Intel’s erratic Core M performance leaves an opening for AMD
extremetech.com
extremetech.com
1. Carrizo isn't shipping yet. Broadwell shipped in September, and Intel's next microarchitecture (Skylake) ships late this year.
2. Carrizo will be fabbed on a 28nm process.[1] Broadwell is 14nm. When it comes to fab tech, AMD is over 3 years behind Intel. Ivy Bridge (22nm) shipped in April of 2012.[2]
3. Carrizo's lowest TDP is 12W.[1] The Core M version of Broadwell has a 3-6W configurable TDP.
I compliment AMD's PR department, but when it comes to performance or efficiency, AMD is not a serious competitor. Intel's fab tech is simply too far ahead. I'd love to see a repeat of the Athlon days, but for the foreseeable future AMD can only compete on price.
1. http://arstechnica.com/gadgets/2015/02/amds-carrizo-system-o...
2. http://en.wikipedia.org/wiki/Ivy_Bridge_(microarchitecture)
Nobody buys desktops anymore. Profits are falling
Mobile is the new competition area. And if Intel wants to sell x86 there they'll have to earn it.
Nobody buys desktops anymore because they haven't gotten faster in any real way in the last 5 years. Probably about 80% of intel's product line can be beaten by a 5 year old intel machine. Sure at the absolute top, there has been a few percent gain, but most people don't seem willing to buy a couple thousand dollar machine to gain a few percent. Sure it may be more power efficient, but I don't think most people buying desktops really care if the CPU draws 30 watts or 80 watts, as its plugged into the wall and the cost savings over years isn't enough to pay for the upgrade.
Intel could be pushing features into the desktop space (say ECC for example) but that would further cannibalize their high profit chips.
A SSD has more impact on your life than the latest CPU. I've seen it numerous times. My neighbors latest SONY i5 laptop doesn't hold a candle to my uber aging ThinkPad simply because of this (and an OS that doesn't accumulate crap over time)
And people bought desktops because that's how it was. Under the slightly fallacious upgrade-ability of towers. But a fanless mini-itx would please anybody in less space. I don't know how important this is for people, do they prefer to have a dedicated area for a large box vs tiny imac / vesa mounted box somewhere.
Because software is constrained by the lack of performance improvement.
If someone managed to double single threaded performance every 18 months for the next 5 years, I can assure you that it would force an upgrade cycle. If for no other reason than than FF/chrome/websites would bloat up enough that people would have to upgrade.
Its not like people sat around for hours waiting for their 286's to boot, or to print word processing documents. Faster technology enables "features" which suddenly people cannot live without which drives the upgrade cycle. There are plenty of technologies which are not yet "mature" and could benefit from more single threaded computing power.
Speech recognition, compression, and on and on, are algorithms which are tuned for available computing power rather than the best compression ratio, or the most accurate recognition.
1) I firmly believe we've reached a plateau, I'm having a hard time imagining what new complex feature people would need that much. That's anecdotal and subjective but even amazing things like computer vision or speech recognition ended up in the gimmick side of things to me. The web is becoming bloated for no reason, old websites had more content, less chrome. Fast web is easy, no javascript, text. ps: bloated websites are enabled by more capacity.
2) Speed is not everything, there's a sweet spot between instantaneous and unusable. Most people aiming for speed are after user conversion, in the fear that response time would lower their numbers. For real usage, speed is not a real factor as long as its predictable and meaningful. I have soft machines with jaggy and unreliable behavior which annoy me more than my old 350MHz machine.
3) Hardware constraints are good. Without limitation nobody would have cared about space or time efficient algorithms (jpeg, mp3, or even zip). See how late Microsoft has been trying to improve storage use of its OS.
All in all it goes both ways.
I'm really open to ideas, but I'm failing to see what I'm doing now that I couldn't do under win2k days (except for HD video, 3D). I used to render cgi animation, compose multitrack, abuse photoshop...
All of that said, a Core i3-5010U has almost the same single-core performance and 40% the multicore while using less than 1/8 the power... I can now just run two systems, one for the server, another for my desktop and still see less power. After getting said CPU in a Brix box for my HTPC, I'm considering such a move for my next home desktop (still a year off though).
For example, Intel will happily sell you an 18 core 2.3GHz Haswell Xeon with a design power of 145W, or a 10 core 3.1Ghz processor that puts out 160W.
[1] AMD SoC architecture reviews showed originality and value, but you needed a different software stack to take advantage of it. Without that, it's just slower, hotter silicon. I think that's how mantle started. Again, it's not a competitive advantage anymore now that Vulcan is a standard.
[2] I may be wrong, but I don't think ARM SoC are really a threat to Intel. They did eat a nice chunk of his lunch though, but Intel's IP and assets seem so huge...
That's why Intel did so much R&D in power efficiency. ARM CPU are sometimes still favorable to Intel ones, in some markets Intel still doesn't have a chance.
PC is dominated by Intel. Smartphones by ARM. Intel invades the tablet market, ARM the notebook.
Personally I would love me an nice ARM based notebook, but other factors like housing, keyboard and display are often better on Intel devices. But I don't think until ARM steps up and produces a high quality device like Google did for their Chromebooks, no manufacture will switch from the thinking that ARM is for cheap devices only.
My god I could already live with a Apple A7 level CPU and modern E-Ink display, which I have in my ebook reader who also can run linux and all tools I really need to work, fast enough to not limit or annoy me. Only thing why I don't use it as daily driver is the missing physical keyboard.
But you have to see, I use to roughly 80% of my time with just two consoles: vim and command window for running scripts and git. So I'm not the norm.
But what about the norm?
Germany has magazine called c't, who just recently published an article how the life with a smartphone as your work pc looks like. Speed wasn't the concern, because their ARM CPUs where powerful enough to run a browser or office software on a full screen monitor without problems. Connectivity was the only problem, which could negated with the right smartphone.
There was also an online article about a reporter test drive, who switched from his notebook to a smartphone and a Bluetooth keyboard for his mobile work setup. He liked it, except it was harder to deploy on the lap and he often needed a table to work. (I think it was engadget, but I can't find at the moment and I run out break time, so please look for yourself.)
TLDR: ARM is the new AMD.
Google should be pushing ARM in notebooks like crazy. If ARM doesn't end up with at least 70% market share in Chromebooks (due to the multiple ARM chip makers, just like Android has the majority of market share in mobile operating systems), then I'll consider that a major competition failure.
Granted it's not all Google's fault for pushing Intel. Intel managed to enter the Chromebook market with some shameless lies:
1) First lie: promoting the new Intel Chromebooks as "Haswell Chromebooks". I have a good memory and I remember at the time most people thought that was basically Core i5 Chromebooks - at least for a while.
2) Second lie: Those chips were Haswell Celeron, but at those device price points of $200-$250, it would've been impossible for Intel to sell them profitably. So they sold them below cost, so people then can have the choice between what is essentially a $110 chip and a $20 chip, yet the devices cost roughly the same ($250). So of course the choice was made for them, and people would rather get the $250 device with the $110-value chip than the $20 one.
It still pisses me off to no end that governments aren't taking action against Intel over this. It's highly monopolistic and anti-competitive behavior. It's no different than Microsoft eating the cost of IE and bundling it with Windows. Intel has also wasted over $8 billion in subsidizing their mobile chips so far, to be able to sell them under-cost, and competitively against ARM. No ARM maker, not even Samsung or Qualcomm, could ever afford to do something like that. So why is Intel allowed to do it to take out the competition from the market in this way?
3) Third lie: After people got used to $250 Celeron (Haswell) Chromebooks - Intel did a bait and switch and switched those more powerful Haswell Celerons with Atom-based Celerons, which are no more powerful than ARM chips, but people now think those "Celerons" are.
It gets worse. Now that people remember that the ~$110 or so "Celerons" were relatively powerful (thanks to the Haswell/Core architecture), they are now starting to charge that much for the Atom-based Celerons. Even though as I said, they are the equivalent of $20-$30 ARM chips. But this is what monopoly can get you (at least in Windows PCs and Chromebooks, where the ARM competition has been all but wiped out).
The lies also continue with the way Intel is now calling its new Atoms - "Braswell", as if they had anything to do with Haswell or Broadwell.
Intel has become (or perhaps has always been, I've only been following this stuff closely for the past few years) an incredibly shady company. And it saddens me that more people don't see it, and that Intel gets away with it stealing ever more market share in nefarious and anti-competitive ways.
> Those chips were Haswell Celeron, but at those device price points of $200-$250, it would've been impossible for Intel to sell them profitably.
If I'm not mistaken, these are mostly trash from production who are not good enough for the regular QA, but still run with clipped wings.
> It still pisses me off to no end that governments aren't taking action against Intel over this.
snort If that is all, you seem to be much happier with your government than I dare to dream of.
When MS started including IE with windows, every other graphical OS included a browser (OS/2 Warp, and early Linux distros). I think the irritation was how deeply they embedded the browser, which does make some sense as it allowed them to create the CHM help files, which is a pretty nice format. As well as some of the push desktop stuff early one, which I kind of liked. The down side is that support for newer browsers on older OS versions was effectively made into a cliff for IE.
I'm glad that Chrome and Firefox carry as much weight as they do. Seeing a roughly 1:1:1 split among the top three is best for the larger community. The down side being how long it takes for an IE version to ship (still), which means that IE while ahead at launch quickly falls behind and becomes our least common denominator for public websites for years.
In particular, while I don't mind using JS transpilers like Babel/6to5 to get new features today, the fact that I can't use these without such tools for the next 4-5 years is kind of sad. As an aside, I really wish there was a common background worker abstraction that worked in both node and the browser... I know I can fork/cluster in node and use webworkers where available, it would still be nice to see a common interface for this stuff.
And once you have just enough RAM and swapping, you realise that the biggest problem is organising the open tabs in browser.
We have just approached the state when computers can finally fully assist us. The problem is in the software. There just isn't a way to manage our shared computer memory.
http://appleinsider.com/articles/14/11/16/after-losing-apple... (there are many articles from other publications saying the same thing)
Personally I would love me an nice ARM based notebook
ARM wins in terms of low heat and low energy consumption, which are huge plusses IMHO. With a bit of effort, they can also win in openness. They will be powering e.g. http://wiki.vero-apparatus.com/ARM64OpenLaptopConcept and http://pyra-handheld.com/
Companies should realize design wins over features. If anyone comes up with a developer oriented laptop that (i) has top notch Linux support (ii) can be upgraded / expanded / serviced (iii) has a good keyboard (iv) has a matte 4:3 screen they'll gain loads of backing. With quick SSDs and cheap RAM, CPU speed is not that relevant.
All of that said, I've been hopeful to see something compelling in an ARM powered laptop for a while... Also, with the likes of the transformer tablet/netbook setups, I'm really surprised nobody has done an interface for a smart phone docking/powering a laptop yet (at least that I'm aware of), carrying the same screen/keyboard/touchpad across multiple generations of mobile device.
Such a beast would have to be able to run at least 2x 1080p displays without much lag and several applications. It would probably need a means of docking with active cooling while docked. It would be an interesting design challenge.
Maybe for mobile, but even AMD is better than ARM for most server tasks. Maybe a new generation of higher performance 64-bit ARM chips will change this...
I really hope AMD pushes the industry ahead once again. Though, they haven't had the best track record when it comes to power and heat.
But that doesn't mean Core-M is a "good" chip. I would stay away from it. The submitted article does a decent job of explaining why. You're much better off with a cheaper (yes, cheaper!) Broadwell Core i5 at 15W (or whatever its TDP is for this generation) than this chip.
Also, correct me if I'm wrong, but first off "Broadwell" only shipped as Core-M in 2014, not any other version. And it didn't do it in September, I think it was more like December, and only in a couple of devices that I think everyone forgot about.
Broadwell "technically" shipped in 2014, but 2015 is actually the year of Broadwell. I imagine the same will happen with Skylake - 99% of Sylake devices will appear in 2016.
Core M allows passive form factors which to me is super important - having a notebook that doesn't sound like an airplane in a distance as it is the case with my rMBP and Zenbook Prime under full load is a major benefit. No noisy interruptions when I need to focus on solving some difficult problem and am most sensitive to external distractions.
Why? If AMD gets acquired, that might just mean that the buyer now holds AMD stock. I can't see how that could possibly affect the license agreement.
You do realize Intel also has to get an x64 license from AMD, right? So unless Intel is good with not making 64-bit chips anymore, I think it will continue to give AMD that license...
The only slight advantage to AMD is that that can be a little more fine grained, but still a crappy thermal design from an OEM is still going to result in a chip that is throttling itself, just like we're seeing with Intel. We might end up closer to the optimal balancing point with AMD, but it's still going to be wildly different performance from manufacturer to manufacturer.
It's purely a function of the cooling available and the target skin temperature. This doesn't mean you can't "trust" it.
It just means you should do what you should have -always- done which is buy machines in metal chassis (and of reasonable size) if you want passive cooling or get a machine with a decent thermal design and active cooling (fans).
Trying to spin this into FUD seems a little desperate.
I've heard technically clueless people complain that their new highend Core-M laptop is slower than their old laptop when they just do work (ie+powerpoint+excel+mail), and they are frustrated.
Right then it doesn't matter that the new laptop only weighs half of the old one and still has much better battery life, work is still done mostly at a desk with power nearby.
That knowledge is rather dated and probably goes back to the Pentium 4 era. Top speed is a function of heat, but also of the process technology and the concurrency available+used. It's best to wrap your head around it by going back to first principles:
> The dynamic power consumed by a CPU is approximately proportional to the CPU frequency, and to the square of the CPU voltage:
> P = C V^2 f
> where C is capacitance, f is frequency, and V is voltage. [1]
Capacitance goes down with smaller transistors, voltage and frequency goes up with higher sequential performance given the same architecture. So, by replacing the need for sequential performance with more concurrency (vectorization/multicore/gpu acceleration), the need for power goes down dramatically. Process technology is another big factor.
On a related note: Comparing ARM with Core M just using some concurrency using benchmarks is not really comparing Apples to Apples - I expect Core M to have a higher sequential performance - so if you want to have the same performance on ARM, you'll need more concurrency in the software.
The Pentium 4 had a very deep pipeline, which got very hot and wasn't fast compared to its clock. The pentium M went back to more instruction level parallelism via shallower parallel pipes (multi core came later).
I don't think ARM is competitive with Intel even with parallel benchmarks; Intel does performance well and doesn't play many tricks with it, ARM's advantage has always been power efficiency and cost.
Yes, parallelism makes more sense in this context. I wanted to express it in a more general way because running multiple programs on different cores at the same time I understand as concurrency, which is also made more efficient by parallel performance rather than sequential - IMO the terms are a little bit fuzzy at times, i.e. it's not that easy to convey the ideas behind parallel programming easily such that there is no ambiguity.
Edit: About ARM vs. Core M in benchmarks: After looking at it again, apparently the two are still quite far apart, you were right about their performance numbers: [1]
[1] http://www.fool.com/investing/general/2014/09/22/intel-corpo...
Even that equation you gave says that if you increase P you will be able to increase f, though the interplay between f and V means that your increases are sub-linear.
> top speed is actually the function of the heat generated
emphasis mine.
> Even that equation you gave says that if you increase P you will be able to increase f, though the interplay between f and V means that your increases are sub-linear.
I don't get your point. Did I write that performance and power is not related?
> I don't think that anybody could reasonably read top speed being a function of heat as top speed only being a function of heat
.. I still don't see it this way. My post wasn't meant to be an attack on OP, rather I wanted to expand some points about chip performance that seemed to me weren't clear. The post I'm quoting however did read uncharitable to me.
More than erratic, I would say it's very "flexible", since it can be used by different devices with different requirements in terms of performance and skin temperature.
They have to do this because there are so many different ways to hold a Yoga Pro 3 that they cannot allow any part of the laptop to get uncomfortably hot in the hand.
http://ark.intel.com/compare/85215,84672
If that's the case, no wonder it gets bogged down by thermal throttling.
It's not a processor that is meant to be doing sustained heavy work (compiling, gaming, simulations), but attend to small burts of CPU loads and then go back to low power mode.
It seems to me it would be more than enough for office work, remote ssh, (non flash) web browsing, e-mail etc.
Anyone buying a thin and light device like the ones mounting this processor, and expecting the performance of a desktop replacement laptop have unreasonable expectations to start with.
I'm always wary of the charts on cpubenchmark, however. To a desktop/laptop user, single core performance is usually much more important than total performance, as they're not using all the cores. Even for my use case, software development, compiling uses multiple cores but doesn't take any significant time these days anyway (C# in VS, JIT for .NET and JavaScript), so I tend to check:
http://cpubenchmark.net/singleThread.html
AMD A4 Micro-6400T APU: 607 Intel Core M-5Y10c @ 0.80GHz: 1,115
Almost twice the (benchmark) speed for the Core M. I'd like some real world usage reviews, but it's still looking like Intel is the best choice if you can afford it.
There isn't, AMD doesn't have anything which comes even remotely close to a 3~6W TDP, which is the Core M target. Furthermore they don't have the fab technology so a 3~6W AMD part would necessarily be even slower and thermally constrained than the Core M.
And it's not about affordability in the case of Core-M. It's about poor performance/price ratio. You get very little for a lot of money, all to get that 5W TDP, which is also misleading because those chips needs to be put into devices that support more like 10W TDP, not 5. That's why most Core-M devices overheat. Intel overpromised on the TDP, and now OEMs get the blame for it.
For the average user, it shouldn't matter what CPU they have in their machine. If their computer is slow, it's because that CPU is busy executing crapware.
Leaving this impression with users is not good for your brand in the long run.
I think Intel optimized too much for battery life and TDP with Core M, and it was also priced and targeted at the wrong highend market.
Imagine if you installed Windows 98 on one of these things. It would fly.
The CPU is perfectly capable of the workloads that users want to throw at it. The software isn't yet.
The difference between Win98 and OS X 10.10 is screen resolution, animations and virtual desktops.
Combined with 8gb of RAM and a very fast SSD, a decent iGPU and I think performance will be very satisfactory indeed.
With less power consumption, it would be even better.
By some chance I have managed to find myself with an AMD Athlon 760k system with 32 GB Ram, and SSD, and a R9-290x GPU. It's pretty much fastest, top-of-the-line components (144 Hz monitor, too!), with a budget AMD chip. This is a 32 nm chip, with 2x2 MB of L2 cache (there's no L3 on the Athlon line these days). But coupled with a fast GPU and crazy amounts of RAM. The CPU is watercooled, too, with a double-sized radiator and 2 120mm fans. And just to be stupid, I picked up a killer bigfoot NIC as well.
I can't say I've noticed any real limitations on the CPU itself. I know it's hampered pretty badly, but I think I'm going to hang on to it as long as it's relevant and see what happens. I don't find myself being limited by that much. It plays Shadow of Mordor at highest detail settings (I bump it down a bit to get 100+ FPS).
It's sad to see them struggle, but Intel has superior technology right now, and everyone depends on technology to survive. I hope they can continue to stay relevant in the next couple of years until they get another home-run platform out of the door.
HSA is very interesting to me, but I don't know how important it will be to others. A few algorithms that I write can burn a lot of time being transferred to GPU and back. A gigantic FPGA is what I need, but I can't afford one of those.
Basically spend some logging how much you stress your CPU and GPU during a normal day developing. If the results are "basically never rises above 5-10%" then the macbook will be fine.
Although my work is mostly based on infra nowadays, i use vagrant & docker to test things locally so i think i should move my worry over to the 8 gb memory part :)