Using 1.8G of RAM in a BioInformatics compute program isn't at -all- difficult. There's a reason I've got systems with MAX physical RAM...these fartknockers are loading multi-gig sized datasets (or generating them into memory on the fly) and they like to use apps that mmap that data:\
Exactly. And -PRESCOTT- cores? What, do they think we're nutty enough to have a desktop system that needs to dissipate 200+ watts of heat? Please. I don't think so.
Where do you get $70/node for fibre? I just bought six half-way decent cards at $800/ea. That's more like $400k. This is why you're seeing a move to things like Lustre. It's cheaper to have a small storage cluster serving the larger cluster than it is to run fibre to the whole thing.
For example: you have 128nodes, a manager node (or 2) and a couple access/job submission hosts (assuming SGE in use here) Total of 132 machines. Let's say you have a multi-TB storage device (say, an EMC Clariion CX500 or so). To run fibre to the whole cluster is going to run 800*132 + $SC (switch cost). 800 * 132 = 105k. Switching is going to run another 100k (or so). 205k for fibre -OR- you can get 4 servers, each with FC, running Lustre, and you have a scalable, REDUNDANT file-store that you can build for under 30k.
Not all Universities place that level of responsibility on the grad students (unless they're CS, EE or CompE grad students). -ESPECIALLY- in Life Sciences. Grad students in research areas are supposed to be going to class and doing research, not spending another 20-40 hours a week co-admining the computing resources. It's one thing to help maintain a specific application (ie. I've got a grad student maintaining the databases for our local NCBI BLAST installation because he's literally the only person using it), it's something entirely different to have a grad student trying to maintain systems. I could comment on other aspects of how your school is doing shit, but if that's the way they want it, more power to 'em.
That said: $100k is fucking cheap, and doesn't buy -that- much computing power (16nodes or so, and no real external storage). Not to downplay 16 nodes, for some folks, that really is all the computing power they need (and in our case, I'll probably be specing out another 32 for purchase in 6 months once the new 32 comes online in 4-8 weeks).
The atmosphere is also warming at a rate unheard of since we started measuring these things and at a rate not found in any ice samples from the last several hundred thousand years
They _ARE_ 4 months late on G5 XServe shipments. I ordered the day the damned machine was launched, and I'm still waiting. If these are awarded and -SHIPPED- before we get ours, I will very likely be speaking to our legal department. I need the machine, and I needed it two months ago.
Certain operations, though, are highly dependant upon each previous result. Physics and chemical simulations are a good example. When you have situations like this, clusters don't do you a lot of good, since only one iteration can be worked on at a time -- leaving most of your cluster sitting there idle.
Umm, bwah? It's only going to be sitting there idle if you're not properly scheduling and qeueing jobs. Also, you -CAN- do the kind of simulations (Physics, chemicals) on a cluster *points at clusters at Chrylser and Shell*. The caveat is that you need to write out the result for the appropriate job to handle (in practice - job run 1 contains step 1, job run 2 step 2, etc). And a cluster is perfectly fine for this.
That all said - a supercomputer like this -IS- generally a better tool for the job if you've got the money. Money, in most places, -IS- an object, so we get the best bang for our buck. *shrug*
Which is forcing me to continue waiting for the one I ordered the day the fucking things were announced. They've gone from giving me a mid to late April ship date to "Sometime in June".
Screw that. Apple is screwing the pooch if they're at all serious about getting into enterprise computing. It's one thing to slip one or two months, but now they're at four, and I wouldn't be suprised to see it go to 6 at this point.
It also doesn't help that there's no version of Windows that takes advantage of 64-bitness yet, or that a HUGE portion of the IT people making purchasing decisions still doesn't trust AMD CPUs.
I don't resemble that remark! And neither do a lot of the HPC geeks. We need bang for buck, primarily, and we're getting with with Opty right now:)
Then you obviously haven't used anything from them recently. Seriously dude, AMDs current crop (Barton XP and Athlon64) are kick ass chips, and the boards are stable. Don't blame AMD for the chipset companies provided half-assed chipsets that predominantly sucked. That's pretty much been fixed, and what do you know, AMD's king of the performance hill.
But to answer your direct question: No, a 2GHz Dothan will -NOT- be able to compete with a 3.6GHz Prescott. But then, you might be able to aircool dothan, or use a lightweight fan if you can't, and you're borderline needing a water-cooler for the highest end prescotts.
Don't you just -love- canned responses? Even though slightly modified to be a "direct" reply to your "problem", it's still a canned response. 'sides, Dell isn't going to offer AMD systems until they're forced to by -corporate- market pressures. It's happening, it's just slow (and they know they lost a decent sized purchase I just made because of it).
This really shouldn't be a suprise to -anyone- who's been paying attention to what's been going on. Prescott is disturbingly hot, and the next-gen chips had no real hope of being much cooler. At most 10-15%, which wouldn't have gotten near their MHz goals. P-M, on the other hand, is a damned good chip in its own right, has better IPC, and is a better CPU, all around, than the P4 line.
Now, what does this mean for those of us in the enterprise space? Are we -really- going to have to wait until 2006 for a new chip iteration from Intel? If that's the case (and I -really- doubt it), AMD would have a disturbingly large (and long) opening in which to pitch its wares...Intel would definately lose marker share in the server arena at that point.
So, multi-core P-M chips for the desktop next year-ish. So we're stuck with the hotplate known as Prescott until then. Guess I'll be sticking with AMD for a while yet:)
Ummm, I've had DB files -much- larger than 2 GB on 32bit linux systems...3 years ago. I don't find it at all implausible that they've got that much data lying around, at all. I know of at -least- one research center with more data on disk than that.:)
While all that is true, you are also ignoring order #5:
5. SCO is to provide and identify with specificity the lines of code that SCO distributed to other parties. This it to include where applicable the conditions of release, to whom the code was released, the date and under what circumstances such code was released.
Now, IANAL, but this tells me that SCO is pretty well fucked. They have to tell the court which lines of code IN LINUX (since they have to provide all lines in Linux which they believe were illegally contributed), SysV, etc were released, and under what licensing terms and conditions they were released. Since SCO (as Caldera) distributed Linux as a product, under the GPL.... The IBM "stuff" is to throw SCO a bone, IMO, although it -is- entirely possible that SCO may be able to find something. They probably won't, but the chance is there, and if they -DO- find something truly damning...well, let's just say that I don't want to follow that line of reasoning at this time:)
*snickerfits* Not quite:) The Orioles are broadcast by one of the local stations, and are -occasionally- picked up by CSN if the game isn't otherwise being televised, the Ravens are always on a broadcast channel, as are, I believe, the Redskins. The Caps are probably the only team that usually gets stuck on CSN, at least in Baltimore, and I'm not that big of a hockey fan. *shrug*
Groklaw has already been cited as a source in at least one of IBM's briefs to Judge Wells. So yeah, they're definately getting direct credit, and due:)
Actually, there's some fairly hefty core changes involved in making the integer pipeline 31 stages instead of 20. Almost enough to account for most, if not all, of the additional transistors in this thing. I'd actually be suprised if the AMD64 extensions were in this chip...most of what I've been hearing/seeing has Tejas being the earliest we could see them.
The Opteron beats Itanium in integer benchmarks, which is why it generally makes for a better overall webserver platform, and general purpose server platform, really. When it comes down to the FPU, Itanic wins, but the price/performance comparison sucks to the point where it's cheaper to buy two or even three opteron systems to get the same FLOPS as a single Itanium box. It's actually kinda funny;)
One important consideration in the G5 versus Opteron comparison is who is standing behind the product. The only tier 1 vendor who has announced Opteron systems is Sun, and those are currently vaporwar
This is -semi-correct-. IBM has also announced (and is apparently also -shipping-) Opteron based cluster systems. I can also say, with authority, that Sun's efforts in getting Opty machines out the door aren't vapor, either. I'm not at liberty be any more specific that that, though (gotta love NDAs...).
In the past at my job we have always purchased systems from tier 1 vendors, first IBM and then Sun. Recently we experimented in cost savings by purchaseing a HPC cluster from a vendor found through the bids system, and it has been nothing but a nightmare. We've decided in the future to purchase only from tier 1 vendors because of this experience, and will probably end up building our next cluster from G5s (we are an educational institution and thus receive a very generous educational discount from Apple), especially with the recent release of IBM's XL Fortran compiler for OS X. Apple's discount isn't -THAT- great, unless you get better than what I'm getting at a major East Coast research school. At the institutional level, we primarily go with Dell and Sun. The -current- HPC system that I run is from a "third party" vendor, and it's been pretty solid for us. No real problems, etc. But then, I don't need much hand-holding from a vendor to get done what needs to be done;)
Slashdot Mirror
Ok. Now how to you distinguish between innocent bystanders (ie. the zombie relay folks) and the fartknockers actually doing the spamming?
You can't.
As nice as it would be, you really need to be -absolutely- sure you've got a spammer before you try to ruin their life with the court system.
Using 1.8G of RAM in a BioInformatics compute program isn't at -all- difficult. :\
There's a reason I've got systems with MAX physical RAM...these fartknockers are loading multi-gig sized datasets (or generating them into memory on the fly) and they like to use apps that mmap that data
Exactly.
And -PRESCOTT- cores?
What, do they think we're nutty enough to have a desktop system that needs to dissipate 200+ watts of heat?
Please. I don't think so.
Where do you get $70/node for fibre?
I just bought six half-way decent cards at $800/ea. That's more like $400k. This is why you're seeing a move to things like Lustre. It's cheaper to have a small storage cluster serving the larger cluster than it is to run fibre to the whole thing.
For example: you have 128nodes, a manager node (or 2) and a couple access/job submission hosts (assuming SGE in use here) Total of 132 machines. Let's say you have a multi-TB storage device (say, an EMC Clariion CX500 or so). To run fibre to the whole cluster is going to run 800*132 + $SC (switch cost). 800 * 132 = 105k. Switching is going to run another 100k (or so). 205k for fibre -OR- you can get 4 servers, each with FC, running Lustre, and you have a scalable, REDUNDANT file-store that you can build for under 30k.
Parallel Virtual Filesystems, man.
Not all Universities place that level of responsibility on the grad students (unless they're CS, EE or CompE grad students). -ESPECIALLY- in Life Sciences. Grad students in research areas are supposed to be going to class and doing research, not spending another 20-40 hours a week co-admining the computing resources. It's one thing to help maintain a specific application (ie. I've got a grad student maintaining the databases for our local NCBI BLAST installation because he's literally the only person using it), it's something entirely different to have a grad student trying to maintain systems. I could comment on other aspects of how your school is doing shit, but if that's the way they want it, more power to 'em.
That said: $100k is fucking cheap, and doesn't buy -that- much computing power (16nodes or so, and no real external storage). Not to downplay 16 nodes, for some folks, that really is all the computing power they need (and in our case, I'll probably be specing out another 32 for purchase in 6 months once the new 32 comes online in 4-8 weeks).
The atmosphere is also warming at a rate unheard of since we started measuring these things and at a rate not found in any ice samples from the last several hundred thousand years
:)
Source? Thanks
They _ARE_ 4 months late on G5 XServe shipments.
I ordered the day the damned machine was launched, and I'm still waiting.
If these are awarded and -SHIPPED- before we get ours, I will very likely be speaking to our legal department.
I need the machine, and I needed it two months ago.
-
J.
Certain operations, though, are highly dependant upon each previous result. Physics and chemical simulations are a good example. When you have situations like this, clusters don't do you a lot of good, since only one iteration can be worked on at a time -- leaving most of your cluster sitting there idle.
Umm, bwah?
It's only going to be sitting there idle if you're not properly scheduling and qeueing jobs. Also, you -CAN- do the kind of simulations (Physics, chemicals) on a cluster *points at clusters at Chrylser and Shell*. The caveat is that you need to write out the result for the appropriate job to handle (in practice - job run 1 contains step 1, job run 2 step 2, etc). And a cluster is perfectly fine for this.
That all said - a supercomputer like this -IS- generally a better tool for the job if you've got the money. Money, in most places, -IS- an object, so we get the best bang for our buck.
*shrug*
Which is forcing me to continue waiting for the one I ordered the day the fucking things were announced.
They've gone from giving me a mid to late April ship date to "Sometime in June".
Screw that. Apple is screwing the pooch if they're at all serious about getting into enterprise computing. It's one thing to slip one or two months, but now they're at four, and I wouldn't be suprised to see it go to 6 at this point.
Fartknockers.
It also doesn't help that there's no version of Windows that takes advantage of 64-bitness yet, or that a HUGE portion of the IT people making purchasing decisions still doesn't trust AMD CPUs.
:)
I don't resemble that remark! And neither do a lot of the HPC geeks. We need bang for buck, primarily, and we're getting with with Opty right now
Then you obviously haven't used anything from them recently. Seriously dude, AMDs current crop (Barton XP and Athlon64) are kick ass chips, and the boards are stable.
Don't blame AMD for the chipset companies provided half-assed chipsets that predominantly sucked. That's pretty much been fixed, and what do you know, AMD's king of the performance hill.
But to answer your direct question:
No, a 2GHz Dothan will -NOT- be able to compete with a 3.6GHz Prescott. But then, you might be able to aircool dothan, or use a lightweight fan if you can't, and you're borderline needing a water-cooler for the highest end prescotts.
Don't you just -love- canned responses?
Even though slightly modified to be a "direct" reply to your "problem", it's still a canned response.
'sides, Dell isn't going to offer AMD systems until they're forced to by -corporate- market pressures. It's happening, it's just slow (and they know they lost a decent sized purchase I just made because of it).
This really shouldn't be a suprise to -anyone- who's been paying attention to what's been going on.
:)
Prescott is disturbingly hot, and the next-gen chips had no real hope of being much cooler. At most 10-15%, which wouldn't have gotten near their MHz goals.
P-M, on the other hand, is a damned good chip in its own right, has better IPC, and is a better CPU, all around, than the P4 line.
Now, what does this mean for those of us in the enterprise space? Are we -really- going to have to wait until 2006 for a new chip iteration from Intel? If that's the case (and I -really- doubt it), AMD would have a disturbingly large (and long) opening in which to pitch its wares...Intel would definately lose marker share in the server arena at that point.
So, multi-core P-M chips for the desktop next year-ish. So we're stuck with the hotplate known as Prescott until then. Guess I'll be sticking with AMD for a while yet
Ummm, I've had DB files -much- larger than 2 GB on 32bit linux systems...3 years ago. :)
I don't find it at all implausible that they've got that much data lying around, at all.
I know of at -least- one research center with more data on disk than that.
Now, IANAL, but this tells me that SCO is pretty well fucked. They have to tell the court which lines of code IN LINUX (since they have to provide all lines in Linux which they believe were illegally contributed), SysV, etc were released, and under what licensing terms and conditions they were released. Since SCO (as Caldera) distributed Linux as a product, under the GPL....
The IBM "stuff" is to throw SCO a bone, IMO, although it -is- entirely possible that SCO may be able to find something. They probably won't, but the chance is there, and if they -DO- find something truly damning...well, let's just say that I don't want to follow that line of reasoning at this time
-
J.
Symetrical.
:)
And yes, yes it is. I wonder how SCO is going to get around that if someone deides to make an issue of it?
don't forget Alabanza! ;)
*crawls back into his hole*
Also, re: Opteron, they've apparently "gotten it", and you -should- be able to run at least some of the models with just a heatsink...not even a fan.
I want to say the rig I've got (dual 244, 6G RAM, 1U rackmount) is fanless, but I'd have to go crack the case and look.
Actually, I like this idea...a _LOT_ .
make Disney -DISNEY- again, dammit.
*snickerfits* :)
Not quite
The Orioles are broadcast by one of the local stations, and are -occasionally- picked up by CSN if the game isn't otherwise being televised, the Ravens are always on a broadcast channel, as are, I believe, the Redskins. The Caps are probably the only team that usually gets stuck on CSN, at least in Baltimore, and I'm not that big of a hockey fan.
*shrug*
Groklaw has already been cited as a source in at least one of IBM's briefs to Judge Wells. :)
So yeah, they're definately getting direct credit, and due
Actually, there's some fairly hefty core changes involved in making the integer pipeline 31 stages instead of 20. Almost enough to account for most, if not all, of the additional transistors in this thing.
I'd actually be suprised if the AMD64 extensions were in this chip...most of what I've been hearing/seeing has Tejas being the earliest we could see them.
Sun is doing an Opteron-optimized build of Solaris, but don't expect it in the short term.
That's about all I can say ATM, but it -is- on the way...
The Opteron beats Itanium in integer benchmarks, which is why it generally makes for a better overall webserver platform, and general purpose server platform, really. When it comes down to the FPU, Itanic wins, but the price/performance comparison sucks to the point where it's cheaper to buy two or even three opteron systems to get the same FLOPS as a single Itanium box. It's actually kinda funny ;)
One important consideration in the G5 versus Opteron comparison is who is standing behind the product. The only tier 1 vendor who has announced Opteron systems is Sun, and those are currently vaporwar
;)
This is -semi-correct-.
IBM has also announced (and is apparently also -shipping-) Opteron based cluster systems. I can also say, with authority, that Sun's efforts in getting Opty machines out the door aren't vapor, either. I'm not at liberty be any more specific that that, though (gotta love NDAs...).
In the past at my job we have always purchased systems from tier 1 vendors, first IBM and then Sun. Recently we experimented in cost savings by purchaseing a HPC cluster from a vendor found through the bids system, and it has been nothing but a nightmare. We've decided in the future to purchase only from tier 1 vendors because of this experience, and will probably end up building our next cluster from G5s (we are an educational institution and thus receive a very generous educational discount from Apple), especially with the recent release of IBM's XL Fortran compiler for OS X.
Apple's discount isn't -THAT- great, unless you get better than what I'm getting at a major East Coast research school. At the institutional level, we primarily go with Dell and Sun. The -current- HPC system that I run is from a "third party" vendor, and it's been pretty solid for us. No real problems, etc. But then, I don't need much hand-holding from a vendor to get done what needs to be done