The Inquirer-Home

Memory vendors fight for the market

Computex 2010 The bandwidth, latency, power and capacity battle
Thu Jun 10 2010, 14:59

AFTER A PROTRACTED MIGRATION from DDR2 to DDR3 due to the time needed for both Intel's Nehalem and AMD's AM3 processor platforms to take off, at least we have a clear standardisation on DDR3 RAM for the next year or two, until DDR4 comes along.

The problem is that, with so many memory module vendors around, and with basically the same few memory chip providers, how much can each module brand differentiate its products? Keep in mind that some of the memory chip vendors like Micron have their own module brands - Crucial in Micron's case - which of course will have preferential in-house treatment to a certain extent.

Well, besides the branding effort, which isn't our focus in this story, there are four key features where the user benefits are clear enough to categorise the modules. First, of course, is capacity, but the module vendor can do little about it, as it is really dependent on the chip makers' generational rollouts. Yes, it does help to get the new ones first and have the first mover advantage, but that doesn't last long usually.

Right now, the dominant modules in performance desktop and mobile markets are 16-chip 2GB sticks, composed of 1Gb dies. However the 16-chip 4GB modules using 2Gb dies are finally making inroads with reduced prices. The vendors who focus on the improved prices here have good sales prospects, since there are users who would like to have 8GB in a notebook or 16GB on their desktop, or 24GB if on a triple-channel Core i7 900 series processor board.

Second, the declared frequency and theoretical bandwidth come into play. DDR3-2500 sounds much better than DDR3-2000, even though any real performance improvements are marginal right now, simply because increasing the other factor, latency, negates the speed-up. But, from a marketing point of view, this is important. Of course, a DDR3-2500 CL9-9-9 module will be noticeably faster than a DDR3-2000 CL9-9-9 one, but try to find the first one for now, and, if you find a pair, find a P55 mainboard which will run them at that speed reliably for the long term, no mean feat.

Third comes latency. Often ignored in previous years, module latency has become possibly the most important performance parameter for DIMMs, as the bandwidth simply has become more than good enough. The new CPUs from both Intel and AMD, with their large caches and tuned pipelined memory controllers sensitive to latency, run overall faster with DDR3-1600 CL6-6-6 memory than with DDR3-2000 CL8-8-8 modules. And yes, the past few Computex shows did focus a lot on this.

Finally, there is power. Saving it, to be precise. Low voltage DIMMs, such as the 1.2 volt GEIL and 1.35 volt G.skill and Crucial offerings we've reviewed over the past few months are all the craze now, as not only do they save some power, but they also reduce heat sufficiently to completely dispose with those heat spreader contraptions, in both desktop and server realms.

Only then do you come to things like PCB design, heat spreader quality and fashion, packaging, warranty and of course, branding and price. Keep in mind that the PCB design can help overclock the DIMMs or, lets just say, push up the bandwidth, by over 10 per cent on its own. However, heat spreader usefulness, despite fancy designs and 'serious' looks, is questionable, with good airflow over bare DIMMs having proven to be more effective in quite a few cases.

At Computex, the efforts by the companies to figure out their 'differentiation' were obvious. Geil, for instance, combined high performance memory modules with some unique configuration combinations. One of these was a six channel DDR3 bundle on the EVGA SR2 mainboard including a 24GB high capacity version with six 4GB DIMMs, as well as Geil's own line of power supply units with patented LCD power and voltage monitoring capabilities. According to its team, this unique innovation is very useful not just for computer publication reviews, but also for general users to know how much power their computer really uses, and how stable the voltages are.

geilevga

On the other hand, G.skill offered some record breaking memory speed, but with unusual latency numbers: 8-11-8 isn't exactly same as 8-8-8, by the way. Gskill has always been known for setting latency records for many years now, however it seems that with the recent DDR3 offerings it does have real competition from others. Nevertheless, there were good G.skill Trident DIMM results - 8-11-8 at DDR3-2500 on an Asus P55 with Coolermaster cooling - from a company which has also diversified into performance SSDs.

gskillcomputex

Other vendors in Taipei focused on other ways of making a difference. While OCZ increasingly focuses on unique SSD solutions, including a compact consumer multi-channel PCIe x4 SSD card, Team and A-Data brought out plenty of USB3-based sticks and storage devices.

In summary, there is still some space for differentiation in the memory module business, even as the vendors diverge to other lines such as SSDs or PSUs for extra revenue or recognition. As the new generation of memory dies should allow even lower latencies, such as CL 6-6-6 at DDR3-1800, as well as true DDR3-1600 4GB 16-chip modules, all at ever lower voltages, the DIMM vendors can play to their own unique strengths using the new dies for the new product lines.

We'll cover more of the new memory modules - especially the low latency, low voltage, and high capacity versions - in upcoming reviews. µ

 

 

Share this:

Comments
fight? yeah right

If by "fighting" you mean "price fixing", especially when it comes to 4GB and larger memory modules, then yeah.

posted by : Leroy Jenkins, 11 June 2010 Complain about this comment
Corection on Speed....

In sony 3D article stated DDR ladder wrong. say have 200 Mhz/s core. then ddr is 400 mhz. ddr2 added second channel, which back slide to ddr, ddr then could use cheaper memory & ddr2 doubles 400 Mhz/s as two channels running at 400 Mhz/s, by hammering on between each other.

ddr3 is 200 Mhz/s times three or 800 Mhz/s on each channel, now true ddr3 with triple channel is coming in or 2400 Mhz/s, yet if core is slower or don't have triple channel, actual speeds are less. DDR4 would be 1,000Mhz/s per channel or only 25% more than ddr3.,000 Mh/s. Boosting Core Might be better increase. You can have as many channels as want, however to limit of controller, main & cpu, in all cases.Point ddr4 just isn't enough,for cost. ddr5 ddr5 be 200+200+200+200+200+200=1,200 Mhz/s then add channels. 2=2.4Ghz/s 3= 3.6 Ghz/s 4=4.8 Ghz/s or good jump with existing channel tech. However after ddr5, New memory that actually sustains faster speed on own will be coming in. so ddr5 might be TOP. Say Core speed of 400 Mhz/s, actualy, might be more ddrs' EG ddr6,7,8., as stomuch to increase, yet finally sandbox controllers will reach limit, too. Outer Limit. SSD Pci-e RAID50 & Dual Channel?While core of Over 1 Ghz/s be ALL New Ballgame.

drashek

posted by : Memory?, 11 June 2010 Complain about this comment
four ddr3 sodimm memory slots

I would love to see notebooks or laptops with 4 ddr3 sodimm memory slots or more. Currently there are only a handful which can take 16gb memory i.e. 4gb x 4 sodimms. Now that 64 bit OS's are gaining traction, more memory really helps. It will not cost much to notebook manufacturers to implement but will go a long way for customers. Also USB 3.0 and hdmi connectors and screen which can show full 1080p resolution is desireable.

posted by : Sam, 10 June 2010 Complain about this comment
Enthusiast is a marketing term for gullible

Let's face it: the returns from investing in "enthusiast" high performance memory are slim and that's where marketing makes a difference. Like Shell and BP will spend millions to convince the public that their brand of gas is better for your engine, memory makers use every trick they can to market what is ultimately a commodity that has only a relatively minor impact on your system's performance.

posted by : tech6, 10 June 2010 Complain about this comment
aboutus
Advertisement
Subscribe to INQ newsletters
Advertisement
INQ Poll

Authorities in several countries raided Megaupload recently, shut down all of its services, seized hundreds of servers and arrested several of its executives on criminal charges.

Do you think the move was justified?