I disagree on the point of more money being throwen at a problem. Look at what linpack is. It is really not a hard problem to scale. It scales to thousands of processors (witness ES and machines like the 1k processor T3E). There is a reason that the HPC Challenge benchmark is being invented.
Having said that I agree with you in that IBM has done some innovative things with BlueGene.
OK. For those of you who are not familiar with Jon "Hannibal" Stokes, he Knows his stuff about CPU architecture and provides a great service to hobbyists and programmers (such as myself, as I do scientific computing).
Depends on what you need to do. If it is 8x10 printing, then go get yourself a Epson R200 ($100), pay for the photo paper (not the premium), and buy the retail ink ($14/cart).
It comes out to $2.50 a print. Premium paper brings the cost to about $4 per print. Go to a store and it is much more expensive.
If you are trying to do it cheaper then this, you are going to be pissed off and not get results worth looking at.
They are making the X1 (no more sales because) they are moving to the X1e (1.5x the clock speed).
The problem with these systems is the scaler unit. It runs at half the clock speed of the vector unit. So as soon as the code stops using the vector unit your screwed (scaler through put = 1/128 of the speed of the vector unit (64 length vector, half the clock speed)).
The XD1 is their low end. They are doing this machine for two reasons. 1. Clusters are eating them alive. Also witness the partnership will Dell. 2. Did you notice the bit about commodity processors compined with special purpose? What they are looking at doing is combining, into the same machine (ie memory space) both AMD CPUs and their custom kit. This all of a sudden means they get a very good scaler unit with the best vector processor in the world. Being that they are in the same machine means you can get the best of both worlds.
All of a sudden you have a machine that really kicks some arse.
SS1 and NASA do different things, that is why there is a price difference.
OK so a shuttle goes into orbit for how long and supports life and experiments for that duration. How long did SS1 stay up for? Not that long (just a couple mins). They didn't even do one orbit.
Not that I am trying to take away from what they did. I shed a tear as I watched this morning.
It is very interesting to note that there is a manufacture that has started to integrate normal and low sensitivity sites onto a single sensor. This allows them to increase the dynamic range of the sensor (closer to the human eye, not quite there). Which is the same problem in essence. The non-linarity of the eye/film allows for a larger dynamic range, with the new sensor we are getting closer.
Michael Reichmann from Luminous Landscape noted that no reviews of film based cameras were being presented on the web, and he thought that indicated the future.
This is a great camera! I want one, but one thing wrong with the story submission. This is not recent news, many people have gone over this before, but a 6MP sensor is enough to get you better then 35mm film.
The 16.7MP of this camera is getting very close to medium format (if not already there).
Again awesome camera!
see not file like-it-is shootout
This guy is one of the best. If you don't believe me check out dpreview or google
OK there are a few things that need to be said about that number.
First you can get cameras that have 25 MP sensors. They are called medium format. Only problem is you will be looking at tens of thousands of dollars (US) up to about $30k.
Second I have a ~6MP Nikon D70. I can print 8x10" just fine and if I had a printer large enough 11x14 with a little bit of interpolation. One just does not need that many pixels to get good prints, and even less for a computer display.
If you don't beleive me go check out http://www.luminous-landscape.com/reviews/sho otout.shtml
and for what 3MP gets you http://www.luminous-landscape.com/reviews/cam eras/ d30/d30_vs_film.shtml
The US Army uses SecureID and Kerbose. They also use a short timeout on the tickets. I have been using it for some time now and as an end user I like it.
The system works on everything from linux, Unicos (Cray), AIX (IBM), Solaris (Sun), and every ones favorite Windows!
I did a video card upgrade, flipped the power on heard a POP and saw some smoke.
Hit the power button and took a look at what could have caused that. Remember those double rows of pins that stuck out of video cards for "upgrades" that never happened? The 5v floppy power connector had attached itself to that. I unhooked the power connect and started the machine. The display came up except it was missing the Red RAMDAC.
Friends used to ask me how I changed the color of my terminal dispay under DOS and Linux.
I love how the stories on/. line up sometimes. If we take this tech and the mechinical computing not too far down the page we can get more computation for our effort. That has to count as over clocking... or at least cool on the geek factor.
The ES-7000 is a machine that is up to 32-way (at the time I was using them about two years ago).
Against our wishes our client purchased one instead of a small cluster. Now aside from the price difference, the stability sucked. We were only running IIS, Apache, SQL Server and Tomcat. The machine needed to be rebooted every day. Yes a million dollar machine running windows needed to be rebooted ever single day.
So my point in relating the story is that MS has a LONG way to go before they are able to really handle supercomputer sized machines. But I do wish them good luck, because they have smart people that will bring some good ideas to the table. On the other hand GO {Linux, FreeBSD, OpenBSD}
The systems that are "only" 256 processors are air cooled. There is a liquid cooled version that is much bigger. I forget how big theoretical, but watch out between IBM and Cray the US will take back the top spots in the Top 500 very soon.
Nah. I don't remember the exact number but the magnitude is correct.
SGI bought Cray for $100M+ SGI sold Cray for $13M to Tera. Tera changed its name to Cray.
now "Cray" offers a system called MTA-2 which used to be Tera's only machine.
BTW the SX-6 is actually a NEC machine that Cray resells in the US because well the Japanese were dumping the SX-6 to try to put Cray out of biz... a lawsuit later and Cray is the distributer for North America.
I assume you are refering to the Altix systems. They are Single System Image systems (SSI), when means they are built like a supercomputer, and not like a cluster. They are big and expensive.
SSI is where all CPUs can see all memory as if it was local. They are also Non-uniform memory access which means all the memory it sees is not as fast as all other memory, but really ALL single systems are like this. For example each CPU can address the entire TB of memory that is in the system, but reading from one memory location might take 100 cycles, and from another might be closter to 1000 cycles.
It all depends on the problem you are trying to solve. I have been doing some work of late that would not complete in my life time on the 108 node cluster that we have. But when programmed for and run on two Cray X1s I should complete inside of a week.
Granted there are many codes (and more every day) that will run on clusters, the big iron will never die.
The problem that you run into with any benchmark is that you only care about the applications that you run and not some stupid benchmark. Having said that Linpack (or or in general FLOPOS) can be used to measure system performance because you can measure all sorts of neat things like: L1/CPU speed Main Memory speed Caching algorithms (ie when your data set is bigger then main memory and you hit swap)
But then you still have to benchmark the video card ect.
So what it basically comes down to is that you need lots of numbers to know how the system is really running, but most people don't know how to use those number and really do not care. They just want a single number that is a general measure of system performance. If you are a speed freak then you will know what those number mean and try to use them when purchasing a system.
The Power 4 is two full CPUs (cores) on a single die, not that crap that Intel puts out called Hyper-Threading where you only have a single full CPU and then some extra logic to quickly swap over to another thread when needed.
Slashdot Mirror
I disagree on the point of more money being throwen at a problem. Look at what linpack is. It is really not a hard problem to scale. It scales to thousands of processors (witness ES and machines like the 1k processor T3E). There is a reason that the HPC Challenge benchmark is being invented.
Having said that I agree with you in that IBM has done some innovative things with BlueGene.
Actually what was amazing about ES is that it came out of no where. Everyone has known about BlueGene for ~2 years now.
OK. For those of you who are not familiar with Jon "Hannibal" Stokes, he Knows his stuff about CPU architecture and provides a great service to hobbyists and programmers (such as myself, as I do scientific computing).
n ing-1.ars
g .ars
m mer-1.ars
m -1.ars
a nd-Computer-Science/6-823Computer-System-Architect ureSpring2002/CourseHome/index.htm
6 17250
OK so if you want to understand pipelines and why longer ones are not always better?
http://arstechnica.com/articles/paedia/cpu/pipeli
No? Well how about caching?
http://arstechnica.com/articles/paedia/cpu/cachin
And then there is the Opteron/Athlon64 goodness
http://arstechnica.com/articles/paedia/cpu/amd-ha
Historical view of the Pentium Architecture?
http://arstechnica.com/articles/paedia/cpu/pentiu
For more great articles from him (and others)
http://arstechnica.com/paedia/
Also dont forget about MIT open courseware to fill in the gaps...
http://ocw.mit.edu/OcwWeb/Electrical-Engineering-
And how can you not have love for
http://slashdot.org/comments.pl?sid=138810&cid=11
Keep up the good work!
That is really awesomely funny... but it scares me. :)
It is about doing research. I am in a PhD program atm. The reason I am is my drive to learn and build things that no one else has done before.
There are plenty of jobs outside of academia if/when you want to leave (speaking of PhDs from physics/math/CS ie technical fields)
Depends on what you need to do. If it is 8x10 printing, then go get yourself a Epson R200 ($100), pay for the photo paper (not the premium), and buy the retail ink ($14/cart).
It comes out to $2.50 a print. Premium paper brings the cost to about $4 per print. Go to a store and it is much more expensive.
If you are trying to do it cheaper then this, you are going to be pissed off and not get results worth looking at.
They did not abandon their custom CPUS!
They are making the X1 (no more sales because) they are moving to the X1e (1.5x the clock speed).
The problem with these systems is the scaler unit. It runs at half the clock speed of the vector unit. So as soon as the code stops using the vector unit your screwed (scaler through put = 1/128 of the speed of the vector unit (64 length vector, half the clock speed)).
The XD1 is their low end. They are doing this machine for two reasons.
1. Clusters are eating them alive. Also witness the partnership will Dell.
2. Did you notice the bit about commodity processors compined with special purpose? What they are looking at doing is combining, into the same machine (ie memory space) both AMD CPUs and their custom kit. This all of a sudden means they get a very good scaler unit with the best vector processor in the world. Being that they are in the same machine means you can get the best of both worlds.
All of a sudden you have a machine that really kicks some arse.
SS1 and NASA do different things, that is why there is a price difference.
OK so a shuttle goes into orbit for how long and supports life and experiments for that duration. How long did SS1 stay up for? Not that long (just a couple mins). They didn't even do one orbit.
Not that I am trying to take away from what they did. I shed a tear as I watched this morning.
It is very interesting to note that there is a manufacture that has started to integrate normal and low sensitivity sites onto a single sensor. This allows them to increase the dynamic range of the sensor (closer to the human eye, not quite there). Which is the same problem in essence. The non-linarity of the eye/film allows for a larger dynamic range, with the new sensor we are getting closer.
Michael Reichmann from Luminous Landscape noted that no reviews of film based cameras were being presented on the web, and he thought that indicated the future.
But the problem with that (as has been pointer out in other posts) is full frame sensors have been out for a long time.
This is a great camera! I want one, but one thing wrong with the story submission. This is not recent news, many people have gone over this before, but a 6MP sensor is enough to get you better then 35mm film.
The 16.7MP of this camera is getting very close to medium format (if not already there).
Again awesome camera!
see
not file
like-it-is
shootout
This guy is one of the best. If you don't believe me check out dpreview or google
OK there are a few things that need to be said about that number.
o otout .shtml
m eras/ d30/d30_vs_film.shtml
First you can get cameras that have 25 MP sensors. They are called medium format. Only problem is you will be looking at tens of thousands of dollars (US) up to about $30k.
Second I have a ~6MP Nikon D70. I can print 8x10" just fine and if I had a printer large enough 11x14 with a little bit of interpolation. One just does not need that many pixels to get good prints, and even less for a computer display.
If you don't beleive me go check out
http://www.luminous-landscape.com/reviews/sh
and for what 3MP gets you
http://www.luminous-landscape.com/reviews/ca
The US Army uses SecureID and Kerbose. They also use a short timeout on the tickets. I have been using it for some time now and as an end user I like it.
The system works on everything from linux, Unicos (Cray), AIX (IBM), Solaris (Sun), and every ones favorite Windows!
I did a video card upgrade, flipped the power on heard a POP and saw some smoke.
Hit the power button and took a look at what could have caused that. Remember those double rows of pins that stuck out of video cards for "upgrades" that never happened? The 5v floppy power connector had attached itself to that. I unhooked the power connect and started the machine. The display came up except it was missing the Red RAMDAC.
Friends used to ask me how I changed the color of my terminal dispay under DOS and Linux.
then select and paste
I love how the stories on /. line up sometimes. If we take this tech and the mechinical computing not too far down the page we can get more computation for our effort. That has to count as over clocking... or at least cool on the geek factor.
The ES-7000 is a machine that is up to 32-way (at the time I was using them about two years ago).
Against our wishes our client purchased one instead of a small cluster. Now aside from the price difference, the stability sucked. We were only running IIS, Apache, SQL Server and Tomcat. The machine needed to be rebooted every day. Yes a million dollar machine running windows needed to be rebooted ever single day.
So my point in relating the story is that MS has a LONG way to go before they are able to really handle supercomputer sized machines. But I do wish them good luck, because they have smart people that will bring some good ideas to the table. On the other hand GO {Linux, FreeBSD, OpenBSD}
Better watch out for the IRS. You have to claim barter on your taxes!
The systems that are "only" 256 processors are air cooled. There is a liquid cooled version that is much bigger. I forget how big theoretical, but watch out between IBM and Cray the US will take back the top spots in the Top 500 very soon.
Nah. I don't remember the exact number but the magnitude is correct.
SGI bought Cray for $100M+
SGI sold Cray for $13M to Tera.
Tera changed its name to Cray.
now "Cray" offers a system called MTA-2 which used to be Tera's only machine.
BTW the SX-6 is actually a NEC machine that Cray resells in the US because well the Japanese were dumping the SX-6 to try to put Cray out of biz... a lawsuit later and Cray is the distributer for North America.
I assume you are refering to the Altix systems. They are Single System Image systems (SSI), when means they are built like a supercomputer, and not like a cluster. They are big and expensive.
SSI is where all CPUs can see all memory as if it was local. They are also Non-uniform memory access which means all the memory it sees is not as fast as all other memory, but really ALL single systems are like this. For example each CPU can address the entire TB of memory that is in the system, but reading from one memory location might take 100 cycles, and from another might be closter to 1000 cycles.
It all depends on the problem you are trying to solve. I have been doing some work of late that would not complete in my life time on the 108 node cluster that we have. But when programmed for and run on two Cray X1s I should complete inside of a week.
Granted there are many codes (and more every day) that will run on clusters, the big iron will never die.
Good info about memory heirarchy!
The problem that you run into with any benchmark is that you only care about the applications that you run and not some stupid benchmark. Having said that Linpack (or or in general FLOPOS) can be used to measure system performance because you can measure all sorts of neat things like:
L1/CPU speed
Main Memory speed
Caching algorithms (ie when your data set is bigger then main memory and you hit swap)
But then you still have to benchmark the video card ect.
So what it basically comes down to is that you need lots of numbers to know how the system is really running, but most people don't know how to use those number and really do not care. They just want a single number that is a general measure of system performance. If you are a speed freak then you will know what those number mean and try to use them when purchasing a system.
The Power 4 is two full CPUs (cores) on a single die, not that crap that Intel puts out called Hyper-Threading where you only have a single full CPU and then some extra logic to quickly swap over to another thread when needed.