Nvidia Takes The GPU Technology Conference On The Road ... - Seeking Alpha

Rethink Technology business briefs for September 26, 2017.

First stop: Beijing, and some announcements about Xavier and Tensor RT3

Source: Nvidia

As part of a marketing push for Nvidia's (NVDA) AI technology, Nvidia is staging a series of GPU Technology Conferences in various parts of the world, including Beijing, Munich, Tel Aviv, Taipei, Washington, and Tokyo.

The conferences are primarily technical, aimed at attracting AI researchers and software developers, especially in the areas of autonomous vehicles and robotics. By this, Nvidia hopes to pave the way for early adoption of Nvidia products by fledgling AI startups.

Nvidia has also been doing things like giving away Tesla V100 accelerators, and it has a contest for the “best AI startup” at the Beijing event for which the prize will be a DGX-1 “AI supercomputer.”

The conferences are also a source of information and news that might not have been shared at previous GTC events, such as the main GTC event held last May. Today in Beijing, CEO Jensen Huang gave a keynote that also provided useful information about key software and hardware products.

Huang announced TensorRT 3, a software package designed to accelerate inference processing. TensorRT 3 enables use of the TensorFlow AI framework on Nvidia's GPUs. Nvidia makes a bold claim that TensorRT 3 delivers the “world's fastest inferencing.” Google (GOOG) (GOOGL), which has spent considerable time, effort, and money on its Tensor Processing Unit, may disagree. I would really love to see a head-to-head competition between the two.

Huang also announced that Alibaba (NYSE:BABA), Baidu (NASDAQ:BIDU), and Tencent (OTCPK:TCEHY) are all deploying Tesla V100 GPU accelerators in their cloud services. This certainly suggests that AMD's so-called “collaboration” with Baidu doesn't pose a threat to Nvidia.

In addition, Huawei, Inspur, and Lenovo (OTCPK:LNVGY) are incorporating Tesla V100 GPU accelerators into their datacenter offerings via the HGX reference standard.

Huang indicated that first availability of the Xavier SOC, the core of Nvidia's next generation Drive PX hardware platform, will begin in early 2018 with sampling to select partners. General availability will be in 2018 Q4.

This is a somewhat slower rollout of Xavier than we were led to believe. Originally, Xavier was to begin sampling by the end of 2017.

I believe the delay means that Xavier will be fabricated on a 10 nm process (probably by TSMC (TSM), rather than the 16 nm TSMC process originally announced. By late 2018, the 10 nm process will have matured sufficiently that it would provide an economical and higher performance process for Xavier.

Nvidia is part of the Rethink Technology Portfolio and is a recommended buy.

Did Tesla drop Nvidia for... Intel?

Just last week, there were reports that Tesla (TSLA) was dropping Nvidia as a supplier of chips for its Enhanced Autopilot computer in favor of a custom chip reputedly designed by Jim Keller, who had led chip design at Advanced Micro Devices (AMD). My take on the reports was that AMD was probably developing a “semi-custom” chip in concert with Keller's staff at Tesla.

Today we have another report in the same vein, this time from Bloomberg, that Tesla has dropped Nvidia as the processor supplier of its infotainment system, primarily the large touch screen, in favor of Intel (INTC).

I have to say, this particular rumor doesn't make a lot of technical sense. I can believe Tesla replacing Nvidia for the infotainment system. There are any number of ARM processor vendors that would be able to stand in, including Qualcomm (QCOM). But powering what amounts to a sophisticated tablet is something that I would have thought an ARM processor would be better suited for.

It may be that Intel has offered a more compelling package of “connected car” features, including Wi-Fi and cellular modems, than Nvidia could offer. And Intel may have promised to build the chips on its much hyped, but very delayed 10 nm process.

Perhaps, with the adoption of an x86 architecture for its Autopilot computer, Tesla has decided to standardize on x86. For what amounts to a mobile device, that seems like the wrong answer, so I'm a little skeptical.

Morgan Stanley estimates Tesla's 2018 vehicle production to be 231 K

Adam Jonas of Morgan Stanley (MS) predicts that there will be 531K Tesla cars on the road by the end of 2018. That sounds impressive, except that 300K of those will have already been built by the end of 2017, according to Jonas.

So, in effect, Jonas is saying that Tesla will only produce about 231K cars for 2018, including Model S and X and the Model 3. If Tesla were to reach its goal of 5000 Model 3 cars/week production by the end of the year, then we would expect at least 250,000 Model 3 cars to be produced in 2018.

So Jonas apparently doesn't believe the bullish prognostications of Elon Musk. Good for him. One of the areas I haven't been particularly optimistic about is Tesla's ability to make good on its schedule. Never has, and probably never will.

I've expected that Tesla's production ramp would be protracted, but Jonas is even more pessimistic than I am. I had expected Tesla to build at least 200,000 Model 3s next year, even in my pessimistic scenario.

Does this mean that Tesla is doomed? Well, it certainly doesn't help. Tesla really needs to get profitable in 2018, and that means controlling costs while ramping sufficient Model 3 volume to cover its operating expenses. Reasonable people have questioned whether it's even possible for Tesla to do this.

I can only say that I hope so. But investments shouldn't be based on hope, and I continue to rate Tesla a sell.

Disclosure: I am/we are long NVDA, TSM, QCOM.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

http://ift.tt/2hwKWMM