I interviewed a candidate a couple of years ago who was, at the time, working on Intel's internal self-driving car team. It sounded like a totally directionless, amateurish also-ran catastrophe. Years behind the competition. I wonder if this product line was in a similar situation.
RealSense cameras are/were pretty good. Lower latency than Microsoft Kinect Azure (~70ms vs 300ms), decent depth quality and a well-maintained open source SDK.
The problem with most of these "pet VP" projects is that there's never any clear link to the external market.
Who are they benchmarking against? What are their performance targets?
As a result, you get a team who's only deliverable is "PowerPoints that go up and to the right, compared to last week."
And the project soldiers on, because to can it and admit failure would embarrass someone. If it's the CEO, even worse.
Which is maybe explainable for a brand new technology (say, Palm with handwriting recognition), but less so when there are serious competitors out there with known performance levels.
About a dozen years ago I worked for a company that partnered with Intel on a big project. Ultimately the project failed. The biggest problems I saw were:
1) Too many cooks. Different groups from several locations had their fingers all over it. So there were too many expensive, useless features.
2) Too many engineers making product decisions. Engineers often are interested in neat technical features, not in whether a product should include those features. And often engineers don't really care how much those features cost in terms of design time and effort.
3) No real sense of urgency. Consequently there were other, hungrier, companies that ate Intel's lunch.
None of this was new even then. All of this was understood by (other) Intel people way back in the early 1980s. The problem is just how do you herd cats (or engineers or chip architects)? It's very difficult.
You may need to have a small, focused startup to accomplish something. Then that startup gets acquired by the big behemoth. See e.g. Cisco back about 20 years ago. I imagine the same thing happens now, but I don't follow the industry very closely any more.
Also, Cisco at least acquired somewhat related companies. WTF was Intel doing buying companies like McAfee?
Sometimes the spunky competition can even be internal. Again, going back a ways, Intel USA marketing was pushing the architects and engineers to do the Netburst architecture (clock speed was all-important). Which had a lot of problems. Fortunately Intel had what amounted to a startup in Israel. They came up with the Core architecture that saved Intel's ass.
This seems to be pretty typical of companies that have a sustaining cash cow. They're simultaneously compelled to try to find new avenues for growth while unable to really establish, operate, and sustain a commitment to see it through. Intel has x86, Google has AdWords, and they both have a graveyard of alternative product roadmaps around them.
That said, I'm glad to see Intel gaining focus as I don't see it as a generally good thing for them to be floundering in the marketplace.
It’s hard being the new product in the BCG matrix. Incumbents seldom remember how meager the traction looked in the beginnings of the lifecycle of their current cash cows.
even more so when it takes some years before the product gets traction.
Twitter was not an overnight success. The first few years it was ridiculed as a big "completely useless" failure by journalists: the same people which now cannot live without it.
> Twitter was not an overnight success. The first few years it was ridiculed as a big "completely useless" failure by journalists: the same people which now cannot live without it.
My opinion:
In fact Twitter is still amazingly useless and is a prime example of a product succeeding despite its flaws instead of because something.
And no, I'm not this negative towards everything simple. Snapchat for example has a point, I can see the point of it even if I use it even less than Twitter.
Twitter succeed because Google botched Google+ after first using all the oxygen in the market together with Facebook.
But Google also has YouTube, Gmail, Gsuite, Android and Chrome - all are hugely influential cash cows. Google is a very successful multi-business company. It’s nothing like Intel.
Intel tries to innovate and diversify but they are strangled by their stagnant workforce and internal politics. Two of their biggest acquisitions, Mobileye and Habana Labs chose to remain independent companies to a large degree under separate branding.
They source what’s good for them from the mothership, and other than that maintain autonomy. Intel acquisitions which did not do that, instead fully integrating into Intel, mostly failed.
Or is it simply that we're more aware of the failures because anything a big company does gets press coverage? If you're small, you can fail quietly (and they do all the time).
These cameras had some frustrating faults. My friend's master's project relied on their facial recognition cameras. Unfortunately my friend was black and they didn't work at all on his face. He needed his white/asian friends to help him test his application.
We've banned this account for repeatedly posting unsubstantive comments and breaking the site guidelines. Creating accounts to do that will eventually get your main account banned as well, so please don't do that.
The documentary 'Coded Bias' goes in to some depth on this problem in facial recognition and on the wider problems of the social impact of biased algorithms.
But isn't it due to the ML models rather than the camera? Facial recognition models were/are? disastrously biased.
Amazon's Rekognition detected darker skinned women as men or children in our application in 2017(we used it to prevent profile photo having children in our privacy focused dating app), Amazon rep said they were working to fix that but doesn't seem like it as of 2019[1].
Not that it's that simple, but it probably contributes...I'd also be curious of the difference in RSU/equity, Apple may be in a better position there as well.
I've been working with Realsense depth cameras for a while, and was mostly pleased with the software, hardware and the ecosystem. Intel developers and engineers seems to be trying hard.
But at the same time I was wondering how, for a behemoth like Intel, that's been making huge profits by selling high volume products with also a high profit margin, such a tiny market, with reasonable priced cameras, is going to satisfy its shareholders.
I guess the answer is here, and won't take long for the whole program to get the axe.
Wondering what will I do if I keep on working on this project
One thing that "old Intel" used to do a lot was trying to find use cases for more performance. This used to be a strong fear internally, that at some point silicon would get "good enough" and you wouldn't need to buy better silicon.
Gelsinger was very much part of that a decade or two ago, but I guess he's getting more rational and accepted that yes, to run Office pretty much anything will do. At the same time, the need for more silicon, even for basic tasks will remain strong as silicon just ages out (in part thanks to software, see Windows 11 requirements).
So not completely surprising to see Realsense and other similar niche initiatives get cut in that context.
Also, most of them were also kept alive by the previous CEO who pushed for the "maker" / IoT thing as the "next big(ish) thing". So part of that is also a classical CEO succession story.
The announcemen leaked weeks ago was such a disaster given it was wrong. We heard that they were cancelling RealSense and that created a real supply chain problem for about 24 hours.
It was an educational moment for me on the importance of owning the story, lest you have to spend weeks digging out of a wrongly dug hole.
I'm glad their not being shut down entirely but I hope all the robotics companies out there remember when Apple bought and shut down PrimeSense's sensors and so don't become too dependent on any particular 3D sensor.
42 comments
[ 4.6 ms ] story [ 46.7 ms ] threadReally curious on the decision making process throughout the line's lifetime.
Who are they benchmarking against? What are their performance targets?
As a result, you get a team who's only deliverable is "PowerPoints that go up and to the right, compared to last week."
And the project soldiers on, because to can it and admit failure would embarrass someone. If it's the CEO, even worse.
Which is maybe explainable for a brand new technology (say, Palm with handwriting recognition), but less so when there are serious competitors out there with known performance levels.
About a dozen years ago I worked for a company that partnered with Intel on a big project. Ultimately the project failed. The biggest problems I saw were:
1) Too many cooks. Different groups from several locations had their fingers all over it. So there were too many expensive, useless features.
2) Too many engineers making product decisions. Engineers often are interested in neat technical features, not in whether a product should include those features. And often engineers don't really care how much those features cost in terms of design time and effort.
3) No real sense of urgency. Consequently there were other, hungrier, companies that ate Intel's lunch.
None of this was new even then. All of this was understood by (other) Intel people way back in the early 1980s. The problem is just how do you herd cats (or engineers or chip architects)? It's very difficult.
You may need to have a small, focused startup to accomplish something. Then that startup gets acquired by the big behemoth. See e.g. Cisco back about 20 years ago. I imagine the same thing happens now, but I don't follow the industry very closely any more.
Also, Cisco at least acquired somewhat related companies. WTF was Intel doing buying companies like McAfee?
Sometimes the spunky competition can even be internal. Again, going back a ways, Intel USA marketing was pushing the architects and engineers to do the Netburst architecture (clock speed was all-important). Which had a lot of problems. Fortunately Intel had what amounted to a startup in Israel. They came up with the Core architecture that saved Intel's ass.
Time to hunker down into the foundry business.
That said, I'm glad to see Intel gaining focus as I don't see it as a generally good thing for them to be floundering in the marketplace.
Twitter was not an overnight success. The first few years it was ridiculed as a big "completely useless" failure by journalists: the same people which now cannot live without it.
My opinion:
In fact Twitter is still amazingly useless and is a prime example of a product succeeding despite its flaws instead of because something.
And no, I'm not this negative towards everything simple. Snapchat for example has a point, I can see the point of it even if I use it even less than Twitter.
Twitter succeed because Google botched Google+ after first using all the oxygen in the market together with Facebook.
Intel tries to innovate and diversify but they are strangled by their stagnant workforce and internal politics. Two of their biggest acquisitions, Mobileye and Habana Labs chose to remain independent companies to a large degree under separate branding. They source what’s good for them from the mothership, and other than that maintain autonomy. Intel acquisitions which did not do that, instead fully integrating into Intel, mostly failed.
I haven't tried them, but Nerian's sensors look like some of the best ones on the market.
https://news.ycombinator.com/newsguidelines.html
a.k.a. please don't feed the trolls
https://news.ycombinator.com/newsguidelines.html
Looks like shadowban over here.
Amazon's Rekognition detected darker skinned women as men or children in our application in 2017(we used it to prevent profile photo having children in our privacy focused dating app), Amazon rep said they were working to fix that but doesn't seem like it as of 2019[1].
[1] https://abcnews.go.com/beta-story-container/Technology/wireS...
Both are working on custom silicon and cutting edge hardware.
Not that it's that simple, but it probably contributes...I'd also be curious of the difference in RSU/equity, Apple may be in a better position there as well.
But at the same time I was wondering how, for a behemoth like Intel, that's been making huge profits by selling high volume products with also a high profit margin, such a tiny market, with reasonable priced cameras, is going to satisfy its shareholders.
I guess the answer is here, and won't take long for the whole program to get the axe.
Wondering what will I do if I keep on working on this project
Gelsinger was very much part of that a decade or two ago, but I guess he's getting more rational and accepted that yes, to run Office pretty much anything will do. At the same time, the need for more silicon, even for basic tasks will remain strong as silicon just ages out (in part thanks to software, see Windows 11 requirements).
So not completely surprising to see Realsense and other similar niche initiatives get cut in that context.
Also, most of them were also kept alive by the previous CEO who pushed for the "maker" / IoT thing as the "next big(ish) thing". So part of that is also a classical CEO succession story.
Little did they know the software community would do this work for them by making slower and slower software.
Because it was everything they could have hoped for.
[1] https://en.wikipedia.org/wiki/Andy_and_Bill%27s_law
[2] https://en.wikipedia.org/wiki/Wirth%27s_law
It was an educational moment for me on the importance of owning the story, lest you have to spend weeks digging out of a wrongly dug hole.
Best RealSense replacements? - https://news.ycombinator.com/item?id=28225065 - Aug 2021 (80 comments)
Intel is giving up on its AI-powered RealSense cameras - https://news.ycombinator.com/item?id=28218354 - Aug 2021 (14 comments)