Print this page
Published in AI

Nvidia GTC 19 Keynote failed to impress

by on20 March 2019


AI is the elephant in the room

First of all, I want to apologise to all the hard-working people that worked so many hours on the production and material of the keynote, this is not about you. Everyone did a great job, but the content picked was mostly just utterly boring.

The Editor in Chief of Anandtech, Ryan Smith, was sitting through the whole two hours and 44 minute keynote and following his live blog sounded like a better idea. The content announced at this years flagship GTC 19 keynote is causing pain to the new Nvidia. Mellanox - Nvidia if you like, as it all aims for a very expensive data center that needs a fast interconnect and costs an arm and a leg.               

Nvidia did talk about graphics, as the supreme leader mentioned how good the RTX cores are and of course, he wants to pack 40 GPUs in one package and sell it for a lot of money. This is what Nvidia has transformed into for the most part, selling extremely expensive but good technology. This is what the Street likes, and it increases the value of your company, until it doesn’t.

Raytraced Quake 2 is definitely something that we and a few other grandpas would love to play and relive our happy days. Jensen and his team at Nvidia might have forgotten that our buddy Daniel Pohl a researcher from Intel showed this more than 11 years ago with Quake 3 and 4 running on CPU. To give Nvidia some credit, its light demo looks cool but so does everything 11 years after. Let us not even touch that Nvidia's is calling it raytracing but is just a tiny bit of raytracing wrapped in the rasterization. 

Jensen’s AI obsession

Jensen only want to talk about AI, neural networks and deep learning as honestly most people still don’t understand what this is. This is what helps Nvidia’s stock price go up, and Crypto helps dragging it down. This is how Wall Street sees Nvidia.

Jensen likes talking about cars, but surprise, surprise, many competitors are catching up and taking a huge chunk of this very complex business. Just a few years back Nvidia was convincing us that it owned the self-driving market and that the cars will drive itself very soon. They did a great job discussing and, like Elon Musk, not calling it Level 4 or god forbid Level 5 self driving. Once Tesla launches its self driving this is kind of a Level 3, a far cry from total autonomous self driving.

Tesla even makes its own SoC for its self driving and there are other players too.

Nvidia wants to talk robots, and this is a trend that outlived drones. No one wants to talk much about drones that didn’t went as well as IDC predicted a few years ago. Giving Fudzilla some credit we could see through all that some four years ago as it didn’t sound like a valid business and something that people need. Nvidia is replacing quite old Tegra X1 with Denver cores and ancient Cortex A57 into robots. This doesn’t sound like a winning strategy to us as there are other players, more experienced in this market too.

Competition is closing in

Once you walk out under Nvidia marketing cap, you begin to realize that competition is closing in on the firm. Intel has Xe graphics that is essentially a GPU supercomputing capable solution too and it wants to succeed. I'm not sure whether AMD can ever jeopardize Nvidia in the GPU world but the GPU market is shrinking if you ask John Peddie Research that's been the case for a few years now.  

Think about Intel integrated with enough performance to play 1080P games at decent settings, this is Nvidia’s worst nightmare but it is just around the corner.

There are many other players including Xilinx that have great automotive and AI solutions and that don’t have an issue with latency - something that GPUs are doomed to have. This, of course, depends on the use case scenario.

It was not surprising to read that analysts following Nvidia were very optimistic after Nvidia meeting. This is what Jensen is good at. Of course, the company is doing well, but not as great as Jensen is able to convince everyone.

The AI elephant in the room would have to have a 7nm name, maybe an Ampere in the name, but this obviously is coming at a later date.



 

Last modified on 20 March 2019
Rate this item
(0 votes)