ocularb0b writes "Cray has announced the CX1desktop supercomputer. Cray teamed with Microsoft and Intel to build the new machine that supports up to 8 nodes, a total of 64 cores and 64Gb of memory per node. CX1 can be ordered online with starting prices of $25K, and a choice of Linux or Windows HPC. This should be a pretty big deal for smaller schools and scientists waiting in line for time on the world's big computing centers, as well as 3D and VFX shops."
I remember seeing an HP server like that. On the left side was the large hard disk drive - on the right were the server processors - both kept off the floor but mounted into the frame where drawers would have been. On top of the desk was the monitor.
Those boxes are just blade systems with up to 8 blades with up to 2 quad core CPUs each, so a total of 64 cores per blade system. Certainly not not "64 cores per node" where Cray calls a blade a "node".
There's two relevant ways to parse that fragment. There's one where the "and" in "64 cores and 64G of memory per node" creates a single coordinated constituent, such that it can be paraphrased as "there are 64 cores per node and there are 64 Gb per node." There's a second, the one that I think you favor and that seems correct pragmatically, which may be paraphrased as "there are 64 total cores, and each node in the machine can have 64 Gb."
Structural ambiguity happens all the time in natural language.
Well... My netbook has 2 GB of memory, 160 GB of storage, gigabit networking and thinks it has two 32 bit cores. It's a veritable late 80's, early 90's supercomputer that fits in my backpack. And I bought it cheap.
Well... My netbook has 2 GB of memory, 160 GB of storage, gigabit networking and thinks it has two 32 bit cores. It's a veritable late 80's, early 90's supercomputer that fits in my backpack.
Even in the mid 90's, GHz processors, and gigs of RAM/hard disk were still largely uncommon. I think you're talking late 90's before that started to become relatively common.
I continue to be stunned at what you can buy as an entry level box nowadays for a really cheap dollar amount. My local "white box" PC store will sell you a dual-core 5GHz (or whatever) 64-bit AMD machine for under $300 -- add a little RAM and disk space and you've got a helluva system for not very much money.
How many home PCs nowadays have TB's of storage? I know several people who do -- I remember when home users didn't have gigabytes, terabytes would have been unimaginable.
"As someone who remembers punching holes in 5.25" floppies so you could turn 'em over and another 360K by using both sides"
I remember poking values in memory to upgrade my 140K disks to 160K and then punching the side of the disk (index holes were not needed in the brilliant Wozniak design) to be able to flip it over.
Boy... We are old. I bet I have icons in my desktop that would not fit in an Apple II floppy disk.
Well, Microsoft had to do something to create demand for the next version of Windows. Not much of a market for an OS where people need to book time at their neighborhood super collider when they need to edit a document.
Well, with all the sloppy inefficient programming, feature bloat, and generally craptastic work that goes into the ongoing, illogical, disuseful, nightmare that is MS Word, you will need one of these puppies just to run Word and Windows 7 anyway.
Vista's MINIMUM memory requirement is 512 megs.
Windows 2000's recommended minimum was 64 megs.
Personally, I don't find Vista any more useful than Win2k. More stable, yes, but I don't see how upping the RAM req by an order of magnitude was required to make Win2k more stable. All it needed was better programming and better testing.
I think what we have going now is the kind of thing that happened when gas was cheap: SUVs. When gas is expensive (viz Europe and Japan) the average car gets Really Small and Efficient. When RAM was really expensive, programming was tight and efficient. Now that RAM is measured in gigs and drives in terabytes, there is no incentive to do efficient programming or wrangle in feature creep and bloatware.
Eventually we will hit some physical / cost limit on RAM, and then good programming will become a requirement. OF course, by then, there won't be anyone left who knows how to do that...
Well, with all the sloppy inefficient programming, feature bloat, and generally craptastic work that goes into the ongoing, illogical, disuseful, nightmare that is MS Word [...]
Feature bloat for sure, but how do you know it's sloppily and inefficiently programmed? Have you seen the source? From what I recall of people commenting on leaked Microsoft code the quality was generally considered pretty good.
Is there a reason microsoft would be the prefered OS for this type of machine? I would think the type of people requiring such hardware would be quite capable of running some kind of *nix OS to perform their operations and see the advantages in doing so, like a familiar OS. I imagine MS has invested a decent amount of cash to be the logo broadcasted on the cray site, is there a reason why they want this market? This seems like it would be a very niche market for them.
There are projects to provide unified process space and inter-node IPC like Mosix, bproc, etc. Generally, these aren't used much in HPC. Having a "bunch of individual machines networked together" works pretty well when you also consider that the network might be 20Gb 4x DDR InfiniBand sending frames from point to point at ~2us. I'm just saying...
Chances are, the GP built an HPC cluster and used a typical SPMD approach with something like MPI or PVM for communications and a centralized job manager/schedul
Perhaps to enhance their marketing, they can offer the computer in CrayOn colors (like Apple's iMac colors). Cray Gray, Big Iron Gray, Super Computing Gray, Gray, Gray Passion, etc..
Remember, you can order any color - as long as it is gray.
For example Blender [blender.org]'s renderer's scale on a system like this? Of course something like MentalRay might scale easily but has anyone any hands on experience?
One might argue if you are throwing away $25,000 on a system like that you might use software that costs, but then again, Blender has made tremendous progress these last years..
Blender has made a lot of progress, but it is still way behind Maya and even Lightwave. I've not been using Blender in the past couple releases, but it used to have some issues on my Quad Core Power Mac and using more than 4GB of Ram. I think this has been addressed now though. But I've never run into the problem of RAM or processor speed being the problem, but video ram when modeling an object. I have created scenes that will even grind a decent 256MB video card into the ground. Sure, it would be nice to render a bit faster, but for $20 - $60 a month, I do as much rendering as I want at Respower.
But let's look at cost. For $25k I can buy about 75 commodity boxes that are dual core, 2GB of Ram each & networking gear. That's about 150 Cores and 150GB of Ram. Put Linux on there and you can run ScreamerNet (you get to put the LW rendering engine on 999 machines per license) or one of a number of Maya distributed rendering programs. End result are going to be more frames being processed at one time. (for animation)
If I went the Mac Mini route, that's about 40 Mac Minis, which is still 80 Cores, 80GB of Ram total and with ScreamerNet or Xgrid....
Now the downsides are, 40 - 80 computers take up a lot of space and probably would eat up more power/cooling costs. But then again, if a couple boxes kick the bucket or hiccup, the other 35 - 75 are still processing. You only loose a percentage of total output.
Where it maybe nice is for folks who are rendering a single frame, like for a large poster. The 64 cores would make quick work of most jobs, but for animation, you're better off going with with a farm.
I have used blender in a 16 processor machine without problems. If you have big renders it should not be a problem since there is not really any interprocess communication.
There's no need to buy a Ferrari if you use it twice a year, just rent it.
Most of the supercomputing locations where I worked at are very shy about their occupation rates. I think it is probably very low except at very active universities. All other places are wating their money buying hardware which will become useless while is not used.
See Powua http://www.powua.com/ [powua.com] as a general implementation or PurePowua http://www.purepowua.com/ [purepowua.com] as a more specialized one, in this case XSI rendering.
Not even close. The heavy lifting for 3D games is done on the GPU, and I'm not aware of any games (except perhaps games that utilize multiple monitors, like flight simulators) that can make use of more than one GPU.
So a single game could potentially drive many monitors, but not do more visually on a single display.
However, this thing could do some amazing real-time raytracing, but again, no games have been designed for such hardware yet.
A number of modern games can make use of 2+ cores, but 8 isn't going to happen with any efficiency. Note also that this is a cluster in a single box -- those 8 nodes are each different computers on a very fast local network. That means a different OS image per node, and each process on its own node. For lots of supercomputing applications, this is the norm -- each node does its share of the work and they talk over the network. But no games support this; they all expect to run on a single computer.
Also, for gaming performance, I imagine you'd want dual graphics cards -- which this box doesn't support. (It does include "visualization node" options, which have a single Quadro FX card each.)
Still, for something like a desktop render farm, this might make sense -- except I imagine the customers for such would be more interested in options with better price/performance.
A modification to an engine (this has already been done to quake 3 and 4) to use raytracing, would lend itself well to this hardware. Raytracing is very SMP-friendly.
Raytracing is also very cluster friendly. One of my favorite cluster benchmarks / demos is showing how the Persistence of Vision Raytracer runs on a single node, two nodes, three, four... (my cluster is only four nodes, so I don't know how well it scales after that.)
For what it's worth, based on that benchmark my current cluster would have placed in the Top 100 in 1993.
Not even close. The heavy lifting for 3D games is done on the GPU, and I'm not aware of any games (except perhaps games that utilize multiple monitors, like flight simulators) that can make use of more than one GPU.
So a single game could potentially drive many monitors, but not do more visually on a single display.
Actually, you can configure the Cray CX-1 with "visualization nodes" [cray.com] that contain GPUs, not just CPUs.
I know you're being facetious, but the limiting factor in the output of a bot on a botnet is its connection speed, not its processing power. A '486 can saturate a 10mbit connection without taking a severe performance hit. Seeing as most of us don't quite have gigabit internet connections at home, this thing wouldn't be any more valuable to a herder than your neighbour's $500 laptop.
Far2Paranoid: Knew this guy in HS
Far2Paranoid: Built a box with 2x 350Mhz Pentium2, back in '98
Far2Paranoid: The trick was, filled his bathtub w/ glycerin
Far2Paranoid: Took apart a mini-fridge and used the coils to cool the glycerin to ~40F
Far2Paranoid: Then sunk the box so he could OC the CPUs to 1.3Ghz
Far2Paranoid: Coolest shit I've ever seen.
AlbinoChpmnk: If this was sitting in his tub, how did he shower?
Far2Paranoid: After what I just said, what
Nice Specs (Score:5, Funny)
Re: (Score:3, Funny)
Overpriced? (Score:4, Funny)
That thing looks mean! I'd pay 25k to be the only person in the office with one of those.
Parent
Yet... (Score:2, Funny)
Re:Yet... (Score:5, Funny)
No, but at least it can run Vista with most of the bells and whistles turned on.
Parent
You'll need one hell of a desk (Score:5, Insightful)
Re:You'll need one hell of a desk (Score:5, Funny)
Parent
Re: (Score:3, Interesting)
I remember seeing an HP server like that. On the left side was the large hard disk drive - on the right were the server processors - both kept off the floor but mounted into the frame where drawers would have been. On top of the desk was the monitor.
Re: (Score:2)
for the rest of the world (Score:5, Informative)
Parent
for Britian (Score:5, Informative)
That's 9 stone 8 lbs
Parent
Re:for Britian (Score:5, Informative)
Parent
Re:You'll need one hell of a desk (Score:5, Funny)
My girlfriend weighs that much, so I'm pretty sure my desk can handle it.
Parent
Re:You'll need one hell of a desk (Score:5, Funny)
Parent
Re:You'll need one hell of a desk (Score:4, Funny)
it's a static load.
That's what SHE said!
Parent
Re:You'll need one hell of a desk (Score:5, Funny)
you're not supposed to inflate her with water
Parent
desktop supercomputer (Score:2)
The question is, is it more "oxy" or "moron"?
Summary is incorrect (Score:2, Insightful)
Re: (Score:3, Informative)
"supports up to 8 nodes, a total of 64 cores and 64Gb of memory per node"
8 [nodes] x (2 [cpu] * 4 [cores]) = 64 total cores.
I do not see where it says 64 cores per node.
Natural language is ambiguous (Score:3, Insightful)
There's two relevant ways to parse that fragment. There's one where the "and" in "64 cores and 64G of memory per node" creates a single coordinated constituent, such that it can be paraphrased as "there are 64 cores per node and there are 64 Gb per node." There's a second, the one that I think you favor and that seems correct pragmatically, which may be paraphrased as "there are 64 total cores, and each node in the machine can have 64 Gb."
Structural ambiguity happens all the time in natural language.
Re: (Score:2)
Desktop? Where's the notebook? (Score:5, Funny)
Re:Desktop? Where's the notebook? (Score:5, Insightful)
Well... My netbook has 2 GB of memory, 160 GB of storage, gigabit networking and thinks it has two 32 bit cores. It's a veritable late 80's, early 90's supercomputer that fits in my backpack. And I bought it cheap.
Parent
Re:Desktop? Where's the notebook? (Score:5, Interesting)
Even in the mid 90's, GHz processors, and gigs of RAM/hard disk were still largely uncommon. I think you're talking late 90's before that started to become relatively common.
I continue to be stunned at what you can buy as an entry level box nowadays for a really cheap dollar amount. My local "white box" PC store will sell you a dual-core 5GHz (or whatever) 64-bit AMD machine for under $300 -- add a little RAM and disk space and you've got a helluva system for not very much money.
How many home PCs nowadays have TB's of storage? I know several people who do -- I remember when home users didn't have gigabytes, terabytes would have been unimaginable.
Cheers
Parent
Re: (Score:3)
"As someone who remembers punching holes in 5.25" floppies so you could turn 'em over and another 360K by using both sides"
I remember poking values in memory to upgrade my 140K disks to 160K and then punching the side of the disk (index holes were not needed in the brilliant Wozniak design) to be able to flip it over.
Boy... We are old. I bet I have icons in my desktop that would not fit in an Apple II floppy disk.
Horsepower (Score:2, Funny)
Well, Microsoft had to do something to create demand for the next version of Windows. Not much of a market for an OS where people need to book time at their neighborhood super collider when they need to edit a document.
Probably makes one hell of a spam node too!
Re:Horsepower (Score:5, Insightful)
Vista's MINIMUM memory requirement is 512 megs.
Windows 2000's recommended minimum was 64 megs.
Personally, I don't find Vista any more useful than Win2k. More stable, yes, but I don't see how upping the RAM req by an order of magnitude was required to make Win2k more stable. All it needed was better programming and better testing.
I think what we have going now is the kind of thing that happened when gas was cheap: SUVs. When gas is expensive (viz Europe and Japan) the average car gets Really Small and Efficient. When RAM was really expensive, programming was tight and efficient. Now that RAM is measured in gigs and drives in terabytes, there is no incentive to do efficient programming or wrangle in feature creep and bloatware.
Eventually we will hit some physical / cost limit on RAM, and then good programming will become a requirement. OF course, by then, there won't be anyone left who knows how to do that...
RS
Parent
Re:Horsepower (Score:4, Insightful)
Feature bloat for sure, but how do you know it's sloppily and inefficiently programmed? Have you seen the source? From what I recall of people commenting on leaked Microsoft code the quality was generally considered pretty good.
Parent
Re: (Score:3, Funny)
Windows 2000's recommended minimum was 64 megs.
.
The real-world hardware requirements for a Windows OS have always been those of a mid-priced system at the time of its release.
Tell me why an OS shouldn't be making use of resources as they become available and cheap.
I have never understood the Geek's obsession with RAM.
You would think he had been raised under the warm glow of a vacuum tube and threaded core for his Mom as a child.
The 8 GB 64 Bit Vista Pre
This is not meant to flame (Score:2, Insightful)
Re: (Score:3, Informative)
More like Apple (Score:5, Funny)
Remember, you can order any color - as long as it is gray.
Yeah, so? (Score:3, Interesting)
How well would for example... (Score:5, Interesting)
For example Blender [blender.org]'s renderer's scale on a system like this? Of course something like MentalRay might scale easily but has anyone any hands on experience?
One might argue if you are throwing away $25,000 on a system like that you might use software that costs, but then again, Blender has made tremendous progress these last years..
Re:How well would for example... (Score:4, Interesting)
Blender has made a lot of progress, but it is still way behind Maya and even Lightwave. I've not been using Blender in the past couple releases, but it used to have some issues on my Quad Core Power Mac and using more than 4GB of Ram. I think this has been addressed now though. But I've never run into the problem of RAM or processor speed being the problem, but video ram when modeling an object. I have created scenes that will even grind a decent 256MB video card into the ground. Sure, it would be nice to render a bit faster, but for $20 - $60 a month, I do as much rendering as I want at Respower.
But let's look at cost. For $25k I can buy about 75 commodity boxes that are dual core, 2GB of Ram each & networking gear. That's about 150 Cores and 150GB of Ram. Put Linux on there and you can run ScreamerNet (you get to put the LW rendering engine on 999 machines per license) or one of a number of Maya distributed rendering programs. End result are going to be more frames being processed at one time. (for animation)
If I went the Mac Mini route, that's about 40 Mac Minis, which is still 80 Cores, 80GB of Ram total and with ScreamerNet or Xgrid....
Now the downsides are, 40 - 80 computers take up a lot of space and probably would eat up more power/cooling costs. But then again, if a couple boxes kick the bucket or hiccup, the other 35 - 75 are still processing. You only loose a percentage of total output.
Where it maybe nice is for folks who are rendering a single frame, like for a large poster. The 64 cores would make quick work of most jobs, but for animation, you're better off going with with a farm.
Parent
Re: (Score:3, Informative)
Must be a nice keyboard, and an amazing power cord (Score:4, Informative)
Supercomputing is on demand, in the cloud (Score:3, Interesting)
Re: (Score:3, Insightful)
Not even close. The heavy lifting for 3D games is done on the GPU, and I'm not aware of any games (except perhaps games that utilize multiple monitors, like flight simulators) that can make use of more than one GPU.
So a single game could potentially drive many monitors, but not do more visually on a single display.
However, this thing could do some amazing real-time raytracing, but again, no games have been designed for such hardware yet.
Re:Gaming? (Score:4, Informative)
A number of modern games can make use of 2+ cores, but 8 isn't going to happen with any efficiency. Note also that this is a cluster in a single box -- those 8 nodes are each different computers on a very fast local network. That means a different OS image per node, and each process on its own node. For lots of supercomputing applications, this is the norm -- each node does its share of the work and they talk over the network. But no games support this; they all expect to run on a single computer.
Also, for gaming performance, I imagine you'd want dual graphics cards -- which this box doesn't support. (It does include "visualization node" options, which have a single Quadro FX card each.)
Still, for something like a desktop render farm, this might make sense -- except I imagine the customers for such would be more interested in options with better price/performance.
Parent
Re: (Score:3, Informative)
A modification to an engine (this has already been done to quake 3 and 4) to use raytracing, would lend itself well to this hardware. Raytracing is very SMP-friendly.
Re: (Score:3, Interesting)
Raytracing is also very cluster friendly. One of my favorite cluster benchmarks / demos is showing how the Persistence of Vision Raytracer runs on a single node, two nodes, three, four ... (my cluster is only four nodes, so I don't know how well it scales after that.)
For what it's worth, based on that benchmark my current cluster would have placed in the Top 100 in 1993.
Re: (Score:3, Informative)
Actually, you can configure the Cray CX-1 with "visualization nodes" [cray.com] that contain GPUs, not just CPUs.
Re: (Score:2)
Re: (Score:3, Informative)
Or you could just buy the Cray for the same price and forget about the extra overhead of 8 separate boxes.
BTW, you can also order these from the factory with RHEL.
Re:Yes, but only for a short time (Score:5, Insightful)
it says it runs windows. that's just what the herders need, a few crays in their herd.
Parent
Re:Yes, but only for a short time (Score:5, Insightful)
I know you're being facetious, but the limiting factor in the output of a bot on a botnet is its connection speed, not its processing power. A '486 can saturate a 10mbit connection without taking a severe performance hit. Seeing as most of us don't quite have gigabit internet connections at home, this thing wouldn't be any more valuable to a herder than your neighbour's $500 laptop.
Parent
Re:Yes, but only for a short time (Score:5, Insightful)
He could use it to crack passwords or something.. lots of processors and memory is pretty handy for that
Parent
Re: (Score:3, Funny)
Yes, but it's easier if just you hit the "Turbo" button.
Re: (Score:3, Funny)