PNNL supercomputer fastest open system in U.S.

PNNL Press release ^ | 8/26/2003 | PNNL

[justlurking]

Photos of the system:

http://www.pnl.gov/news/2003/03-33_photos.htm

[Mrs Zip]

ping

[TechJunkYard]

Heh. Some of those pictures remind me of an old AS/400 F95 that I used to "own".

[biblewonk]

I wonder how much power it takes to run it.

[Coral Snake]

The Coral Snake Ping.

Calling all Anti Commies and Anti Pirates

The government should have done this when

More open to the world idiocy from the folks of the DOA and the "Free Software" movement!!!

[shadowman99]

Linux users - Ignore the trolls. They are baiting you.

Rockin' on without Microsoft
			Posted by Coral Snake to Bush2000; Golden Eagle; TheEngineer On News/Activism 08/22/2003 1:22 AM CDT #278 of 346 I have been studying all the tech threads that we have been involved with since I started posting here and have determined that there is more than flame wars going on against us. These Linux trolls have now gone NUCLEAR and are now using the moderators against us in an attempt to get us banned on Free Republic. Their stratagy seems to be to use either FR posting policy banned profanity or personal attacks against us and getting us to respond with our own garbage so they can sic the moderators on US. I am proposing that we now go nuclear also. Rather than responding to any further profanity or personal attacks by these Linux blowhards WE should be the first one to the ReportAbuse button when they occur, in effect either shutting them up or getting THEM banned on FR instead. This message is being sent to The Members of the Coral Snake Ping list ONLY and should not be repeated over regular FR channels. The Linux troll justlurking made a BIG mistake when he threatened that we would be history on FR. In war you DO NOT blab your stratagy to the planet. Because he did I can now propose this identical preemtive counter stratagy against THEM. Post Reply | Private Reply | To 261 | View Replies

[Coral Snake]

Sorry Penguin BOTS:

That stratagy has been abandoned for a new one that I gave the members of my ping list through FReep Mail as this one was intended to be given. You only posted some DEAD intellegence, Typical.

[Coral Snake]

Sorry Penguin BOTS:

That stratagy has been abandoned for a new one that I gave the members of my ping list through FReep Mail as this one was intended to be given. You only posted some DEAD intellegence, Typical.

[stainlessbanner]

ping em

[kevkrom]

Wow... you guys need a "strategy" for posting your anti-Linux diatribes? Silly me, I just offer my opinions when I think they're relevant (or amusing)...

[stainlessbanner]

It's a "stratagy", kevkrom.

[Liberal Classic]

Interesting post. We are due to receive our Itanium2 nodes soon. Of course, we're not even getting a 20th of this cluster. Out of the three vendors we compared, HP, IBM, and SGI, Hewlett-Packard had the best Itanium2 nodes. This isn't to say that the SGI and IBM nodes were bad, just that the HP ones seemed to have the best memory performance. HP has had a lot of experience with Merced and Itanium1, and it shows.

[stainlessbanner]

At the William R. Wiley Environmental Molecular Sciences Laboratory, this new supercomputer operates 9,200 times faster than a PC.

[Coral Snake]

No we don't need a stratagy to post against Linux or the GPL. The whole anarchistic "herding cats" development model behind it and the GPL having a communistic "poison pill" provision to protect it when BSD does not more than speak for themselves. The stratagy is only for dealing with penguin bot trolls who attack our exposes.

[rdb3]

The Penguin Ping.

Wanna be Penguified? Just holla!

Got root?

[justlurking]

This isn't to say that the SGI and IBM nodes were bad, just that the HP ones seemed to have the best memory performance.

That's because HP acquired Convex back in 1995. The former Convex engineers showed them how to design a scalable memory system.

[SteveH]

I think this article de-emphasizes the potential drawbacks of the PNNL approach. For example, what overall good is a supercomputer which can be programmed only for the sciences mentioned in the article? A better approach might be a more generic application target, especially considering the expense involved.

What is the expense? That is not explicit.

What in particular characterizes this supercomputer versus others? Hint: it uses 64 bit processors... very nice, very sexy, but relatively expensive to procure and maintain. Why not P-4s with hyperthreading? It costs a lot of money to be on the bleeding edge...

It's interesting that the Quadrics interconnect (in this case) might not be "fast enough" to keep up with the output of the individual processor nodes...

[justlurking]

There's not a lot of technical information in this article, but it's probably safe to say that the cluster is at least partly relying on PVM or MPI to distribute the application. It's enormously expensive to build a large parallel machine with a single shared memory, so techniques have been developed to partition applications into pieces that can be processed independently without significant interconnect bandwidth. There are a lot of applications that can be distributed in this manner. Environmental and molecular science is just a subset of them.

The new 64-bit processors (Itanium and AMD's Hammer) have two important features that can't be duplicated with the P4: more registers and a larger addressable virtual memory space. The former results in faster code (presuming a compiler that knows how to use it), because more data can be retained in registers without accessing memory and/or thrashing the cache. The latter makes it easier (and faster) to tackle large problems without bumping up against the process virtual memory limit in the P4 (and other 32-bit processors), requiring creative techniques to manage the data.

[Redcloak]

I wonder how much power it takes to run it.

No problemo. They've got their own nuclear reactors just up the road at Hanford.