Slashdot Mirror
SGI Rolls Out "Personal Supercomputers"
CWmike writes "They aren't selling personal supercomputers at Best Buy just yet. But that day probably isn't too far off, as the costs continue to fall and supercomputers become easier to use. Silicon Graphics International on Monday released its first so-called personal supercomputer. The new Octane III system is priced from $7,995 with one Xeon 5500 processor. The system can be expanded to an 80-core system with a capacity of up to 960GB of memory. This new supercomputer's peak performance of about 726 GFLOPS won't put it on the Top 500 supercomputer list, but that's not the point of the machine, SGI says. A key feature instead is the system's ease of use."
Can you imagine a Beowulf cluster of those? :-)
Or you can save a good chunk of change and buy PS3s at $300 for 8 cores. That's 200 cores for under $8K. And open source too.
and one Xeon does not make a super-computer.
What do you think it's for?
Wouldn't most people who would NEED a supercomputer be able to build one much more cheaply using a dozen workstations? It's hard to see how this SGI system might be sold (except perhaps as a replacement for an overburdened business-office server).
Those are my principles, and if you don't like them... well, I have others.
Hell, a Mac Pro is even cheaper with 8 cores. Who in their right mind is going to spend $8000 for a single core pc?
I know you're joking, but from the article:
I've seen the term 'personal supercomputer' so many times over the past 20 years. It's just baloney marketing. What you have on your desktop RIGHT NOW is more capable than some of the original CDC machines. So what?
Yeah, but will it run linux?
Picture here: http://www.ubergizmo.com/tags/octane-3
Those are my principles, and if you don't like them... well, I have others.
Who was the idiot who thought that it would be a good idea to call this the "Octane III"? This has almost no resemblance to the SGI Octane systems of that past, which were graphics workstations running Irix with MIPS processors. I think the only thing that makes them similar is the price range.
This goes right up there with Honda constantly recycling their product names; passport, odyssey, pilot, and more recently insight.
Damn_registrars has no butt-hole. Damn_registrars has no use for a butt-hole.
"They aren't selling personal supercomputers at Best Buy just yet."
Sure they are, it just depends on what era supercomputer you are comparing that commodity computer to. A modern desktop machine is insanely fast with inconceivable amounts of ram and disk storage if you think back a couple (several) decades. Best-buy will never sell super-anything, it's not their game. But the computers we take for granted are insanely capable machines based on the problems tacked in the past by supercomputers.
Now get off my lawn,
Sheldon
(who did his master's thesis on a 16MHz machine)
Isn't this basically the failed business model that put them under the first time?
Well... It's 8k, and the CPUS, according to a quick google, are ~2k apiece new (?! For one CPU?). So presumably you can get the full 80-core experience for 168k.
For comparison, a fast commodity rig might cost, I dunno, 1.5k? Times 80, and you're at 120k? So this thing, fully decked out, is possibly 40% more expensive than an equivalent commodity setup? If it's commensurately faster -- which is easy to believe as the processors are on the same mobo instead of strung across a network -- then it could be a net win to use this machine, maybe?
Anyway, it's hard to say. I'm using ballpark numbers and the results are the same order of magnitude, so it might go either way. The point is that the price doesn't seem completely absurd, at first glance at least...
This is a super computer not some thing you would use for normal every day computer activities. I think it would cater more to people that are doing protein bending or other extremely processor intense activities. Most of people wanting to use one probably have access to a University or Government Super Computer so that only leaves the self-employed or small business research market. I would not be surprised if even more of these appear from other companies because that market is one of the few that is growing.
As someone who has a whole Sun Enterprise 5500 rack in his room, There is indeed a great difference between server class hardware and commodity gear, where shall we start.
Multiple power supplies, varied in number depending on your load out but hot swappable and configured as such that 1-2 of them can die before your system goes down. Along with diagnostic interface and usually visible indicators going 'part failure, replace asap'.
Same with cpus, hot swappable cpu/memory boards are a must, so long as a single cpu remains functioning the system should still run albeit at a lower capacity.
While I've already mentioned psu redundancy, the AC power outlets it uses would usually have redundancy also, with two separate connections to different circuits or ups etc.
Anyway, no commodity hardware does this, only high end, high availability stuff has this, and you will pay through the nose for it. If this octane has these features, it is very cheap for what it is.
What will a home user do with an 80 core, 1TB RAM sysetm? Ray tracing? Protein folding? Local weather prediction? All things really high on the list for personal computers.
Still, you'd never need to heat your house again.
Finally had enough. Come see us over at https://soylentnews.org/
Aw crepe, if these become commonplace M$ might rewrite Windows using dot net, and of course Sun would write a knockoff in Java. By then Linux will have 8 different windowing toolkits necessary for the basic apps and 29 sound systems. Oh well, I guess it's back to 0x7C00...
Then again, I'm buying up Marvel SheevePlugs as fast as I can afford 'em. With built-in 1000TX networking and a Kingston SOC chip delivering approximately the same performance as a 1GHz Intel CPU, I figure I can network 'em together and have a scalable (Beowulf) supercomputer for a lot less money and only a modest investment in elbow-grease. The uBoot environment is already smart enough for TFTP boot and root over NFS (which is how IBM does the magic, IIRC). All I need is a monotasking kernel to serve to my nodes and I'm in business. For now, I'll settle for the standard Linux 2.6 kernel and take the modest performance hit.
Ouch! I never understood the need for all of this specialized "server-class" hardware when cheap-o commodity hardware and a little elbow grease works just as well. Maybe most people don't want to put the work into it, considering the huge jump in price between retail consumer and server pricing, I've never been able to justify shelling out those kind of bucks.
I can't decide if you're a troll or an amateur. If you're the latter, I meant that in the nicest possible way.
If you don't understand that there's a difference between "cheap-o commodity hardware" and server hardware, you don't understand nearly as well as you think you do.
I'm assuming you're a troll but I'll bite anyway, modern Macs all run on 64-bit Intel CPUs, early Intel Macs did have 32-bit CPUs though.
/Mikael
Greylisting is to SMTP as NAT is to IPv4
So, SGI is finally putting Erwin into *real* SGI machines?
Colorless green Cthulhu waits dreaming furiously.
Bottom of the Top500 as of June is about Max 17 Tflops, Peak 37 Tflops. So if this is really 0.6 Tflops well... maybe one of these would make one node of a many-noded supercomputer.
Interesting to compare to a rack of Apple Xserves. Each rack is 8 cores (same cpus as the Octane III it seems). Again about $50k for 80 cores. Looks like sgi is aiming at that segment.
Can anybody with Xserve experience say how these would compare? I see Apple has something called Xsan too.
You forget service and support! Sun will make sure that for the serviceable life of your machine they will have replacement parts on hand and technical support for your machine. Imagine a commodity system looses a motherboard, will you be able to get the exact one three years down the line? And with pretty much every board maker located in Taiwan will they give you proper tech support in a timely manor? Will they ensure you get matching memory and CPU's? That's the other strong point of server class hard ware that is thoroughly supported by the vendor.
whole Sun Enterprise 5500 rack in his room
Does it run 24 hours/day (as the high availability options would suggest)? How much power does it consume? Why are you burning all that power all day?
I'm hoping for an answer like: "I'm modeling a quantum electrodynamic system at 100 hours of processing per microsecond." Or maybe "I'm trying to find a zero of the Zeta function off the critical line." Or even "I'm trying to factor a big number." Please don't tell me "It sits idle 23.9 hours a day and then I play Spacewar on it once in a while."
Although we all know the likelihood of each of these answers.
Hardware fails, and commodity hardware fails more. Using it on a large scale for systems with a high-availability requirement needs engineering/operations skills and resources that many companies simply do not have, and even at the companies that theoretically have the resources, they often fail.
Designing and building applications to withstand server failures can be extremely difficult, especially if you have to scale, and properly testing them can border on the impossible. Sometimes you're better off just getting hardware with a sufficiently low failure rate and dealing with the rare outages as they come up.
I read TFA, apologies if I missed it, but, what operating system does it run? Please, for the love of all that is good, let the answer NOT be IRIX.
O how the mighty have fallen. On a brighter note, I'm looking forward to the eight core xeons coming out this year....
EXACTLY. Its just a box. It woudl be very easy to build on of these yourself i my opinion. And a bunch of ARM chips on mini arm motherboards and you have yourself a large capable web cluster. etc.
With that much personal computing power, just put virtualization on there and run as many VMs as you want to your heart's delight! No point in arguing the superiority of various OSes. Just make a VM of one and run it at the same thing as VMs of other OSes. If you want, spin up a few VMs and say you have a Beowulf cluster, which puts the whole idea on its head. Have your own virtual datacenter inside a single computer! If you want to be all MBA about it, create your own "cloud" inside. This is like Legos but for grown nerds (or more like second set of Legos since grown nerds still play with Legos).
EvilCON - Made Famous by
When someone says server class think redundancy. If a something breaks on your cheapo server it will be down until it is repaired costing the company money in lost productivity, or even sales. With a "server class" system a singular point of failure will not kill it and the part can usually be hot swapped so NO down time. Trying to build redundancy into a chepo server would be a lot of work easily costing more to build then a "server-class". Lets say you did manage to build a server on the cheap with redundancy there will be no one to fix it or diagnose it if the builder is unavailable so unless you plan on never vacationing I would stick to the out of the box solutions.
Knowledge = Power
P= W/t
t=Money
Money = Work/Knowledge so the less you know the more you make
You can't fool me, Kilgore Trout!
Do what thou wilt shall be the whole of the Law
Here's just a brief search for personal supercomputers of days gone (not too far) by. Most if not all are cheaper than the SGI. Being older they may not stack up spec-wise, and the definition will always be changing anyway. More than one claim to be 'first', and to SGI's credit they only claim it's 'their' first.
http://tech.slashdot.org/article.pl?sid=08/11/23/068234
http://www.researchchannel.org/prog/displayevent.aspx?fID=569&rID=4263
http://aslab.com/products/workstations/marquisk942.html
http://www.reghardware.co.uk/2006/06/07/tyan_unveils_typhoon/
http://www.hpcwire.com/features/Cray_Unveils_Personal_Supercomputer.html
"I may be synthetic, but I'm not stupid." -- Bishop 341-B
As someone who has a whole Sun Enterprise 5500 rack in his room
Do all the blinking lights impress the ladies? Oh wait, this is slashdot, no ladies have probably seen the inside of your room :P
The definition of a super is a system at the top order of magnitude of speed and memory. Since the current record is two petas, a super would be one hundred teras. A one tera computer is a super of a decade ago.
The 5550 is a quad core. So for 80 cores, you only need 20 procs.
That makes the 80 core experience $8000 + ($2000 * 19) = $46,000.
For commodity hardware, you'd have to be getting your computers for $575 for single core machines, or $1150 for dual core. And that doesn't count the discount you'd probably get on the CPUs for buying 19 at once.
Although you might get a similar discount on commodity machines, but it's certainly not guaranteed....
Then there's all the hardware necessary to connect up 40 or 80 machines. Have you priced out a 40+ port Gigabit switch lately? Plus the 40 Cat6 cables, plus all the screwing around trying to get the clustering working, since the target market for this will probably rarely, or never have done it before...
Then there's the extra wiring your house/business is going to need to plug in 40 computers.
Sure, this thing will probably need a 50 amp circuit if it's loaded with CPUs, but a single high current circuit is much easier and safer than a 40 outlet octopus that would be needed for a cluster.
"City hall" in German is "Rathaus" Kinda explains a few things......
I can't find a price for the configuration with a bunch of atom processors, so it is hard to say whether it is an economical way to create a certain amount of processing power. I also wouldn't have the numbers to put a price on the possible savings due to lower electricity and cooling requirements. None of this matters unless you are running something that can use that many processors efficiently.
Ironically, those that don't know what it is tend to be like "You have a towering monolith in your room, but where are the monkeys?" (2001: a space odyssey reference)
Big deal. I have a 1 GHz processor that is almost a decade old, and a 1.2 GHz processor that is six months younger than the other one. The new one has a number more than the CUBE of the old one! And they're only six months apart!!1one!
"Trolls they were, but filled with the evil will of their master: a fell race..." -- J.R.R. Tolkien on Olog-hai
The system code named DESCH (named for Joseph Desch, who led the secret WWII project which developed a decoder for the Nazi Enigma encrypted messages.) collects terabytes of raw data gathered by the USAF Gotcha's synthetic aperture radar equipped UAV's orbiting over an area of interest in a war zone. The system images a 5km dia "city sized" view and processes the result into 3D image maps while recording to disk for review. The 400 Megapixel per second streaming images allow zooms into areas of interest, observation of minute changes and the ability to track personel and vehicals in the urban battle field. http://www.wpafb.af.mil/news/story.asp?id=123165818/
Let me know when you assemble what is essentially a blade chassis (it's really more comparable to the Onyx/Origin, Superdome, or Sun's E$numK systems given that you have the option of using PCIe nodes instead) with Infiniband interconnects in a deskside form factor, plus redundant hot-swappable components and the ability to run the entire thing as a single OS image yourself.
"The more corrupt a society, the more numerous are its laws." -Tacticus
Why would I buy the SGI at 726 Gigaflops instead of a single Nvidia Tesla card and get 1 Teraflop? I could buy four cards and get four teraflops for a lot less money than the SGI. There probably is something else going for the SGI, what is it?
Since apparently the definition of a supercomputer is a machine capable of 1 gigaflop, SGI was scooped by Apple 10 years ago!
The price for a fully decked out system is $58,000 (according to an earlier post), however the floor space / cooling system alone would justify the cost increase. This would make an excellent virtual server system provided your servers did not require a massive amount of data storage locally. I wonder if we could get one with a fiber channel connector so it would interface with a large disk array?
Would someone please enlighten me as to how this is better than a cluster of commodity servers? Or a cluster of workstation class machines? Or a cluster of commodity servers with a workstation class machine as the head node? I'm not seeing it; and, the SGI looks pricey.
Actually, windows is not on the supported list of OSes for SOME MODELS... RHEL, Suse, etc are.
The marketing makes it sound like you get a single computer with 80 cores, which is not the case. Basicly you get one machine that's the controller. In the case there are room for a number of sleds, each it's own diskless server. They are intended to boot over bootp or PXE or some other network boot. By themselves, they do nothing. Only the basic $8k box is a personal computer. Once you add the first sled it becomes a 'cluster in a box.'
Because of the clustering it's completely useless as a windows workstation. Vista/Win7 is not going to cluster crap for you. Windows server CAN cluster, but 'common people' (those not on /.) don't go that far for their DESKTOP computers.
So this is only practical if you have the knowledge and software to set this up as a cluster. First thing that comes to mind are the guys at Pixar. Map it out on the workstation, save it to a network share, then the blades pick it up and render. Actually, they'd probably have a distributed rendering system, so it would be even more seamless.
I wouldn't mind using it for web development. Program on the workstation, one sled runs apache, the other runs mysql on a ramdisk :) Another running memcached. Yeah, I know. Overkill.
Do not meddle in the affairs of sysadmins, for they are subtle, and quick to anger.
I've had it described recently that computers are cheap, people are expensive. It's worth shelling out for high end equipment to prevent your employees from wasting their time due to hardware issues. Any given consumer PC is likely to be stable, but when stringing together several hundred for a simulation, you're going to have problems. When you lose a week's worth of work because one of your (165) nodes was causing silent data corruption because of a cheap, onboard gigabit ethernet controller, those expensive server grade NICs and infiniband/myranet cards start looking better and better.
On the other hand, graduate researchers are dirt cheap too, so universities like to go for the cheaper, less robust hardware.
I've seen the term 'personal supercomputer' so many times over the past 20 years. It's just baloney marketing. What you have on your desktop RIGHT NOW is more capable than some of the original CDC machines. So what?
What you have on your desktop RIGHT NOW is most likely more powerful than the Cray Y-MP by a factor of three, if you've got a quad-core Core2 Duo; those babies push +1Gflop.
It's also 1/50th to 1/100th as capable as this supercomputer (or more- I don't know the relative performance between a current desktop processor and current Xeon.) Yes, it's relative, and relatively speaking, this is most certainly a supercomputer. In terms of memory, the maximum amount of ram you can put into a consumer-available motherboard is around 64GB, maybe 128. This has a maximum of 10 times that.
80 xeon cores, 1TB of memory, and you call it a "marketing ploy"? And you got modded up "insightful"? May the hand of metamoderation come on down from high.
Please help metamoderate.
It may not be enough to simulate global weather patterns, but surely it's powerful enough to predict the weather in my parents' basement. Drafty today? Time to pull out my electric "7 of 9" blanket and nuke a few frozen cheeseburgers.
Seriously though, I doubt that there is much demand for personal supercomputing.
You're not getting Xeon FSB communication between any arbitrary two of the processors.
The machine's organized into boards with 2 CPU sockets on each, which can support a quad-core XEON. I suppose the two Xeons on the processor board could communicate at FSB speeds.
However, between boards, for interconnect you get either dual gigabit ethernet (2Gb/s) or QDR Infiniband (10Gb/s), a far cry below your claimed 170Gb/s.
In other words, the comparison to a cheaper network of workstations is really quite a fair one.
--PeterM
Can you make a Hackintosh out of it is the better question!
Single OS instance? I don't suppose you would consider TFTP/NFS to be solutions to this (the standard for IBM's BG and for Beowulf clusters)? I sure do - although I don't think Redmond's products will work quite that way I know that *NIX operating systems do. Think of each node as a fat client (although I still use them as thin clients, not needing the local SSD disks for anything more than swap space).
Form factor doesn't intrigue me. If that's important, I recommend you build your own case with lots of pretty transparent covers and neon lights/LED's. I'm sure you could come up with something straight out of an Irwin Allen show; I'll settle for nailing a bunch of power-strips to a hunk of plywood, personally. Don't forget - 10W per active node. Okay, at 1,000 nodes I'll be drawing 10KW; but for now, I'm running a LOT less than the 200W your average desktop box draws and getting lots more FLOPS per watt.
...none of them are.
Seems like this still wouldn't be the best solution even for what SGI used to be commonly used for--CG. If you are running 3ds Max (or some other rendering software) it seems you would still be better off with a dozen cheap, independent rendering nodes than one of these things decked out. This is because the nodes do not need to communicate with each other (only between the slave and master nodes, and that's only at the start of rendering each frame and at the end of rendering a frame) so there is no need for a high-speed connection between all of the nodes or to run it under a single OS environment.
If 1000TX had anywhere near the latency of Infiniband, maybe. There's no doubt that clustering has its point, but you cannot possibly believe that individual nodes communicating across 1000TX (or 1000FX, for that matter) are going to compete with a chassis like this in this segment. No, I don't consider TFTP/NFS/PXE booting to be solutions in this manner, since nodes in a cluster are not the same thing as a single image, and they have no access to peripherals which may be in the chassis (if one goes with the PCIe expansions).
The form factor matters not because of aesthetics, but because fast interconnects and redundant components matter for some applications. It is not a competitor to desktops.
"The more corrupt a society, the more numerous are its laws." -Tacticus
Excuse me, but...this doesn't compare to a single ATI Radeon or Nvidia GPU card. How is this super?
Even if you're talking DP FLOPS, I think GPU computing has got to be more cost effective than this.
"It's the height of ridiculousness to say for those 9 lines you get hundreds of millions."
Not sure why you would say that unless you aren't aware that PS3s are being used to make HPC clusters for scientific computing:
http://www.top500.org/blog/2007/10/27/rough_guide_scientific_computing_playstation_3
Wouldn't an NVIDIA Tesla based system give you a lot more horsepower for a lot less money, and a lot lower power consumption?
Donte Alistair Anderson Roberts - hi son!
Karma: Chameleon
Killer Access Point? Is that like a wifi Killer NIC?
Botnets are useless for number crunching. There may be many CPUs involved but the communication between processors is dead slow and unreliable. The ability of botnets to send lots of traffic from a huge number of Internet connections in different locations makes them ideal for spamming and DDoSing.
Supercomputers have few or no Internet connections and have no more potential than office PCs for spamming or DDoSing. They do have many powerful CPUs that have no problem communicating with each other as fast as they can, which makes them ideal for number crunching. The use of a supercomputer to a black hat would be for breaking encryption (or they could be useful to comic book villains for simulating nuclear explosions), but a rack of PS3s might be faster than a botnet, and definitely more difficult to trace.
"When information is power, privacy is freedom" - Jah-Wren Ryel
But... when I bought my iMac G4 Steve Jobs told me it was a supercomputer!
It Is the Nature of Information to Transgress Artificial Boundaries
For my budget, yes, I need to make quite a few concessions. Gig copper comes with the plugs and it's as fast as I'm likely to be able to afford; but I can still handily outrun SGI's offering in the 'bang for buck' category clear up until I hit around a hundred, maybe even a couple hundred nodes.
My old uni did just that. The down time with 128 PCs was pretty high. Sure you could use it we less "cores" but... then its not a 128 core machine then is it.... The real killer was paying sometime just to get the broken down ones fixed.
We went with blades next. Up front costs was more. After 2 years it was cheaper over all than the cheapo version.
The Grey Goo disaster happened 3 billion years ago. This rock is covered in self replicating machines!
just imagine how fast compiles will be done on this system.....
make -j80 ...
Linux forever
Ok, redundancy and high availability is an issue. But isn't it still a lot cheaper to just duplicate the entire box (or hell, 3 commodity boxes are probably still a lot cheaper than one of these...) and use software which can fail over gracefully?
Proud member of the Ferengi Socialist Party.
I had difficulty finding pricing quickly enough to make a simple slashdot post about it. The pricing is non-obvious.
“Common sense is not so common.” — Voltaire
I just priced out hardware for a 64 core MPI parallel computer system.
The commodity hardware came in at ~ $7800.00.
The SGI base system only comes with one [*ONE!!!*] processor at
near the same price....
They must be offering one hell of a warranty to ask that price
for essentially a high end workstation. (A nice colorful SGI case maybe.)
As far as 'Easy' goes; Well, most of the apps that use supercomputing
these days require parallel processing. Maybe the parallel API they are
using is something special. If the software is provided for free
it could be worth it. Probably not....
960GB of RAM should be enough for anybody!
I am anarch of all I survey.
Except that a gigabit network would be too slow for the sorts of workloads you'd normally use a system like this for, so you want 40+ infiniband links. Each Infiniband HCA is about $650, plus the switches: 2x 24-port switches at $3500 each. That adds another $20,000. Of course, that's neglecting all the fiber cable you need for the actual connections. For reference, the quick google search price results: HCA (card) Switch
Not a sentence!
First of all, this is a shared memory system, meaning it isn't really comparable to a cluster-supercomputer using distributed memory. Building a big shared memory system isn't usually something done by hobbyists or regular sysadmins.
If you need a shared memory system, you need a shared memory system and no amount of cluster nodes can do the job for you.
Instead this is more like an entry, low end, version of some of SGI's other offerings such as the Altix 4700 which supports up to 1024 cores in a shared memory system.
That you can get an 80 core shared-memory system with 240GB of memory for $53k is news to me, but then I haven't been watching the prices of these systems lately. Also the price point is certainly attractive enough that it may make sense to buy one of these rather than convert your shared-memory code to MPI, even if such a transition is possible for your purpose. Given complex enough algorithms the transition could easily cost you much more than that in manpower.
Actually, I just realised my posts probably reads a bit like SGI astroturfing....
will they give you proper tech support in a timely manor?
I don't know about a timely manor, but all my tech support comes from a stately manor, where we can spend a few hours hunting quail before retiring to sip brandy and admire the house's portraits.
... and then they built the supercollider.
Well, not specifically a genuine Cray 1. However, the cheap-as-chips generic Windows PC that I bought at the local electrical warehouse pisses all over the performance of that archetypal 'Supercomputer'. All the figures trotted out these days for the Top 500 are just so much willy-waving. Ridiculous.
Squirrel!
If a botnet relied on a consistent connection to any kind of single centralized server, it would run for just a few hours (maybe longer, depending on where exactly it's located) before the IP was blocked and a SWAT team is dispatched to the server's location. Folding@Home wouldn't work very well if ISPs and private firewalls were trying to block it, virus scanners were trying to remove it, and the server was only online for short periods from differing locations. It might sort-of work if some technical problems can be overcome (and depending on exactly what kind of calculations are being done), but it wouldn't work well.
"When information is power, privacy is freedom" - Jah-Wren Ryel
I realize Gb network would be too slow.
But would commodity hardware have an expansion slot quick enough to take advantage of Infiniband?
I don't know much about it, but I assume you'd need at least a PCIe 16x slot, and even that would probably be too slow.
Considering a lot of the cheaper commodity hardware (I'm looking at you, Compaq!!) doesn't have anything other than 32-bit PCI expansion slots, or maybe a PCIe 1x on their newer stuff, it's pretty much impossible to use commodity hardware to come in cheap enough for this, even if all the network hardware was free.
"City hall" in German is "Rathaus" Kinda explains a few things......
Although Octane III is a far cry from the MIPS/IRIX systems which the
product name harks back to (and I should know, I have 60 of them), it
is a decent design nonetheless, even though I'm sure the aesthetics
will not appeal to many. Strangely for me, it looks rather like my PC
from the front (Centurion Plus 534 case).
I recently had to spec out a 24-core renderfarm for a small design
company in Spain, consisting of three Dell rack servers, each dual
quad-core i7 XEON 2.93GHz, 32GB ECC RAM, etc., which came to
16000 UKP + tax (list price). From the information I have at the moment,
an Octane III with a similar spec (ie. six CPUs, 96GB RAM) would provide
nearly 50% better price/performance, which is very respectable indeed.
I'll see if I can obtain some pricing, find out whether this system
does compete to a decent degree. It absolutely does with 80 cores, but
a more mid-range spec will be of greater interest to the companies I
typically deal with.
Ian.
SGI Guru. Email: mapesdhs@yahoo.com | Tel: +44 (0)131 476 0796
Set it up to constantly crawl the web and download all the latest pr0n, then sort and categorize it all before finally prioritizing and assembling it into a nice presentation mode.
Specs suggest that it is just a cluster under your desk (and a hot one if I can notice). No fast interconnects - i.e. it's just another custom beowulf...
To slashdot - overload a? This comment will not be saved until you click the Submit button below. You must wait a little bit before using this resource; please try again later.