DDR3 Roundup: New Elpida Kits from OCZ, Mushkin and Corsair

Memory by thorgal @ 2009-08-09

Sometimes there´s more to a memory module than meets the eye : Corsair Dominator and Mushkin Redline modules have always looked better than nice, and now they have the performance to match. We put these two against the OCZ Blade, a real high end kit based upon Elpida´s -hyper- chips, and against our 8 previous triple channel kits, in order to get a good view of the Core i7 memory landscape to date.

  • next

Introduction

Introduction

During our last memory roundup we talked about 8 different triple channel memory modules from 8 different manufacturers. As always, some were better than others, but in the end most - if not all - of them deserve their place in your PC setup.

We did have a couple of minor issues however with two of the memory kits, the Corsair Dominator and the Mushkin "Radioactive" modules to be precise. Although there was physically nothing wrong with the modules - both did their rated specifications without any issues - both proved a nightmare to overclock. At first we thought it was due to our motherboard, but after testing them with the "king" of the memory tuning boards (the DFI X58), we were able to rule out that possibility. What was wrong then ? Did we just get two unlucky OC kits ?

Whatever it was, both manufacturers were quick on their feet and provided us with two new memory kits straight from their labs, just a couple of days before our roundup went on air. At the time we didn't have the time to include them in our roundup, but we'll try and make up for it in this review.

Two more 1600 Mhz kits

This is what came into our lab:

Madshrimps (c)


The Corsair Dominator is, apart from its very different packaging, an identical kit to the one we already tested - at first look at least. The Mushkin kit comes in the same package as the "Radioactive" kit we reviewed before, but as you'll soon see the appearance is very different.

Both these kits are reportedly equipped with a new chip revision made by Elpida, a revision B if you like. In the past months we got to know some of these "rev. B" chips as the "Elpida Hyper" chips - a special binning if you like, geared towards the (ultra) enthusiast. Elpida "hypers" have seen their share of positive and negative press, but more on that later.

A high end 2ghz kit to join the 1600Mhz bunch

To see what the new Elpida chips are really made of, we got hold of the true high end in today's memory chips : a 2Ghz "cas 7" kit, kindly provided to us by OCZ for testing. This is the box it comes in :


(click for larger version)


Let's see if these Revision B and Hyper chips are really all they're told to be : will they offer top performance and most importantly, will they withstand the many days of stress testing and benchmarking that is necessary for this review...
  • next
Comment from Massman @ 2009/08/09
I know I've said it already and you don't really like people saying it that much, but this is a HIGH-QUALITY review, Thorgal!!
Comment from jmke @ 2009/08/10
not many (if any) out there match his testing method and thoroughness
Comment from thorgal @ 2009/08/10
Thanks guys, but no need for such praise.

I hope our readers will enjoy the article, and hope that it helps them decide which memory kit to get. There's a kit out there for every budget
Comment from Mike89 @ 2009/08/12
I have a question. The article references two revisions of the Corsair ram, 2.1 and 2.2. I just purchased some of this memory (3 2 gig sticks kit) and the revision I have is 3.2. So where does that fit in this article?
Comment from thorgal @ 2009/08/12
Quote:
Originally Posted by Mike89 View Post
I have a question. The article references two revisions of the Corsair ram, 2.1 and 2.2. I just purchased some of this memory (3 2 gig sticks kit) and the revision I have is 3.2. So where does that fit in this article?
Hi Mike, welcome.

Corsair has a way to number their revisions by the kind of chips used in the modules. A 2.X revision means we're talking Elpida based ram, 3.X revisions means a "Micron" chips based ram kit.

For a complete list look at the excellent ramlist from i4memory.com
Comment from Mike89 @ 2009/08/13
So are the 3.2 chips not as good as the 2.2 chips in the review?

That link doesn't really cover this memory.

On the sticks of my Corsair, everything is exactly written the same as in the picture shown on the review. The ONLY difference in wording is the revision number.
Comment from leeghoofd @ 2009/08/13
PN ends with V3.X stands for Micron IC's used... if they are better or worse you have to tell us... might be that yours scale good too with some added voltage... (be carefull though)
Comment from Mike89 @ 2009/08/13
Personally I don't plan on pushing it. I'm running an I7 920 @ 3.5 gig (with turbo on) running the ram at 700 @8-8-8-24. I'm quite happy with that as my system is fast, stable and cool. I'm a conversative overclocker, I want some extra speed but don't have a burning desire to see how far I can go before I burn it up! Heh heh. Going higher than what I'm getting most likely isn't going to give me anything more than I have that I would notice anyway except a lot of heat and shorter life. I'm not one of those guys that has to hit a certain number just to say I did. My overclock and subsequent system performance has put a big smile on my face and that's good enough for me.

I was just curious about the difference in the revision number and why the review didn't mention it being it was the same ram (just different revision).
Comment from leeghoofd @ 2009/08/13
Since you run them 200mhz below specs you can try to stabilise Cas 7-7-7-20, for that little bit extra...
Comment from thorgal @ 2009/08/13
As Leeghoofd said, at 700Mhz they should be able to do 7-7-7 timings.

And indeed, whether they are better or worse than an Elpida based kit I really can't predict. I tested Crucial ram in the roundup as well, you probably can compare them with your sticks, as Crucials are also micron based (they are a subgroup of Micron). Low voltage Micron parts are D9JNL chips, and in general Micron ram is known for scaleability with added voltage. Since Intel wants us to limit memory voltage to 1.65V on core i7, I didn't really test the scaleability of D9JNL yet beyond 1.65V.
Comment from Kougar @ 2009/08/17
What programs would you guys recommend and used for RAM stability testing for this article?
Comment from jmke @ 2009/08/17
I think Thorgal uses SP2004 Orthos, but not sure;
if you have a C2D or i7 system you could use linpack too
Comment from leeghoofd @ 2009/08/17
Don't say the linpack word, bah bah bah, it has proven over and over again that it needs a 64bit OS and there are loads of instabilities with these programs (usually shell and not core related) I can pass IBT or LinX and yet get a reboot or freeze on Hyperpi. One of the reasons we discovered B2B as it failed even a simple superpi yet passed the IBT's here ( maxmem selected )

A very good ram test for I7 : Hyperpi 32mb 8 threads. For all platforms also a good test is HCI memtest in windows.
Comment from thorgal @ 2009/08/17
I use the multithreaded wprime (you can see it in the screenshots) and occasionally hyperpi 0.99b for i7, as some motherboards dislike superpi 32M for unknown reasons.
Comment from jmke @ 2009/08/17
all depends on what you call "stable", if everything runs fine (apps, games) and only HyperPi crashes after several hours.... no big deal
Comment from leeghoofd @ 2009/08/17
Hyperpi only takes 20 mins max John, but is really a good way to stress the IMC and rams. My point is you can be linpack stable, but this doesn't warrant RAM stability...

The linpacks have been hyped due to the fact that ya CPU tends to get hotter than with prime or Orthos... For them stability issues some examples : With the v2.3 Intel Burn Test there's an error in the shell as it will always fail very high ram setting (4 - 8 thread setting), no matter if you OC or not... Linx had some issues too not correctly loading the cores...seems to be corrected with 6.02 release.

Indeed stability is personal, but in this respect of a certified ram test you need to run Memtest in Dos ( test 5 and 7 ), Prime blend or large, Hyperpi32 or HCI Memtest. Believe me linpack really doesn't cut it !!

For a total stability test try Folding at home. If anything is wrong with ya rig it will produce errors or will just reboot ya rig...
Comment from jmke @ 2009/08/17
or just run at stock speeds
Comment from thorgal @ 2009/08/17
I quit folding some time ago, but in the pre-i7 times folding was less taxing than prime in my book, so I started using prime. Anyway, when your rig is folding stable you can mostly sleep on both ears
Comment from leeghoofd @ 2009/08/17
And it's for a good cause no wasted cycles in silly test programs... but GPU folding is far faster these days...

Stock is for MAC users...
Comment from jmke @ 2009/08/17
Quote:
Originally Posted by leeghoofd View Post
Stock is for MAC users...
and people who care about integrity of their data
Comment from Kougar @ 2009/08/17
Quote:
Originally Posted by leeghoofd View Post
Don't say the linpack word, bah bah bah, it has proven over and over again that it needs a 64bit OS and there are loads of instabilities with these programs (usually shell and not core related) I can pass IBT or LinX and yet get a reboot or freeze on Hyperpi. One of the reasons we discovered B2B as it failed even a simple superpi yet passed the IBT's here ( maxmem selected )

A very good ram test for I7 : Hyperpi 32mb 8 threads. For all platforms also a good test is HCI memtest in windows.
Okay, I need help with the acronyms... B2B? And HCI Memtest?? Do you mean the memtest built into the Windows Vista/7 CD?

Thanks, I will give HyperPI a try!

Quote:
Originally Posted by thorgal View Post
I use the multithreaded wprime (you can see it in the screenshots) and occasionally hyperpi 0.99b for i7, as some motherboards dislike superpi 32M for unknown reasons.
Good info to know, I will try this one as well. Thanks!

Quote:
Originally Posted by jmke View Post
all depends on what you call "stable", if everything runs fine (apps, games) and only HyperPi crashes after several hours.... no big deal
Stable as in rock stable, stock stable, etc. Anything less than completely stable just leads to data corruption and unexplained errors.

Quote:
Originally Posted by jmke View Post
or just run at stock speeds
And waste a ton of performace? Is like buying a Tesla Roadster and capping it to 55mph. And I've seen stock systems still be unstable for various reasons, regardless.

Quote:
Originally Posted by leeghoofd View Post
And it's for a good cause no wasted cycles in silly test programs... but GPU folding is far faster these days...
One Core i7 920 folding four Linux SMP clients (each inside one of four 64bit Ubuntu Virtual Machines) takes about 19 hours for all four SMP programs to cycle once @ 4.2GHz. So say 20 hours = 4x1920 is 7,680 points every 20 hours. PPD results are better than many GPUs.

But yes, even better to use a GPU or two with the Core i7 as it doesn't affect performance much. This is partly why I am asking, been having nonstop issues with the GPU client, suspecting the GPU core or compute shaders on my GTX 260 might not stable, but have not been able to prove it. Could just be more F@H GPU code or CUDA driver issues...

Quote:
Originally Posted by jmke View Post
and people who care about integrity of their data
That's what my RAID'd NAS is for. Lost to much data to RAID arrays on my desktop, Intel's RAID is no protection at all.
Comment from leeghoofd @ 2009/08/17
B2B : Back to Back Cas 2 delay (think Asus, MSI and now gigabyte biosses got it)

Read up here : B2B investigated Massman Style

And here : B2B on Gene II

HCI Memtest : download here : hci memtest

I never went deep into the PPD stuff, but my 285GTX folds about 3-5 cores per nite... my 8 cores on the I7 can't keep up with that ! Maybe the CPU get's other heavier assignments, dunno. Not an expert on the matter
Comment from Kougar @ 2009/08/17
Ah, I had read that article on Back to Back CAS delay. Gigabyte has not released any BIOS's for the EX58-UD5 in months and I could not find a setting for it, was it under a different name? (I see they just released one, will update and check again).

Thanks for the links, will download the HCI program and play with it.

I'd recommend http://fahmon.net/ Just point it at the GPU directory and the CPU F@H directory to get an idea on your PPD figures... Two Windows SMP clients only fold half as much as four Linux SMP clients, Folding@home code is not optimized very well. Kinda hard to compare cores as the GPU folds quite a few different types, but your GPU should fall between 6,000 to 8,000PPD... same as my Core i7 920.
Comment from jmke @ 2009/08/17
Quote:
Originally Posted by Kougar View Post
And waste a ton of performace?
what TON are you referring to? 2.5% at the most; OCing memory doesn't improve performance noticeably...

Quote:
That's what my RAID'd NAS is for. Lost to much data to RAID arrays on my desktop, Intel's RAID is no protection at all.
not quite, if your system is unstable and you can't SAVE the data CORRECTLY, you can have all the RAID, NAS , SAN, etc in the world, it won't restore the corrupted DATA due to a SYSTEM crash and INSTABILITY.

choosing between ~2% "potential" performance increase, or ROCK STABLE, doesn't crash, doesn't bluescreen, always works, always saves data. IS A NO BRAINER
and to be honest, any Dual Core system at 3ghz is plenty fast for anything you throw at it; the only way at this moment to increase performance noticeably is SSD; that will make a difference;
OCing your CPU/VGA/MEM is not worth the risk of an unstable system, for a few % performance increase; In my opinion; to each his/her own
Comment from Kougar @ 2009/08/17
leeghoofd, nice catch! That new just released BIOS does unlock the B2B memory setting... doesn't say what it is set to by default though. I guess I'll be spending a bit of time playing with it.

JMke, you are assuming I was referring to OCing the memory, I am not. Not worth the time, headache, risks involved for the tiny gain. All I am doing is tightening the timings, but I wanted something a bit more accurate than Prime95 Blend to double check my settings.

A 1,560Mhz CPU overclock is "not a few % increase" especially when referring to programs that use all four cores / eight threads.
Comment from jmke @ 2009/08/17
that's why I finished my post with "in my opinion"; I have zero applications which use more than 2 cores, and the apps I do have, few need more than one core. I noticed no real difference between C2D E8200 running at stock vs one overclocked to ~4ghz; running two identical systems, side by side, one overclocked to 4ghz, the other at stock speed; I couldn't tell which one was which until you start up a CPU dependent benchmark.

while I agree with you that single threaded performance increases are the way forward to make our systems fasters, we have run into bottlenecks where a faster CPU does no longer overcome the slowdown, "in ye old days" one could overclock a 300mhz CPU to 600Mhz and the difference in overall computing would be night and day. Today, if we take a 2.66ghz CPU to 4ghz, there is an increase in performance... but nowhere near comparable to the past; we have reach a point of diminishing returns;

the single biggest factor at this point is storage, make that super speedy and you'll be able to feed that overclocked CPU with data it can actually do something with; your i7 @ 4.2Ghz is a monster, but with normal storage you're throwing peanuts at it, by switching to SSD/ramdisk you'll be providing it with a real meal and the outcome will be an eye opener.
Comment from Kougar @ 2009/08/18
Well it sounds like we mostly agree, even I can't notice if my system is at stock or if it is partly overclocked unless I start up an intensive program or several programs at once.

With things like VMWare and its four virtual machines I easily notice the smoother loading and responsiveness between stock and overclocked states. Game levels also tend to load faster even with a basic HDD, oddly I'm usually the first to load in 24 player TF2 servers or 8 player L4D games, but not more disk intensive games. Folding@home is only minimally (at best) affected by disk performance and to a slightly larger extent RAM performance, it's almost completely CPU / GPU bound.

That said an SSD upgrade has been on my list for that very reason, I've just been waiting for the new G2's to be stocked so I can get one for 20% off with Bing cashback. In any case, a system is only as fast as it's slowest bottleneck. It wouldn't make much sense to buy an SSD for a Pentium 4, or a system with 1GB of RAM. Just been waiting for the new 2nd gen drives and some price cuts.

---

I'll post what I found about Gigabyte's B2B in the appropriate thread. Memtest HCI seems to be stuck at a 2GB limit so three instances must be run simultaneously for it to work. I don't understand what Wprime does that Superpi and Hyperpi don't already?
Comment from Massman @ 2009/08/18
Wprime stresses all cores as you can manually set the number of threads.
Comment from leeghoofd @ 2009/08/18
So does Hyperpi Wprime is more of a raw CPU test, while Hyperpi and co stresses the rams and IMC more...

HCI memtest is limited inmax ram usage (think it's just the freeware edition) so yes I run several instances too and even attribute affinity for the cores... you run 2 2048 ones and one with rest of the unused memory ( auto )
Comment from Massman @ 2009/08/18
Wprime really stresses the cpu 100%, if HyperPi is just 8xSuperpi it doesn't do that.
Comment from jmke @ 2009/08/18
Quote:
It wouldn't make much sense to buy an SSD for a Pentium 4, or a system with 1GB of RAM.
you'd be surprised by the performance boost you'll get by equipping and older system with an SSD that maxes out the storage controller; instead of a slow PC, the system will become responsive and a pleasure to use again
Comment from leeghoofd @ 2009/08/18
@Massman : He wanted to have a good ram test PJ , not a CPU test !

@ Boss : And for the SSD, it's absolutely true, even when benching an old S478 Pentium 4 2.0Ghz, with Vista 32bit as OS and the SSD as boot drive, the load times are scary fast... SSD's also make laptops scream !!
Comment from leeghoofd @ 2009/08/18
@Massman : He wanted to have a good ram test PJ , not a CPU test !

@ Boss : And for the SSD, it's absolutely true, even when benching an old S478 Pentium 4 2.0Ghz, with Vista 32bit as OS and the SSD as boot drive, the load times are scary fast... SSD's also make laptops scream !!
Comment from Massman @ 2009/08/18
Yes, I answered to this question

Quote:
Originally Posted by Kougar View Post
I don't understand what Wprime does that Superpi and Hyperpi don't already?
Comment from Kougar @ 2009/08/18
Thanks a bunch for the info and the help everyone! I think I have a better idea of what all these programs are good for now.

Regarding the SSD... I'd have bought one last week during the Bing cashback sale if there were any in stock, but looks like Intel won't get them re-shipped back into etailers until the end of this month at the earliest.
Comment from jmke @ 2009/08/18
it will be the hardware piece which will give you the biggest speed boost this year

 

reply