DDR2 prices fell 60% in four months
Monday 23rd April 2007, 11:11:00 AM, written by Arun
DDR2 memory, which is the current mainstream PC memory standard and also widely used by low-end GPUs, is facing significant pricing pressure and the dynamics at play seem unlikely to be changing anytime soon. Between this, the NAND Flash collapse and the plummeting microprocessor prices, it should come as no surprise that some analysts aren't expecting much growth in the semiconductor industry this year.
The latest DRAM and NAND quotes are available at DRAMeXchange, although sadly historical data is only available for paying subscribers. Basic data for DRAM is given in this news piece's first link, and NAND historical prices since January 2006 can be found in this View from Silicon Valley article. The primary reason behind the recent DRAM price declines can also be seen in the EETimes article quoted in that piece: "To some degree, three big NAND suppliers—Samsung, Hynix and Micron—have separately scaled back their fab production targets, moving a big chunk of their respective capacity to more-profitable DRAM."
Basically, since NAND was weak, a substantial chunk of capacity was moved to DRAM. The good news with that strategy is that NAND prices have now recovered slightly, although they're still lower than in December. The bad news is that DRAM prices have collapsed even more rapidly than NAND prices improved, as the market demand could not come close to the newly increase supply (so much for the 'Vista Effect'!). That's why makers tried to shift part of their capacity to different kinds of flash (NOR, NAND SLC, etc.) instead but looking at the DRAM market today, that obviously wasn't enough.
So, what's happening now? Shock, horror, Samsung is moving some capacity back to NAND. With the prices of the higher-end capacities going up by more than 40% since February and new process nodes ramping up, that's not very surprising. However, NAND pricing remains relatively weak, and this is especially problematic for some of Samsung's competitors who are on less advanced process nodes and will thus have to sustain even lower margins. At the same time, many companies keep investing in aggressive capacity expansion plans, so long-term demand growth doesn't seem like it'll change anything.
So, how does all of this effects GPUs and logic chips in general? For obvious price reasons (GDDR3 is a fair bit more expensive than DDR2), this is the kind of memory used in most low-end (sub-$99) GPUs, and as such, it should positively affect the prices in the lower-end parts of the market. Vista Premium basically forces discrete GPUs to have 128MiB+ of memory, and at the current pricing levels, that would cost less than $5.
Arguably, this could reduce the pricing gap between IGPs and ultra-low-end discrete GPUs, although the impact probably won't be too significant. The lower level of integration and the extra PCB remain important cost considerations, and TurboCache/HyperMemory GPUs with as little as 32MiB of memory didn't really kill the IGP market, now did they?
As for logic chips in general (that is, basically all non-memory chips), it will be interesting to see how this affects their pricing power in certain markets. Arguably, if the goal is to have a complete bill of materials (BOM) of less than $25, and the price of the memory goes down, then the relative revenue proportion of the logic chip(s) can increase. This principle might also apply to handheld music/video players with NAND Flash if consumer demand for higher capacities doesn't go up as fast as flash prices go down.
The latest DRAM and NAND quotes are available at DRAMeXchange, although sadly historical data is only available for paying subscribers. Basic data for DRAM is given in this news piece's first link, and NAND historical prices since January 2006 can be found in this View from Silicon Valley article. The primary reason behind the recent DRAM price declines can also be seen in the EETimes article quoted in that piece: "To some degree, three big NAND suppliers—Samsung, Hynix and Micron—have separately scaled back their fab production targets, moving a big chunk of their respective capacity to more-profitable DRAM."
Basically, since NAND was weak, a substantial chunk of capacity was moved to DRAM. The good news with that strategy is that NAND prices have now recovered slightly, although they're still lower than in December. The bad news is that DRAM prices have collapsed even more rapidly than NAND prices improved, as the market demand could not come close to the newly increase supply (so much for the 'Vista Effect'!). That's why makers tried to shift part of their capacity to different kinds of flash (NOR, NAND SLC, etc.) instead but looking at the DRAM market today, that obviously wasn't enough.
So, what's happening now? Shock, horror, Samsung is moving some capacity back to NAND. With the prices of the higher-end capacities going up by more than 40% since February and new process nodes ramping up, that's not very surprising. However, NAND pricing remains relatively weak, and this is especially problematic for some of Samsung's competitors who are on less advanced process nodes and will thus have to sustain even lower margins. At the same time, many companies keep investing in aggressive capacity expansion plans, so long-term demand growth doesn't seem like it'll change anything.
So, how does all of this effects GPUs and logic chips in general? For obvious price reasons (GDDR3 is a fair bit more expensive than DDR2), this is the kind of memory used in most low-end (sub-$99) GPUs, and as such, it should positively affect the prices in the lower-end parts of the market. Vista Premium basically forces discrete GPUs to have 128MiB+ of memory, and at the current pricing levels, that would cost less than $5.
Arguably, this could reduce the pricing gap between IGPs and ultra-low-end discrete GPUs, although the impact probably won't be too significant. The lower level of integration and the extra PCB remain important cost considerations, and TurboCache/HyperMemory GPUs with as little as 32MiB of memory didn't really kill the IGP market, now did they?
As for logic chips in general (that is, basically all non-memory chips), it will be interesting to see how this affects their pricing power in certain markets. Arguably, if the goal is to have a complete bill of materials (BOM) of less than $25, and the price of the memory goes down, then the relative revenue proportion of the logic chip(s) can increase. This principle might also apply to handheld music/video players with NAND Flash if consumer demand for higher capacities doesn't go up as fast as flash prices go down.
Tagging
financial ± dram, nand, ddr2, pricing
