Do you still think AMD will have the edge in 2006? 2007? 2008?
Get real. Intel makes more than AMD is worth every 3 months. P4 was a blunder. Don't expect future chips to be as bad.
Intel has the fab capacity to supply Apple without any problem. AMD is always fab limited. AMD is correctly attacking a niche (servers). This is where they will shine.
Some statistics that I have seen show something on the order of 80% of the dollar revenue of game publishers comes from consoles with PCs getting the remaining 20% (check out the financial stats of a publicly traded company like EA).
Now put yourself in the shoes of a game developer...What is going to make you the most money? A 20% dollar market or an 80% market?
Compound the fact that developing games for the PC is far more expensive -- there are all sorts of different hardware configs and not always stable driver headaches to deal with. This means longer testing times and therefore more expense. Remember that developer time is the most expensive resource in a game company.
Constrast that to a console which has a very consistent and well documented hardware platform. It's easier (read cheaper) to develop for a couple of configurations (xbox/ps).
Now even a 20% PC market is huge, so there will probably always be PC games. Another reason is that most development and content creation is done on PCs. But do not under estimate the enormous advantages of the console from a game developer's perspective. The PC market will increasingly shrink and fall into a niche when it comes to gaming.
Not to mention the big cost advantages of the console with their subsidized hardware model. Yes there are people willing to pay $400 for a graphics card, but these people are definitely in a niche market. The bulk of the money is made in the low-end to mid-range segment.
What possible motivation is there for Microsoft to move to a different ISA? Almost all of the world's software is written for x86. Moving to a different ISA would seriously weaken Microsoft's monopoly position because all of the tools and infrastructure (drivers, compilers, debuggers etc...) are build around x86. Microsoft would commit financial suicide if this happened not only because the ISVs would be pissed, but for another reason - There simply isn't enough capacity to supply NON-x86 silicon to meet global demand (computers would start costing $5000 again - not a good thing for Microsoft).
Intel supplies 80% of the worlds microprocessors. Nobody else has the fab capacity to do this. It takes a few years to put up a fab and each one costs about 3 billion dollars (which keep getting even more expensive for each process generation). IBM/AMD/others simply don't have the capital structure to build all the fabs necessary to meet global demand. Who is going to put up the capital? Microsoft? Why? The already enjoy monopoly pricing. How would moving to a different ISA give them any kind of a return on investment given the costs?
Building microprocessors is so expensive that DEC-Alpha, HP-PA RISC and SGI-MIPS have all thrown in the towel yeilding to Intel. SPARC will be the next to go. If Apple goes with Intel, expect POWER to be the next to die (sorry but console profits are tiny).
Here's my prediction. In 10 years, everything will run on x86. Yes, even PDAs and cell phones. The risk to Intel is that more competitors start building x86 processors (AMD is too small). Maybe Samsung or some other company with the capital requirements and semiconductor expertise necessary to build out fabs.
In fact, I think that Microsoft is more at risk.
They are at risk on the desktop/laptop side from Apple and Linux. Both of which run x86 (Admittedly, the Apple thing is still a rumor).
Let's say the market moves away from the PC to a more server/thin-client model. Microsoft gets cut out of the picture (think Google already running their software services on x86-Linux and upstarts like RIMM/Blackberry running their software on the clients). Google runs on Intel. RIMM does not, but Intel is already a leader in the handheld space (think ARM).
Now let's consider the diversification argument by listing the markets that require computing power.
PCs/Desktops? No question here. Intel has a dominant market share in both processors and chipsets.
Graphics? Intel is #1 in graphics marketshare with their integrated graphics solution.
How about notebooks? Intel is the clear leader in that rapidly growing space. Transmeta tried to compete and got crushed. AMD is all but invisible with their one-trick opteron pony. The market is moving here and Intel is there to capitalize.
PDAs? Intel inside with StrongARM.
Cell phones? Intel inside (with their flash chips).
Servers? Intel is again very big here with over a 90% share of x86 servers.
Embedded? No Intel there. The reason? There's no money there...
Consoles? No Intel there. There's very little money to be made with silicon. All of the value is in the software. NVIDIA only made about 30 million on XBOX (Intel made something similar). ATI is reported to make about 2$ per console on XBOX360. To put it into perspective, Intel makes about 600 million every single week.
Diversification has worked beautifully for companies like IBM and GE. 10 years ago, I do not think that IBM would have been able to dump its PC business without significantly damaging themselves.
IBM is valued less than Intel by about 30%. GE has been around for 100 years, Intel is only 30.
In the 80s Intel was concentrated on making memory chips but were getting killed by the Japanese, and in a heartbeat, diversified into microprocessors. Until the demand for microprocessors diminishes, Intel will keep making them. It's hard to stop when you still have 60% gross margins. If there's value and growth in silicon, Intel is well positioned to take advantage with their enormous fab capacity and their army of skilled engineers.
Nowhere does it say x86. This could be an agreement for Intel to get into the PPC business, which would be a great supplier coup for Apple
Unlikely.
The global PC market produces 200 million PCs a year. Apple sells 3 million of those.
The incremental revenue for Intel is a rounding error. Why would they devote an entire design team to supply such a tiny market when they can get a much higher return on investment with x86? In case you haven't heard - microprocessor development is ridiculously expensive. You need huge volumes to justify the capital and R&D expenditures.
Hmm...I wonder why IBM microelectronics loses money every single year, and AMD barely makes any money even with a superior architecture?
The benefits for Apple are huge. They get far cheaper processors relative to IBM which means Apple can increase profit margins. I'm suprised they haven't gone with Intel sooner - they've probably spent all this time perfecting the software.
Intel can supply chips at will (think 200M units per year vs. Apple needing 3M). IBM has had problems keeping up with the demand for even this small market. They are not cost competitive and cannot deliver in the volumes that Intel can. AMD can't either. Unless you have the capital to invest 5-7 Billion dollars a year on Fabs, you just can't beat Intel on a manufacturing basis which is why Intel is bigger than AMD, Apple and IBM combined (on a market cap basis).
Re:Microsoft Will Fail - Tales From The Inside
on
Gates on Google
·
· Score: 1
>> And I don't even know what the URL for Microsoft's search engine is. Do you? Does anyone? I never hear anyone talk about it, never hear anyone refer to it, mention it, use it.
When they integrate search into Longhorn, that's when the threat becomes more real. Google is doing a pre-emptive strike with Google-Desktop-Search, but microsoft counters with their own.
I think Microsoft is losing currently, but I wouldn't be so quick to count them out. Msn.com is one of the most frequently visited websites by default. Many non-techies don't bother to change the default home page on their IE browsers. Owning the OS gives them tremendous leverage.
Waterloo is a great school and their co-op program is more valuable than any traditional education you would get at a typical North American Ivy school.
The co-op program allows you to explore what you really would like to do in the real world, instead of some abstraction in the classroom. You get the best of both worlds - a solid academic background and a real taste of the engineering world.
"In the long run, we're all dead." - John Maynard Keynes
Yes, I am familiar with Keynes and his famous quote.
When there is a sharp and rapid supply/demand imbalance, it can take a very long time for the economy to stabilize into equilibrium. Even the most zealous free-marketer will agree that if I jump off a cliff there isn't enough time for someone to setup shop and sell me a parachute on the way down...
The Federal reserve can soften the pain with interest rate adjustments so that the equilibrium can be reached faster. This is probably the best we can do - it's a blunt instrument but it helps.
The fact that it can take time to stabilize doesn't alter the fact that change happens and can happen quickly. Change is normal. Change can be good.
I'm replying because you quoted my favorite author Heinlein. Great quote BTW:)
That's real nice rhetoric. I'm not sure whether it falls under the "Appeal to Pity" or "Complex Cause" fallacy. Let's call greed and laziness exactly what it is.
You're reading way too much into what I said. I'm not appealing for anything. I'm just stating a value-neutral fact. All I said was that corporations need to make a profit - this is a pretty well established fact. Given this fact, if you can produce the EQUIVALENT good or service using lower labor costs - you will go for the lower labor costs because increasing profitability is the main function of a corporation. Another fact that is pretty hard to dispute. It's like saying force = mass * acceleration. It's a value neutral fact - whether is it good or bad, I'll leave to the philosophers.
As a consultant, I've observed many cases where my client companies have decided to cut the wages of the peon while executives and board members received dividend and bonus increases, and released reports that complained about the cost of doing business as validation for the "team" to make sacrifices.
In this example, you have to ask yourself if the labor that has been cut is scarce? If it is and the scarce peons decide to leave, then the company is screwed isn't it? Because it will no longer be able to function in a similar manner and this is a sign of poor management.
However, if the labor is plentiful and there are other equivalently skilled people in the labor pool willing to do the work for less, then perhaps they are just maximizing profits with no longer term risk to their business.
Why would the team ever agree to sacrifice if they could find better work elsewhere? It must be that the costs of labor are declining because they are unable to find a better deal. Or - the managers are just moronic and don't realize the scarcity of labor (if it does exist), in which case the people would leave for work elsewhere and the profitability of the firm will suffer and the market will eliminate them.
Everybody acts in their own selfish interests whether they like to admit it or not. However, what many people don't realize is that maximizing self interest often involves trade and cooperation.
You'll find few if any "workers" who are willing to do the same for less in a market with rules or law. In a market with rule and law this argument clearly makes no sense. The 12 yr old in a country with no law, stitching up a tshirt for export to an entirely different country.
You have to ask yourself this...What is the alternative for the 12 yr old in a poor country? In many cases they are working because it is the best alternative for them. If you slap tarriffs on imports from countries that allow this, suddenly, instead of making some money, the child makes no money and has to continue to live in abject poverty. Who are you to take away that income from them to protect American workers who already enjoy a vastly superior standard of living? This is not a zero sum game - the standard of living is raised for the poorer workers and the standard of living is raised for the average american through either lower prices or higher profits. There is a loser in the equation (the non-competitive worker), but that's how the world works. Business models change, things that worked yesterday don't necissarily work today (The buggy whip industry has been in decline for over a century)
Also the markup price for the tshirt is clear in its final price. Have you seen some of the retail prices for clothing in the US? 40-50-60 US dollars for a shirt
No one is forcing you to pay $50 for a shirt. If you don't like the price, don't buy - that will affect the supply/demand equation and the price will come down. But the fact that they CAN charge that much means that there are enough people willing to pay that price. The profit goes to the corp. which is most likely a large public corporation whose shares are mostly owned by financial institutions that act as intermediaries that manage money for most of middle class America. So in effect, that extra profit is returned to society through stock price appreciation and/or dividends.
Prices of goods and services in the US are up across every single industry every single year running. Even a pack of gum is up!
Yup, it's called inflation. There are many causes of inflation, but I'm not going to spew out what you can read about in any basic economics book. It's a bad thing - the Federal reserve is on top of things to make sure it doesn't become rampant.
If we are a services oriented industry then why are service oriented jobs being outsourced? If we are not manufacturing anything and not providing a service to anyone or anything. What industry is exactly left in the US? You're argument confuses me here, if we don't have enough qualified people in the US, we don't have enough people willing to work below minimum wage. Americans are obviously lazy people even though on average Americans work more than any other country in the world.
We are a service oriented economy. Check out the GDP section of the CIA factbook on America, which shows that services make up almost 80% of our GDP. The US is still dominant in many industries (aerospace, defense, biotech, medical etc...) because we have the most innovative people on earth that continue to come up with new and valuable ideas consistently and relentlessly. Industries come and go, but for the past 150 years or so, it's been the US determining what comes and what goes.
I love America - I moved here from Canada years ago and for good reason. I'm a former H1B. You would think that Canada has a similar standard of living to America, but you'd be mistaken. It's much better here.
The world bends over backwards to provide goods and services to appease the demands of the all mighty American consumer. I didn't say anything about Americans being lazy. I just made a point that in some industries, there are people willing to do the same work for less money and corporations will be motivated to seek out those workers.
Tell me ONE THING we can innovate from an information standpoint that cannot be copied overseas and mass produced dirt cheap.
The internet has made the world hypercompetitive. You come up with a good idea and it is copied a short time later. That time window is your opportunity to make a profit, until you have to come up with another idea. The implications of this are either that the rate of new ideas increases, or that the ideas slow down or stop because the time/profit window is too small to be worthwhile. Sometimes the ideas are so complicated that it takes a long time to copy. The Asians have copied lots of semiconductor ideas, but my industry (micrprocessors) is still dominated by Americans. The same can be said for much of aerospace, pharacuticals, biotech and defense.
Financial services, legal services and healthcare services make up a tremendously large portion of our economic output. Protecting your methods and intellectual property is paramout to success. I don't think Goldman Sachs shares their proprietary trading methods, for example.
Electronic circuit design was my first career after college. I watched manufacturing being outsourced in the 80s. By the late 80s, it was clear that the engineering work would also be outsourced. I retooled myself to be a software developer and have been doing that for more than 10 years. Now, the same thing is happening to this line of work.
Jobs leave the US because the US workers are no longer competitive with foreign workers. If I am a corporation, I need to produce goods and/or services that will yield a profit otherwise the corporation will not exist. If I can deliver the same good/service with a lower cost of labor, I will do it - or my competitor will and will eat my lunch. What are the implications of this? Sometimes, workers lose their jobs to others willing to do the same for less. This results in the production of cheaper goods and services, but a potential loss of income for the non-competitive worker. Income is reduced, but prices are also reduced (due to a more efficient corporation). Dislocated workers can either choose to give up, or find new work. In the case of manufacturing outsourcing, the country has moved from a manufacturing/industrial base to a more service oriented economy. The jobs that used be stable and high paying (manufacturing) were taken over by more efficient workers overseas. The labor pool shifted towards services.
This is progress. Someday, engineering may be outsourced to people willing to do it for less. That means, that in order to stay competitive we need to INNOVATE. Find stuff that is so novel and valuable that only we can do it. This is also progress. The economy does this all of the time - think back to when the US was a agricultural based economy, and then BOOM, the industrial revolution obsoleted many manual laborers.
This phenomenon has been labelled by economists as Schumpeterian creative destruction. This disequilibrium that is created is painful and can last years, but we all benefit in the long run, as goods and services are created more efficiently leading to lower prices and hence higher standards of living - provided that the labor pool is willing to look for work in more valuable areas (farming -> industry -> services -> whatever_is_next).
>> What can be said to someone who refuses to pay his debts, who actively OPPOSES balanced budgets?
I know it's hard for a socialist to understand the basics of capitalism but bear with me.
Case 1: Let's say I borrow $100,000 at 5% interest for a term of 30 years. My monthly payment is $536.82 At the end of the 30 year term, I will have paid out $193,256 in total
Case 2: Let's say, I didn't want to take on any debt and I just put $100,000 up front. My monthly payment is $100,000 in the first month, but $0 for every month afterwards - sounds good huh? At the end of the 30 year term, I will have paid out $100,000 in total.
Wow, a 93% savings over 30 years? Sounds fantastic, let's get rid of the debt!
Not so fast... There's opportunity cost.
Let's say that as a productive person, I have the ability to earn 6% on capital. If I spend $100,000 up front... I don't have $100,000 anymore to earn a return on... At the end of 30 years, I will have paid $100,000.
If I take out my 5% loan... I pay $535.82 in the first month, but I have $99464.18 left over. Here's the KEY...I can earn 6% (amortized for that month) on that balance during that month. Next month, I have another $535.82 to pay, but I still have the balance left over to invest at a HIGHER rate of interest (6%).
So at the end of this 30 year exercise, I will have paid out $193,256 but I will have MADE $115,838 due to my 6% return on the unvested money, which means that I have SAVED $22,582. So my $100,000 worth of services have actually cost only $77,418 because I'm so godamm productive. This is a much better deal than paying up front without a deficit.
See now how lower interest rates motivate taking on MORE debt when your economy can grow faster?
You only want to reduce deficits when the rate of interest EXCEEDS the rate at which your economy can grow capital. I suppose that Canadians are so unproductive that they can't take on debt. Such is life in a socialist democracy.
>> Unlike loans, taxes have NO interest rate, NO maintinence fees, and do NOT have to be paid back.
Taxes redistribute wealth according to whatever the government wants. There is an opportunity cost to taxes. That is to say - what could I have done with the taxes if I had NOT sent it to the govt? In almost all cases, the citizen, privately deploying that money would be more productive than the government deploying that money.
>> Why do you so fervently believe that debt, in a society that can afford not to pay that extra overhead, is a good idea? Doesn't taking money from the future like that run counter to your ideals of smaller government?
By taking on more debt - you can REDUCE taxes with the same level of spending. Let the foreigners fund the govt. and let the citizens be more productive with the added capital. Growing the economy at a rate that is HIGHER than the interest rate means we come out ahead. See how that works? It's more productive to grow that capital privately - so take on debt to capture that advantage.
I guess we'll just have to disagree. That's why I left Canada and you still choose to live there.
Enjoy your high income taxes, your GST, your bloated and anti-capitalist government regulations, your smaller pay checks, your higher unemployment, your iPod tax, your 25% download tariff, your 15% webcast tariff, your 10% gamer tariff, your digital media tariff, your government-imposed limited media choices, your government imposed limited financial institutions and of course, your complete and total dependence on the prosperity and security of your far more productive neighbors to the south (you're welcome).
>> Seriously, you don't borrow money just because you can!!! You DO have to pay that money back someday, and the fact that the idea of not doing so is so popular is evidence that something is seriously wrong with the ideology you profess!!!
You're looking at absolute debt. The correct metric to look at is debt-to-GDP. If you get a large salary raise, you can move to bigger house - ie, take on more debt. Or, using your logic, trade down to a smaller house because I was irresponsible for borrowing for the first house in the first place - I need to "make up" for that wreckless borrowing.
Debt to GDP (ie - economic output/productivity) has been very stable. Guess what? when the economy grows due to the ingenuity and productivity of it's citizens, you can borrow more because your debt-to-equity is lowered - AND, on better terms (ie - lower interest rates).
>> Once again, look at what happened to Japan when the bubble burst. Their economy is very much like the american economy, in that it's artificially inflated by debt. The only difference is that their bubble has already burst
The Japanese bubble burst because of rampant asset inflation (example - the emperor's palace was worth more than the entire state of California). Their central bank was inept and still is to some extent. Price stability was not in their lexicon.
The US had an equity bubble already, but that was burst by the central bank taking rates to 6.5 percent in 2000. Problem solved. The US is nothing like Japan at all. We have sustainable business cycles and have had extremly robust economic growth for 200 years. We have competent central bankers. Again, there's a very good reason why the US is the largest, richest economy in the world, and has been for a good 150 years overtaking the British empire in the 1800s.
Actually, "Libertarian" is the word you're looking for. I don't buy into that Christian right garbage.
>> Dangerous - The government paying back it's debts is not an indication of overtaxation, it's an indication that the populous was undertaxed for a period of time, which caused the government to have to take a debt. Your ideas that defecits are ok fly in the face of the fact that the majority of our taxes go to paying MAINTINENCE on the debt.
Interest rates on govt. debt (particularly in the US) have been declining for the past 20 years. Therefore, why should we cut back debt? Lower interest rates imply that we should take on more debt. This is simple finance 101. If interest rates on treasuries rise - THEN we should cut back spending and reduce deficits. Interest rates reflect the risk others see in investing in a nation. Obviously the risk is low as rates have been decling for 20 years or so. Probably because the American population is the most productive, economically stable, wealthiest, largest economy in the world. There's a very good reason for that. There's also a reason why more than 1/2 of Canadian GDP is a result of trade with the US. You see, citizens are responsible for generating wealth and therfore generating tax revenue. Private interests are far more efficient than government interests with respect to growing capital. The more capital that is left to the private citizens, the faster it grows. Therefore, the more tax revenue the govt can pull in. Hence, the lower interest rates (an implicit representation of foreigner's trust in the ability of the US to grow). If the US started implementing European style socialism, you'd see rates skyrocket due to slower economic growth, and then we'd really be in trouble.
>> Selfish - This is pretty damned obvious. "Oh, they're overtaxing us! Waah!! They should give back that money instead of paying down the federal debt!"
The question you should be asking is - what is the most efficient way I can deploy money (capital). I can either give it to a central bureaucracy and trust it to invest. Or, I can let the market decide how to allocate capital. No one can argue that the market is more efficient. The central planners operate with limited information and therefore necissarly cannot allocate the capital in an efficient manner. The true meaning of freedom is to let the people decide what to do with their money - through supply and demand. Where do you think "demand" comes from? It is the collective wishes of the populace at large.
>> Stupid - When you take out a mortgage, you pay it back, or you lose your house. They have a word for people who live in defecit as you'd advocate: Bankrupt. Eventually the house gets paid off, and that's what people like you completely fail to recognise
Umm...then why are interest rates on US treasuries so low? Why have they been on a relentless 20 year downtrend? Are foreign investors (like Canadians) that pour trillions of dollars into the US treasury bond market stupid? They don't invest because they're our "buddies". They invest to make a profit. I'd say they're pretty damn happy with their investment and they continue to invest and drive bond prices higher and yields lower.
>> Yes. We build hospitals in norther communities too. We also give money to college students, build roads for people who couldn't afford to build them on their own, and maintain police forces, fire departments, and democratic institutions, all with taxpayer dollars.
I'm on the more radical (libertarian side), but perhaps a logical rethinking of the role of government is required. I'd recommend reading "Anarchy, State and Utopia" by Nozick to give a logical justification for a minimalist state (one that does not include funding schools, hospitals etc... Only a military system and justice system for enforcement of laws that protect individual liberties). It will never happen in my lifetime, but I believe it is a b
>> Comparing the US to Canada with a single number like that is ignorant
I grew up in Canada and have lived in the US for the past 9 years. I believe that I have a valid perspective on the situation and can comment qualitatively as well as quantitatively. If you believe that the average Canadian is wealthier or even on par with the average American, you are living in a fantasy world. By the way, as much as Canadians would like to think they are culturally distinct, the line is very thin. Per capital GDP is only one number, but it's an important one - it measures economic output. How about another economic metric like unemployment? Or net worth? Or income? Look these up on the web, and you'll see how relatively poor Canadians are vs. Americans.
>> the Federal Government of Canada has had a balanced budget for the better part of a decade
So what? That just says that the Canadian government is taxing too much. Perhaps the government should give back the surplus it has been running the past few years, so that the average Canadian can have more in his/her pocket to spend. Does having a balanced budget suddenly mean that Canadians can afford bigger houses and fancier cars? The government does not create wealth, the citizens do. The more control of capital that the citizens have, the better off the citizens are. Perhaps the Canadian govt. should look into *GASP* a deficit! Deficits are not all bad. Everybody that has a mortgage has a deficit. Sometimes, you need to borrow - sometimes, it's financially prudent because the opportunity cost of paying it ALL UP FRONT is too high vs. other uses of that money (like giving it back to the citizen to be more productive).
>> Broadband is an infastructure which the market can't decide it wants -- Either someone brings it in or they don't get it. Geektown NWT, with 5000 people willing to pay for broadband won't have broadband just because they want it, they won't get it until it is provided
So you're saying it's justified to redistribute wealth to appease the wishes of a group that does not have the financial means to attain their demands? You're saying it's ok to take money from someone that doesn't give a shit about broadband (money that this person made from his/her own productivity) and re-direct it? That is theft. It is immoral. It is wrong. It is also bad for the economy as it distorts demand, resulting in capital being allocated inefficiently. If you live in the NWT and you want broadband, you should have to pay the market price for it. If the is cost prohibitive to deliver broadband to the NWT, so be it! That's the market. Don't live in the NWT if you want cheaper broadband, or pay up!
>> The same was true when the interstate highway system was installed in the US half a century back
I believe that a private highway system would have been far more efficient. There is just too much economic incentive out there for private capital to invest in interstate highways. These would have been built by private enterprise in the absence of government and funded on a usage basis (like tolls). Look at the railways in the 1800s. They were for the most part private, and were built out by private enterprise.
>> A lot of the systems folks that I've been hearing from and things that I've been reading have suggested that, like it or not, multi-core systems are the way of the future. The argument is that the clock-speed aspect of Moore's law has been slowing down for the past couple of years and that we've seen single processors that are as fast as they'll go with current chip design and fabrication technology. (Barring fundamental breakthroughs, of course.) Hence parallelism and multi-core systems.
You are exactly right. The burden of extracting parallelism will shift squarely on the programmer in the coming years. The hardware has been extracting ILP (instruction level parallelism) for you automatically for several years now.
But, the cost of extracting additional parallelism out of a single thread is extremely power-prohibitive.
Intel released a paper on silicon scaling outlining the problem (google for "limits to binary logic switch scaling - a Gedanken model). There is a minimum amount of energy required to implement a switch and a maximum rate at which heat can be dissipated from a solid. In the past 30 years, computer users have benfitted from both an increase in switching speed AND an increase in density. Because of thermal limits, we are entering an era where you will either have to scale switching speed (frequency) OR density. Given the choice - Intel/AMD and others are choosing density, because you can still get performance with MORE transistors (through parallelism) even though the frequency remains constant.
>> This isn't a suprise. The free market is good, but not as good as pre-existing infastructure and a government mandate
Government mandates eliminate choice and result in an inefficient allocation of capital. What about Canadians that don't give a shit about broadband - they have to pay anyways because of higher taxes. Let the market decide. Not some govt. bureaucrat. It's policies like this that made me leave Canada for the US about 9 years ago.
If people don't want broadband, then they should have the CHOICE not to fund it. Just as it is in the US.
Let's take a look at one measure of wealth - per capita GDP and compare Canada vs. the US:
http://www.cia.gov/cia/publications/factbook/ran ko rder/2004rank.html
>> using up extra real estate on the die of the Athlon in order to get SMT isn't really worth it in terms of the performance it would bring. Even on the Pentium the benefits aren't all that hot and it's only in specific types of code that you see any impresive speed gains
The real estate used is only about 5% on a P4. If you get more than 5% return in performance (as you do in many cases), then it's a win. It's really the complexity of it all that kills it for AMD - they can't afford the engineering resources to put something like that in. SMT was actually implemented on Willamette (the first P4) way back in 2000, but it was disabled until the engineers could get it to work a few years later.
It may interest you to know that many resources on an out-of-order machine are often idle. On an Athlon, you may get about 1-1.5 instructions per clock throughput on average when the peak is 3. Adding another thread can more fully utilize those idle resources. Beyond 2 threads, the resources are probably saturated (1.5 x 2 = 3).
Then you move to dual core, where your execution resources double again.
So that raises the question, why not build a wider processor with more resources but with a single thread and let the hardware find the parallelism for you INSTEAD of dual-core/SMT where the burden is on the programmer? Short answer - it takes too much power for hardware to extract paralellism.
This isn't some cop-out by "lazy" circuit designers. This is fundamental (ie - the laws of physics work that way).
Sorry coders - your lives will get harder and harder as the years go on and many-threaded processors sell a 100M units in volume each quarter. Better get to work on making parallel programming easier, or you won't see the performance gains that you've come to expect from Moore's Law.
>> No DVD player in the Dell, nor FireWire, nor a modem, nor a stack of bundled software, nor 90 days of free telephone support. Nor is it small, or silent. Laptop technology, which is what the Mini uses, is more expensive.
Subtract out the 19" LCD (approx $300) and you have a $229 system. Now add back in the DVD, firewire, modem and blah blah blah and it is still much cheaper.
It isn't small form factor, but nothing is really stopping a company like Dell from offering such a system. If it turns out that this form factor is what people want and the Mac Mini is a hit, then I'm sure Dell will have a cheaper offering. Apple just doesn't have the manufacturing infrastructure to compete with Dell on price. Hell, even giants like IBM and HP can't compete with Dell on price.
>> Until AMD has the capacity to mostly replace Intel, Dell just has to smile and say "Do you want HypeThreading with that?"
At 3 billion dollars per fab, it'll be a cold day in hell before that happens. AMD barely makes money as it is. Every 3 months, Intel makes more money than all of AMD is worth (market cap wise).
>> Next year, when AMD's new 65nm fab is up and running and Charter (and IBM?) start fabbing AMD CPUs too, THEN things will get interesting.
You're assuming Intel is going to stand still and keep mis-executing on 65nm and let AMD walk all over them. Yes, they've mis-executed the past few years but would you bet against them in the future given their 15 year track record with the Pentium brand?
What happened to Jerry Sander's maxim - "Real men have fabs". AMD is at a serious cost disadvantage compared with Intel on the manufacturing front.
AMD is a one-trick pony with the Athlon/Opteron. Where is their answer to Centrino when the entire market is shifting towards notebooks? Voltage-scaled Athlons/Opterons don't really cut it in that market as evidenced by Intel's 88% notebook share.
>> To my understanding, in the Itanium, you don't execute both paths indefinitely, only so long as the predicate register is "pending".. As soon as the register is no longer pending, all instructions bound to the false-predicate are skipped over.
This is equivalent to branch-resolution delay in a branch-predicted machine. If you mispredict, you're only wasting resources until the branch is resolved upon which time you get back on the right path. You only waste if you guess wrong as opposed to Itanium where you waste the resolution delay all of the time. Of course you could add prediction to Itanium, but then now you have the same latency problem of the misprediction so where is the benefit? It's a cute idea, but it doesn't work in practice - see SPECint scores for the Itaniums.
>> predicate-execution is done via the consolidation of super-block code chunks into hyper-block code chunks...
Yes, but the code expansion is still there. The combinatorial explosion also affects the compiler. There are only so many paths that the compiler can evaluate before it has to give up because it would take too long to consolidate into a block.
Last time I checked, it took 7 hours to compile GCC optimized on an Itanium.
>> Larger cache is a fact of life.. And with multi-megabyte caches finding there way into home-PC's, asking for a larger instruction cache isn't the end of the world to me. Not to mention, last I checked, it was the data-cache with the greater contention.
Larger caches relative to what? If your code expands 50% because of your ISA, then a competing ISA (ie - x86) can have live with a 50% smaller cache. This means more area on the chip is available for other more productive resources. I have already tried to make the case that the benefit of code expansion (predication) is pretty much zero and even negative vs. branch-prediction.
It's not only cache size, it's also memory bandwidth which is very expensive. It's instruction fetch bandwidth. Execution bandwidth... etc.. etc.. You pay extra every single time.
>> Does the latest x86 allow you to load a potentially invalid memory address and not throw an exception if the memory is never used?
There are non-faulting prefetch instructions. However, the usefulness of software prefetching is limited given a highly effective hardware prefetcher. I speak from experience as I hold several HW prefetcher patents:)
>> Another danger is that by finding other work to do, additional memory requests might be queued up, providing contention with the critical-path memory-fetch
Oldest loads are given priority in OOO machines. Critical block forwarding is also a common feature. Everything in the design is made to insure the fastest possible execution.
>> However, I still contend that if two architectures are identical in throughput (seconds / instruction), the one with the shorter pipeline has the advantage (to a point, of course). Each additional stage adds overhead and adds to the penalty of branch mispredictions (which thereby merits extra architectural CPU resources to beef up the predictor).
If two architectures have the same instructions/second, then they have the same performance. Long pipe or short pipe, they take exactly the same amount of time to run the program.
My argument is this - you cannot attain the same instructions/seconds (performance) with a 5 stage pipeline as you can with a 15 stage pipeline (designed correctly). Deepening the pipe improves performance up to a point, after which it degrades performance. Many papers have been written outlining the tradeoffs of pipeline depth vs. performance.
>> I find it odd that you seem perfectly happy with a JIT / branch-prediction, yet are worried about throwing resources at caches or register files. Obviously, I'm biased in the other direction, so to-each-his-own philosophy I guess.
>> The Itanium executes all possible execution paths (that are serialized; no branches) until one or more paths are resolved. Thus in a two-way logical branch, if it takes 20 instructions to resolve the dependency. If 10 instructions were the true-path and 10 instructions were the false path, you would have no risk of having "branch predicted" the wrong path. Put that into a 15+ stage plus architecture and a wrongly chosen path starts hurting.
Executing both paths isn't free. It's way more power. Let's take a 2-way branch --- you're then using TWICE the compute resources than are necessary ALL OF THE TIME until the branch is resolved. Furthermore - your code size blows up - in fact for highly complex conditionals, you get a combinatorial explosion. This means that your caches need to be MUCH larger to hold more code.
Contrast this to a branch-predicted machine which is much more efficient if you have a good predictor. You only waste compute resources if you predict wrong as opposed to ALWAYS wasting resources with predication. History has shown that good predictors are possible to build and the costs of mispredicts is much smaller than the cost of predicated code (since we mispredict so rarely).
>> Additionaly, the Itanium has nice memory pre-fetching instructions with delayed consequences. I know the Pentium Pro added a couple of pre-cach loading instructions; don't know if they've incorporated the equivalent though.
x86 has prefetch instructions. Furthermore, dynamic hardware prefetchers are far more effective than software prefetches because the hardware knows more about the memory access pattern (a dynamic run-time type of information).
>> Additionally, the register-rolling capability of the Itanium was interesting. Theoretically providing an optimal for-loop, where excluding setup-tear-down, every iteration of the loop took as little as 1 logical clock-tick. (At least as far as the software is concerned; the hardware had room to grow in EPIC-bundles / clock).
A terrible idea. It blows up the size of the register file thus limiting the frequency of the entire chip. It's a bad tradeoff - make EVERYTHING slower so you can unroll loops. They are bad on SPARC (which I have designed) and they are bad on EPIC.
>> I'm not sure how you can say this. Let me restate this: As the speed of the Memory approaches the speed of the CPU, the number of NOP-cycles is reduced in the CPU. What I meant was that if memory speed is the bottle-neck, and we can't cache enough of it to appreciably overcome that bottleneck, then we should look for valueable tradeoffs between clock-speed and work-done-per-instruction. The P4 dramatically traded off clock-speed for work-done-per-instruction and thereby produced less overall performance than an architecture with fewer stages. (Granted, it's an green-apples to red-apples comparison of the AMD-64 and the P4). Thus, if an archetecture has two equivalent performance characteristics but of different clock-speeds, but the memory speed is drastically slower, then the slower CPU-speed should be chosen so that the impeedence mismatch is reduced. With the exception of critical-paths-fitting-in-cache applications, general system performance should be improved.
I'm not sure what you mean here. Computer performance is measured in time.
The time it takes to fetch something from DRAM cannot be improved due to the laws of physics. It doesn't matter how fast you wiggle the clock, 70 ns to DRAM is about as good as you're going to get.
Performance = dynamic_program_length_in_instructions * CPI / frequency
Performance = time to execute program CPI = cycles per instruction
Program length can be reduced statically with a compiler and other dynamic tricks.
Less work per clock means a lower CPI but it also implies a higher frequency. CPI is a function of frequency.
For the parts of the program that are not memory-bound, you can speed up with a higher frequency (assuming
>> Most steps of computational algorithms are inherently dependent on the results of previous steps, and are thus not parallelizable. single-threaded CPU's have gotten VERY good at parallelizing individual instructions
This makes no sense. If the instructions are dependent, they are dependent, there's nothing you can do to reduce latency other than increasing frequency. The effect of increasing frequency doesn help the parts of your instruction that includes some delay through an RC path since RC-delay is independent of switching speed.
>> The compilers aren't well suited for helping the CPU out, so things like the Itanium were supposed to exploit such parallelism. But the loss of backward compatability (and the Itanium's focus on floating point) spelled the death nell for that architecture.
Itanium is stupid because there is lots of information that is not available at compile-time rendering the compiler useless (ie - branch prediction and caching behavior). Dynamic information is needed and is VERY important for extracting parallelism. That's why out-of-order CPUs routinely trounce in-order VLIWs like Itanium. Yes, Itanium has stellar FP performance, but that's only because their caches and memory busses are overdesigned relative to the competition. Level the playing field with comparable memory systems. IE - put a 6-9 MB L2 cache on a P4 or an Athlon and double the memory bandwidth and see who gets higher FP performace - I can assure you that the Itanium will get smoked.
>> Personally, I think the answer is taking a step back from MHZ and pipelining. Go back to a 3, 4 or 5 stage pipeline with MASSIVE read-ahead decompilation of instructions (similar to transmeta).
Sounds like Itanium. See above. Using a dynamic compilation technique like Transmeta doesn't work as well as say - doing an out-of-order machine because the delay from finding out something new dynamically to reacting to it is way too long for software to handle. Hardware is much more effective for this. Things are already decompiled dynamically by hardware (P4's trace cache and AMD's pre-decoded instruction cache).
3,4 5 stages? Do you work for Motorola? There are many ways to compute the optimal pipeline depth. I can assure you that 3, 4, 5 is very non-optimal. Start with the Sprangle/Carmean paper and follow up with some IBM papers.
>> Get lots of high-speed cache on board with as little latency as possible (current large-cache architectures have HUGE latancies). By lowering the CPU MHZ, you reduce the latency to the all-important main-memory.
Frequency has nothing to do with latency. If it takes say 10 nanoseconds to access a cache (due to the physics of propagating electrons across a distance - RC delay) then it takes 10 nanoseconds. 10 cycles at 1 GHz is just as long as 1 cycle at 100MHz. Hint - cache access time gets longer for larger caches because the cache is BIGGER and the signal needs to travel a larger distance. Until you can implment wormholes in chips, I don't see the problem getting solved.
What you really want are smarter dynamic prefetching algorithms into a cache heirarchy so the latency can be hidden.
>> Advance the state-of-the-art in power-consumption
Absolute power is very optimized already. You might not think so because of the prevalance of 100W CPUs but there are very good reasons they're not 200W instead.
What architects are focusing on now is power-efficiency. Performance per watt is the new thing to optimize (as opposed to performance per mm^2 which is still a constraint).
>> This is going to be a huge industry. If they could produce, say, a 800MHz CPU which ran on 1W or 0.5W of power and had sensible float performance, it could easily sell exceedingly well.
800MHz is a useless number. It says nothing about performance.
Anyway, if there is such a huge market, don't you think Intel or AMD or 30 or so other semiconductor companies would have something to deliver to that market? What's so special about Transmeta other than the fact that they used to employ Linus? Their "competetive advantage" was their code morph technology which has failed miserably in the marketplace. It's also technically not that great of an idea. Now they're staking their future on their chip making ability? Sorry, in this case they are beat cold - Intel and AMD have 100x the engineering budget of Transmeta. Not to mention Motorola (Freescale), Texas Instrument, IBM etc... IE - companies that know how to deliver silicon to the marketplace and make money.
Slashdot Mirror
Do you still think AMD will have the edge in 2006? 2007? 2008?
Get real. Intel makes more than AMD is worth every 3 months. P4 was a blunder. Don't expect future chips to be as bad.
Intel has the fab capacity to supply Apple without any problem. AMD is always fab limited. AMD is correctly attacking a niche (servers). This is where they will shine.
Some statistics that I have seen show something on the order of 80% of the dollar revenue of game publishers comes from consoles with PCs getting the remaining 20% (check out the financial stats of a publicly traded company like EA).
Now put yourself in the shoes of a game developer...What is going to make you the most money? A 20% dollar market or an 80% market?
Compound the fact that developing games for the PC is far more expensive -- there are all sorts of different hardware configs and not always stable driver headaches to deal with. This means longer testing times and therefore more expense. Remember that developer time is the most expensive resource in a game company.
Constrast that to a console which has a very consistent and well documented hardware platform. It's easier (read cheaper) to develop for a couple of configurations (xbox/ps).
Now even a 20% PC market is huge, so there will probably always be PC games. Another reason is that most development and content creation is done on PCs. But do not under estimate the enormous advantages of the console from a game developer's perspective. The PC market will increasingly shrink and fall into a niche when it comes to gaming.
Not to mention the big cost advantages of the console with their subsidized hardware model. Yes there are people willing to pay $400 for a graphics card, but these people are definitely in a niche market. The bulk of the money is made in the low-end to mid-range segment.
This is a silly argument.
What possible motivation is there for Microsoft to move to a different ISA? Almost all of the world's software is written for x86. Moving to a different ISA would seriously weaken Microsoft's monopoly position because all of the tools and infrastructure (drivers, compilers, debuggers etc...) are build around x86. Microsoft would commit financial suicide if this happened not only because the ISVs would be pissed, but for another reason - There simply isn't enough capacity to supply NON-x86 silicon to meet global demand (computers would start costing $5000 again - not a good thing for Microsoft).
Intel supplies 80% of the worlds microprocessors. Nobody else has the fab capacity to do this. It takes a few years to put up a fab and each one costs about 3 billion dollars (which keep getting even more expensive for each process generation). IBM/AMD/others simply don't have the capital structure to build all the fabs necessary to meet global demand. Who is going to put up the capital? Microsoft? Why? The already enjoy monopoly pricing. How would moving to a different ISA give them any kind of a return on investment given the costs?
Building microprocessors is so expensive that DEC-Alpha, HP-PA RISC and SGI-MIPS have all thrown in the towel yeilding to Intel. SPARC will be the next to go. If Apple goes with Intel, expect POWER to be the next to die (sorry but console profits are tiny).
Here's my prediction. In 10 years, everything will run on x86. Yes, even PDAs and cell phones. The risk to Intel is that more competitors start building x86 processors (AMD is too small). Maybe Samsung or some other company with the capital requirements and semiconductor expertise necessary to build out fabs.
In fact, I think that Microsoft is more at risk.
They are at risk on the desktop/laptop side from Apple and Linux. Both of which run x86 (Admittedly, the Apple thing is still a rumor).
Let's say the market moves away from the PC to a more server/thin-client model. Microsoft gets cut out of the picture (think Google already running their software services on x86-Linux and upstarts like RIMM/Blackberry running their software on the clients). Google runs on Intel. RIMM does not, but Intel is already a leader in the handheld space (think ARM).
Now let's consider the diversification argument by listing the markets that require computing power.
PCs/Desktops? No question here. Intel has a dominant market share in both processors and chipsets.
Graphics? Intel is #1 in graphics marketshare with their integrated graphics solution.
How about notebooks? Intel is the clear leader in that rapidly growing space. Transmeta tried to compete and got crushed. AMD is all but invisible with their one-trick opteron pony. The market is moving here and Intel is there to capitalize.
PDAs? Intel inside with StrongARM.
Cell phones? Intel inside (with their flash chips).
Servers? Intel is again very big here with over a 90% share of x86 servers.
Embedded? No Intel there. The reason? There's no money there...
Consoles? No Intel there. There's very little money to be made with silicon. All of the value is in the software. NVIDIA only made about 30 million on XBOX (Intel made something similar). ATI is reported to make about 2$ per console on XBOX360. To put it into perspective, Intel makes about 600 million every single week.
Diversification has worked beautifully for companies like IBM and GE. 10 years ago, I do not think that IBM would have been able to dump its PC business without significantly damaging themselves.
IBM is valued less than Intel by about 30%. GE has been around for 100 years, Intel is only 30.
In the 80s Intel was concentrated on making memory chips but were getting killed by the Japanese, and in a heartbeat, diversified into microprocessors. Until the demand for microprocessors diminishes, Intel will keep making them. It's hard to stop when you still have 60% gross margins. If there's value and growth in silicon, Intel is well positioned to take advantage with their enormous fab capacity and their army of skilled engineers.
Nowhere does it say x86. This could be an agreement for Intel to get into the PPC business, which would be a great supplier coup for Apple
Unlikely.
The global PC market produces 200 million PCs a year. Apple sells 3 million of those.
The incremental revenue for Intel is a rounding error. Why would they devote an entire design team to supply such a tiny market when they can get a much higher return on investment with x86? In case you haven't heard - microprocessor development is ridiculously expensive. You need huge volumes to justify the capital and R&D expenditures.
Hmm...I wonder why IBM microelectronics loses money every single year, and AMD barely makes any money even with a superior architecture?
The benefits for Apple are huge. They get far cheaper processors relative to IBM which means Apple can increase profit margins. I'm suprised they haven't gone with Intel sooner - they've probably spent all this time perfecting the software.
Intel can supply chips at will (think 200M units per year vs. Apple needing 3M). IBM has had problems keeping up with the demand for even this small market. They are not cost competitive and cannot deliver in the volumes that Intel can. AMD can't either. Unless you have the capital to invest 5-7 Billion dollars a year on Fabs, you just can't beat Intel on a manufacturing basis which is why Intel is bigger than AMD, Apple and IBM combined (on a market cap basis).
>> And I don't even know what the URL for Microsoft's search engine is. Do you? Does anyone? I never hear anyone talk about it, never hear anyone refer to it, mention it, use it.
When they integrate search into Longhorn, that's when the threat becomes more real. Google is doing a pre-emptive strike with Google-Desktop-Search, but microsoft counters with their own.
I think Microsoft is losing currently, but I wouldn't be so quick to count them out. Msn.com is one of the most frequently visited websites by default. Many non-techies don't bother to change the default home page on their IE browsers. Owning the OS gives them tremendous leverage.
Hopefully, Google can make the OS less relevant.
You're not a snob, you're correct.
:)
Waterloo is a great school and their co-op program is more valuable than any traditional education you would get at a typical North American Ivy school.
The co-op program allows you to explore what you really would like to do in the real world, instead of some abstraction in the classroom. You get the best of both worlds - a solid academic background and a real taste of the engineering world.
--
Waterloo Computer Engineering Grad 1996
"In the long run, we're all dead." - John Maynard Keynes
Yes, I am familiar with Keynes and his famous quote.
When there is a sharp and rapid supply/demand imbalance, it can take a very long time for the economy to stabilize into equilibrium. Even the most zealous free-marketer will agree that if I jump off a cliff there isn't enough time for someone to setup shop and sell me a parachute on the way down...
The Federal reserve can soften the pain with interest rate adjustments so that the equilibrium can be reached faster. This is probably the best we can do - it's a blunt instrument but it helps.
The fact that it can take time to stabilize doesn't alter the fact that change happens and can happen quickly.
Change is normal.
Change can be good.
I'm replying because you quoted my favorite author Heinlein. Great quote BTW :)
That's real nice rhetoric. I'm not sure whether it falls under the "Appeal to Pity" or "Complex Cause" fallacy. Let's call greed and laziness exactly what it is.
You're reading way too much into what I said. I'm not appealing for anything. I'm just stating a value-neutral fact. All I said was that corporations need to make a profit - this is a pretty well established fact. Given this fact, if you can produce the EQUIVALENT good or service using lower labor costs - you will go for the lower labor costs because increasing profitability is the main function of a corporation. Another fact that is pretty hard to dispute. It's like saying force = mass * acceleration. It's a value neutral fact - whether is it good or bad, I'll leave to the philosophers.
As a consultant, I've observed many cases where my client companies have decided to cut the wages of the peon while executives and board members received dividend and bonus increases, and released reports that complained about the cost of doing business as validation for the "team" to make sacrifices.
In this example, you have to ask yourself if the labor that has been cut is scarce? If it is and the scarce peons decide to leave, then the company is screwed isn't it? Because it will no longer be able to function in a similar manner and this is a sign of poor management.
However, if the labor is plentiful and there are other equivalently skilled people in the labor pool willing to do the work for less, then perhaps they are just maximizing profits with no longer term risk to their business.
Why would the team ever agree to sacrifice if they could find better work elsewhere? It must be that the costs of labor are declining because they are unable to find a better deal. Or - the managers are just moronic and don't realize the scarcity of labor (if it does exist), in which case the people would leave for work elsewhere and the profitability of the firm will suffer and the market will eliminate them.
Everybody acts in their own selfish interests whether they like to admit it or not. However, what many people don't realize is that maximizing self interest often involves trade and cooperation.
You'll find few if any "workers" who are willing to do the same for less in a market with rules or law. In a market with rule and law this argument clearly makes no sense. The 12 yr old in a country with no law, stitching up a tshirt for export to an entirely different country.
You have to ask yourself this...What is the alternative for the 12 yr old in a poor country? In many cases they are working because it is the best alternative for them. If you slap tarriffs on imports from countries that allow this, suddenly, instead of making some money, the child makes no money and has to continue to live in abject poverty. Who are you to take away that income from them to protect American workers who already enjoy a vastly superior standard of living? This is not a zero sum game - the standard of living is raised for the poorer workers and the standard of living is raised for the average american through either lower prices or higher profits. There is a loser in the equation (the non-competitive worker), but that's how the world works. Business models change, things that worked yesterday don't necissarily work today (The buggy whip industry has been in decline for over a century)
Also the markup price for the tshirt is clear in its final price. Have you seen some of the retail prices for clothing in the US? 40-50-60 US dollars for a shirt
No one is forcing you to pay $50 for a shirt. If you don't like the price, don't buy - that will affect the supply/demand equation and the price will come down. But the fact that they CAN charge that much means that there are enough people willing to pay that price. The profit goes to the corp. which is most likely a large public corporation whose shares are mostly owned by financial institutions that act as intermediaries that manage money for most of middle class America. So in effect, that extra profit is returned to society through stock price appreciation and/or dividends.
Prices of goods and services in the US are up across every single industry every single year running. Even a pack of gum is up!
Yup, it's called inflation. There are many causes of inflation, but I'm not going to spew out what you can read about in any basic economics book. It's a bad thing - the Federal reserve is on top of things to make sure it doesn't become rampant.
If we are a services oriented industry then why are service oriented jobs being outsourced? If we are not manufacturing anything and not providing a service to anyone or anything. What industry is exactly left in the US? You're argument confuses me here, if we don't have enough qualified people in the US, we don't have enough people willing to work below minimum wage. Americans are obviously lazy people even though on average Americans work more than any other country in the world.
We are a service oriented economy. Check out the GDP section of the CIA factbook on America, which shows that services make up almost 80% of our GDP. The US is still dominant in many industries (aerospace, defense, biotech, medical etc...) because we have the most innovative people on earth that continue to come up with new and valuable ideas consistently and relentlessly. Industries come and go, but for the past 150 years or so, it's been the US determining what comes and what goes.
I love America - I moved here from Canada years ago and for good reason. I'm a former H1B. You would think that Canada has a similar standard of living to America, but you'd be mistaken. It's much better here.
The world bends over backwards to provide goods and services to appease the demands of the all mighty American consumer. I didn't say anything about Americans being lazy. I just made a point that in some industries, there are people willing to do the same work for less money and corporations will be motivated to seek out those workers.
Actually, what killed the agricultural market
Tell me ONE THING we can innovate from an information standpoint that cannot be copied overseas and mass produced dirt cheap.
The internet has made the world hypercompetitive. You come up with a good idea and it is copied a short time later. That time window is your opportunity to make a profit, until you have to come up with another idea. The implications of this are either that the rate of new ideas increases, or that the ideas slow down or stop because the time/profit window is too small to be worthwhile. Sometimes the ideas are so complicated that it takes a long time to copy. The Asians have copied lots of semiconductor ideas, but my industry (micrprocessors) is still dominated by Americans. The same can be said for much of aerospace, pharacuticals, biotech and defense.
Financial services, legal services and healthcare services make up a tremendously large portion of our economic output. Protecting your methods and intellectual property is paramout to success. I don't think Goldman Sachs shares their proprietary trading methods, for example.
Electronic circuit design was my first career after college. I watched manufacturing being outsourced in the 80s. By the late 80s, it was clear that the engineering work would also be outsourced. I retooled myself to be a software developer and have been doing that for more than 10 years. Now, the same thing is happening to this line of work.
Jobs leave the US because the US workers are no longer competitive with foreign workers. If I am a corporation, I need to produce goods and/or services that will yield a profit otherwise the corporation will not exist. If I can deliver the same good/service with a lower cost of labor, I will do it - or my competitor will and will eat my lunch. What are the implications of this? Sometimes, workers lose their jobs to others willing to do the same for less. This results in the production of cheaper goods and services, but a potential loss of income for the non-competitive worker. Income is reduced, but prices are also reduced (due to a more efficient corporation). Dislocated workers can either choose to give up, or find new work. In the case of manufacturing outsourcing, the country has moved from a manufacturing/industrial base to a more service oriented economy. The jobs that used be stable and high paying (manufacturing) were taken over by more efficient workers overseas. The labor pool shifted towards services.
This is progress. Someday, engineering may be outsourced to people willing to do it for less. That means, that in order to stay competitive we need to INNOVATE. Find stuff that is so novel and valuable that only we can do it. This is also progress. The economy does this all of the time - think back to when the US was a agricultural based economy, and then BOOM, the industrial revolution obsoleted many manual laborers.
This phenomenon has been labelled by economists as Schumpeterian creative destruction. This disequilibrium that is created is painful and can last years, but we all benefit in the long run, as goods and services are created more efficiently leading to lower prices and hence higher standards of living - provided that the labor pool is willing to look for work in more valuable areas (farming -> industry -> services -> whatever_is_next).
>> What can be said to someone who refuses to pay his debts, who actively OPPOSES balanced budgets?
I know it's hard for a socialist to understand the basics of capitalism but bear with me.
Case 1:
Let's say I borrow $100,000 at 5% interest for a term of 30 years.
My monthly payment is $536.82
At the end of the 30 year term, I will have paid out $193,256 in total
Case 2:
Let's say, I didn't want to take on any debt and I just put $100,000 up front.
My monthly payment is $100,000 in the first month, but $0 for every month afterwards - sounds good huh?
At the end of the 30 year term, I will have paid out $100,000 in total.
Wow, a 93% savings over 30 years? Sounds fantastic, let's get rid of the debt!
Not so fast... There's opportunity cost.
Let's say that as a productive person, I have the ability to earn 6% on capital.
If I spend $100,000 up front...
I don't have $100,000 anymore to earn a return on...
At the end of 30 years, I will have paid $100,000.
If I take out my 5% loan...
I pay $535.82 in the first month, but I have $99464.18 left over.
Here's the KEY...I can earn 6% (amortized for that month) on that balance during that month.
Next month, I have another $535.82 to pay, but I still have the balance left over to invest at a HIGHER rate of interest (6%).
So at the end of this 30 year exercise, I will have paid out $193,256 but I will have MADE $115,838 due to my 6% return on the unvested money, which means that I have SAVED $22,582. So my $100,000 worth of services have actually cost only $77,418 because I'm so godamm productive. This is a much better deal than paying up front without a deficit.
See now how lower interest rates motivate taking on MORE debt when your economy can grow faster?
You only want to reduce deficits when the rate of interest EXCEEDS the rate at which your economy can grow capital. I suppose that Canadians are so unproductive that they can't take on debt. Such is life in a socialist democracy.
>> Unlike loans, taxes have NO interest rate, NO maintinence fees, and do NOT have to be paid back.
Taxes redistribute wealth according to whatever the government wants. There is an opportunity cost to taxes. That is to say - what could I have done with the taxes if I had NOT sent it to the govt? In almost all cases, the citizen, privately deploying that money would be more productive than the government deploying that money.
>> Why do you so fervently believe that debt, in a society that can afford not to pay that extra overhead, is a good idea? Doesn't taking money from the future like that run counter to your ideals of smaller government?
By taking on more debt - you can REDUCE taxes with the same level of spending. Let the foreigners fund the govt. and let the citizens be more productive with the added capital. Growing the economy at a rate that is HIGHER than the interest rate means we come out ahead. See how that works? It's more productive to grow that capital privately - so take on debt to capture that advantage.
I guess we'll just have to disagree. That's why I left Canada and you still choose to live there.
Enjoy your high income taxes, your GST, your bloated and anti-capitalist government regulations, your smaller pay checks, your higher unemployment, your iPod tax, your 25% download tariff, your 15% webcast tariff, your 10% gamer tariff, your digital media tariff, your government-imposed limited media choices, your government imposed limited financial institutions and of course, your complete and total dependence on the prosperity and security of your far more productive neighbors to the south (you're welcome).
>> Seriously, you don't borrow money just because you can!!! You DO have to pay that money back someday, and the fact that the idea of not doing so is so popular is evidence that something is seriously wrong with the ideology you profess!!!
You're looking at absolute debt. The correct metric to look at is debt-to-GDP. If you get a large salary raise, you can move to bigger house - ie, take on more debt. Or, using your logic, trade down to a smaller house because I was irresponsible for borrowing for the first house in the first place - I need to "make up" for that wreckless borrowing.
Debt to GDP (ie - economic output/productivity) has been very stable. Guess what? when the economy grows due to the ingenuity and productivity of it's citizens, you can borrow more because your debt-to-equity is lowered - AND, on better terms (ie - lower interest rates).
>> Once again, look at what happened to Japan when the bubble burst. Their economy is very much like the american economy, in that it's artificially inflated by debt. The only difference is that their bubble has already burst
The Japanese bubble burst because of rampant asset inflation (example - the emperor's palace was worth more than the entire state of California). Their central bank was inept and still is to some extent. Price stability was not in their lexicon.
The US had an equity bubble already, but that was burst by the central bank taking rates to 6.5 percent in 2000. Problem solved. The US is nothing like Japan at all. We have sustainable business cycles and have had extremly robust economic growth for 200 years. We have competent central bankers. Again, there's a very good reason why the US is the largest, richest economy in the world, and has been for a good 150 years overtaking the British empire in the 1800s.
>> Ah, a "Conservative".
Actually, "Libertarian" is the word you're looking for. I don't buy into that Christian right garbage.
>> Dangerous - The government paying back it's debts is not an indication of overtaxation, it's an indication that the populous was undertaxed for a period of time, which caused the government to have to take a debt. Your ideas that defecits are ok fly in the face of the fact that the majority of our taxes go to paying MAINTINENCE on the debt.
Interest rates on govt. debt (particularly in the US) have been declining for the past 20 years. Therefore, why should we cut back debt? Lower interest rates imply that we should take on more debt. This is simple finance 101. If interest rates on treasuries rise - THEN we should cut back spending and reduce deficits. Interest rates reflect the risk others see in investing in a nation. Obviously the risk is low as rates have been decling for 20 years or so. Probably because the American population is the most productive, economically stable, wealthiest, largest economy in the world. There's a very good reason for that. There's also a reason why more than 1/2 of Canadian GDP is a result of trade with the US. You see, citizens are responsible for generating wealth and therfore generating tax revenue. Private interests are far more efficient than government interests with respect to growing capital. The more capital that is left to the private citizens, the faster it grows. Therefore, the more tax revenue the govt can pull in. Hence, the lower interest rates (an implicit representation of foreigner's trust in the ability of the US to grow). If the US started implementing European style socialism, you'd see rates skyrocket due to slower economic growth, and then we'd really be in trouble.
>> Selfish - This is pretty damned obvious. "Oh, they're overtaxing us! Waah!! They should give back that money instead of paying down the federal debt!"
The question you should be asking is - what is the most efficient way I can deploy money (capital). I can either give it to a central bureaucracy and trust it to invest. Or, I can let the market decide how to allocate capital. No one can argue that the market is more efficient. The central planners operate with limited information and therefore necissarly cannot allocate the capital in an efficient manner. The true meaning of freedom is to let the people decide what to do with their money - through supply and demand. Where do you think "demand" comes from? It is the collective wishes of the populace at large.
>> Stupid - When you take out a mortgage, you pay it back, or you lose your house. They have a word for people who live in defecit as you'd advocate: Bankrupt. Eventually the house gets paid off, and that's what people like you completely fail to recognise
Umm...then why are interest rates on US treasuries so low? Why have they been on a relentless 20 year downtrend? Are foreign investors (like Canadians) that pour trillions of dollars into the US treasury bond market stupid? They don't invest because they're our "buddies". They invest to make a profit. I'd say they're pretty damn happy with their investment and they continue to invest and drive bond prices higher and yields lower.
>> Yes. We build hospitals in norther communities too. We also give money to college students, build roads for people who couldn't afford to build them on their own, and maintain police forces, fire departments, and democratic institutions, all with taxpayer dollars.
I'm on the more radical (libertarian side), but perhaps a logical rethinking of the role of government is required. I'd recommend reading "Anarchy, State and Utopia" by Nozick to give a logical justification for a minimalist state (one that does not include funding schools, hospitals etc... Only a military system and justice system for enforcement of laws that protect individual liberties). It will never happen in my lifetime, but I believe it is a b
>> Comparing the US to Canada with a single number like that is ignorant
I grew up in Canada and have lived in the US for the past 9 years. I believe that I have a valid perspective on the situation and can comment qualitatively as well as quantitatively. If you believe that the average Canadian is wealthier or even on par with the average American, you are living in a fantasy world. By the way, as much as Canadians would like to think they are culturally distinct, the line is very thin. Per capital GDP is only one number, but it's an important one - it measures economic output. How about another economic metric like unemployment? Or net worth? Or income? Look these up on the web, and you'll see how relatively poor Canadians are vs. Americans.
>> the Federal Government of Canada has had a balanced budget for the better part of a decade
So what? That just says that the Canadian government is taxing too much. Perhaps the government should give back the surplus it has been running the past few years, so that the average Canadian can have more in his/her pocket to spend. Does having a balanced budget suddenly mean that Canadians can afford bigger houses and fancier cars? The government does not create wealth, the citizens do. The more control of capital that the citizens have, the better off the citizens are. Perhaps the Canadian govt. should look into *GASP* a deficit! Deficits are not all bad. Everybody that has a mortgage has a deficit. Sometimes, you need to borrow - sometimes, it's financially prudent because the opportunity cost of paying it ALL UP FRONT is too high vs. other uses of that money (like giving it back to the citizen to be more productive).
>> Broadband is an infastructure which the market can't decide it wants -- Either someone brings it in or they don't get it. Geektown NWT, with 5000 people willing to pay for broadband won't have broadband just because they want it, they won't get it until it is provided
So you're saying it's justified to redistribute wealth to appease the wishes of a group that does not have the financial means to attain their demands? You're saying it's ok to take money from someone that doesn't give a shit about broadband (money that this person made from his/her own productivity) and re-direct it? That is theft. It is immoral. It is wrong. It is also bad for the economy as it distorts demand, resulting in capital being allocated inefficiently. If you live in the NWT and you want broadband, you should have to pay the market price for it. If the is cost prohibitive to deliver broadband to the NWT, so be it! That's the market. Don't live in the NWT if you want cheaper broadband, or pay up!
>> The same was true when the interstate highway system was installed in the US half a century back
I believe that a private highway system would have been far more efficient. There is just too much economic incentive out there for private capital to invest in interstate highways. These would have been built by private enterprise in the absence of government and funded on a usage basis (like tolls). Look at the railways in the 1800s. They were for the most part private, and were built out by private enterprise.
>> A lot of the systems folks that I've been hearing from and things that I've been reading have suggested that, like it or not, multi-core systems are the way of the future. The argument is that the clock-speed aspect of Moore's law has been slowing down for the past couple of years and that we've seen single processors that are as fast as they'll go with current chip design and fabrication technology. (Barring fundamental breakthroughs, of course.) Hence parallelism and multi-core systems.
You are exactly right. The burden of extracting parallelism will shift squarely on the programmer in the coming years. The hardware has been extracting ILP (instruction level parallelism) for you automatically for several years now.
But, the cost of extracting additional parallelism out of a single thread is extremely power-prohibitive.
Intel released a paper on silicon scaling outlining the problem (google for "limits to binary logic switch scaling - a Gedanken model). There is a minimum amount of energy required to implement a switch and a maximum rate at which heat can be dissipated from a solid. In the past 30 years, computer users have benfitted from both an increase in switching speed AND an increase in density. Because of thermal limits, we are entering an era where you will either have to scale switching speed (frequency) OR density. Given the choice - Intel/AMD and others are choosing density, because you can still get performance with MORE transistors (through parallelism) even though the frequency remains constant.
>> This isn't a suprise. The free market is good, but not as good as pre-existing infastructure and a government mandate
n ko rder/2004rank.html
Government mandates eliminate choice and result in an inefficient allocation of capital. What about Canadians that don't give a shit about broadband - they have to pay anyways because of higher taxes. Let the market decide. Not some govt. bureaucrat. It's policies like this that made me leave Canada for the US about 9 years ago.
If people don't want broadband, then they should have the CHOICE not to fund it. Just as it is in the US.
Let's take a look at one measure of wealth - per capita GDP and compare Canada vs. the US:
http://www.cia.gov/cia/publications/factbook/ra
>> using up extra real estate on the die of the Athlon in order to get SMT isn't really worth it in terms of the performance it would bring. Even on the Pentium the benefits aren't all that hot and it's only in specific types of code that you see any impresive speed gains
The real estate used is only about 5% on a P4. If you get more than 5% return in performance (as you do in many cases), then it's a win. It's really the complexity of it all that kills it for AMD - they can't afford the engineering resources to put something like that in. SMT was actually implemented on Willamette (the first P4) way back in 2000, but it was disabled until the engineers could get it to work a few years later.
It may interest you to know that many resources on an out-of-order machine are often idle. On an Athlon, you may get about 1-1.5 instructions per clock throughput on average when the peak is 3. Adding another thread can more fully utilize those idle resources. Beyond 2 threads, the resources are probably saturated (1.5 x 2 = 3).
Then you move to dual core, where your execution resources double again.
So that raises the question, why not build a wider processor with more resources but with a single thread and let the hardware find the parallelism for you INSTEAD of dual-core/SMT where the burden is on the programmer? Short answer - it takes too much power for hardware to extract paralellism.
This isn't some cop-out by "lazy" circuit designers. This is fundamental (ie - the laws of physics work that way).
Sorry coders - your lives will get harder and harder as the years go on and many-threaded processors sell a 100M units in volume each quarter. Better get to work on making parallel programming easier, or you won't see the performance gains that you've come to expect from Moore's Law.
>> No DVD player in the Dell, nor FireWire, nor a modem, nor a stack of bundled software, nor 90 days of free telephone support. Nor is it small, or silent. Laptop technology, which is what the Mini uses, is more expensive.
Subtract out the 19" LCD (approx $300) and you have a $229 system. Now add back in the DVD, firewire, modem and blah blah blah and it is still much cheaper.
It isn't small form factor, but nothing is really stopping a company like Dell from offering such a system. If it turns out that this form factor is what people want and the Mac Mini is a hit, then I'm sure Dell will have a cheaper offering. Apple just doesn't have the manufacturing infrastructure to compete with Dell on price. Hell, even giants like IBM and HP can't compete with Dell on price.
>> Until AMD has the capacity to mostly replace Intel, Dell just has to smile and say "Do you want HypeThreading with that?"
At 3 billion dollars per fab, it'll be a cold day in hell before that happens. AMD barely makes money as it is. Every 3 months, Intel makes more money than all of AMD is worth (market cap wise).
>> Next year, when AMD's new 65nm fab is up and running and Charter (and IBM?) start fabbing AMD CPUs too, THEN things will get interesting.
You're assuming Intel is going to stand still and keep mis-executing on 65nm and let AMD walk all over them. Yes, they've mis-executed the past few years but would you bet against them in the future given their 15 year track record with the Pentium brand?
What happened to Jerry Sander's maxim - "Real men have fabs". AMD is at a serious cost disadvantage compared with Intel on the manufacturing front.
AMD is a one-trick pony with the Athlon/Opteron. Where is their answer to Centrino when the entire market is shifting towards notebooks? Voltage-scaled Athlons/Opterons don't really cut it in that market as evidenced by Intel's 88% notebook share.
>> To my understanding, in the Itanium, you don't execute both paths indefinitely, only so long as the predicate register is "pending".. As soon as the register is no longer pending, all instructions bound to the false-predicate are skipped over.
...
:)
This is equivalent to branch-resolution delay in a branch-predicted machine. If you mispredict, you're only wasting resources until the branch is resolved upon which time you get back on the right path. You only waste if you guess wrong as opposed to Itanium where you waste the resolution delay all of the time. Of course you could add prediction to Itanium, but then now you have the same latency problem of the misprediction so where is the benefit? It's a cute idea, but it doesn't work in practice - see SPECint scores for the Itaniums.
>> predicate-execution is done via the consolidation of super-block code chunks into hyper-block code chunks
Yes, but the code expansion is still there. The combinatorial explosion also affects the compiler. There are only so many paths that the compiler can evaluate before it has to give up because it would take too long to consolidate into a block.
Last time I checked, it took 7 hours to compile GCC optimized on an Itanium.
>> Larger cache is a fact of life.. And with multi-megabyte caches finding there way into home-PC's, asking for a larger instruction cache isn't the end of the world to me. Not to mention, last I checked, it was the data-cache with the greater contention.
Larger caches relative to what? If your code expands 50% because of your ISA, then a competing ISA (ie - x86) can have live with a 50% smaller cache. This means more area on the chip is available for other more productive resources. I have already tried to make the case that the benefit of code expansion (predication) is pretty much zero and even negative vs. branch-prediction.
It's not only cache size, it's also memory bandwidth which is very expensive. It's instruction fetch bandwidth. Execution bandwidth... etc.. etc.. You pay extra every single time.
>> Does the latest x86 allow you to load a potentially invalid memory address and not throw an exception if the memory is never used?
There are non-faulting prefetch instructions. However, the usefulness of software prefetching is limited given a highly effective hardware prefetcher. I speak from experience as I hold several HW prefetcher patents
>> Another danger is that by finding other work to do, additional memory requests might be queued up, providing contention with the critical-path memory-fetch
Oldest loads are given priority in OOO machines. Critical block forwarding is also a common feature. Everything in the design is made to insure the fastest possible execution.
>> However, I still contend that if two architectures are identical in throughput (seconds / instruction), the one with the shorter pipeline has the advantage (to a point, of course). Each additional stage adds overhead and adds to the penalty of branch mispredictions (which thereby merits extra architectural CPU resources to beef up the predictor).
If two architectures have the same instructions/second, then they have the same performance. Long pipe or short pipe, they take exactly the same amount of time to run the program.
My argument is this - you cannot attain the same instructions/seconds (performance) with a 5 stage pipeline as you can with a 15 stage pipeline (designed correctly). Deepening the pipe improves performance up to a point, after which it degrades performance. Many papers have been written outlining the tradeoffs of pipeline depth vs. performance.
>> I find it odd that you seem perfectly happy with a JIT / branch-prediction, yet are worried about throwing resources at caches or register files. Obviously, I'm biased in the other direction, so to-each-his-own philosophy I guess.
It's not about philosophy - It'
>> The Itanium executes all possible execution paths (that are serialized; no branches) until one or more paths are resolved. Thus in a two-way logical branch, if it takes 20 instructions to resolve the dependency. If 10 instructions were the true-path and 10 instructions were the false path, you would have no risk of having "branch predicted" the wrong path. Put that into a 15+ stage plus architecture and a wrongly chosen path starts hurting.
Executing both paths isn't free. It's way more power. Let's take a 2-way branch --- you're then using TWICE the compute resources than are necessary ALL OF THE TIME until the branch is resolved. Furthermore - your code size blows up - in fact for highly complex conditionals, you get a combinatorial explosion. This means that your caches need to be MUCH larger to hold more code.
Contrast this to a branch-predicted machine which is much more efficient if you have a good predictor. You only waste compute resources if you predict wrong as opposed to ALWAYS wasting resources with predication. History has shown that good predictors are possible to build and the costs of mispredicts is much smaller than the cost of predicated code (since we mispredict so rarely).
>> Additionaly, the Itanium has nice memory pre-fetching instructions with delayed consequences. I know the Pentium Pro added a couple of pre-cach loading instructions; don't know if they've incorporated the equivalent though.
x86 has prefetch instructions. Furthermore, dynamic hardware prefetchers are far more effective than software prefetches because the hardware knows more about the memory access pattern (a dynamic run-time type of information).
>> Additionally, the register-rolling capability of the Itanium was interesting. Theoretically providing an optimal for-loop, where excluding setup-tear-down, every iteration of the loop took as little as 1 logical clock-tick. (At least as far as the software is concerned; the hardware had room to grow in EPIC-bundles / clock).
A terrible idea. It blows up the size of the register file thus limiting the frequency of the entire chip. It's a bad tradeoff - make EVERYTHING slower so you can unroll loops. They are bad on SPARC (which I have designed) and they are bad on EPIC.
>> I'm not sure how you can say this. Let me restate this: As the speed of the Memory approaches the speed of the CPU, the number of NOP-cycles is reduced in the CPU. What I meant was that if memory speed is the bottle-neck, and we can't cache enough of it to appreciably overcome that bottleneck, then we should look for valueable tradeoffs between clock-speed and work-done-per-instruction. The P4 dramatically traded off clock-speed for work-done-per-instruction and thereby produced less overall performance than an architecture with fewer stages. (Granted, it's an green-apples to red-apples comparison of the AMD-64 and the P4). Thus, if an archetecture has two equivalent performance characteristics but of different clock-speeds, but the memory speed is drastically slower, then the slower CPU-speed should be chosen so that the impeedence mismatch is reduced. With the exception of critical-paths-fitting-in-cache applications, general system performance should be improved.
I'm not sure what you mean here.
Computer performance is measured in time.
The time it takes to fetch something from DRAM cannot be improved due to the laws of physics. It doesn't matter how fast you wiggle the clock, 70 ns to DRAM is about as good as you're going to get.
Performance = dynamic_program_length_in_instructions * CPI / frequency
Performance = time to execute program
CPI = cycles per instruction
Program length can be reduced statically with a compiler and other dynamic tricks.
Less work per clock means a lower CPI but it also implies a higher frequency. CPI is a function of frequency.
For the parts of the program that are not memory-bound, you can speed up with a higher frequency (assuming
>> Most steps of computational algorithms are inherently dependent on the results of previous steps, and are thus not parallelizable. single-threaded CPU's have gotten VERY good at parallelizing individual instructions
This makes no sense. If the instructions are dependent, they are dependent, there's nothing you can do to reduce latency other than increasing frequency. The effect of increasing frequency doesn help the parts of your instruction that includes some delay through an RC path since RC-delay is independent of switching speed.
>> The compilers aren't well suited for helping the CPU out, so things like the Itanium were supposed to exploit such parallelism. But the loss of backward compatability (and the Itanium's focus on floating point) spelled the death nell for that architecture.
Itanium is stupid because there is lots of information that is not available at compile-time rendering the compiler useless (ie - branch prediction and caching behavior). Dynamic information is needed and is VERY important for extracting parallelism. That's why out-of-order CPUs routinely trounce in-order VLIWs like Itanium. Yes, Itanium has stellar FP performance, but that's only because their caches and memory busses are overdesigned relative to the competition. Level the playing field with comparable memory systems. IE - put a 6-9 MB L2 cache on a P4 or an Athlon and double the memory bandwidth and see who gets higher FP performace - I can assure you that the Itanium will get smoked.
>> Personally, I think the answer is taking a step back from MHZ and pipelining. Go back to a 3, 4 or 5 stage pipeline with MASSIVE read-ahead decompilation of instructions (similar to transmeta).
Sounds like Itanium. See above. Using a dynamic compilation technique like Transmeta doesn't work as well as say - doing an out-of-order machine because the delay from finding out something new dynamically to reacting to it is way too long for software to handle. Hardware is much more effective for this. Things are already decompiled dynamically by hardware (P4's trace cache and AMD's pre-decoded instruction cache).
3,4 5 stages? Do you work for Motorola? There are many ways to compute the optimal pipeline depth. I can assure you that 3, 4, 5 is very non-optimal. Start with the Sprangle/Carmean paper and follow up with some IBM papers.
>> Get lots of high-speed cache on board with as little latency as possible (current large-cache architectures have HUGE latancies). By lowering the CPU MHZ, you reduce the latency to the all-important main-memory.
Frequency has nothing to do with latency. If it takes say 10 nanoseconds to access a cache (due to the physics of propagating electrons across a distance - RC delay) then it takes 10 nanoseconds. 10 cycles at 1 GHz is just as long as 1 cycle at 100MHz. Hint - cache access time gets longer for larger caches because the cache is BIGGER and the signal needs to travel a larger distance. Until you can implment wormholes in chips, I don't see the problem getting solved.
What you really want are smarter dynamic prefetching algorithms into a cache heirarchy so the latency can be hidden.
>> Advance the state-of-the-art in power-consumption
Absolute power is very optimized already. You might not think so because of the prevalance of 100W CPUs but there are very good reasons they're not 200W instead.
What architects are focusing on now is power-efficiency. Performance per watt is the new thing to optimize (as opposed to performance per mm^2 which is still a constraint).
>> This is going to be a huge industry. If they could produce, say, a 800MHz CPU which ran on 1W or 0.5W of power and had sensible float performance, it could easily sell exceedingly well.
800MHz is a useless number. It says nothing about performance.
Anyway, if there is such a huge market, don't you think Intel or AMD or 30 or so other semiconductor companies would have something to deliver to that market? What's so special about Transmeta other than the fact that they used to employ Linus? Their "competetive advantage" was their code morph technology which has failed miserably in the marketplace. It's also technically not that great of an idea. Now they're staking their future on their chip making ability? Sorry, in this case they are beat cold - Intel and AMD have 100x the engineering budget of Transmeta. Not to mention Motorola (Freescale), Texas Instrument, IBM etc... IE - companies that know how to deliver silicon to the marketplace and make money.
Transmeta is as good as dead.