Hacker News Reader

Comment by PaulHoule | Hacker News Reader

That's not entirely true.

Intel has invested in plenty of things that haven't worked out, for instance, Itanic, Ultrabooks, Realsense, etc.

Also companies like Intel and Microsoft have bought into the Clayton Christensen theory of "disruption" to the point where they really are throwing their existing customers under the bus. That is, Windows 8 was perceived as a tablet operating system, Intel only sells high-end chips under the Xeon brand, etc.

Had Intel stuck more piggishly to making better PC's, we might have mainstream water-cooled desktops running at 5Ghz+ and more reasons for people to upgrade.

View on HN

According to Wikipedia[1], Intel set aside "$300m over 3-4 years" to invest in Ultrabooks, so $100m/year. Contrast that to ~$8B they spend on R&D each year[2]. So it was a relatively small bet for them.

I'm not saying tech companies never invest in anything outside of their core products. It's just often those projects are given low internal priority precisely because their (presumed) impact to the bottom line will be small.

[1] https://en.m.wikipedia.org/wiki/Ultrabook [2] https://ycharts.com/companies/INTC/r_and_d_expense

edit: typos

I'll just note that a startup of 10 - 100 people that raised $300M over 3-4 years would be a big deal.

Innovation isn't about ideas, its about finding ways to execute against those ideas in a way that others haven't or ideally can't.

Sure, but important difference is that 10-100 people startup that raised $300m behaves way differently than an internal department of a major tech firm that has decided to invest into a side project: authority to make decisions, willingness to take on risk, speed of innovation - everything is very different between the two.

Cisco and it's spinning out approach comes to mind.

They probably got this right.

Amazon is another company that is innovating disruptively despite it's size.

I think your definition of innovation makes sense only in the context of corporate or SV innovation.

In a more general context, innovation takes on the other meaning.

And small bets rarely materialize into anything useful. They end up being great marketing, and worth the cost in the marketing spend so others think you are being innovative while you continue to do the same stuff you always have.

> Also companies like Intel and Microsoft have bought into the Clayton Christensen theory of "disruption" to the point where they really are throwing their existing customers under the bus.

They could've fooled me. Clayton's theory wouldn't recommend changing Windows to adapt to the mobile form factor. It also wouldn't recommend Intel investing what are now probably over $10 billion in Atom and in making x86 work for mobile.

Quite the opposite. It would tell Microsoft to create a new operating system that is dedicated for those form factors, and it would tell Intel to adopt ARM on their own. Both should also be done under an isolated division, so they have no "conflicts of interest" with the incumbent products, whether it's about product features (like say Intel refusing to make Atom too strong so it doesn't encroach on Core territory), or setting up expensive (and unsustainable) infrastructure to compete against the leaner competitors.

I would also say both Microsoft and Intel failed in these markets because they were also late to try and become players in those markets, which is what the disruptive theory always warns incumbents about. It may be difficult to remember now, but even Windows 8 was considered "late" for the mobile market. So was Windows Phone 7 compared to iOS and Android. Same goes for Atom "catching up" to ARM chips in performance/low power (though not in price, a typical failure of a disrupted incumbent). If the "Wintel" monopoly wasn't so strong, you'd already see Celerons and Pentiums be replaced by ARM chips in notebooks and hybrids and whatnot, because those chips are the same ones that failed against ARM in mobile.

This is inline with what Stanford professor John Ousterhout says about the evolution (and necessary death) of software :

"The most important component of evolution is death Or, said another way, it's easier to create a new organism than to change an existing one. Most organisms are highly resistant to change, but when they die it becomes possible for new and improved organisms to take their place. This rule applies to social structures such as corporations as well as biological organisms: very few companies are capable of making significant changes in their culture or business model, so it is good for companies eventually to go out of business, thereby opening space for better companies in the future. Computer software is a counter-example to this rule, with ironic results. Software is incredibly malleable: it can be updated with new versions relatively easily to fix problems and add new features. It is easier to change existing software than to build new software, so software tends to live a long time. To a first approximation, software doesn't die (compare this to the hardware in a computer, which is completely repaced every few years). At the same time, it is difficult to make major structural improvements to software once it has been shipped, so mistakes in early versions of the program often live forever. As a result, software tends to live too long: just good enough to discourage replacement, but slowly rotting away with more and more problems that are hard to fix. I wonder if the overall quality of computer software would improve if there were a way of forcing all software to be replaced after some period of time."

https://www.quora.com/What-are-the-most-profound-life-lesson...

> Had Intel stuck more piggishly to making better PC's, we might have mainstream water-cooled desktops running at 5Ghz+ and more reasons for people to upgrade.

Can you think of what those reasons might be? I can't. My office computer is an i5-3570 and it sits idle most of the time because it's just not that hard to have Outlook and Firefox open. A/V work and AAA gaming drive upgrades now like they always have, what other reasons would there be for Intel to push beyond their current efforts like you suggest?

I think that higher resolution displays are probably the leading reason behind gamer upgrades today. Pushing 3-4x the pixels is a lot of overhead compared to the decade before UHD. Another reason really is power consumption. There's been a lot of effort to reducing the power envelope in CPUs, this investment would have happened even without a gamer segment.

Servers are probably Intel's single biggest market today, and the companies running these miles of racks of servers are demanding more cores and lower per-core power. That goes in line with mobile improvements too. Intel is a perfectly reasonable cpu option for tablets today, within 3-4 years they'll probably have something suitable for phones.

I've always liked the concept of a phone with everything that you dock into a workstation, or a laptop form factor to use, but as a phone it's still great. There's a lot of possibility there. More than that, with the likes of services that go beyond echo, we may see the watch form factor take over from phones... relying on larger systems in your home/office/car to fill the void. Who knows where things will go.

>> we might have mainstream water-cooled desktops running at 5Ghz+ and more reasons for people to upgrade.

> Can you think of what those reasons might be?

Some applications that I would want that might need to wait for high-cpu (some high-memory and/or high-disk) to become more mainstream, since I want them all to run locally until FTTH reaches my part of the boonies:

A scanner I can literally wave a piece of paper at, have it process all of the ultra-high speed video frame grabs into a high-dpi scan, then reconstitute the scan into an appropriate LibreOffice document or PDF, with better OCR and better vectorization of raster pictures. Then I auto-file all my invoices, bills, statements, receipts, etc.

A personal digital assistant that transcribes, annotates, indexes and categorizes all of my conversations across all mediums (voice, text, chat, social, etc.). This doesn't need to be ML-fancy at first (except for the voice recognition piece), just lots of memory+cpu+disk to drive intensive, continuous indexing.

A millimeter-accurate 3D positioning system, using differential GPS off of a geodetic mark on my property, with relays of signals to indoors. This would drive my robotic chicken tractor, curb edger, and vacuum cleaner. I could keep a detailed inventory of household items with this, then query the house computer over a voice link the next time I forgot where I set down $thing (some RFID and 3D positioning integration needed here). Outside, it keeps track of exactly where various pipes, outlets, irrigation heads, etc. are located.

A software-defined radio that scans for chatter on interesting bands like police and fire, performs voice recognition on them, and pings me when it hears keywords about my neighborhood, street, nearby streets, or region when combined with other keywords (like "tornado").

A house computer that can tell whether or not I'm in the house or not, if I'm asleep, getting ready to sleep, or still undergoing my morning routine, at my computer, at my tablet, on my phone, etc. And do this for all occupants and visitors. Lots of visual recognition from lots of camera input. For myself, I want this information to drive what information is presented to me, when, and in what context. Bloomberg Radio after I've been up for 10 minutes, and after I've put on my headset, before my first call (incoming or outgoing).

A process that continuously combs through my Emacs Org agenda, and compares against my current "state". If I'm driving away from the house, and confirm a query from the computer that I'm going out to buy groceries and run errands for the day (the query is launched by continuously-computed probability cones narrowing possible agenda matches against my GPS position until a threshhold is met), and no one else is in the house, then automatically put the house HVAC system into energy-conserving mode.

A dumb robotic mechanics driven by updateable software in my desktop computer to fold my laundry, turn over my compost bed, rotate stop valves monthly/quarterly (otherwise they freeze in place, the primary challenge the automatic flood sensors face today), change HVAC filters, open and file incoming snail mail, etc.

Intensive AR to turn nearly all parts of my house into smart surfaces. "What's inside that drawer?" "What's behind that wall?" "What's inside that freezer?"

If we are talking only about pure desktop applications constrained to the physical desktop/laptop computer itself, then yes, you may have a point. However, when we include extending the reach of those systems to our environment, and there is an explosion of data to crunch, then today's typical consumer desktop computer (4-8 GB RAM, 2-3 GHz x86_64, 2-4 TB at the top end) doesn't have the capacity to manage all of those demands simultaneously.

I love this. This is all great. But with the exception of the scanner, all this can be done across an internet connection, and so therefore all of this will be done across an internet connection. For entrepreneurs building an ANI (or AGI), it makes little sense to let anyone else get their hands on the code, even in binary form, and then to have to support that code running on arbitrary PC configurations.

I hear you, and I hope to see these applications arrive in both cloud and on-prem versions, perhaps from different origins/vendors. That signals the software ecosystem is strong, vigorous and robustly healthy enough to support many different solution angles. Let users choose what best fits their needs, and let the diversity seed further innovation in a virtuous spiral.

I sometimes muse if on-prem will get a second look as containerization, self-hosted AWS Lambda-like microservices, and similar advances become more mainstream on servers, and we'll see more hybrid solution ecosystems evolve. I strongly suspect we will look back in decades hence and lament that governments are Why We Can't Have Nice Things In The Cloud; while the US has been particularly egregiously public, we shouldn't be surprised at other nation state actor reveals in the future, either. If I'm on point about that, then we'll probably see hybrid cloud+on-prem (COP?) instead of an overwhelming dominance of one or the other for the near future at least.

I don't know about you, but I'm not going to willingly send all my personal conversations and a mass of data about my living space over the internet to be analyzed by a third party...

So this will never happen in my household unless I control the data, and it is stored/analyzed locally.

I would argue that it is not a trivial upgrade of the capacity of the hardware (8 GB Ram, 2-3 Ghz x86_64) but the optimization and work on the underlying algorithms and hardware. Furthermore, the cost associated with introducing the new products might not be currently sustainable.

An ASIC can perform some of the computational tasks you mentioned (such as image recognition) with much better thermal/performance levels; it might be hard to make it a commercial success.

These applications absolutely depend upon more algorithmic and software engineering work, and it definitely is not a matter of just adding more hardware; if it was, I'd have the rack or two (or more?) of hardware it could take today. It's hard to predict markets, but in the Open Source world there are at least primordial forms of every application I mentioned, or at least the elements of them. It's fun to tinker with them at the very least, while the hardware keeps improving.

It's like back in the 80's it was neat to imagine GB-scale disks with ubiquitous, continuously-indexed search across all files on the system. We could even implement it, and tinker with it on a toy scale for practical speeds, or on a system-wide scale at slow demo speeds. It took better hardware before that feature was really practical.

I'm wondering if ASIC-based products will come out for "settled" niches that large swathes of the workstation market agree upon for a kind of standardized co-processing set of sockets/slots in workstations, and workstation-embedded FPGA's as software-reprogrammable become available for niches that are not as settled, but not quite amenable to GPU/CPU processing. I don't know enough yet about hardware at that level to discern what field of application could possibly penetrate mainstream computing that would need those devices at such a low level, though, that couldn't be addressed with PCIe cards today, if not through CUDA/OpenCL for accelerators/co-processors and general-purpose CPUs.

- voice recognition on the computer - better search (apple's OSX is impressive... PC... nosomuch) - AI-based autocorrect of entered commands that learns who you are - better data visualization - better web visualization - gesture recognition without a mouse - eye tracking without a mouse

Here's Intel's problem (and my apologies, I'm regurgitating a study I read a couple years ago and I can't find a link.)

Basically, Intel needs to sell huge numbers of chips in order to make their numbers, ie giant investments in fabs, work. While Xeon may generate all the profits, they still need to sell millions of low end pc chips to pool costs. If that stops happening, even assuming they sell the same xeons for the same prices, the R&D costs to make those fabs eat them.

"Investment in (new) fabs" only makes sense if volumes are increasing or manufacturing processes are changing. Volumes are projected as flat for the foreseeable future, and as far as anyone can tell process tech is near stagnating as well.

Investments also are to research and build out that fab tech to stay ahead of Samsung tmsc et al

They could have been a major ARM vendor if they had gotten over their NIH attitude, seen the writing on the wall, and kept on with StrongARM/XScale.

In what way did Itanium not work out? It isn't in every PC sold, but it was never intended to be. It was intended to replace Sparc/power/alpha. While they didn't get Sun or IBM on board, they did get HP, and the chip is profitable. It's not going to drive the future of the company, but it's not like they didn't make money on the investment... it turned it's first profit 7 years ago and has been making money for them since.

https://commons.wikimedia.org/wiki/File:Itanium_Sales_Foreca...

It was intended to be the x86 sucessor.

No, it was intended to be the "Big Iron" CPU replacement. They never had any intention of bringing it to the desktop which is why they delayed as long as they did in x64. They wanted 64-bit to be a long-term differentiator between server class and desktop class CPUs. Hence the forecast being Servers sold a year...

"No, it was intended to be the "Big Iron" CPU replacement"

All the computer press was pretty sure it would replace x86. Microsoft released Windows port for Itanium, which was unseen of them.

They planned to keep x86 as 32-bit which means gradual phase out (complemented by x86 emulation on Itanium).

Reply on news.ycombinator.com