a faster CPU frequency spread the results between memory speeds, giving an edge to faster memory modules, or would it show about the same spread between modules of different speeds? Or, to put it concretely, would the 1600 memory trail farther behind 1866 modules, if an OC'ed CPU would be used?
I have no means of testing it, but your article has led me to believe that the difference would be very minor, insignificant even, especially in games.

I feel disappointed that I came across your article AFTER I purchased the faster 1866 4GB memory kit, when I could have ordered a 1600 8GB kit for a slightly higher price, or just save some money by ordering a 4GB 1600 kit. I believe that I ordered the RAM shortly before or after the article was published. Bummer! :)

Sometimes, you just can't win. :D

It would give the same spread regardless of frequency as the memory bus is not linked to the CPU overclocking any more, because the CPU uses the multiplier exclusively. If you push the base clock 2/3/4/7.8MHz, then both CPU and memory clock increase in relation, but the memory ratios available remain the same, so the spread still remains the same.

It's not like before where running the CPU and memory "in sync" would yield better efficiency because SB uses an internal ringbus that cannot be manipulated afaik. It used to be the case with separate northbridges that getting the lowest 'tRD' yielded a faster performance, but that hasn't been the case since we got Nehalem.
schmidtbag 12th January 2011, 13:19 Quote
HA!
finally a 2nd source that shows having 16gb will actually start slowing things down! i'm so sick of people bragging over having ridiculous amounts of memory and that it somehow manages to greatly improve their overall performance when they probably don't even use up 2/3 of it.

i'm also sick of people acting like higher frequency memory = better memory, because these tests show that isn't true in all situations.

thanks for the post bit-tech, memory is one of the most misunderstood components in the entire computer.
Farting Bob 12th January 2011, 16:36 Quote
I just ordered my sandy bridge goodness, including GEIL 8GB (2x4GB) 1600Mhz goodness for a very nice £91. RAM hasnt been this cheap in a while, sure beats the 2GB of DDR2-663 i have right now...
Glix 12th January 2011, 18:45 Quote
Quote:
Originally Posted by Eiffie
Do I sense the topic of a awesome article about everything you want to know regarding a page file? Unless it's already been done before and is fairly recent with new and updated info, then a link will do. :)

Linky - Tomshardware Ram Article Nov 2010

Skip to Swap file part. :P
Or:
Quote:

Conclusion

We would advise against disabling the swap file if you have less than 12 GB of RAM, at least without thoroughly testing all of your programs first. There is a performance increase when you disable it, but you have to factor in the risk of possible data loss.

I don't know anyone with 12gb of ram. :)
Fingers66 12th January 2011, 19:03 Quote
I've seen a post on another forum from someone who claimed to have spoken to Intel about RAM voltages (note that I cannot verify if this is true).

\Quote (point 1 is about vcore, point 2 about RAM, point 3 about BCLK overclocking):
"I've spoke to Intel this morning and got the following answers from them:-

1. Anything over 1.40v on air is risky, upto 1.50v on water should be OK but risky.
2. Anything over 1.50v on memory is not recommended but should not kill a CPU.
3. Clocking on BCLK is highly un-recommended due to onboard graphics.

So the biggest factory they say is vcore, pushing above 1.40v is a risk, therefor I'd say don't go beyond 1.35v. BCLK clocking is a no no for sure.

I've asked for documents regarding 1.50v as motherboards manufacturers seem to think 1.65v is the safe limit, though Intel are saying 1.50v is more advisable.

I am going to update all our CPU descriptions with this information.

Anyone with 1.6-1.7v memory should run the memory at 1.50v as in theory it will run fine at those voltage as all memory manufacturers are now on new IC's. "

/Quote

Bindi - I specifically want to know about the 1.5v (Intel) vs 1.65v (manufacturers) advice. What is the word on the ground where you are? Are Intel just being cautious/conservative or are manufacturers just wanting to shift stock of 1.65v RAM?
Bindibadgi 13th January 2011, 02:24 Quote
The rule is within 0.5V, so it'll be within 0.5V of System Agent voltage (not vcore, since the System Agent houses the memory controller. I did say vcore above/before and that was wrong).

I think (iirc) the SAVcc is 1.05V which means 1.55V OK, but you'd get away with 1.65V.

So, yes, it goes outside the rule of thumb, but have you ever heard of anyone blowing up a Nehalem, Lynnfield or Clarkdale CPU memory controller?? Or even an Intel CPU for that matter? I actually don't think I have and I've seen guys put 2V through DDR3 with a 980X (also a 32nm CPUs) in OC tournys. I would say they are hardwearing enough to handle it, so if you already own memory don't worry, however if you are buying new then aim lower. ;)
Fingers66 13th January 2011, 09:49 Quote
Cheers Bind. ;)
Shifter549 14th January 2011, 11:48 Quote
Quote:
Originally Posted by wuyanxu

i won't be so sure with that.

i have ran out of memory on my 8GB system, when trying to transcode videos while play MW2, with a VMware of my project in the background. so page file is still needed.

8GB should be standard memory ammount for Sandy bridge. i think Intel should have restricted so that anything less than 8GB memory in its DIMM slots, and it won't boot. that's a sure way to move this forward. i've had 8GB since 2007, waiting for everyone else to catch up.

So Intel should have forced Joe Average, who only runs internet explorer and windows media player, to spend the extra dosh on 8GB of memory in his PC from PC World, because extreme users like us use more than 4GB occasionally? I don't think so... As a side note, why not work on making things more efficient and optimised, rather than have operating systems hog over a gig of ram before you actually open anything?

I actually have 8GB of ram as well, so I do see your point, but I strongly disagree that EVERYONE should be forced to have extreme amounts of memory or no working PC. And although everyone's entitled to their own opinion, as far as we know, we could be all using DDR4 by the time 8GB is needed to run a simple web browser - that would render Joe Average's initial outlay for 8GB of DDR3 rather pointless, don't you think? Not everyone's made of money :)

Great article though, peeps, and I love the thorough testing! Thanks for a good read :)
wuyanxu 14th January 2011, 12:49 Quote
due to the lack of progress with memory capacity, we are still stuck with 32bit operating system as lowest denominator that results in programmers having to compile their application in 32bit, sacrificing efficiency and speed for compatibility.

only way to move forward is to have Microsoft ditch 32bit operating systems on fully powered system (exclude netbook editions of Windows, such as starter) and for Microsoft to consider that, the hardware must be demanding the larger memory support, hence 8GB requirement. at the end of the day, it's more of chicken and egg.



yes, average Joe doesn't need 8GB. but they also don't know which comes first, chicken or the egg? choosing 4GB (or 3GB on some stupidly spec'd machines) would usually have 32bit operating system installed by uneducated OEMs. (especially with Win7's decision to print physical memory capacity instead of OS addressable capacity on 32bit machines) the requirement for 8GB memory on Sandy Bridge will ensure 64bit system to be installed as standard.

hopefully speed up adaptation of 64bit, making newer applications much better. (eg. BF3 already require Dx11, why not take it further and require 64bit operating system, it's going to need it with 64 player multiplayer) Crysis Wars multiplayer for example, launch in 64bit mode means cannot join multiplayer server, says incorrect version. final example: if Supreme Commander is native 64bit application, then we won't see crashes during 2000 unit battles.



although the price is a concern. so best way is to perhaps limit the 8GB requirements to i5 and i7's. if average Joe thinks power isn't needed, then he can always opt for i3, which is designed for low end anyway, with 4GB requirement to ensure 64bit usability.
perplekks45 17th January 2011, 11:34 Quote
Quote:
Originally Posted by wuyanxu
due to the lack of progress with memory capacity, we are still stuck with 32bit operating system as lowest denominator that results in programmers having to compile their application in 32bit, sacrificing efficiency and speed for compatibility.

only way to move forward is to have Microsoft ditch 32bit operating systems on fully powered system ...

As they have done with Server 2008 R2. ;)
I don't think I've used a 32bit OS since Vista came out.
Eiffie 17th January 2011, 21:42 Quote
[QUOTE=Glix]
Quote:
Originally Posted by Eiffie
Do I sense the topic of a awesome article about everything you want to know regarding a page file? Unless it's already been done before and is fairly recent with new and updated info, then a link will do. :)

Linky - Tomshardware Ram Article Nov 2010

Skip to Swap file part. :P

Thank you!!!
jimz360 21st January 2011, 03:56 Quote
Great stuff, now just need to know if you fancy doing 4 x 2gb & 2 x 4gb benchmarks
timevans999 23rd January 2011, 09:33 Quote
Quote:
Originally Posted by jimz360
Great stuff, now just need to know if you fancy doing 4 x 2gb & 2 x 4gb benchmarks

Yeah typical using bloody crysis to test what 16 gig might do for you. Why would you think Crysis would be any better with loads of ram I give up on these gaymofos!
jimz360 23rd January 2011, 19:45 Quote
Quote:
Originally Posted by timevans999
Quote:
Originally Posted by jimz360
Great stuff, now just need to know if you fancy doing 4 x 2gb & 2 x 4gb benchmarks

Yeah typical using bloody crysis to test what 16 gig might do for you. Why would you think Crysis would be any better with loads of ram I give up on these gaymofos!

Who said "Test Crysis"? Who said "16gb"? Your giving up on these gaymofos? Are you actually going to offer any advice, or are you here just to talk crap? I think these "gaymofos" would be better of without you.
Gradius 29th January 2011, 23:02 Quote
My! So much LOVE here!

Ok, ok, you can go back and kill each other.
simbr 15th February 2011, 01:17 Quote
The graphs labels are messed up in the Custom PC printing of this review - three of the graphs are labelled Average Memory Bandwidth, and the the actual avg b/w table the legend states that "lower is better".
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums