That's a peak speed, not a STREAM speed. Some of these machines (like the NEC SX-6) have peak speeds that are *much* higher. STREAM is an attempt at showing how a system performs on a somewhat more realistic workload.
Well, they did make the referential comparison in their press release, but yes, they just published the numbers. I also find it interesting that they claim to have set the world record for "microprocessor-based systems" when a T3E (which is a bunch of Alphas) has done quite a bit better on STREAM, at least according to a site referenced a few posts up...
The T90 is/was a pretty impressive machine, technologically - they really pushed some boundaries. Unfortunately, they may have pushed them a bit too far, since they ended up being fairly unreliable (at least as far as Cray's go).
The X1 should be coming very soon - according to some of my sources, I may not have very long to wait...I'm pretty sure they've got hardware already...
Additionally, results show that the upcoming Itanium 2-based SGI system has not
only outperformed the IBM eServer p690 and Sun Microsystems Sun FireTM 15K
high-end microprocessor-based systems, it has also surpassed memory bandwidth
performance on the CRAY C90TM, the CRAY SV1TM and the Fujitsu VPP5000 CMOS
vector-based supercomputers.
Interesting...Looks like a T932 has got about a 3x performance on it, and the NECs (understandably, since they are the most modern) get like 5x. Still pretty impressive for a MPP machine, I would think. Were you able to find stats on MPP systems (such as the T3E or SP) anywhere?
...interesting that SGI chose the Cray C90 - a system released in *1991* - to compare against. It's nice to know that it's only taken them 10+ years to catch up.:)
They also mention the SV1, which is a "low-end" Cray. I'm curious how the new X1 (nee SV2) does on the STREAM suite.
It's good to see that their "scalable linux" work seems to be doing pretty well! I'm sure it was much easier for them to use the IA-64 port of Linux than to port IRIX...
They *are* on the mass market now. Cray made an agreement with NEC to resell them in the US. In return, they dropped the dumping lawsuit stuff that was keeping them out.
I agree that Americans need a more heterogeneous set of supercomputers these days. Vector computing has "gone out of style", but it's still very very useful for a lot of applications. We may see vectors return somewhat with this reselling plan, and with the soon-to-be-released Cray SV2.
you're right about the first one. For the second one, imagine gripping a steering wheel. You may also know it as: "Calling Ralph on the big white phone"
what I'm sure he ment to say was for SETI, where I seem to recal that MIPS chips blew away anything intel. At least last time I looked (which was a while ago)
The SV1 was built as an upgrade path for the J90 users, so they were somewhat compatable as far as board swapping, etc. goes.
Now keep in mind that the J90/SV1 is Cray's "budget" line...The SV2 (due out next year) is supposed to be a successor to both the T90 AND the T3E (its both vector and Mass Parallel).
I'm curious to see what happens with the Tera multithreading systems as well. The first few years I imagine they will just be bought as computing research machines. (so that people can see what they do)
I don't know about you, but I'm reading/. right now, not \. (backslashdot?) :)
And, while I'm at it - I agree. There are jerks who ruin it for others, but the whole peer review of work thing makes forged results a really dangerous career move.
1) It's called the NSF. A very very big general fund. (The article says that 92% of research is still gov't funded...) You're only supposed to use it on research needs, either materials, or for paying grad students to help you out (but not professors)
2) They don't do royalties. Instead, your donation back to them is this: They get rights to use the technology you developed for free and forever.
The royalties idea is an interesting one - research perpetuating research. Unfortunately, because of the huge scope they deal with, *enforcement* of the royalties would cost more than you'd make because of legal fees I think...
I have to agree with you. A lot of people benefit from this. Besides, without the Bayh-Dole Act, how would any of this government-funded research (still 92%) get put to market? The Government would own all the IP.
The real benefit of this act is that it gives the researchers the ability to make some money off of their own ideas. Better them then some suit who has no clue whats going on, right? The professor I work for (I'm a graduate student) has made a fair hunk of cash off of a research project that he did as a graduate student. I've seen how much work he put into it, and along with how much it has helped out the research community, he more than deserves it.
There are several other professors in our department that have taken their ideas to the corporate world as well, most of them successfully. One of them (ArborNetworks) was recenty featured here.
The problem is not with the Bayh-Dole act, but with the ridiculous deals like the one at Berkeley mentioned in the article. They're seriously abusing the system. The goal of the act was to allow businesses to emerge out of the research community. The goal is not to have the researchers bought and directed by a company, when they don't even know what's going to come out the other end of the research!
If you want research lackeys to do what you want and to give up all their IP rights to you, hire your own.
There's no need to go calling names. I'm sorry that my humor is too subtle for you. In fact, the guy at the desk next to me is doing research on fast software decoders of instruction sets. And between being a CS grad student and having a math minor, I most certainly know what a DAG is. And how to use it.
My point was that the reporter treated the usage of "Directed Acyclic Graph" as if the Transitive guy said "Our patented UberCool(tm) Technology".
...the subject program in the form of what Transitive calls "directed acyclic graphs."
Wow! What innovative technology! I wonder when they will patent this so-called "directed acyclic graphs". And they picked such a cool name! It sounds so mathematical!
Okay, enough laughing at the expense of clueless reporters...
Even while under SGI, the hard-core CRAY lines kept the name (like the T3E, T90, and SV1).
Now that they've been "un-bought", Cray gets to put its name back on the list as an independent company.
As a side note, SGI sold the Cray division because it was "unprofitible" and a fiscal liability. Yet, Cray Inc. made a profit last quarter, and SGI has lost about $2/share for the last several quarters in a row, and just layed off another 1/3 of their workforce. Oh, and Cray's stock price is higher. Go fig.:)
It's a totally different environment. This runs on commodity hardware with network interfaces (in this case, Myrinet), whereas Unicos/mk runs on the mostly-custom (except for processor and memory) Cray T3E. The inter-node interface for the T3E is totally different - it uses E-registers, which essentially allow a machine to do all of its communication through loads and stores to memory instead of having to context switch and send network packets. So, the two are like apples and oranges - not very comparable because of their environments. To the first order, unless you've got some pretty dull software people, the scalability of a system is 99% hardware dependent.
That being said, as far as this hardware - a commodity cluster system - goes, it seems that this is a pretty decent set of tools and optimizations.
And, I quote:
"This is related to the software used in their ASCI Red supercomputer, and eliminates several scalability problems to allow hundreds of nodes for algorithms which can't be parallelized for Beowulf-type clusters."
This is a pretty big over-statement. From exploring their site, it seems pretty clear that, while they made a few scalability enhancements (like cutting out the TCP/IP stuff, etc), they're main goal was to make large commodity cluster systems (Beowulf or not) more usable. They made a lot of good progress in this area by porting over several tools from their learning experience with ASCI Red.
I also found it funny that their "commodity machine" had a custom-made myrinet switch. I think it must be hard to resist the "if we don't have it, we'll build it" mentality of a National Lab. Very cool.
Oh, and I'm not sure when the source was put up, but from what I can tell, the site hasn't been updated in almost a year.
For pure whack-job value, I'm definitely partial to the TimeCube (which is discussed in one of the questions). I would love to see his face when someone demonstrates that he's wrong. I mean, come on, everyone knows time is a dodecahedron...
Slashdot Mirror
The copyright is on the photo. It's a similar thing to any other photo credit - i.e. the AP copyrights it's news photgraphs.
That's a peak speed, not a STREAM speed. Some of these machines (like the NEC SX-6) have peak speeds that are *much* higher. STREAM is an attempt at showing how a system performs on a somewhat more realistic workload.
The T90 is/was a pretty impressive machine, technologically - they really pushed some boundaries. Unfortunately, they may have pushed them a bit too far, since they ended up being fairly unreliable (at least as far as Cray's go).
The X1 should be coming very soon - according to some of my sources, I may not have very long to wait...I'm pretty sure they've got hardware already...
Additionally, results show that the upcoming Itanium 2-based SGI system has not only outperformed the IBM eServer p690 and Sun Microsystems Sun FireTM 15K high-end microprocessor-based systems, it has also surpassed memory bandwidth performance on the CRAY C90TM, the CRAY SV1TM and the Fujitsu VPP5000 CMOS vector-based supercomputers.
Interesting...Looks like a T932 has got about a 3x performance on it, and the NECs (understandably, since they are the most modern) get like 5x. Still pretty impressive for a MPP machine, I would think. Were you able to find stats on MPP systems (such as the T3E or SP) anywhere?
They also mention the SV1, which is a "low-end" Cray. I'm curious how the new X1 (nee SV2) does on the STREAM suite.
It's good to see that their "scalable linux" work seems to be doing pretty well! I'm sure it was much easier for them to use the IA-64 port of Linux than to port IRIX...
I agree that Americans need a more heterogeneous set of supercomputers these days. Vector computing has "gone out of style", but it's still very very useful for a lot of applications. We may see vectors return somewhat with this reselling plan, and with the soon-to-be-released Cray SV2.
"Kazaa, Verizon Propose Compulsory Music Listening"
I got an image of being *forced* to listen to whatever music I download...
you're right about the first one. For the second one, imagine gripping a steering wheel. You may also know it as: "Calling Ralph on the big white phone"
what I'm sure he ment to say was for SETI, where I seem to recal that MIPS chips blew away anything intel. At least last time I looked (which was a while ago)
(its in the back - kinda bad picture)
Cray2 - sometimes called "the world's most expensive fish tank".
T90
Now keep in mind that the J90/SV1 is Cray's "budget" line...The SV2 (due out next year) is supposed to be a successor to both the T90 AND the T3E (its both vector and Mass Parallel).
I'm curious to see what happens with the Tera multithreading systems as well. The first few years I imagine they will just be bought as computing research machines. (so that people can see what they do)
Hate to break it to you...but the SV1 *is* the first Cray to have a cache. Specially designed, of course. :)
And, while I'm at it - I agree. There are jerks who ruin it for others, but the whole peer review of work thing makes forged results a really dangerous career move.
2) They don't do royalties. Instead, your donation back to them is this: They get rights to use the technology you developed for free and forever.
The royalties idea is an interesting one - research perpetuating research. Unfortunately, because of the huge scope they deal with, *enforcement* of the royalties would cost more than you'd make because of legal fees I think...
The real benefit of this act is that it gives the researchers the ability to make some money off of their own ideas. Better them then some suit who has no clue whats going on, right? The professor I work for (I'm a graduate student) has made a fair hunk of cash off of a research project that he did as a graduate student. I've seen how much work he put into it, and along with how much it has helped out the research community, he more than deserves it.
There are several other professors in our department that have taken their ideas to the corporate world as well, most of them successfully. One of them (ArborNetworks) was recenty featured here.
The problem is not with the Bayh-Dole act, but with the ridiculous deals like the one at Berkeley mentioned in the article. They're seriously abusing the system. The goal of the act was to allow businesses to emerge out of the research community. The goal is not to have the researchers bought and directed by a company, when they don't even know what's going to come out the other end of the research!
If you want research lackeys to do what you want and to give up all their IP rights to you, hire your own.
My point was that the reporter treated the usage of "Directed Acyclic Graph" as if the Transitive guy said "Our patented UberCool(tm) Technology".
Laughter is fun. You should try it sometime.
I got a 9.0 in my Microarchitecture class last semeseter...go grad school...
Wow! What innovative technology! I wonder when they will patent this so-called "directed acyclic graphs". And they picked such a cool name! It sounds so mathematical!
Okay, enough laughing at the expense of clueless reporters...
Now that they've been "un-bought", Cray gets to put its name back on the list as an independent company.
As a side note, SGI sold the Cray division because it was "unprofitible" and a fiscal liability. Yet, Cray Inc. made a profit last quarter, and SGI has lost about $2/share for the last several quarters in a row, and just layed off another 1/3 of their workforce. Oh, and Cray's stock price is higher. Go fig. :)
That being said, as far as this hardware - a commodity cluster system - goes, it seems that this is a pretty decent set of tools and optimizations.
And, I quote: "This is related to the software used in their ASCI Red supercomputer, and eliminates several scalability problems to allow hundreds of nodes for algorithms which can't be parallelized for Beowulf-type clusters." This is a pretty big over-statement. From exploring their site, it seems pretty clear that, while they made a few scalability enhancements (like cutting out the TCP/IP stuff, etc), they're main goal was to make large commodity cluster systems (Beowulf or not) more usable. They made a lot of good progress in this area by porting over several tools from their learning experience with ASCI Red. I also found it funny that their "commodity machine" had a custom-made myrinet switch. I think it must be hard to resist the "if we don't have it, we'll build it" mentality of a National Lab. Very cool. Oh, and I'm not sure when the source was put up, but from what I can tell, the site hasn't been updated in almost a year.
For pure whack-job value, I'm definitely partial to the TimeCube (which is discussed in one of the questions). I would love to see his face when someone demonstrates that he's wrong. I mean, come on, everyone knows time is a dodecahedron...
I definitely need one of These. Don't even have to make any big mods! Probably could be used to heat the house as well...