Slashdot Mirror
96 Processors Under Your Desktop
Roland Piquepaille writes "A small Santa Clara-based company, Orion Multisystems, today unveils a new concept in computing, 'cluster workstations.' In October, you'll be able to choose between a 12-processor unit for less than $10,000 or a 96-processor system for less than $100,000. These new systems are powered by Efficeon processor from Transmeta and are running Fedora Linux version 2.6.6. Apparently, this new company has friends in the industry. You already can read articles in CNET News.com ("A renaissance for the workstation?"), the New York Times ("A PC That Packs Real Power, and All Just for Me," free registration, permanent link) and the Wall Street Journal ("Orion Sees Gold in Moribund Workstations," paid registration). The company is targeting engineers, life scientists and movie animators. It's too early to know if the company can be successful, but I would certainly have to get one of these systems under my desk. In this overview, I've picked the essential details from the three stories mentioned above."
Any ideas?
I thought '96 processors under your desktop! That would be the Pentium at 133MHz!
Seriously, why 96? Why not 64 or 128?
-
If you keep throwing chairs, one day you'll break windows....
One thing perplexes me:
Chips on the same board communicate using Gigabit Ethernet, while board-to-board communication takes place on 10 Gigabit Ethernet.
Wouldn't same board communication be more frequent, hence needing the faster connection?
Better yet, why not 10GBe for both?
I really admire this guy; although the ventures he took part in haven't gone anywhere financially, they were pretty cool. Transmeta, OQO, and now this! Go Colin Hunter!
96 processors running Fedora? I want one!
Actually, I would be willing to bet that the university I'm at could use a few of these things. After all, we've got undergrads doing BLAST database work, just to teach them about it. Having been through that hell myself, it'd be a lot easier if you didn't have to have a cluster to do the work by computer. For those who don't know, BLAST is a genetic sequencing database that allows for comparison with an extracted gene (retrived through polymerase chain reaction) with a known, sequenced gene in their database.
Haec merda tauri est. Ceterum censeo Carthaginem esse delendam.
So I think they know the difference at least...
UNIX? They're not even circumcised! Savages!
Now we just need to see a company come up with a complete LAN solution where every machine acts as a SETI@Home type client in a cluster, giving SMB's a supercomputer that works while it's various workstations comprising it are idling.
Personally, I think most of these will still end up as servers for groups of people instead of individual "workstations." But the logistics of a normal 100-workstation cluster are pretty bad - a large server room, enormous air conditioning unit, a massive power supply, and lots of cabling to be done. This new thing can probably share an existing server room with other computers.
Granted, it's probably just a bit smaller and more power efficient than previous "blade" servers, but maybe presenting it as something brand new is a good marketing angle.
How many stories a week does this frigging guy get into Slashdot?
His business plan:
1) Sell Ad Space on "News" Website
2) Shovel In Content From Online Articles
3) Submit To Slashdot Daily
4) Tout "Slashdot Coverage" To Advertises
5) Profit!
And looking at his site, it works fine and dandy indeed.
Then again, is he just doing a service to us?
Monolithic systems usually out-perform clusters intended for similar workloads. However, this isn't intended as a normal workstation, it's a cluster architecture shrunk to fit under your desk.
This may find a home under an artist's table for quick rendering. However, I think it'll be more useful as a prototyping tool for the people who write the applications that run on top of clusters.
tasks(723) drafts(105) languages(484) examples(29106)
First, I see that Transmeta are finally making 1.5GHz Efficeons, which is a good sign, they looked to be stuck at 1GHz for so long.
This merely looks like 12 computers on a single motherboard with a GigE switch connecting them together. Each computer is highly optimised of course, just a processor, memory, support chipset (GigE, IDE).
I do have to wonder how it compares with something similar made with Opterons or Pentium-Ms. Opteron has the advantage of being able to do SMP so the per-system processing power would be much higher, each board could have 4 low power 2GHz Opterons which will probably be close to the 12 Efficeons in terms of computing power and power consumption.
But still, this is a cool system. I wonder how fast it can do a kernel compile?
"why not get a huge server..."
Reading the article, the answer appears to be: politics. eg this from IDC: "There are probably plenty of engineers in the world who would love to have their own cluster so they don't have to wait for the machines in the lab"
you what now? If its about lack of compute power on the network - usually something your project/dept contributes money to - then this comment can only mean those people who have enough money in their budget to go it alone. Most likely these people won't want sysadmins or support contracts either - those things tend to get taken care of only when there's economies of scale.
I've seen this kind of thing happen, its a disaster. 2 years later they'll be whining to central computing that they're desktops arent being replaced when they're no longer contributing to the budget for them. That being said, it seems to happen often enough that Orion could be on a winner.
1-15 submissions.
seriously, if you try enough you can get almost anything on slashdot.
world was created 5 seconds before this post as it is.
Or lack there of. I was imagining one of those SGI deskside Onyx servers when I read the post but these are just wide full towers that pack 96 processors. Quite nice. I am sure the movie industry is all over these babys. The 12 unit is around the size of a Sun pizza box.
How do the Transmeta CPU's do in fp computations? That is obviously the metric to note. I wonder how long it would take to render a movie? Is the USB USB 2? No firewire though.
Get quad opteron. It should get about better performance for same price as the other but without need for clustering, for the small system. For bigger system, you could use myrinet and dual opterons. Oh what the heck. It costs 600-800$ per processor to build a rack of Athlon64 based cluster with Gb ethernet. So this effineon based cluster would be beaten with system costing less than their solution. So it is beaten from two different solutions.
A) Getting single image opteron system if communicationlatencies are important.
B) Getting cluster of AMD64 if price/performance was important.
ONLY thing they bring is density of A for system type of B while costing more than A.
Emacs is good operating system, but it has one flaw: Its text editor could be better.
Probably not, actually. Most programs being assembled with make never have 96 possible concurrent builds... it's usually a couple dozen at most, and quite often just a handful. Also remember that almost everything depends on something else that must be built first. Building Gentoo is actually quite linear. So for the major components of Gentoo, such as glibc and your basic windowing system libraries, you probably wouldn't notice a massive increase in speed. You might notice some benefit when compiling many userspace apps at once, but for the most part, you'd be better off with a few fast CPUs in something like a 4-way Opteron system.
He who laughs last is stuck in a time dilation bubble.
These machines are not designed to do the type of work that multiple people can benifit from at the same time. They are, generically, personal super computers. My degree is in BioChem, and a room full of these vs 1 or 2 supercomputers would have been a godsend. I can't imagine what I could have gotten done if I didn't have to schedule SC time. Being able to just walk into a room of these and sit down would have been amazing.
I browse on +1 so AC's need not respond, I won't see it.
In only a couple of months' time, you'll be able to put 8 Opteron cores (4 dual core CPUs) in a desktop-size case - and this is a rather reliable information. It is also very likely that similarly sized boxes with 8 CPU sockets (and thus possibly 16 cores) will appear next year: infoworld.com article.
"I love my job, but I hate talking to people like you" (Freddie Mercury)
http://www.tomshardware.com/hardnews/20040830_0920 16.html
It says:
"According to Hunter, the Efficeon architecture allows Orion to reach a performance of one flop per Watt - more than would be possible with any competing processor."
I'm familiar with megaflops and gigaflops and teraflops and petaflops, but what is so magical about "one floating point operation per watt"? Is this just a misquote, or does it mean something?
.. pa-ra-bo-la, pa-ra-bo-la, 2 pi R, 2 pi R, where's your latus rectum, where's your latus rectum, 2 pi R
While it's certainly cheaper to pickup 20 Dell PC's for $500 each, an integrated 12-way workstation may signify the beginning of a new desktop computing standard.
I'm certainly not disagreeing with you, but my point to the original poster is that he doesn't need to worry about this in the here and now.
Personally, I don't see this sort of design becoming standard in a Personal Computer any time soon. Too much horsepower for a single user who simply needs more torque. However, I *do* see such designs leading into concepts like a "house computer" where the ability to multitask is more important than raw performance. Just imagine if you could install one computer for ~$2000, and have enough system resources to provide a desktop to a small office building (not to mention your entire extended family).
Such a computer would not only provide a thin client desktop, but also handle multimedia capabilities like PVR, watching movies/TV from the internet, streaming radio stations and purchased music to anywhere in the house, interfacing with digital cameras/camcorders via Bluetooth, etc. It's even possible that such a machine could control aspects of your home via X.10, but I wouldn't count on that being a common use for quite a long time.
Javascript + Nintendo DSi = DSiCade
Ummm...no, not very wrong. One dual-xeon server is ~$2.5K, four dual-xeon servers is ~$10K.
Bigger SMP machines, ie 4-ways, 8-ways, 20-ways, 106-ways cost considerably more. I know that, I have a 20-CPUs Sunfire 10feet behind me. But those aren't the computers I was talking about.
The Orion isn't a 12-way machine. It's 12 different machines in one chassis. I was comparing it to the normal equivalent in the clustering world. Which these days are typically dual-cpu compute nodes.
Also keep in mind that the Efficeon machine is only a cluster, hence it works well for (like you pointed out) heavy-duty algorithms that are easy to compute on clusters. :-)
Some algorithms cannot be computed efficiently on clusters; for those, you rather need a single-image supercomputer, such as the SGI Altix. Unfortunately, many of the examples you provided fall into this category.