NVIDIA’S TEGRA LINE is not doing well. Sales expectations are being guided down, and the chip is losing major contracts in shocking numbers. To make things more palatable to the financial community, Nvidia is playing a very cynical PR game for the technically unaware.
The big news first. Last week at its analyst day, Nvidia guided expectations for its Tegra line down quite a bit. The reason had been making the rounds of the financial community for the prior week or two, Samsung canceled a major contract with Nvidia for Tegra based phones. Nvidia for some reason did not see fit to mention this to the analysts.
Instead, Nvidia was talking about how many design wins it has for Tegra 1 and Tegra 2. Nvidia promotes design wins rather than shipping products for two reasons – there are few shipping products two years after Tegra’s introduction, and it is easy to play the term ‘design win’ with the technically uninformed.
Shipping products is an easy one, the list of Tegra devices for sale is so small it is not a topic that any rational PR or IR person would want to purposefully bring up. Of those products, the ones that matter can be counted on one hand, even if the counter has lost a few fingers in an industrial accident. To say the platform uptake of Tegra has underwhelmed expectations is being overly kind.
There is one bright shining star, however, Microsoft with the ZuneHD music player and its upcoming ‘pink’ phones. While these are wins worthy of shouting from the hilltops, which Nvidia did, they don’t seem to be wins on merit. If you look under the surface, there are far too many aspects that make you wonder for these deals to be entirely legitimate.
If you recall, the Microsoft-Nvidia phone was set to debut at 3GSM/MWC last year. That didn’t happen. From the first public debut of Tegra, to direct conversations with Nvidia Distinguished Engineers at Hot Chips, Nvidia was adamant that the only operating system to run on Tegra was going to be WinCE, nothing more was needed. Ever.
Microsoft had the keynote at 3GSM/MWC last year, and it showed off one shaky prototype of a WinCE 6.5 phone. The product that was meant for the show, namely ‘pink’, was delayed, so Microsoft put on a brave face and showed what it could. Fair enough.
On day two of that show, Nvidia started shouting from the rooftops about how its Tegra platform could run Android really well. If you didn’t know better, you might think a non-disclosure agreement expired and the handcuffs were taken off Nvidia to talk about non-Microsoft operating systems on Tegra. You have to wonder what the deal was that got Nvidia to not talk about the operating system, Linux, that powers the overwhelming majority of ARM devices. If you ask, Tegra won the Microsoft ZuneHD and ‘pink’ phone completely on merit.
Getting back to design wins, Nvidia is playing this card very cynically, so cynically that it can’t be by chance. If you are not versed in the process of hardware development, you might think that a design win means that there will be, or at least there is a high likleyhood, of a product that ships with the claimed chip inside.
In the PC world, this is exactly the case. When Nvidia says it has X design wins for its GPUs in PCs, you can expect the number of PC models with those GPUs to be fairly close to X. If the hardware that Nvidia gave to the PC maker for evaluation diverges from production silicon six months later, that is usually not a problem. A PC will easily tolerate 5W more power use or a few percent less GPU performance. Not a big deal. If things go really wrong, putting a different PCIe graphics card in it’s place is a trivial task.
The mobile space is a different kettle of artfully packaged toxic substances however. Any handset, laptop, widget or device is designed around the CPU or SoC. Even though they are all ARM cored, you can’t swap out a Snapdragon for a Tegra or an OMAP. The thermals, electrical connections, and everything else are so different that it necessitates a new design for a new CPU. On top of that, specifications are enforced with a military zeal, no divergence is tolerated.
Silicon delivery schedules for most chips in this space tend to be hit and miss. Some companies hit schedules on the nose, others don’t. Some deliver the promised specifications, others don’t. Performance is the same, you get what you get, and vendor reliability might vary widely. Trust but verify is the name of the widget game.
It isn’t hard to cherry-pick one or two good chips to send to a phone maker while promising that ‘they are all like this’. OEMs aren’t stupid, so any project that matters will have several possible chips to use, anything less almost guarantees a late or failed device. When production silicon is in hand, a decision is made on which of the possibilities to use. This is the verify part of trust but verify.
The problem for OEMs is that you can’t just pick a existing CPU and go with it, the design and testing process takes months or years. If you wait until production silicon is in hand to make the decision, then when the product comes out, your competitors will be using two or three generations newer silicon. Sales will reflect your poor choices.
OEMs have to make two or three, sometimes more, full designs and have them all waiting when the silicon comes out in production quantities. This essentially gives an OEM the ability to pick from multiple vendors, even if it had to make multiple platforms to do so. The game is more than serious enough to warrant that level of expenditure.
If a chipmaker misses its targets on a desktop GPU by 5W, you just buy a bigger heatsink or possibly specify a bigger power supply. For mainstream consumers, price is key. Price so key that it is almost the only thing that matters, the other factors are rounding errors as far as big PC OEMs are concerned.
In the widget and phone space, if you miss power targets by a few milliwatts, that is a hanging offense. Modern phones have precious little battery life as it is, and most get hot enough to be worrisome. A 20mW miss here and there is enough get a chip vendor thrown out the door with shocking speed.
Phone OEMs test with a fervor that is hard to put into words. Tenths of a mW are looked into, and no variations between specification sheets and delivered silicon are acceptable. If anyone tries to play games, it is on to design number two or design number three, that is why they exist. In the mobile space a design win means nothing more than being in the top handful of candidates.
Here is where Nvidia has problems, huge problems. For the past few years, it has not been able to deliver a single product on time, on spec, or with the claimed reliability. The company is quite good at delivering promises though, and that is what gets it the design wins. A phone OEM will start designing a Tegra based product based on cherry-picked silicon or specification sheets, and things look good.
Nvidia PR and IR shout design wins by the dozen, and uninformed press and analysts simply assume these will become products that carry Nvidia chips. Nothing could be further from the truth. When the silicon is delivered in typical Nvidia style – hot, late, slow, and buggy – it is shown the door. Again and again and again. Ask Samsung.
If you want to see the success of Tegra, the easiest way is to go back and look at Nvidia PR and IR’s claims for design wins, and when those products will be out. Start in mid-2008. Of the dozens of phones and widgets due in Q4/2008, the total count of released devices is single digits. Of the 50+ design wins claimed last week, how many do you think are going to make it to market? Given Nvidia’s guidance, it doesn’t think there will be all that many, especially when SEC rules cover the release.
Nvidia’s fast and loose ways might work in the PC space where tolerances are much more lax. In the phone and related portable device space, that game is simply not acceptable, but Nvidia is culturally unable to change its ways. The Nvidia ‘attitude’ might get its foot in the door, but it almost universally ends up with that same door slammed in its face.
Go back and look at the promised numbers of design wins for Tegra, the record is quite clear. In this device space, a design win does not mean a sale, you must deliver the promised product. Nvidia can’t deliver. Its competitors, also with dozens of design wins, can. Qualcomm, Broadcom, TI and Freescale sell chips by the tens or hundreds of millions because of this. Nvidia doesn’t, not with Tegra.S|A
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018