Blhaflhvfa.

There’s a lot to talk about with regards to NVIDIA and no time for a long intro, so let’s get right to it.

At the end of our Radeon HD 5850 Review we included this update:

“Update: We went window shopping again this afternoon to see if there were any GTX 285 price changes. There weren't. In fact GTX 285 supply seems pretty low; MWave, ZipZoomFly, and Newegg only have a few models in stock. We asked NVIDIA about this, but all they had to say was "demand remains strong". Given the timing, we're still suspicious that something may be afoot.”

Less than a week later and there were stories everywhere about NVIDIA’s GT200b shortages. Fudo said that NVIDIA was unwilling to drop prices low enough to make the cards competitive. Charlie said that NVIDIA was going to abandon the high end and upper mid range graphics card markets completely.

Let’s look at what we do know. GT200b has around 1.4 billion transistors and is made at TSMC on a 55nm process. Wikipedia lists the die at 470mm^2, that’s roughly 80% the size of the original 65nm GT200 die. In either case it’s a lot bigger and still more expensive than Cypress’ 334mm^2 40nm die.

Cypress vs. GT200b die sizes to scale

NVIDIA could get into a price war with AMD, but given that both companies make their chips at the same place, and NVIDIA’s costs are higher - it’s not a war that makes sense to fight.

NVIDIA told me two things. One, that they have shared with some OEMs that they will no longer be making GT200b based products. That’s the GTX 260 all the way up to the GTX 285. The EOL (end of life) notices went out recently and they request that the OEMs submit their allocation requests asap otherwise they risk not getting any cards.

The second was that despite the EOL notices, end users should be able to purchase GeForce GTX 260, 275 and 285 cards all the way up through February of next year.

If you look carefully, neither of these statements directly supports or refutes the two articles above. NVIDIA is very clever.

NVIDIA’s explanation to me was that current GPU supplies were decided on months ago, and in light of the economy, the number of chips NVIDIA ordered from TSMC was low. Demand ended up being stronger than expected and thus you can expect supplies to be tight in the remaining months of the year and into 2010.

Board vendors have been telling us that they can’t get allocations from NVIDIA. Some are even wondering whether it makes sense to build more GTX cards for the end of this year.

If you want my opinion, it goes something like this. While RV770 caught NVIDIA off guard, Cypress did not. AMD used the extra area (and then some) allowed by the move to 40nm to double RV770, not an unpredictable move. NVIDIA knew they were going to be late with Fermi, knew how competitive Cypress would be, and made a conscious decision to cut back supply months ago rather than enter a price war with AMD.

While NVIDIA won’t publicly admit defeat, AMD clearly won this round. Obviously it makes sense to ramp down the old product in expectation of Fermi, but I don’t see Fermi with any real availability this year. We may see a launch with performance data in 2009, but I’d expect availability in 2010.

While NVIDIA just launched its first 40nm DX10.1 parts, AMD just launched $120 DX11 cards

Regardless of how you want to phrase it, there will be lower than normal supplies of GT200 cards in the market this quarter. With higher costs than AMD per card and better performance from AMD’s DX11 parts, would you expect things to be any different?

Things Get Better Next Year

NVIDIA launched GT200 on too old of a process (65nm) and they were thus too late to move to 55nm. Bumpgate happened. Then we had the issues with 40nm at TSMC and Fermi’s delays. In short, it hasn’t been the best 12 months for NVIDIA. Next year, there’s reason to be optimistic though.

When Fermi does launch, everything from that point should theoretically be smooth sailing. There aren’t any process transitions in 2010, it’s all about execution at that point and how quickly can NVIDIA get Fermi derivatives out the door. AMD will have virtually its entire product stack out by the time NVIDIA ships Fermi in quantities, but NVIDIA should have competitive product out in 2010. AMD wins the first half of the DX11 race, the second half will be a bit more challenging.

If anything, NVIDIA has proved to be a resilient company. Other than Intel, I don’t know of any company that could’ve recovered from NV30. The real question is how strong will Fermi 2 be? Stumble twice and you’re shaken, do it a third time and you’re likely to fall.

Chipsets: One Day You're In and the Next, You're Out

Presently, NVIDIA’s chipset business is far from dead. They are in nearly every single Apple computer on the market, not to mention countless other OEMs. I’m not sure how much money NVIDIA is making from these chipsets, but they are selling.

NVIDIA won Apple's chipset business, Intel was not happy

Long term I don’t see much of a future for NVIDIA’s chipset business. NVIDIA said that they have no interest in pursuing an LGA-1156 chipset given Intel’s legal threats. Even if NVIDIA had a license to produce DMI chipsets, I’m not sure it makes sense.

NVIDIA's Advantage: A single chip GeForce 9400M instead of a dated Intel solution

Once the ‘dales hit, every single mainstream CPU from Intel is going to come with graphics on-package. Go out one more generation and Sandy Bridge brings the graphics on-die. AMD is doing the same thing starting in 2012.

It’s taken longer than expected, but there’s honestly no need for a third party chipset maker anymore. Most of the performance differentiation in chipsets has been moved onto the CPU die anyway, all that’s left are SATA, USB, and a bunch of validation that no one likes doing. NVIDIA is much better off building a discrete GeForce 9400M GPU at low cost and selling that. There’s much less headache involved with selling discrete GPUs than selling chipsets, plus graphics is NVIDIA’s only value add when it comes to chipsets - everyone knows how to integrate a USB controller by now. I’d say the same about SATA but AMD still has some AHCI silliness that it needs to sort out.

NVIDIA committed to supporting existing products in the channel and continues to poke fun at AMD with lines like this:

“On AMD platforms, we continue to sell a higher quantity of chipsets than AMD itself. MCP61-based platforms continue to be extremely well positioned in the entry CPU segments where AMD CPUs are most competitive vs. Intel”

As successful as NVIDIA’s AMD chipsets are today, AMD is telling us that nearly all OEM designs going forward use AMD chipsets. Again, NVIDIA’s chipset business is quite healthy today, but I don’t see much of a future in it - not that it’s a bad thing.

The only reason NVIDIA’s chipset business has lasted this long is because AMD and Intel couldn’t get their houses in order quickly enough. AMD is finally there and Intel is getting there, although it remains to be seen how well the next-generation of Atom platforms will work in practice.

A pair of Ion motherboards we reviewed

The main reason Ion got traction in the press was because it could play Blu-ray content. If Intel had done the right thing from the start and paired Atom with a decent chipset, NVIDIA would never have had the niche for Ion to fit into.

Final Words

Is NVIDIA in trouble? In the short term there are clearly causes to worry. AMD’s Eric Demers often tells me that the best way to lose a fight is by not showing up. NVIDIA effectively didn’t show up to the first DX11 battles, that’s going to hurt. But as I said in the things get better next year section, they do get better next year.

Fermi devotes a significant portion of its die to features that are designed for a market that currently isn’t generating much revenue. That needs to change in order for this strategy to make sense.

NVIDIA told me that we should see exponential growth in Tesla revenues after Fermi, but what does that mean? I don’t suspect that the sort of customers buying Tesla boards and servers will be lining up on day 1. I’d say best case scenario, Tesla revenues should see a bump one to two quarters after Fermi’s launch.

Nexus, ECC, and better double precision performance will all make Fermi more attractive in the HPC space than Cypress. The question is how much revenue will that generate in the short term.

Nexus enables full NVIDIA GPU debugging from within Visual Studio. Not so useful for PC gaming, but very helpful for Tesla

Then there’s the mobile space. NVIDIA could do very well with Tegra. NVIDIA is an ARM licensee, and that takes care of the missing CPU piece of the puzzle. Unlike the PC space, x86 isn’t the dominant player in the mobile market. NVIDIA has a headstart in the ultra mobile space much like it does in the GPU computing space. Intel is a bit behind with its Atom strategy. NVIDIA could use this to its advantage.

The transition needs to be a smooth one. The bulk of NVIDIA’s revenues today come from PC graphics cards. There’s room for NVIDIA in the HPC and ultra mobile spaces, but it’s not revenue that’s going to accumulate over night. The changes in focus we’re seeing from NVIDIA today are in line with what it’d have to do in order to establish successful businesses outside of the PC industry.

And don’t think the PC GPU battle is over yet either. It took years for NVIDIA to be pushed out of the chipset space, even after AMD bought ATI. Even if the future of PC graphics are Intel and AMD GPUs, it’s going to take a very long time to get there.