Memory Market Overview: May 2005

What has been happening in the memory market lately? We are going to discuss the situation with the memory prices, the changes in the market shares, new established joint projects and of course, new products that might be of interest to hardware enthusiasts and overclockers, as well as to ordinary users.

by Anna Filatova
05/11/2005 | 02:30 PM

When something bad has been happening for a long time this tendency should end sooner or later. This is exactly what happened in the beginning of April, when the players in the memory market evaluated the current situation and arrived at the conclusion that this is the lowest the memory could drop. Therefore, considering that there is no room for further price reduction, they stated the upcoming stabilization in the memory market. So, the first week of the past month started with the bull speculation. As a result, the price for 256Mbit DDR400 chip grew up from $2.43 to $2.55. And that was it: the strategic situation in the memory market with pretty low demand didn’t change that much. Moreover, the manufacturers will definitely do their best to stimulate the demand growth keeping the prices at the minimum acceptable level at the same time.


So, no wonder that in the second week of the month the prices moved back to $2.4 per chip. Then the price drop continued for another while until it hit $2.34, but that was mostly inertial movement, I should say. It is true: the $2.4-$2.3 for a single 256Mbit DDR400 chip seems to be the bottom for the today’s market. So, it is not at all surprising that the memory prices stabilized around $2.35-$2.39 for a chip like that, as this price point is closely connected with the production cost of these chips.

As a result, it looks like the prices will hardly fall really far below this point, but as for the growth, it is not just possible: it is inevitable. Back-to-school season is approaching and this year I assume it will play its role in the market situation, so that the prices will finally grow up in Q3. So, this seems to be the best time to upgrade your system memory if you need to. No matter what but contemporary mass applications (operating systems and office packages in the first place) still do not require more than 512MB of RAM anyway.

Of course, this situation has its influence on the monthly results of the memory manufacturers. Summing up the results of the Q1 2005, Nanya alone indicated that their average sales price for memory chips dropped down by 24%. Of course, there are still two ways to improve sales revenues: you either reduce the production cost, or resort to some trick to raise the average price back up.

Just like its competitors, Nanya is trying to move in two directions at the same time. The growing yields and production cycle improvements that allow minimizing the production time stimulate the reduction of the production cost of the memory chips. At the same time, the average sales price should get higher because the share of more expensive DDR2 memory in the company’s sales volume is growing, too. Nanya is traditionally super-optimistic here: they believe that in Q3 already this memory type will win 60% of the entire memory market.

However, Nanya certainly does have something to back up these high ambitions: the company has been remaining the leader of the Taiwanese market for a long time already. For instance, in Q1 2005 the leader in the Taiwanese market with the highest sales revenues appeared Inotera enterprise – a joint venture of Infineon and Nanya. So, Nanya managed to get its well deserved half of the revenue and demonstrate very attractive business results in general.

In fact, I would partially agree with Nanya’s opinion on DDR2. Of course, it is highly unlikely to reach 60% market share in Q3 (maybe it will reach 60% of the Nanya’s sales), but the tendency has every right to exist. Samsung’s demarche is also likely to push it this way. As you probably know, they dropped the price of their DDR2-533 memory modules by 20% in the beginning of April, so that it hit $52 per unit. As a result, the price of the chips used to manufacture the above mentioned memory modules also went down.

The remarkable thing is that about a week later, in the middle of April, many other manufacturers reduced the prices of their DDR2-533 512Mbit chips, too. I think that this fact can be regarded as a clear indicator that a lot of large companies in Japan, Korea and other countries managed to reach pretty high yields with the current 0.11micron production technology. As we know, 90nm process is coming in next: in the end of the month Elpida announced the beginning of pilot production of new 90nm DDR2 chips, with the mass manufacturing scheduled for Q3 2005.

Keeping in mind that they will manage to achieve acceptable production yields by the end of the year at best, and that most competitors in this market make progress at about the same pace, we can hardly hope to see some real effect from the introduction of the new 90nm production technology this year. Although end-users do not have anything to worry about: the scale will still balance in favor of the supply, rather than demand. Moreover, the manufacturers still have the whole lot of opportunities to polish off the current 110nm process even more, which should definitely affect the production cost as well. For example, today Powerchip provides the production yields of about 50-60%, which should reach 82-85% by the end of the year, according to the company.

So, the DDR2 prices should keep going down throughout the entire year. The possible price growth in Q3, which we have just mentioned, will not be dramatic and will simply serve as an additional stimulator for the growing DDr2 market share.

Winding up the discussion of the memory prices, I would like to say the following. The remarkable thing is that in the end of the month Hynix officially admitted that they participated in the agreement between all the major memory market participants aimed at locking the memory prices at a certain level. In fact, if you have been following our memory reports, you should remember that Hynix has been accused of this for quite a while, which has almost led to bankruptcy. The other “plotters” were: Micron, Samsung, Infineon… Samsung seems to remain the only one right now who hasn’t yet suffered any financial losses because of that case, although last year they have already reserved about 100 million dollars for this matter. All other companies have already been fined. And in April Hynix had to pay 185 million: they were pronounced guilty.

But why should we keep talking about the sad stuff? Hynix also had some good news to share. To begin with, they finally started putting into life the plans that have been in discussion since the early 2004: they started the construction of a 200/300mm fab for DRAM and flash memory manufacturing in China, a joint venture of Hynix and STMicroelectronics. The Koreans will own 2/3 of this business. And since Hynix likes dumping and does it really well, next year they will acquire one more powerful instrument for that.

By the way, another alliance has also shaped up by now: the Japanese Elpida and the Taiwanese Powerchip. The Japanese got presence in the Powerchip’s board of directors and at the same time increased their share in this company. Hopefully, this thing will not end up in scandals like the cooperation of the Infineon and ProMOS. By the way, since we came to speak about Infineon. In April the company announced the production of 512Mbit 800MHz GDDR3 samples with the intention to start their mass production in H2 2005.

The Germans continued doing one of their favorite things this month: fitting as many chips onto a single PCB as possible. This time they were experimenting with the memory modules for notebooks. Infineon demonstrated 2GB SO-DIMM and 1GB MicroDIMM modules. In fact, this is a great approach from the power saving point of view, especially, since using only one PCB for all memory chips allows saving a lot of power consumed by the memory subsystem. According to Infineon, they could save up to 30%! And we all know that any type of “saving” is a much more important thing for notebooks than performance.

Well, little by little we have come to the memory modules discussion. Most of the news here is connected with the memory solutions based on the recently introduced 800MHz DDR2 chips from Elpida. This time Corsair announced their product: XMS2 6400. And they not just announced it but also released into the mass market - at least we could see them selling in Japanese stores in the middle of April already for about $380 for a set of two 512MB modules. Of course, the price is kind of extreme, but these memory modules are targeted for extreme users anyway, as they cannot be used in any today’s system working in the nominal mode anyway.

The April announcement of the Corsair XMS2 5400UL – 675MHz DDR2 memory modules looks much more adequate, I should say. They are intended for use in NVIDIA nForce4 SLI Intel Edition based platforms because they do support this speed officially. Moreover, the timing settings are also more adequate here: 3-2-2-8.

However, the level of adequacy of the today’s solutions is inversely proportional to the DRAM chips pricing, so… Another April solution joined the HyperX memory modules from Kingston: the 750MHz of the PC2-6000 pseudo-standard. A pair of 1GB modules with 4-4-4-12 timings sells for $498. No comments necessary, I assume. There is also not much I could comment on regarding the new Kingmax product line aka Mars, where not very fast timings (to put it mildly) correspond to DDR2-667 already…

Of course, all these announcements have to do with the nForce4 SLI Intel Edition launch, because otherwise these products would have hardly had any market at all. OCZ couldn’t stay aside when things like that were going on around them, although they approached the matter in a more global way. They extended their Gold product family (lower latency solutions) onto DDR2. For the first time OCZ announced DDR2 Gold Series Modules for all performance segments: from 533MHz to the notorious 800MHz. We wouldn’t call the timings really low, but it would be strange to expect something else at these speeds. 667MHz modules support 4-4-4-8, while 800MHz ones – 5-5-5-8.

Corsair XMS2 5400UL seems to look the most attractive of all here. Although OCZ wouldn’t be themselves, if they hadn’t thought of a way to stick out somehow. In April they announced their remarkable PC-5000 EL Platinum modules designed to work with DFI mainboards on nForce4 chipset. This was quite an interesting move, I should say, especially since it is becoming a pretty frequent situation that a company announced memory modules that are guaranteed to work with only one selected mainboard they were tested it. I sincerely hope though, that this tendency will not become overwhelming in the near future.

In conclusion to our memory module announcements discussion I would like to mention a totally different type of products: 4GB Registered PC2-3200 DDR2 RDIMM. These solutions were announced by Micron in the beginning of April. 36 memory chips fit into a single module (with the help of Stakpak and Staktek brand name technologies) allow equipping your system with up to 16GB of RAM. This is a very acute product for server systems in the first place, where the requirements to memory subsystem grow up pretty fast, unlike the desktop segment.

And in the end of our monthly coverage I have to say a few words about our usual hero: Rambus Company. After the fiasco they suffered in late 90s, this company rose from the ashes just like a proper phoenix. Only in April the company managed to sign a long-term license agreement for DDR and DDR2 standards with NEC (which is actually not a very big surprise, as NEC has always been known for very loyal attitude to Rambus), to convince three courts to continue trials against memory manufacturers who wanted to terminate them, and to even launch a new technology. Here I am talking about something similar to processor Hyper-Threading, the so-called “micro-threads”. The idea is actually very clear from the mere name of this technology already: it is about more effective use of the bus bandwidth and faster access to the chip core.