Does NASA realise you can make that kind of money by simply working?!
Maybe it's to stimulate young minds? When you're a teenager yet to go to University or College, $50k is a lot of money.
The other thing is, some people will take part purely for the fun of it. If you had to spend money on hardware, $50k might help you recoup your costs, and may provide an incentive for the more economically-challegend amongst us.
Offering a cash prize, however small, gets you in the news. There's no such thing as bad publicity, as they say.
I'm "between jobs" just now, so $50k to me would be very useful, although if I had the brains, I'd probably enter just for the fun of it.
Well, no. My father is a film (and english, but that's irrelevant) professor at a pretty decent college--he does this kind of thing all the time. If he's writing a paper--or having his students write papers--about a film, and wants to discuss one scene, it's great to have that scene be *right there*.
Isn't that what hypertext (and more specifically, HTML) is for? It's nice and simple and pretty portable. Heck, there's even an Open Standard for video called MPEG.
Well, they named their cool new Opteron workstations which run Linux and Solaris "Java" workstations. Lots of people wonder why they'd buy a workstation to "run Java" when what they want to run is Linux, Windows or Solaris.
We suggested they convert one of their existing NT machines to NT Terminal Server on their LAN. Install the application and the Oracle database on that, bingo. Instant legacy application support, that should work for a long time to come.
It would be interesting to see whether it would run well under WINE. Old Delphi stuff seems to work quite well, but that's Borland, not M$.
If your aunt is that stupid, the only way she'd be at university would be in a jar of formaldehyde.
She's from a generation where women weren't expected to think for themselves.
Last decade, politicians made it very easy for almost everyone to go to university. They turned all the old colleges into universities and now you don't even have to pass any exams to get in. It used to be you needed 'A's and 'B's (my day) but now people with 'D's and 'E's are getting in. The idea was to keep as many people as possible out of the Unemployment statistics and to give degree status to people studying Leisure and Tourism and Hotel Catering.
My wife teaches English, and the knock-on effect on the school pupils has been dreadful. None of them have any ambition at all to do well, because they know they can continue their studies with E grades.
My aunt is doing a computing course at university. While she was at college, before uni, she'd never seen Linux, so I sent her a KNOPPIX CD.
She refused to run it. She thouth "KNOPPIX" meant "no pics" as in "no graphics."
I explained what it was but she still refused to try it, because all the r4d d00dz on her course told her that Linux was an abomination, unclean, and not to be let near a Windows PC. Windows was the best OS in the world, and that dodgy Linux thing might mess up your PC.
Other attempts at Linux advocacy have also failed. I don't bother now.
The itanic is effectively a very big, clumsy, over-engineered Digital Signal Processor. This is demonstrated by the fact that it is very good at floating-point intensive code (e.g. simulation) and very poor at everything else.
Unfortunately, one of their very best processors, the i860, never gained wide acceptance either and was condemned to niche markets.
If only the Motorola 68000 had been ready on time, IBM would not have chosen the 8086 for the PeeCee and the world would be a better place.
The itanic is based on ideas that were popular amongst supercomputer designers in the late 1970s. However, when RISC CPUs came along, and transistor budgets increased, things like superscalar processors, out-of-order execution, register renaming, branch prediction etc. were all developed as time went on. These all address issues that are very hard, or impossiblem for a compiler to deal with. intel bloody-mindedly ignored this (who can guess why) and persevered with this 1970's folly, which was only every really applicable to a very small market niche anyway.
Look what's happened.
itanic has all but sunk and intel has been forced to implement AMD's 64-bit extensions to the 386 architecture. The instruction set wars are over. All moder CPUs are some sort of highly-superscalar RISC internally, some with instruction set translators to implement crufty old instruction sets (i.e. 386) and some with translators that can be reprogrammed like Transmeta's offerings.
But it's still not clear to me that a big system built around one die that uses transistors really efficiently is going to be less expensive than eight smaller systems that don't use their transistors as efficiently.
You have to take into consideration floor space, rack space, cooling and power requirements, etc. Of course, you can take a bunch of cheap Dell PeeCees and rack them. You could also, probably in less space, take 8 of these systems and use less power and pump out less heat and get vastly more throughput, supposing it does turn out to be as good as they say. Who knows how much a Niagara system is going to cost, but I'll bet it will be competitive with other small blade systems.
My Pentium 4 processor has 2 threads. Linux treats them as 2 processors, and makes full use of them.
Solaris makes full use of them too. In real world applications, intel's "Hyperthreading" only buys you a few percent up to 30% at most in the best cases and often results in a performance decrease. Hyperthreading is a very cheap and over-hyped implementation of SMT. It is a hastily cobbled together afterthought tacked on to a disasterous architecture to try to mitigate its ludicrously long pipeline which spends most of it's time being filled and flushed due to branches and context switches.
An 8-core chip that shares the cache, VM infrastructure, and memory interface between all cores is going to work best for CPU-intensive tasks that are not also I/O or memory-intensive and can be partitioned into multiple threads easily.
Wrong. Niagara is designed to have very high memory bandwidth and to hide most of the latency through context-switching to different threads. It is designed for highly multi-threaded applications such as web serving.
Not photorealistic rendering, for example, that requires too much data. And it won't handle separate-process loads as well as 8 blades would.
It is not designed for floating-point workloads such as rendering or simulation. It will handle separate processes very well.
So, watch all of those parameters: price, power, memory bandwidth, how cache and VM are done, PC board real-estate, etc. How they are combined will tell you whether that chip is a win or lose for your application.
It's designed for efficient, high-density, highly multi-threaded blade applications.
For such workloads, it's a much more efficient use of the available transistors on the CPU die than conventional CPUs. Conventional CPUs spend 75% of their time idling waiting for memory access. The Pentium IV is much worse and an extreme case as I mentioned above. Niagara does things properly and hides latency by switching thread contexts when a thread is waiting for memory access.
From the lack of non-Sun-supplied buzz regarding this technology, it would appear that many people aren't finding it very exciting.
More like none of Sun's competitors have anything which comes remotely close.
Notice how nearly a year after Sun announced this, intel finally admitted that clock frequency (i.e. gigahertz) isn't everything and that they'd be bringing out dual core processors?
Niagara has 8 cores each capable of 0-clock cycle latency switching between 4 different thread contexts.
Who else has working hardware and an OS to go that can do this?
Mark my words. I am certain that Sun will absolutely not open source solaris.
Well, I've had a 4-year tour of duty at Sun which has just come to an end so I know a thing or two. The Solaris source definitely will be coming out very soon now.
CDE is still in Solaris for compliance with all the old UNIX standards so that Solaris can still qualify for US Government contracts and for other organisations that demand strict adherence to the standards.
Sun publically stated at the time of Solaris 9, nearly 2 years ago, that GNOME (in the form of JDS nowadays) is the official Solaris desktop.
Of course we wouldn't depressurize all the way... just so that the last 8 hours of his or her life is excrutiating.
Actually, that's not what happens. As pressure falls and so the amount of available oxygen, the victim goes into a kind of a hysterical stupour laughing at everything. Sounds like fun to me! I'm going to die! Hahahahahahahh!
Well, if you saw Jeremy Clarkson in the RAF training thingymabob that's what happened to him.
Depending on the OS being used, the amount of memory space useable by an application varies between 3.5GB and 2GB on the x86 architecture.
I realise that the 80286 and 8086 had very different addressing modes. The 286 had a very primitive form of protected mode, but my point was about the 64k segement size and how inconvenient it was to the programmer, hurting program reliability and performance due to increased complexity.
Further more, my point was that with the segmented 36-bit addressing of the Pentium II and later proecssors, intel has, in effect, brought back a segmented memory hach with many of the drawbacks and failings of the old memory models on the processors of 25 years ago
Compare this with all the processors of the last 15 years which have 64-bit address registers (some only capable of physically addressing 40 bits of memory) which engineer out these problems in a simple and elegant way.
The reason intel didn't extend the x86 architecture in this manner earlier (c.f. AMD64) was to artificially restrict the market of the Pentium range to try to persuade people to buy itanic.
Now, we could go on to the relative merits of extending the x86 architecture to 64-bits or starting with a clean slate (c.f. Alpha, MIPS, PA-RISC, UltraSPARC), however, the instruction set wars are over, as AND has demonstrated with the AMD64 architecture (Opteron etc.) getting all the performance of the big RISC chips with effectively a 64-bit RISC with an x86 ISA translation layer on chip (yes, I know Pentium II, K6 and Cyrix M1 started this off years ago, but Opteron was the first 64-bit one).
The virtual addressable space will still be 4GB, however the 36-bit address bus will allow for a theoretical limit of 64GB.
That's true, however it's a dirty hack. The extra memory is only addressable in 4GB (or 2GB) segments. Therefore, if you have an application that needs more than 4GB in a contiguous chunk, you are out of luck. You could probably fake it with operating system calls and complicated wrpper functions to hide the memory address arithmetic. However, you loose a lot of performance.
It's a bit like the old days of the 8086 and 80286 where memory was in 64k segments and there were segment registers used in conjunction with index registers to calculate addresses. As you can imagine, writing programs with datastructures larger than 64k was complicated and bug-prone due to the added complexity. It also slowed the program down significantly due to the extra calculations required.
So you see, a much better, cleaner, more efficient solution to the problem is to have a flat 64-bit address space, like the AMD64 architecture.
Of course, intel tried very hard to pretend that no-one "needs" 64-bits on the desktop for many years while they tried to peddle the dreadful itanium for servers and workstations. They hoped that naieve users would believe them long enough until they could get itanium PCs out in the mass market. Then along came AMD with a much better processor...
Slashdot Mirror
You're right.
Somehow I feel I have nothing to add to the discussion except a petty attempt at whoring for karma.
Maybe it's to stimulate young minds? When you're a teenager yet to go to University or College, $50k is a lot of money.
The other thing is, some people will take part purely for the fun of it. If you had to spend money on hardware, $50k might help you recoup your costs, and may provide an incentive for the more economically-challegend amongst us.
Offering a cash prize, however small, gets you in the news. There's no such thing as bad publicity, as they say.
I'm "between jobs" just now, so $50k to me would be very useful, although if I had the brains, I'd probably enter just for the fun of it.
Money isn't everything, but is sure helps :-)
Isn't that what hypertext (and more specifically, HTML) is for? It's nice and simple and pretty portable. Heck, there's even an Open Standard for video called MPEG.
BBC Radio 4 is the jewel in the crown of British (and world) speech radio.
OK, it isn't perfect, but the Today programme (and PM, its 17:00 counterpart) is excellent.
Well, they named their cool new Opteron workstations which run Linux and Solaris "Java" workstations. Lots of people wonder why they'd buy a workstation to "run Java" when what they want to run is Linux, Windows or Solaris.
It would be interesting to see whether it would run well under WINE. Old Delphi stuff seems to work quite well, but that's Borland, not M$.
No, the 1960s weren't that long ago, and we are talking about puritanical right-wing Scotland here.
Not if it comes from IBM. We only hate the stuff from Sun, and we also like .NET in the form of Mono, because Miguel started that so it must be cool.
She's from a generation where women weren't expected to think for themselves.
Last decade, politicians made it very easy for almost everyone to go to university. They turned all the old colleges into universities and now you don't even have to pass any exams to get in. It used to be you needed 'A's and 'B's (my day) but now people with 'D's and 'E's are getting in. The idea was to keep as many people as possible out of the Unemployment statistics and to give degree status to people studying Leisure and Tourism and Hotel Catering.
My wife teaches English, and the knock-on effect on the school pupils has been dreadful. None of them have any ambition at all to do well, because they know they can continue their studies with E grades.
She refused to run it. She thouth "KNOPPIX" meant "no pics" as in "no graphics."
I explained what it was but she still refused to try it, because all the r4d d00dz on her course told her that Linux was an abomination, unclean, and not to be let near a Windows PC. Windows was the best OS in the world, and that dodgy Linux thing might mess up your PC.
Other attempts at Linux advocacy have also failed. I don't bother now.
intel tried wierd, complicated and over-engineered architectures in the past that also died a death.
Unfortunately, one of their very best processors, the i860, never gained wide acceptance either and was condemned to niche markets.
If only the Motorola 68000 had been ready on time, IBM would not have chosen the 8086 for the PeeCee and the world would be a better place.
The itanic is based on ideas that were popular amongst supercomputer designers in the late 1970s. However, when RISC CPUs came along, and transistor budgets increased, things like superscalar processors, out-of-order execution, register renaming, branch prediction etc. were all developed as time went on. These all address issues that are very hard, or impossiblem for a compiler to deal with. intel bloody-mindedly ignored this (who can guess why) and persevered with this 1970's folly, which was only every really applicable to a very small market niche anyway.
Look what's happened.
itanic has all but sunk and intel has been forced to implement AMD's 64-bit extensions to the 386 architecture. The instruction set wars are over. All moder CPUs are some sort of highly-superscalar RISC internally, some with instruction set translators to implement crufty old instruction sets (i.e. 386) and some with translators that can be reprogrammed like Transmeta's offerings.
You have to take into consideration floor space, rack space, cooling and power requirements, etc. Of course, you can take a bunch of cheap Dell PeeCees and rack them. You could also, probably in less space, take 8 of these systems and use less power and pump out less heat and get vastly more throughput, supposing it does turn out to be as good as they say. Who knows how much a Niagara system is going to cost, but I'll bet it will be competitive with other small blade systems.
Yes, two cores on a die. Niagara has 8.
Solaris makes full use of them too. In real world applications, intel's "Hyperthreading" only buys you a few percent up to 30% at most in the best cases and often results in a performance decrease. Hyperthreading is a very cheap and over-hyped implementation of SMT. It is a hastily cobbled together afterthought tacked on to a disasterous architecture to try to mitigate its ludicrously long pipeline which spends most of it's time being filled and flushed due to branches and context switches.
An 8-core chip that shares the cache, VM infrastructure, and memory interface between all cores is going to work best for CPU-intensive tasks that are not also I/O or memory-intensive and can be partitioned into multiple threads easily.
Wrong. Niagara is designed to have very high memory bandwidth and to hide most of the latency through context-switching to different threads. It is designed for highly multi-threaded applications such as web serving.
Not photorealistic rendering, for example, that requires too much data. And it won't handle separate-process loads as well as 8 blades would.
It is not designed for floating-point workloads such as rendering or simulation. It will handle separate processes very well.
So, watch all of those parameters: price, power, memory bandwidth, how cache and VM are done, PC board real-estate, etc. How they are combined will tell you whether that chip is a win or lose for your application.
It's designed for efficient, high-density, highly multi-threaded blade applications.
For such workloads, it's a much more efficient use of the available transistors on the CPU die than conventional CPUs. Conventional CPUs spend 75% of their time idling waiting for memory access. The Pentium IV is much worse and an extreme case as I mentioned above. Niagara does things properly and hides latency by switching thread contexts when a thread is waiting for memory access.
More like none of Sun's competitors have anything which comes remotely close.
Notice how nearly a year after Sun announced this, intel finally admitted that clock frequency (i.e. gigahertz) isn't everything and that they'd be bringing out dual core processors?
Niagara has 8 cores each capable of 0-clock cycle latency switching between 4 different thread contexts.
Who else has working hardware and an OS to go that can do this?
$5 will not get you very far nowadays.
Well, I've had a 4-year tour of duty at Sun which has just come to an end so I know a thing or two. The Solaris source definitely will be coming out very soon now.
Sun publically stated at the time of Solaris 9, nearly 2 years ago, that GNOME (in the form of JDS nowadays) is the official Solaris desktop.
Actually, that's not what happens. As pressure falls and so the amount of available oxygen, the victim goes into a kind of a hysterical stupour laughing at everything. Sounds like fun to me! I'm going to die! Hahahahahahahh!
Well, if you saw Jeremy Clarkson in the RAF training thingymabob that's what happened to him.
Depending on the OS being used, the amount of memory space useable by an application varies between 3.5GB and 2GB on the x86 architecture.
I realise that the 80286 and 8086 had very different addressing modes. The 286 had a very primitive form of protected mode, but my point was about the 64k segement size and how inconvenient it was to the programmer, hurting program reliability and performance due to increased complexity.
Further more, my point was that with the segmented 36-bit addressing of the Pentium II and later proecssors, intel has, in effect, brought back a segmented memory hach with many of the drawbacks and failings of the old memory models on the processors of 25 years ago
Compare this with all the processors of the last 15 years which have 64-bit address registers (some only capable of physically addressing 40 bits of memory) which engineer out these problems in a simple and elegant way.
The reason intel didn't extend the x86 architecture in this manner earlier (c.f. AMD64) was to artificially restrict the market of the Pentium range to try to persuade people to buy itanic.
Now, we could go on to the relative merits of extending the x86 architecture to 64-bits or starting with a clean slate (c.f. Alpha, MIPS, PA-RISC, UltraSPARC), however, the instruction set wars are over, as AND has demonstrated with the AMD64 architecture (Opteron etc.) getting all the performance of the big RISC chips with effectively a 64-bit RISC with an x86 ISA translation layer on chip (yes, I know Pentium II, K6 and Cyrix M1 started this off years ago, but Opteron was the first 64-bit one).
...or Sun (UltraSPARC) or DEC (ie Alpha) or HP (PA-RISC) or IBM (POWER).
It runs 64-bit Solaris 10 very well indeed.
That's true, however it's a dirty hack. The extra memory is only addressable in 4GB (or 2GB) segments. Therefore, if you have an application that needs more than 4GB in a contiguous chunk, you are out of luck. You could probably fake it with operating system calls and complicated wrpper functions to hide the memory address arithmetic. However, you loose a lot of performance.
It's a bit like the old days of the 8086 and 80286 where memory was in 64k segments and there were segment registers used in conjunction with index registers to calculate addresses. As you can imagine, writing programs with datastructures larger than 64k was complicated and bug-prone due to the added complexity. It also slowed the program down significantly due to the extra calculations required.
So you see, a much better, cleaner, more efficient solution to the problem is to have a flat 64-bit address space, like the AMD64 architecture.
Of course, intel tried very hard to pretend that no-one "needs" 64-bits on the desktop for many years while they tried to peddle the dreadful itanium for servers and workstations. They hoped that naieve users would believe them long enough until they could get itanium PCs out in the mass market. Then along came AMD with a much better processor...
One man's Troll is another man's Funny.