Why Nvidia graphics cards also sold out in first couple weeks?

Marketing ploy? The new 980 Ti is just released and all versions are sold out now on all channels including nvidia.com, amazon and neweggs. This doesn’t make sense, if your selling something, why don’t you make sure you have enough supply in the launch date? This happened with 980 and 970 too. Products like Apple never have this kind of issue. I am wondering if this is some kind of “hunger marketing”?

Constant pressure to get their product to market before a competing product hits.

Apple has less competition issues as most Apple users won’t drop the brand because Samsung put out a newer widget.

Gamers looking to be on the cutting edge don’t have the same loyalty. If AMD’s newest greatest card hit the market before Nvidia’s you can count on Nvidia’s sales taking a hit because of it.

As soon as video card manufacturers are confident their card is going to work in the current landscape they rush to get it to market. They’ll make more as soon as they can.

Their production may be lagging, or they may be simply releasing product slower than demand would otherwise dictate. Either way they can maintain a high price point, and high demand, versus a glut of product and its associated discounting.

Remember they are fulfilling worldwide demand. Production and shipping of so many units takes a lot of time. They’d rather get the early crowd and at a premium while they can and then back-fill the orders.

The Apple Watch sold out before it shipped.

It’s not sold out; they just don’t have any stock yet.

The 980Ti is a high end graphics card.

Usually its important to get the NEW card out as fast as possible, to sell the smaller cards.

Only a small percent of customers usually buy the high end cards anyhow.

There are multiple reasons, and some good answers so far.

Ultimately, though, what it comes down to is that big GPUs are hard. They have a huge die area; far bigger than anything in a phone, and bigger than anything else in a PC. In fact, the largest GPUs are at the “reticle limit”, which is basically the size of the lens that projects the die pattern onto the silicon wafer. You can’t get larger than that.

Yields–the fraction of dies that function–tend to be low in the beginning. There are lots of defects, and because the dies are so big, they are much more likely to be affected by a defect. These issues tend to be ironed out over time, but it takes a while.

So there’s a balance between shipping before the competitors vs. having enough product to meet demand. GPU makers do build up some supply, but it’s often not enough. They do better on the low end SKUs, partly because they’re smaller (and less sensitive to defects), and partly because some of the issues that cropped up on previous high-end SKUs have been resolved.

Incidentally, GPU makers are able to salvage some defective dies by disabling bad subunits. Both the 970 and 980 Ti do this. But there are limits to it, and supply is still constrained early on.

Aren’t GPUs used for mining bitcoins? Perhaps the bitcoin miners snapped up the entire supply.

Not anymore. Miners use custom ASICs now. At one point, there was a bit of a shortage of AMD products due (possibly) to miners snatching them up, but that isn’t a problem now.

They are still using optical projection for these?

I remember when the photographic world went nuts over the chip makers’ switch to purely digital etching and needing the space the old optical units occupied.
Legend has it that the basic deal was: You want it? Take it - But get it OFF the property.
The reason for the excitement?

Those lenses would cost at least $10K to duplicate, and their focal lengths were good for large format (where lenses start at $2K).

Yeah. No other way to get the detail they need at reasonable speeds. The projection takes place under water to increase the effective numerical aperture. And they use multiple masks for each layer, which (with the right patterning) can resolve features below the wavelength of light. They use a 193 nm laser but achieve features as small as 14 nm.

There have been efforts to achieve higher resolution more directly. One is to use a shorter-wavelength laser. But going lower than 193 nm requires new, even more expensive lenses. It hasn’t worked out so far. Another approach is to use an electron beam scanned across the surface, and dispense with lenses completely. But this is slow (since it’s scanline based instead of all at once), and the high energy beam damages the silicon.

In the short term, 193 nm lasers will continue to work. They still have a few remaining tricks. It’s not clear what will happen past around 10 nm, though.