I’ve mentioned before my paradigm of ‘problem space’ vs technical advancement. Basically the entire problem space of computing is a bell curve. Easy problems that need a computer are few, then it rises through a large number of problems that benefit from medium levels of compute power. Finally, at the very high end, there are a very few problems that take all the compute power you can create and then some. In the beginning, the largest computers could only take on the smaller problems. As compute power and technology improved, the scope of the problem space that could be address by computers expanded. We saw the explosive growth of companies like IBM, Control Data, Cray, Burroughs, Sperry and Univac. All making main frame and “supercomputer” scale hardware.
As they reached the far end of the bell curve, other competitors came along making smaller cheaper hardware that used newer technology. They began to ‘nibble off’ the lower demand areas of the compute curve. Eventually they expanded into the middle ground. At that point we had the growth of the “Mini-computer” makers. HP, DEC Digital Equipment Corporation with the PDP-11 and VAX, Sun Microsystems. Eventually Intel and the micro-processor gave rise to the dominance of the Personal Computer at the low end of the compute scale.
IBM made a successful transition into smaller machines (and services, increasingly) but the others did not. As each class of machine could take on ever larger problems, the space ‘left over’ that needed “supercomputers” became less and less at the far right of the bell curve. Cray survived, but only after transitioning to a design using thousands of microprocessors in an integrated cluster. Sperry, Burroughs, and Univac all merged. Parts eventually becoming parts of Honeywell while the remainder survives as Unisys. Increasingly focused on software and services. Other companies, like RCA and GE made computers for a while, then stopped. Some, like CDC, gave up and stopped entirely.
All the time, the “little boxes” consumed ever more of the “problem space”. Adding features, like color animated displays and sound that take an ever larger compute load. Shrinking in size. The Personal Computer was large enough to provide many kinds of “server” functions and increasingly computer rooms were filled with racks of microprocessor driven machines. Now the Tablet (and even the cell phone) are continuing the trend to ever smaller boxes doing ever more of the large compute load tasks. This trend is not ending yet.
So Tandem, DEC, and other mini-computer makers went away (via merger into HP) while other makers of PC scale machines and even those rack mounted “server” PCs, found ever tougher going. Compaq was also absorbed into HP. Others simply went out of business (it’s a long list of CP/M and DOS based machine makers). Apple was in decline at one point, then transitioned to the small phone and pad / tablet market. Folks rarely buy a full sized “box” computer these days, they buy laptops or tablets. This trend to fewer and smaller “Personal Computers” is showing up in sales data today. Sales of Personal Computers, in the sense of desk side or desk top boxes, are down. Tablets and smart phones are up.
When Intel was The Big Dog of micro-processors, many folks wanted to compete with them. One of those was AMD Advanced Micro Devices. They are one of several CPU processor makers, but Intel is the leader. When a market slows down, the leader has more room to survive than the guy competing on price with smaller economies of scale. AMD has spent LOADS of money to keep making big micro-processors to compete in the desktop and laptop markets. The special purpose (smaller lower power and dedicated embedded processor) chips used in cell phones and tablets were largely made by others. AMD is fighting for a larger share of the shrinking market. ARM chips are the “hot chip” in tablets and cell phones. Folks buying an iPhone or iPad are getting an ARM chip, not an Intel / AMD architecture chip…
So what has AMD stock done lately?
It’s now just a $2 “perpetual option” on not going out of business…
Here is a 10 year chart of INTC Intel, vs AMD Advanced Micro Devices, vs ARM Arm Holdings.
Here we can clearly see that Intel has a large diverse sales base and isn’t particularly winning, nor losing, over time. Stable, with wobbles. So it can be traded on a modestly fast chart for “rolling” movements inside those ranges. ARM has had a rocket ride up (on the explosion of demand in tablets and phones) but now is stabilized. AMD was a “Golden Child” a while back, looking to carve off a large chunk of Intel business. Unfortunately, they grabbed that nettle just as the tablets were poised to take over. At the time of highest stock price, they OUGHT to have bought out ARM Holdings. (Frankly, I was at Apple a long time back when the first design evaluations of the ARM chip were underway. At that point, Apple ought to have bought them. But they didn’t ask me…)
AMD was a great ride, but “buy and hold” just doesn’t work. Things move in “spike and drop” unless you see management preparing for the NEXT tech cycle. Also visible are the roughly annual pops and runs. Trading those runs with a 1 year chart, knowing that to buy at the end of a ‘rip’ is a mistake and selling the ‘dip’ is also a mistake, would make a lot of money. There’s a reason the market aphorism is “buy the dips, sell the rips”… Once that big blow off top was reached, shorting was the right thing to do. PSAR (those red dots on the price chart) tell you when to buy / sell by changing sides of the price lines. Also note RSI at 80, then MACD just plunges. All the time that MACD is below zero, it’s a short, not a long. In the last three years, MACD is wobbling each side of zero, mostly below. Some few “long side trades” but mostly a short. DMI red on top also says when to be out. Blue on top was strong during that “spike” run up.
Oct. 18, 2012, 4:15 p.m. EDT
AMD Reports Third Quarter Results and Announces Restructuring
Fourth Quarter Actions to Target Cost Savings of More Than $200 Million Through 2013
Restructuring sometimes works; more often it is the kiss of death in a dying business.
AMD AMD -16.79% today announced revenue for the third quarter of 2012 of $1.27 billion, a net loss of $157 million, or $0.21 per share, and an operating loss of $131 million. The company reported a non-GAAP net loss of $150 million, or $0.20 per share, and a non-GAAP operating loss of $124 million. AMD is also announcing a restructuring plan designed to reduce operating expenses and better position the company competitively.
First off, notice that AFTER that long plunge from the top to down ‘near nothing’, we got a 16%+ drop in one day. That’s why you never ride a loser down. Once price is under the SMA stack and dropping, you exit. Also note that they are focused on cutting costs, not creating a new RISC chip to compete with ARM Holdings…
But at least they seem to recognize what is the core problem:
“The PC industry is going through a period of very significant change that is impacting both the ecosystem and AMD,” said Rory Read, AMD president and CEO. “It is clear that the trends we knew would re-shape the industry are happening at a much faster pace than we anticipated. As a result, we must accelerate our strategic initiatives to position AMD to take advantage of these shifts and put in place a lower cost business model. Our restructuring efforts are designed to simplify our product development cycles, reduce our breakeven point and enable us to fund differentiated product roadmaps and strategic breakaway opportunities.”
Doing that with less money, making losses, and a stock price about to hit “Penny Stock” levels is not good. But at least they know they screwed up… a half decade ago…
So that’s the charts. What about behind the tech?
Chips and Changes
Some couple of years back I figured a rough time when Moore’s Law would fail (for memory devices based on electricity at least). The number of electrons being used to store “one bit” had become small enough to be roughly counted. A couple of hundred thousand. You can only “divide by two” so many times before you have ONE electron, indivisible. We’re reaching the point where fundamental limits of physics are preventing much shrinkage or higher speed of storage cells or switches.
ARM RISC chips to erode x86 market share
Posted on July 18, 2011 – 18:18 by Aharon Etengoff
I’d say “has eroded”… but better late than never…
A prominent industry analyst has confirmed that ARM’s RISC-based chips remain on track to claim a sizable chunk of the lucrative notebook PC market by 2015.
According to IHS principal analyst Matthew Wilkins, RISC-powered ARM processors will ship in nearly one out of every four notebook PCs (22.9%) by 2015.
Unsurprisingly, the projected jump in market share for ARM is directly linked to Microsoft’s upcoming RISC-friendly Windows 8 operating system.
“Over the next generation, billions of PCs were shipped based on x86 microprocessors supplied by Intel and assorted rivals - mainly AMD. However, the days of x86’s unchallenged domination are coming to an end as Windows 8 opens the door for the use of the ARM processor, which already has achieved enormous popularity in the mobile phone and tablet worlds.”
Indeed, ARM support will enable the full-fledged Windows PC operating system to work on highly integrated chips that are more space- and power-efficient than traditional x86 microprocessors.
So it’s going to get worse under Windows 8…
The ARM chip is a RISC chip. Reduced Instruction Set Computer. The Intel / AMD chips are CISC. Complex Instruction Set Computer. There has been a very long fight between RISC and CISC. The pendulum shifts from time to time based on which tech moves the cost curve or performance curve furthest. Right now, CISC is reaching a hard limit on clock rate and chip size.
So “what comes next”?
We’ve already gone to “more cores” in CPU chips from Intel and AMD. My laptop has a ‘4 core’ chip. Supercompters have gone to thousands of CPUs working in massively parallel arrays. (My Cray was only 4 CPUs, but they were very fast ones for the day. Now the laptop has about the same processor power, also with 4 CPUs…) So one might speculate that more processors, and RISC ones, packed in ever smaller packages and cheaper…
Chipmaker takes to Kickstarter to become the Raspberry Pi of parallel computing
Summary: Chip company Adapteva hopes to launch a Raspberry Pi-style parallel computing revolution via a Kickstarter funding campaign, but getting enough developers could be difficult.
Established chip company Adapteva has launched a Kickstarter fundraising campaign to create a low-cost parallel chip board for supercomputing — a kind of Raspberry Pi for parallel programming.
However, there need to be more developers skilled in parallel programming for future applications to get the most out of the chips they run on. This is because in the mid-2000s, chip companies stopped being able to economically deliver significant clock rate increases on CPU cores and were forced to move to a multicore strategy.
If you can’t make the clock faster, and are having trouble shrinking the die size by orders of magnitude, going to more CPUs in a “cluster” (even if on one small chip) is the way to go.
That article spends some time moaning over how few folks have software that runs on multiple CPUs. They need to look at all the (large) software libraries written by massively parallel supercomputer companies… This is a known technology. On Linux, there is a message passing kernel (several variations) that move processes to other boxes. That same code will work on multi-core machines. (Mosix is one example). It is a start, but not an end point.
So my point?
Expect to see ever more parallel processing inside your computers. Windows will have more trouble using that effectively than will Linux. AMD will struggle coming up with a non-CISC competitor to the ARM chip family. ( Were I at AMD, I’d look to fund that startup with an option to buy the company…) Expect to see Intel continue along as a Ho Hum company making OK stuff in a wide range of products, but no big stock move. It’s the folks taking advantage of the new move into ARM / RISC chips and ever more parallel processing who will win.
Me? I built a Beowulf Cluster from scratch once “just for fun”. Didn’t do much with it (other than put ppmake in play so distributed compile of complex code was sped up). I’m currently looking mostly at secure computing Linux solutions, but along the way I’m keeping track of those Linux Distributions that are set up for cluster or parallel computing. Mosix, as of the 2.6 kernel, became a ‘for profit’ company and stopped development of the free public code. That has moved into the LinuxPMI project. This is focused on process migration over a network to separate machines; but that’s an easier problem than closely coupled CPU sharing. Still, if you can share over a network, you can share faster inside the box. The Linux kernel already has support for ‘multi-core’ CPUs (up to 8 is easy, I’ve seen up to 64 listed as a settable flag). So it’s ready to go for the next few years (decade?).
Who will lose the most from this? Likely the mini-computer / mainframe makers and to lesser extent the supercomputer makers. The supercomputers have already been pushed out into the long tail of ‘near infinite’ computes demand problems. But a “PC” with 64 CPUs in it will likely run most any engineering codes pretty easily, and handle computing for accounting for a large number of companies… It will also make a damn fine graphics heavy tablet ;-)
Postscript on Bittorrent
There are interesting things you can learn by accident. I’ve started using Bittorrent to download Linux releases. (And promptly became mildly addicted… now having a few hundred GB of them ;-)
It has a nice display of how many folks are “seeds” making the release available for download; and, for most, resolves thier IP address to a location and displays a flag. For slightly older ARM releases, the major locations of seeds are in Germany, Ukraine, Sweden, and some in Russia. Finland also shows up for some releases. For newer releases, France comes into play. For releases that have heavy US seeding, they tend to be Intel / AMD focused.
The implication here is interesting. It implies that northern and eastern Europe are ahead of the curve on the use of Linux on ARM chips. ( It’s also possible that they are playing ‘catch up’ and the USA guys already did their download a couple of years ago when I wasn’t looking…) Still, the number of seeds for x86 is ‘way high’ when compared to the number for “ARM”. The further implication here is that there’s plenty of room for Some Guy in the USA to get good on the ARM chips and have a niche. Guess I really DO need to order that Raspberry Pi now. ;-)
As a side note, the x86 chip releases were more likely to have a Latin American seed. Argentina and Columbia both showed up. Their were also many entries that were only an IP number. Some of those mapped to China and other Asian sites. Only once do I remember seeing a Japanese flag show up. (Then again, I’ve started with the older, smaller, releases. They “expire” sooner. So I might well find that the ‘trendy’ Japanese show up on the newest releases of the hottest topics, or mostly on Japanese language releases). Still, it’s a bit fun to notice ‘who and where’ are looking at any particular release level and technology base.
For example, one ARM release had only 2 seeds, one German and one Swedish. The same era x86 release of the same Debian code had several hundred seeds. Of those I connected with (only a dozen) many were U.S., French, Latin American, etc. Though at that scale, Bittorrent does some selection for ‘near and fast’, that will bias the ‘data’ of the connections. But just the scale of present interest in x86 vs ARM implies there’s a lot of folks not ready for the transition to RISC massively parallel cores just yet…
One other note to make: The technology of “Go Fast” tends to be invented in the supercomputer world, then moves ‘down scale’ over time. So pipeline architecture and ‘look ahead’ and parallel computing of threads and throwing away the path that wasn’t taken; first in supercomputers, now in PC chips. Supercomputers went massively parallel in the last dozen or so years, now it’s time to reach the mini and personal computer spaces.