In the midst of a global DRAM shortage, Digitimes reports that the market prices for graphics memory from Samsung and SK Hynix have increased by over 30% for August. This latest jump in memory prices is apparently due to the pair of DRAM manufacturers repurposing part of their VRAM production capacities for server and smartphone memories instead. As Digitimes’ sources report, this VRAM pricing is expected to increase further in September, impacting graphics card and gaming notebook manufacturers. Consumers have already felt the pain through skyrocketing DDR4 prices, and TrendForce/DRAMeXchange expects the upward trend of PC DRAM chips to continue to 2018.

Generally speaking, this production prioritization is not new. Late last year, the top three memory suppliers, Samsung (55% market share), SK Hynix (35% market share), and Micron (10% market share) shifted production capacity to prioritize servers and smartphones, causing the initial spike in PC DRAM prices. Overall, DRAMeXchange attributes the tight supply to lack of short term capacity expansion, as well as yield issues with new processes. The research firm had also noted that capacity expansion will be rather subdued as manufacturers try to keep commanding the higher margins of an undersupply environment.

In light of recent GDDR6 announcements by Micron and SK Hynix, these supply/price issues could have knock-on effects for both current and upcoming graphics cards. Additionally, as both Samsung and SK Hynix are the only HBM2 suppliers, HBM2-equipped cards may be adversely constrained by supply. Earlier this month, an SK Hynix executive stated that customers were willing to pay 2.5 times more for HBM2 over HBM1; this sentiment may soon be put to the test. The situation with Micron is a little less clear, as they not only have their unique GDDR5X memory, but also may not have raised VRAM prices. If they haven't, they may have an opportunity on their hands.

Related Reading:

Source: Digitimes

POST A COMMENT

55 Comments

View All Comments

  • JCB994 - Thursday, August 17, 2017 - link

    Many processes on a new device cannot be run on current toolsets. New devices, new processes, new tools. I worked in DRAM 20 years ago and went from 40nm to 19nm in 4 years. Photo tools, Etch chambers, CVD chambers, Diffusion tubes, etc all did multiple upgrades in that time. Production is lost in downtime, ramps, and lower yields until we figured out the process. Reply
  • ddriver - Thursday, August 17, 2017 - link

    Are you talking about a particular plant or a particular manufacturer? A particular plant is understandable if they had other facilities doing intermediate modes.

    Or maybe a scenario with a resource-constrained maker, that lacked the funds to go for a gradual update cycle. Thus exploiting old equipment for as long as possible, then replacing it with new equipment due to lack of additional fab space. But generally a very bad business model and not something that is practiced at those deep pocket multi billion dollar corporations.

    The point is nobody removes his primary product and money making equipment to replace it with untested tooling and process which are not production ready. The way big boys do it is they figure out the process first, and only then they begin to scale it up, either in new facilities or replacing production lines of secondary importance.

    There are additional indications that shortages are very much intentional. The US and its puppets seem to be VERY CONCERNED with China entering the market. If there is a legit shortage, then more production will be a very welcomed thing. Yet the CFIUS went out of the line just to make sure that China does not purchase SanDisk IP from WD out of fear that IP will jump start flash production in China. Micron and Hynix were also told to hold off on any licensing deals in efforts to prevent the introduction of another major (and perhaps most importantly non-US-puppet) flash maker.

    But the Chinese keep on pushing even without relying on the licensing on IP, and now we have all those analysts who spread gloom and doom how DISRUPTIVE it would be for China to enter the flash market with 100k wafers a month. But again, why is that a bad thing? I mean it is not like they are complaining about China filling in a bunch of shortages, including cheap labor, why is it so important to protect the flash shortage?

    I mean unless it is deliberate, and China offering enough supply to meet demand is not utterly detrimental to the plan of the big boys who engineered this shortage to reap profits on it. I mean they did lost money staging the oversupply that was necessary to justify the shortage and ridiculous price jacking. China entering the market would be a big blow to that plan.
    Reply
  • CaedenV - Thursday, August 17, 2017 - link

    I have worked in warehouses before. Big as they are, floor-space can be limited, especially in established businesses. Starting a new production line often means tearing something out to put something else in. This means decreased production on everything before the increase in production on the new thing hits. It is all logistics.

    Plus, with as volatile as the phone market has made chip production, it makes things very difficult to choose what to make. A new Sammy or Apple phone launch means a large ramp-up in production in certain chips, followed by a sharp drop-off in demand after the product hits shelves. So just how much floor space do you convert to these temp bursts in demand, compared to the stable day-to-day demand for other products? It is all time v money v space. No easy decisions, and a miss-step can mean loosing money on every chip produced if the market falls out from under you.
    Reply
  • ddriver - Thursday, August 17, 2017 - link

    It is quite simple really, use idle inventory as buffer and to hide the capacity penalties of refitting the output of a production line.

    A maxim I generally abide to - "you don't have enough until you have left-over". And it is not eggs, it won't spoil if it sits around in a few weeks, nor is it unique stuff that cannot find a target market if the original intended target has been saturated.

    It is much, much faster to change what a production line makes than it is to upgrade tooling.
    Reply
  • ddriver - Thursday, August 17, 2017 - link

    And just to clarify, working in the warehouse, or in logistics, or pretty much in 99.99% of the industry doesn't really grant anyone extra insight into the actual reasons and motivations behind executive decision making.

    So "I worked and the industry and therefore know better" is kind of a rather moot point.
    Reply
  • Samus - Thursday, August 17, 2017 - link

    That all depends. Apparently you didn't collude with your competitors...so you just weren't doing business properly! ;) Reply
  • Stochastic - Thursday, August 17, 2017 - link

    The days of affordable computers may be coming to an end temporarily. What a shame. Reply
  • Adramtech - Thursday, August 17, 2017 - link

    Not true, prices per bit are falling drastically. Reply
  • Impulses - Thursday, August 17, 2017 - link

    DRAM prices has always been quite volatile, don't get so hung up on any 1-2 year period or even 3-4 years, it's always yo-yo'd like crazy... Frankly the current price hikes are still nothing compared to what's gone down in the not too distant past. Reply
  • milli - Thursday, August 17, 2017 - link

    This. Around 16 years ago there was a huge price hike. Memory prices went up more than two fold, even three fold. The current price increase is nothing compared to that. Reply

Log in

Don't have an account? Sign up now