If clients are reduced to consumption terminals where special purpose IP blocks (e.g. video decoding) neutralize the difference between x86 and ARM, there is less justification for the price premium of a powerful general purpose CPU.

If powerful clients have large local storage and compute-oriented apps that are optimized for low-latency human interaction on large desktop monitors or VR workspaces, then there would be good reason to pay for a general purpose CPU.

A powerful desktop can also serve as a cloud to mobile devices, but a cloud server cannot offer the low-latency response of a local desktop. Why discard the one form factor that can serve both roles?

Is the problem Moore's Law or the lack of desktop software innovation and business models? If Hollywood can promote business models via PC hardware DRM (e.g. SGX), can the software industry support hardware for client-oriented business models?

The devices aren't necessarily disappearing; we just don't need new ones anymore.

I feel like CPU speeds have plateaued. Not so long ago getting a new machine after a few years meant getting a new CPU that was vastly quicker whereas all of the laptops I am looking at are only marginally faster than my ~3 year old laptop.

My hard drive requirements are not huge so it looks like I am essentially buying a new laptop just to get more RAM.

However, for many devices it difficult to wade through specs working out if you are actually getting NVMe (or whether you are just getting M2 SATA at the slower speed instead).

So in Android with dex, UWP with .NET Native and iOS with bitcode, compiled at store servers the client systems can have multiple CPU offerings.

But that brings some headaches when profiling the applications across devices.

Way easier than converting between Assembly languages.

Half way between cloud and mobile means you are not needed for the most part. I think that VR rendering will also be eaten by mobile with high end GPUs.

The only reason I need desktop is for compiles, but compiles could easily be done on the cloud.

If you're calling a laptop "mobile" then ah - okay, but that's not the general use of the term.

My hope setup is already this for the most part: laptop + 4K monitor + wireless keyboard + wireless mouse. And I write applications and do CAD and 3D computer graphics. If I can live this way now, it means that most everyone else can too. Just need to continue to shrink that laptop down until it is just the size of a smart phone.

Surface Pros too of course, but I agree this is getting close to a laptop.

With rehashing of 3D tv in the last few years, it was obvious that it was more of a product gimmick with manufacturers than real value.

With VR, you're seeing big companies pushing into the space from a spectrum of angles, ranging from re-imagining the workspace to 3D gaming. The bigger question is whether or not VR will become commonplace before augmented reality advances enough to replace it.

Even in gaming, I expect VR games will be only moderately successful.

One thing I've expected to see more of is Second Life. I thought for sure all the VR hype would reinvigorate Second Life for things like online classes and virtual meetings.

http://www.voxiebox.com/

I saw a voxiebox demonstrated at the NIAC symposium last year, and I was quite impressed. Being able to display a real-time 3D picture of my face as I made expressions was pretty cool. However, with the price point and mechanical manufacturing required for it, I don't think it will ever become mass-market in its current form. The state of the technology reminds me of the PDP-11 before PCs became a common thing; the basics are in place but it's just not quite there yet.

I don't want to invest in a voxiebox, but I think I would bet strong on whatever replaces it.

Intel derives much of its competitive advantage from its ability to manufacture.

This means that a lot of Intel's advantage will only express itself when the market grows really large. There have been a number of times when other companies beat Intel to a new technology (introducing, say, the 64 Bit chip several months ahead), but when the market took off and the cycle began, only Intel was able to deliver in sufficient quantity.

I don't know what the total market for a particular chip is likely to be, or how fast it will get there.

Intel's big advantage has been their ability to reliably and continually improve their manufacturing process, and for this to reliably and continually deliver performance improvements (the tick/tock strategy).

There have been a number of times when other companies beat Intel to a new technology (introducing, say, the 64 Bit chip several months ahead), but when the market took off and the cycle began, only Intel was able to deliver in sufficient quantity.

This is entirely untrue. AMD delivered a better consumer-focused 64bit architecture, and for an entire generation the Athlon outperformed the Pentium4. Yes, Intel sold plenty, but AMD did too.

That was the last time Intel made a misstep in their x64 product line.

Nowdays, desktop chips are facing some challenges. TSMC, Samsung and other are able to deliver plenty of ARM chips to market for the phone and tablet chips. Nvidia is increasingly eating the compute market.

This wasn't the case during the x64 transition, which was the example you used. To quote Wikipedia:

In commercial terms, the Athlon "Classic" was an enormous success not just because of its own merits, but also because Intel endured a series of major production, design, and quality control issues at this time. In particular, Intel's transition to the 180 nm production process, starting in late 1999 and running through to mid-2000, suffered delays. There was a shortage of Pentium III parts.[citation needed] In contrast, AMD enjoyed a remarkably smooth process transition and had ample supplies available, causing Athlon sales to become quite strong[1]

However, that was during the early 2000s. Since 2007, Intel has continually improved their processes and architectures on a reliable timeline[2], and haven't had any significant problems delivering those continual improvements.

[1] https://en.wikipedia.org/wiki/Athlon#Athlon_.22Classic.22

[2] https://en.wikipedia.org/wiki/Tick-Tock_model

I assume you meant to say "Since 2007, AMD has continually improved..."

So I stand corrected in that I didn't realize AMD had closed the gap.

However, my point was about the relative manufacturing ability of Intel and its competitors. Am I to understand your claim that AMD's manufacturing is (approximately) as good as Intel's at the moment?

I don't have any new information, but its a very, very hard problem. And a brief google search seems to indicate that AMD has faced more recent challenges: http://www.wired.com/2012/03/amd-global-foundries/

No, Intel is the one with the reliable continual improvement process. That is their advantage, not ability to manufacture large volumes.

The middle of the road is shrinking fast.

Therefore, Intel's competitive advantage of running their own fabs could keep them ahead in the server space as well.

Of course, the business relationship side is not quite as clear as the technology side if you want to ensure consistent delivery of a lot of parts -- "sell us a bunch of CPUs for this machine designed as a stopgap until we don't need your CPUs anymore", though maybe that is solvable by dealing with AMD.

All that said... I'm perfectly happy with x86 on the desktop and have yet to see any indication that ARM can compete on the types of jobs I still do with "desktop"-class (whether they be actual desktops or laptops) machines, like code compiling, RAW photo editing, video editing, etc.

This in no way eats the x86 market but in fact grows the ARM market.

Apple could help a great deal here, too, if it decides to turn the iPad Pro, into something more like an iOS-based Remix OS-like notebook. In other words, make the iPad Pro more useful by giving it an actual notebook keyboard, and redesign the OS interface a little more to accommodate a desktop environment. Keep the price at least $300 less than a Macbook Air.

Is it Google's doing ? or just Intel competing to the max, and desperately ?

Cloud, servers.. I get it, Intel is well established there, but IoT ? Do they have a foothold in this area ? When I look on their page I don't see anything promising.. and when I think of IoT, I imagine some low power ARM SOCs like Raspbery Pis combined with Arduinos rather then anything Intel sells today.

http://www.intel.com/content/www/us/en/internet-of-things/pr...

And as far as Linux OS's - nobody will buy an OS that can't be ported to ARM and be done so reasonably, so i'm not sure Wind River helps Intel that much in selling chips.

I don't see how that can save a company though.

They used to sell $200 processors like popcorn.

But the "internet of watching things", a world full of connected cameras is a different story. But it ain't a great slogan :)

That's the SigFox definition of IOT but not necessarily typical for all cases. There are several usecases out there which require slightly higher bandwidth and data rate, but not quite broadband.

One quick example is over the air firmware updates. Take MSP430. If we were to build a smart meter using a MSP430, our firmware probably be couple of kilobytes.

Likewise, the size of the data can vary upstream as well.

But do you think those usecases have a big impact when coming to evaluate how big of cloud the IOT will require ? if so, please share a bit, so we could grasp the scale of things.

Anyway, I think they should keep focussing on making the computer chips and focus on the temperature problem.

Except there are two things wrong with that theory:

1) "IoT" is mainly a new name for something that has already existed - a market that has already been dominated by ARM: the embedded/microcontroller market. ARM has much better expertise in this market, better relationships, and better ecosystem, as well as a full range of products.

2) If Intel's chips weren't competitive in the mobile market (especially when you compare all three metrics of performance, power, and price), then why would they be in the embedded IoT market, which plays even more to ARM's strengths?

The IoT is much more than just the embedded hardware. It's the potential that lies within connecting embedded hardware. Extracting and refining data from embedded machines/sensors/nodes. The whole ecosystem of apps/services that can potentially be created on top of these new data streams.

While the embedded/microcontroller tech hasn't changed drastically, the cloud/edge processing & analytics parts of the tech stack are new. The "services" business models are also new.

And these are the kind of devices Intel has decided to overlook 15, 20 years ago and now ARM is eating its cake.

What makes it an even bigger ARM turf than otherwise.

I haven't seen any of that yet, just hype. Of course every company wants to collect more data on consumers and is hoping IoT will enable them. None has provided a compelling reason for a "consumer" to buy an IoT device. Look at how Nest was all hype and then had a thermostat fail at its primary function when it lost the net connection. That's a huge step backward, and it's not going to get any better when the data collection remains the priority.

(Disclaimer: I work on the MessageSight messaging engine used in Watson IoT)

It's really more complex then that.

x86_64 has a lot of backwards compatibility. Even low power chips made by Intel typically consume 2-5x the wattage of ARM counter parts. Intel's very low power line (matches ARM) doesn't actually have the 64bit extension and is functionally a i586 chip from circa 1999-2003. Modern x86_64 chips have a whole section of die space dedicated to emulation, re-ordering, re-naming, and caching for us to pretend x86 is fast.

Then you have monopoly. Intel is the only company making x86_64 chips (Yes VIA/AMD exist, but collectively they have <10% of the market). They are the only show in town, it's their prices. While ARM simply licenses it's IP to other companies, who then compete with one another and drive prices even lower.

So, whether ARM64 has an power efficiency advantage isn't clear at all when one tries to actually compare them fairly. I frequently see people linearly scaling power/performance curves, or comparing cores that contain ECC, significantly more IO, etc against cores that don't and claiming that the core with 10x the IO bandwidth is somehow less efficient computationally because it consumes 5 watts more to power a PCIe bus.

So, lets get this out of the way, performance is not linear to power. Simply having a clock rate 50% faster has a large impact on power given the rough P=CV^2f equation because often the voltage goes up to support the higher frequencies. Worse designing for a target top end frequency of 4Ghz may entail extra pipeline stages/etc than one targeting 2Ghz in the same process. The result is significantly higher power draw at the same frequency due to the fact that higher frequencies are supported. Then there is leakage current/etc, which will be proportional to the number of transistors, so a design with 15MB of cache is going to waste more on leakage than one with 1MB. The extra 14MB of cache may only contribute another percent or two to the bottom line in many workloads, but frequently the difference between a processors at X performance and one at 1.5X is not due to a single factor but dozens of design tradeoffs that individually only net small percentage gains.

Bottom line, its better to compare implementations, and when ARM vs x86 chips are compared on similar grounds, they are a lot closer than you hear in glib remarks on forums like this. Intel's handicap in mobile (lack of native x86 android apps) and ARM's handicap in desktop/server all really come down to the software (and maybe in the case of some ARM products what one might consider alpha/immature products).

"Instruction sets" are not fast, the implementations are. Also AMD has a nice chunk of marketshare which will only get bigger once Zen is out.

Getting more register names in x86-64 was a huge boost too.

"to pretend x86 is fast" is a terrible choice of words. The fundamental problem is the memory wall. There are only two ways to solve it. Either put the processor directly onto the memory or add complex circuitry to mitigate it's effects. The former is difficult because DRAM and CPUs use different manufacturing processes. Thus neither x86 or ARM can avoid the latter if they want to maximise performance and in turn they become less power inefficient.

What you should have said is "to pretend DRAM is fast". It's not.

So Intel and Microsoft want to go where IBM and Oracle are withering?

@nVidia: how about thinking about affordable high end GPUs? The GPUs got 400% more expensive since 2011.

It's a pitty that AMD bought ATI, and now they can't compete with Intel and nVidia. And we all have to suffer and pay higher prices and get less performance - because no real competition exists any more.

They tried that -- the Pentium 4. The key design goal was to push clock speed as high as possible, and they used some crazy tricks, like 30-some-stage pipelines, a really long instruction scheduling loop with a pretty long lookahead, a double-pumped ALU with staggered 16-bit half-adds, etc. Fascinating from a microarchitecture perspective, but the big lesson was that high clock speeds sacrifice efficiency. The tricks needed to get there cause a lot of performance outliers/bad cases too -- e.g. the instruction scheduling replay system on lookahead conflicts was notorious for "tornados" which would kill IPC. And the branch prediction of the day was somewhat suboptimal for the pipeline lengths involved.

> Why are we stuck in 2006 era CPU singe core performance?

We aren't! The core microarchitecture teams at Intel and their competitors have made a lot of incremental progress -- Intel gets about 15% single-thread performance per generation, for example. No one is evilly scheming and holding back from turning a knob higher. There's just a lot of really hard engineering. Clock speed has topped out due to power limitations so we're at the point of looking for better branch prediction algorithms, cache replacement algorithms, and lots of little tricks everywhere to optimize bad cases. It's hard work (this was my job for a bit).

I'd recommend looking at, e.g., the proceedings of ISCA and MICRO conferences in the 2000-2006 timeframe -- this was when the industry and associated academia figured out that chasing clock speed was a losing battle after a certain point.

Many important things which affect performance have been stagnant for a very long time, especially cache size and RAM latency. But there's been steady progress in raw CPU power, coupled with a significant reduction in TDP.

I'm asking because I recently "upgraded" my desktop to a cheap westmere 6 core xeon [1] as in most benchmarks it is competitive with recent i7s.

[1] x5670 @3ghz, but should easily overclock well into the 4ghz range.

Edit: autocorrect

https://browser.primatelabs.com

Useful information if your workload is single threaded.

Note that their usage of the word browser is nothing to do with www browsers.

Still, even looking at the scores for less broken sub benchmarks, like (possibly) lua and dijkstra show a %50 increase in performance from Westmere to Skylake for the same frequency, which is more than I was expecting. And skylake should potentially clock much higher although I believe the top is currently still 4ghz.

Of course for anything that can take advantage AVX2 (or AVX3 for xeons) skylake would smoke the old xeon.

edit: reword

See Hinton et al., "The microarchitecture of the Pentium 4 processor" [1] for all the nifty details -- pp 8-9, and Fig 7 in particular.

[1] http://www.ecs.umass.edu/ece/koren/ece568/papers/Pentium4.pd...

From section 2 of this:

https://gmplib.org/~tege/x86-timing.pdf

"Pentium F0-F2 can sustain 3 add r, i per cycle for -32768 <= i <= 32767, but for larger immediate operands it can sustain only about 3/2 per cycle."

While this might not make sense at the surface (the next stage of the pipeline won't be ready for the data a half clock cycle earlier), you can minimize logic area by doing 2 quick 16 bit operations. Additionally, for some operations that have dependent instructions, for example a super scalar processor executing two integer instructions at once, if you have an ADD that depends on another ADD, you can forward the result from the first cycle on the first ALU to the second cycle on the second, although I'm not sure if this is actually done.

Another case where this is useful is if the result of an ADD operation needs to be used on a LOAD instruction earlier, you can forward the result a half clock cycle sooner.

Because that's physically impractical (if not impossible) with today's technology. Intel themselves actually have a good article[0] on the topic. Besides, there's way more to CPU performance than clock speed—today's Intel CPUs are several times faster than the Penitum 4, for instance, even at lower frequencies.

> @nVidia: how about thinking about affordable high end GPUs? The GPUs got 400% more expensive since 2011.

What are you talking about? GPU performance is plateauing a little bit, but the cards are only getting cheaper and more energy efficient, especially the high-end ones. Both Nvidia and AMD have announced new high-end cards coming soon that will give you the performance of the current high-end cards (Titan X and Fury X) at a little over half the price. Edit: Apparently they won't be that much cheaper than previous generations, but certainly within the grasp of any enthusiast or professional who really wants/needs them.

Plus, isn't "affordable high end" kind of an oxymoron? "High end" is almost by definition not affordable for most people.

[0]: https://software.intel.com/en-us/blogs/2014/02/19/why-has-cp...

The 980 TI I'm not sure is the best comparison to a GTX 580. It was intentionally a very expensive ultra-high end card. The GTX 590 debuted at $700.

There does seem to be a problem with prices being ridiculously high in most non-US regions, though.

I don't know where you've been but GPU competition between nvidia and AMD has been pretty healthy for the last few years, and AMDs specifically have been quite affordable. (That's not even taking into account they easily paid for themselves with Litecoin mining at one point.) It's only CPU competition that has been dead, but that's been the case ever since the Core line came out. (I guess you could count ARM too for low-end CPU competition, but both Intel and ARM can actually afford to lose that market entirely as their businesses are propped up by other things.)

NVIDIA held the upper hand with GK110 (780/Titan), then AMD dropped the 290X. This forced NVIDIA to readjust their prices and release the 780 Ti as an interim measure, but they held onto their lead position. Then NVIDIA started with the Maxwell series, which drastically improved both performance and efficiency. AMD tried to respond to the Titan X with the Fury X, but NVIDIA cut the legs out from under them by releasing the 980 Ti. Now once again AMD is behind the performance curve - NVIDIA has staked out the high-end space with the 1080 and the midrange with the 1070, and AMD is stuck in the low-end market with the RX480 until NVIDIA can get GP106 out.

AMD's response to the 1080 is going to be the Vega 10 chip, and unfortunately it's unlikely to be released before Q4 2016 at the earliest (more likely Q1 or Q2 2017). NVIDIA already has GP102 coming down the pipe, which is going to be a Titan/1080 Ti, and they'll drop that when they feel it's necessary to spoil AMD's sales.

That's the story of AMD and NVIDIA recently. AMD has put out some cards that were good value for the money - both the 7950/7970 and 290/290X have been very long-lived performers (and will likely continue to perform well thanks to DX12). But they haven't taken the top since 2012, all they can do at this point is force some price reductions. Like it or not they aren't a serious competitor in the GPU space any more than they are in the CPU space. They are a budget choice, not a serious contender.

There is also a rumor about AMD pushing Vega forward to october to release it around the new Battlefield (I doubt it, but it would certainly be nice).

Wouldn't other semiconductor materials allow higher clock speeds and it's just that the switch would be difficult to do incrementally?

Short answer is "no", but other materials may allow smaller manufacturing technologies. "Incrementally" isn't really a factor - every die-shrink is basically a new manufacturing process.

http://arstechnica.com/gadgets/2015/02/intel-forges-ahead-to...

http://arstechnica.com/gadgets/2015/07/ibm-unveils-industrys...

nVidia got their bets right with investing on uses other than gaming for their GPUs now they're eating AMD/ATI in the deep learning field

Affordable and High End are rarely on the same phrase, so there's your answer to when nVidia is going to do that

Example: https://www.cpubenchmark.net/cpu_lookup.php?cpu=Intel+Pentiu...

Other examples: http://www.tomshardware.com/reviews/intel,264-18.html

Not that this matter for normal cars.

That said, I suppose this differs depending on wether or not you think of the gearbox as a part of the engine (CPU) or not.

... but nothing compares, nothing has prepared me, for the sheer awesomeness, of these two links. I now must store the PDFs in a safe repository with GoogleCalendar note to re-read them frequently :->

IMO the only issue is commercial pressure killing integration. Until Vulkan GPUs were underutilized. Audio drivers lying about hardware capabilities. Software layers ever shifting. Absurd use cases: ads invaded html5, half accelerated html video....