The goal is to not be reachable when that so-called electrician (whose visit they scheduled without telling you) working with a soldering iron in the machine room (can you tell this is from experience?) somehow manages to make the full-room UPS go all "tango uniform."
Okay, okay, maybe you can be available for that. But since being available means you'll alsoget the inevitable calls about all kinds of trivial problems that are beneath your station (users forgetting to turn on their monitors, etc.), you've gotta have a pretty major masochistic streak to want this.
A quick glance at the iTunes Music Store reveals that they do have the complete album Come Poop With Me by Triumph The Insult Comic Dog, including the track Together in Pooping. If that's not quite to your liking, you could get one of these songs to express your post-fast-food sentiments:
NOFX - Brain Constipation
Tommy Lee - Mr. Shitty
Warren Zevon - My Shit's Fucked Up
DMX - Keep Your Shit the Hardest
Frumpies - I Just Wanna Puke on the Stereo
The Tony Rich Project - My Stomach Hurts
The Locust - Identity Exchange Program Rectum Return Policy
I have a Power Mac G5 2x2.0GHz here, and this evening I compared XBench numbers on it under Jaguar (10.2.8 G5, build 6S90) and Panther (10.3 build 7B85). The overall score in XBench went from around 180 to around 210, a 16-17% increase in benchmark performance, but some of the subtests showed more significant changes.
The XBench CPU Test score went from 148.72 to 193.29. There was a slight decline in the "Floating Point Basic" category, but performance in "AltiVec Basic" and "vecLib FFT" improved by over 50% and "Floating Point Library" performance also improved by over 20%.
The XBench Thread Test score went from 185.93 to 209.27, with most of that accounted for by an 18% gain in the "Computation" subtest.
The XBench Memory Test score went from 293.70 to 312.41, gaining primarily in the System (vs. Stream) memory subtests, particularly "Allocate" which went up almost 40%. (On my iBook G3-600, Panther improved "Allocate" scores 304%!)
So if my machine - roughly equivalent to a single node of VT's cluster (theirs have more RAM; mine has more disk) - can get a 30% boost on the CPU test, a 12%+ boost on the thread test, and a 6% boost on the memory test, it looks like the planned upgrade to Panther mentioned in a previous article might help it get past the 10-TeraFLOP mark.
(Hypothetically speaking, if VT's code for LINPACK made extensive use of the AltiVec and vecLib bits included in the OS, going to Panther could boost things up into the 12-14 TeraFLOP range. However, I believe they're probably using custom-written libraries built with optimizing compilers, so I don't think the difference will be that profound.)
Wired's target audience isn't nerds, it's nerd wannabe's. A half-dozen years ago I was working in web design with a cow-orker who managed to read that day-glo typesetter's nightmare, and I had to keep telling her that 'twas better to spend her time doing stuff that would make her article-fodder, rather than article-reader.
Well... hmm. The "regular" Mac sitting next to my desk running Panther is a dual G5, so... I'd say there are plenty of similarities. There are also lots of differences in configuration. VT went for more RAM in their nodes than I have on my Mac, while I maxed out disk space and went for the higher-end video card. I've also got a mouse, keyboard and display; clearly they don't want 1100 of each of those taking up space. And I've got Bluetooth and 802.11g, which probably don't make as much sense in a huge cluster. (Like, what're you gonna do, sync 1100 machines to your cell phone at once?)
It's good the company takes privacy seriously, because they're launching a networked application that thus far only runs on an OS that isn't known for keeping private information private.:)
There are other categories of things that permit running Windows on Linux or MacOS X boxes - system partitioners like VMware and Plex86, and emulators like VirtualPC and Bochs X86.
Generally, I try to set things up so the Windows instance doesn't have any ports open to the world, and if at all possible, its "filesystem" is within a file in the real filesystem, so it can't trash anything but itself.:)
Yeah, that's a pretty odd term. "Digital SLR" is better, since the sensor usually does not measure 35mm. The exceptions to that are Canon's 11-megapixel EOS-1Ds which costs about $8,000, and Kodak's 14-megapixel model which costs about $5,000 (but had a lot of problems and delays and generally gets worse reviews than the Canon).
Small market share is a common argument, but it's a red herring.
Compare Apache's webserver market share to that of Microsoft IIS. Compare the number of exploitable vulnerabilities in those products, and the severity of the results.
Compare Sendmail's SMTP server market share to that of Microsoft Exchange. Compare the number of exploitable vulnerabilities in those products, and the severity of the results.
Compare Oracle's (or IBM's) SQL RDBMS market share to that of Microsoft SQL Server. Compare the number of exploitable vulnerabilities in those products, and the severity of the results.
Deduction: Microsoft manages to lead in introducing exploitable vulnerabilities to market segments, with severe results, even in segments where they do not enjoy market share leadership.
Now that's innovation!:)
To be blunt and honest, Microsoft designed and maintained its operating system product(s) in ways that failed to take security (and multiple users, and networking, and...) into consideration for far too long, and now finds itself in the unenviable position of being the only operating system vendor most people have even heard of that doesn't have a properly secure operating system.
-Dan (whose new "cheesegrater" G5 has fewer holes than Windows)
I think we should be more concerned with bringing up the audiovideo quality of videoconferencing.
Well, that's all well and good, but even though there's one-to-one Internet video conferencing technology good enough that TechWeb thought it was a satellite feed when they saw it last week, it still needs extensions to support one-to-many, many-to-one and many-to-many conferencing.
And even then you'd still have the onerous task of convincing everyone to useiChat AV. After all, it's an Apple product, and everyone knows that Windows users aren't going to wanttouse an Apple product!
I'm pretty sure I've seen scenarios where a SAN was hooked to two different computers or whatever, which could conceivably happen over FireWire (though Fibrechannel might be more common).
The catches are:
The thing that's being hooked to both other things being smart enough to realize that it's hooked to two things instead of one, and not get confused.
Actually doing whatever it is you want to do (which starts with knowing what you want to do).
There are probably computers out there with enough horsepower (stop looking at me like that, you may not borrow the dual G5, my precioussss...) to take 2 typical little video streams from webcams and do something with them in real time.
What something is remains unknown, but for example you could slap the images side-by-side, or you could composite one over the other, or you could compare them for differences, or insert 1 frame of stream 2 into stream 1 every 30 frames for subliminal purposes, or whatever.
Reasoning? Was I reasoning? Here I thought I was just stating mathematical facts.:)
If most/all the machines have multiply-add in a single cycle, then that one factor -- how much the benchmark code takes advantage of that capability -- should have a relatively consistent mathematical effect on the scores of various machines. So it does apply to other machines that do that, and I don't recall saying otherwise (I certainly didn't mean to).
There are other factors of course, like the speed of the interconnects, efficiency of SMP within each node, etc.
Most other machines in the top 10 have Rmax-to-Rpeak efficiency in the 60-80% range. If what you say is true, then Virginia Tech tuning their cluster until it gets into that range is a good idea.:)
A single G5 FPU (each CPU has 2) can do 1 64-bit (double precision) FLOPs per cycle, or 2 if and only if those two are a MULTIPLY and an ADD.
Apparently there are a lot of cases where a MULTIPLY and an ADD do come together like that, but I'm not surprised if LINPACK doesn't consist entirely of those pairs.;)
The 17.6 TFLOP theoretical peak assumed a perfect case consisting entirely of MULTIPLY-ADD pairs. In a case assuming no MULTIPLY-ADD pairs, the theoretical peak is 8.8 TFLOPs.
7.4 TFLOPs is only 42% of 17.6 TFLOPs, but it's 84% of 8.8 TFLOPs. I suspect the actual "efficiency" of the machine lies somewhere in the middle.
Slashdot Mirror
Okay, okay, maybe you can be available for that. But since being available means you'll alsoget the inevitable calls about all kinds of trivial problems that are beneath your station (users forgetting to turn on their monitors, etc.), you've gotta have a pretty major masochistic streak to want this.
(Okay, it's not fun. Wake me when it's over.)
- NOFX - Brain Constipation
- Tommy Lee - Mr. Shitty
- Warren Zevon - My Shit's Fucked Up
- DMX - Keep Your Shit the Hardest
- Frumpies - I Just Wanna Puke on the Stereo
- The Tony Rich Project - My Stomach Hurts
- The Locust - Identity Exchange Program Rectum Return Policy
- Turbonegro - Rendezvous With Anus
- Witchy Poo - Anal Satan
- The Strokes - Meet Me In The Bathroom
Hope this helps!I seem to recall some jaded audiophiles griping about how limited digital sound was... ;)
How to write a program in AppleScript:
(Okay, okay, well, Mac users will get the joke.)
The XBench CPU Test score went from 148.72 to 193.29. There was a slight decline in the "Floating Point Basic" category, but performance in "AltiVec Basic" and "vecLib FFT" improved by over 50% and "Floating Point Library" performance also improved by over 20%.
The XBench Thread Test score went from 185.93 to 209.27, with most of that accounted for by an 18% gain in the "Computation" subtest. The XBench Memory Test score went from 293.70 to 312.41, gaining primarily in the System (vs. Stream) memory subtests, particularly "Allocate" which went up almost 40%. (On my iBook G3-600, Panther improved "Allocate" scores 304%!)
So if my machine - roughly equivalent to a single node of VT's cluster (theirs have more RAM; mine has more disk) - can get a 30% boost on the CPU test, a 12%+ boost on the thread test, and a 6% boost on the memory test, it looks like the planned upgrade to Panther mentioned in a previous article might help it get past the 10-TeraFLOP mark.
(Hypothetically speaking, if VT's code for LINPACK made extensive use of the AltiVec and vecLib bits included in the OS, going to Panther could boost things up into the 12-14 TeraFLOP range. However, I believe they're probably using custom-written libraries built with optimizing compilers, so I don't think the difference will be that profound.)
Wired's target audience isn't nerds, it's nerd wannabe's. A half-dozen years ago I was working in web design with a cow-orker who managed to read that day-glo typesetter's nightmare, and I had to keep telling her that 'twas better to spend her time doing stuff that would make her article-fodder, rather than article-reader.
Well... hmm. The "regular" Mac sitting next to my desk running Panther is a dual G5, so... I'd say there are plenty of similarities. There are also lots of differences in configuration. VT went for more RAM in their nodes than I have on my Mac, while I maxed out disk space and went for the higher-end video card. I've also got a mouse, keyboard and display; clearly they don't want 1100 of each of those taking up space. And I've got Bluetooth and 802.11g, which probably don't make as much sense in a huge cluster. (Like, what're you gonna do, sync 1100 machines to your cell phone at once?)
Ol' Ralph actually appears to have been involved in a company or two with clue (although said clue may have arrived after his departure).
Hey, what could possibly go wrong?
But then again, I only had to pay $20 for it, since I just got that new Power Mac, so I shouldn't complain.
I already had an NT 4.0 instance running under Bochs, and on the G5, while not overly fast, it's at last usable.
... 'cos I just installed Panther on my Mac, and it was making noises about FreeBSD *5*.
Generally, I try to set things up so the Windows instance doesn't have any ports open to the world, and if at all possible, its "filesystem" is within a file in the real filesystem, so it can't trash anything but itself. :)
Yeah, that's a pretty odd term. "Digital SLR" is better, since the sensor usually does not measure 35mm. The exceptions to that are Canon's 11-megapixel EOS-1Ds which costs about $8,000, and Kodak's 14-megapixel model which costs about $5,000 (but had a lot of problems and delays and generally gets worse reviews than the Canon).
Compare Apache's webserver market share to that of Microsoft IIS. Compare the number of exploitable vulnerabilities in those products, and the severity of the results.
Compare Sendmail's SMTP server market share to that of Microsoft Exchange. Compare the number of exploitable vulnerabilities in those products, and the severity of the results.
Compare Oracle's (or IBM's) SQL RDBMS market share to that of Microsoft SQL Server. Compare the number of exploitable vulnerabilities in those products, and the severity of the results.
Deduction: Microsoft manages to lead in introducing exploitable vulnerabilities to market segments, with severe results, even in segments where they do not enjoy market share leadership.
Now that's innovation! :)
To be blunt and honest, Microsoft designed and maintained its operating system product(s) in ways that failed to take security (and multiple users, and networking, and...) into consideration for far too long, and now finds itself in the unenviable position of being the only operating system vendor most people have even heard of that doesn't have a properly secure operating system.
-Dan (whose new "cheesegrater" G5 has fewer holes than Windows)
Well, that's all well and good, but even though there's one-to-one Internet video conferencing technology good enough that TechWeb thought it was a satellite feed when they saw it last week, it still needs extensions to support one-to-many, many-to-one and many-to-many conferencing.
And even then you'd still have the onerous task of convincing everyone to use iChat AV. After all, it's an Apple product, and everyone knows that Windows users aren't going to want to use an Apple product!
The catches are:
There are probably computers out there with enough horsepower (stop looking at me like that, you may not borrow the dual G5, my precioussss...) to take 2 typical little video streams from webcams and do something with them in real time.
What something is remains unknown, but for example you could slap the images side-by-side, or you could composite one over the other, or you could compare them for differences, or insert 1 frame of stream 2 into stream 1 every 30 frames for subliminal purposes, or whatever.
If most/all the machines have multiply-add in a single cycle, then that one factor -- how much the benchmark code takes advantage of that capability -- should have a relatively consistent mathematical effect on the scores of various machines. So it does apply to other machines that do that, and I don't recall saying otherwise (I certainly didn't mean to).
There are other factors of course, like the speed of the interconnects, efficiency of SMP within each node, etc.
Most other machines in the top 10 have Rmax-to-Rpeak efficiency in the 60-80% range. If what you say is true, then Virginia Tech tuning their cluster until it gets into that range is a good idea. :)
Apparently there are a lot of cases where a MULTIPLY and an ADD do come together like that, but I'm not surprised if LINPACK doesn't consist entirely of those pairs. ;)
The 17.6 TFLOP theoretical peak assumed a perfect case consisting entirely of MULTIPLY-ADD pairs. In a case assuming no MULTIPLY-ADD pairs, the theoretical peak is 8.8 TFLOPs.
7.4 TFLOPs is only 42% of 17.6 TFLOPs, but it's 84% of 8.8 TFLOPs. I suspect the actual "efficiency" of the machine lies somewhere in the middle.
(As for me, I'm happy with just ONE dualie...)
Multimap is, in my experience, better at giving a map with most/all streets named, even in the US, than most US-based mapping sites.
I've got a dual G5 right here too, but no Alphas. :(
...but of course it'll run even better on the Mac you buy next year... ;)
All this foofaraw over the G5 and Athlon 64 is just revisionist history. ;)