network narks

[John Holotko]

Yup I like Dells too have a dual 2.8 G at work and the guy next to me just ordered one. 2 at home, older units.

Most of my machines I assemble myself.  The Dells are not my favorites but, they are good general purpose desktops for everyday use.

[John Holotko]

Tom for a x64 machine you can expect to pay in the ballpark what you paid initially for your machines that you're using now. $2500 -- $3500. The thing is that there isn't much software out there to take advantage of the x64 architecture, so you would be running XP 32 until the new stuff comes out. But you would have a machine that supports it!!! cause the x64 architecture runs older 32bit code!

I myself dual boot my AMD64 machine, I run XP 32 for watching DVD's and flashchat. And boot the machine into GNU/Linux x64 for my scientific work.

I know of a few people who watch DVD's right on their Linux boxes. No need to have to drop into windows when you can watch it right on your X windows desktop.  KDE or GNOME they work great. Of course I would never do such a thing because it would require using unlicensed DVD software which is something I would never ever do,  Too bad they never distruibuted a legal licensed Linux DVD player that a Linux user could just download or buy shrink wrapped and install it.  5 years ago companies said they were going to develop and market such a beast and I have yet to see it hit the market. One company dioes offer a legal Linux DVD player but it is only available to OEM's and primarilly for embedded devices. Shame, the only thing available to the avaerage user is deCSS and the unlicensed stuff. 

[wa1knx]

tom,
        you should be verdi happy with that beast your getting!

john,
        I'm with you, I assemble mine to.  i have a 500mh pentium3 with scsi back in vt. next
summer it gets a new brain board, maybe upgrade some of the other stuff.

frank,
         I used to tell the other engineers - back up your own stuff! don't trust your IT guys!
if they screw up, you'll be able to recover.  our management never accepted "its their fault"
if something messed up!

       as to assembly code, your right. not many do that any more. working on the OS
instead off applications we had to often.  I just applied for a job, where a customer
application was in assembly!! that was a first to hear about

deano

[KE1GF]

Hey Guys don't kill the messenger:

http://bofhcam.org/co-larters/bofh-nutshell/

[KE1GF]

Man I thought it was a joke the person that came up with this stuff is really sick in the head!

[KE1GF]

TOTAL:$369.00 

Way to go Tom, just as others stated. Don't go out and spend a ton of cash on a 32 bit machine.

I have XP x64 but I don't use it yet and probably will not. GNU/Linux or some other unixies is where the 64 bit action is at.

I have ng_spice (SPICE3F5?) optimized compiled for AMD64 to do my Classs-E simulations, but I still use the SwCADIII GUI to make the SPICE netlist on an XP 32 bit machine.

Seeing that SPICE uses double precision floating point numbers, the simulations scream running on the processor in 64 bit mode. Double presision is the word size of the registers on my AMD64 and there are double the number of them (vs 32 bit) with a very large L2 cache and lots of RAM. Less load/store instructions to populate the registers.

Also I have SPIN optimized compiled for AMD64 for my concurrent system simulations.

Yet again SPIN has an advantage on the AMD64 because it can utilize the 4GB of physical RAM (In my machine) and the 100GB++ of swap space (under GNU/Linux) to do the process interaction simulations (gigantic state space vector).

Machines that have a 32 bit pointer are limited to 4GB of virtual address space. Not only is it insane to write the memory management code in the OS to make a 32 bit machine utilize more memory. It causes serious performance degradation and its nearly impossible to work the bugs out.

This constitutes the bulk of my computing needs at the moment.

Inexpensive 32 bit machines are great for doing lots of tasks like reading email, surfing the web, recording, flashchat, etc... Also the hardware support is best for the 32 bit machines, every new piece of boxed hardware works under XP 32, but barely any of it will work with XP x64.

When I was experimenting with operating systems on this AMD64 of mine I found that GNU/Linux has the most momentum as far as hardware support goes. Microsoft XP x64 and SUN Solaris x64 are way behind.

People ask me why I didn't buy a dual core 4000 series processor and there is one specific reason why. 'cause none of the things that I do can purely utilize symmetrical multi processing without performace degredation.  It shines when you have completely independant processes running on the machine aka: multiple users, multiple simutanious web surfers on a threaded webserver, etc.. But for a causal system like a circuit or for exaustive depth first search for a contradiction of an assertation, the uniprocessor shines.

Cray knew this long before I was born. He realized before his car accident (which took his life) that it was impossible to make a physical machine of such magnitude that would parallel the laws of diminishing returns. So he made virtual uniprosseor machines with software, that were constructed out of high performance commodity processors, lots of RAM, and high throughput low latency networks.

Of course Cray's genious went much further than just fast number crunchers. He designed software, hardware, semicondutors, layout, power-supplies, cooling systems, etc... All aspects of high performance computers and to boot his machines looked very sexy too.

I don't know much about antenna modeling software OM but I know it's numerical scientific computation. I know that you depend on your machines to run your antenna modeling simulations, we all depend on computer simulations. I'm curious, for your most advanced antenna and feedline models how much time does it take to run a simulation?

[K1JJ]

Wow, that's quite a mouthful, Bill! 

It highlights how much the new computer technology has passed me by.
I started out in 1978 with a full blown Rat Shark TRS-80 system using four floppies in a line. Thought that was slick. But I JUST barely keep up always lagging behind a few years in my hardware.

Anyway, you axed about the antenna modeling speed. Depends upon how many segments you specify and the complexity of the array. A simple dipole takes less than a second on my old 233 mhz machine. A full blown 32 element Sterba curtain array that uses as many segment as the software will allow may take 20 seconds or so. So, it's pretty fast for the mainstream ham antenna modeling stuff.

I would imagine the newest Nec4 or higher govt stuff would require 1 ghz machines or even 64 bit stuff like you use there.

Sounds like you really found your EE nitche there, Bill. I applaud you.

T

[KE1GF]

Wow, that's quite a mouthful, Bill! 

It highlights how much the new computer technology has passed me by.
I started out in 1978 with a full blown Rat Shark TRS-80 system using four floppies in a line. Thought that was slick. But I JUST barely keep up always lagging behind a few years in my hardware.

Anyway, you axed about the antenna modeling speed. Depends upon how many segments you specify and the complexity of the array. A simple dipole takes less than a second on my old 233 mhz machine. A full blown 32 element Sterba curtain array that uses as many segment as the software will allow may take 20 seconds or so. So, it's pretty fast for the mainstream ham antenna modeling stuff.

I would imagine the newest Nec4 or higher govt stuff would require 1 ghz machines or even 64 bit stuff like you use there.

Sounds like you really found your EE nitche there, Bill. I applaud you.

T

Okay, cause some of my SPICE stuff can take several minutes for maybe 200uS of simulation.

As far as the SPIN (formal verification for my MQP) ones, I don't know I haven't built the models yet. Guessing from what I read they may take days...

Anyway Tom, 6 more credits for my degree in "Computer Science"

[KA1ZGC]

Yet again SPIN has an advantage on the AMD64 because it can utilize the 4GB of physical RAM (In my machine) and the 100GB++ of swap space (under GNU/Linux) to do the process interaction simulations (gigantic state space vector).

100+ G swap, Bill? Damn. Did you actually make (a) swap partition(s) that size (I'm assuming you did), or are you doing that in a swapfile? I'm only asking because of the size, that's a lot of disk space to be dedicating to swap space, even in this day and age. The FS overhead of a swapfile sucks, though you'll never convince MICROS~1 of that.

Tom, I have an AMD64 box here on my desk that our "IT" (ahem ahem) guys bought from some bulk-sales outfit. I've run Redhat Enterprise Linux 4.0 and SuSE Linux 9.3 on it and it leaves the x86 machines in the dust, at least for what few things I've done on them. Haven't tried any MICROS~1 64-bit offerings, but if it's anything like the pain they went through going from 16 to 32, I'd say wait a few years. If you don't mind the unix environment, any Linux or any BSD should do nicely.

Hey Bill (or anyone), if I wanted to reasonably benchmark an x86 vs. an AMD64, both new and both running SuSE 9.3, what's a reasonable benchmark to use? I've got both here on my desk, and I've been curious to see how the numbers compare. Any thoughts?

--Thom
King Abraham One Zebraham George Charlie

[KE1GF]

100+ G swap, Bill? Damn. Did you actually make (a) swap partition(s) that size (I'm assuming you did), or are you doing that in a swapfile? I'm only asking because of the size, that's a lot of disk space to be dedicating to swap space, even in this day and age. The FS overhead of a swapfile sucks, though you'll never convince MICROS~1 of that.

Tom, XP x86 runs with no swap file on the machine.

Under linux it's a 100GB swapspace on my RAID1+0, buffered reads I get on average of 110MB/Sec transfer and from cached reads its 1700MB/Sec. I don't know if I'll even need it, it's easy enough to turn on and off. The machine is part of my MQP, it's just a lab rat right at the moment. It's all imaged up waiting for evil impending sceintific doom. MUHAHAHA

Tom, I have an AMD64 box here on my desk that our "IT" (ahem ahem) guys bought from some bulk-sales outfit. I've run Redhat Enterprise Linux 4.0 and SuSE Linux 9.3 on it and it leaves the x86 machines in the dust, at least for what few things I've done on them. Haven't tried any MICROS~1 64-bit offerings, but if it's anything like the pain they went through going from 16 to 32, I'd say wait a few years. If you don't mind the unix environment, any Linux or any BSD should do nicely.

For a 32 Bit OS there's no performance boost on an AMD64 machine. Actually if you read the benchmarks for common applications, the P4 actually beats the AMD64 most of the time. When you compare lets say Linux for pentium vs AMD64 you're comparing apples and oranges. For double presision integer or fp it mops the floor with x86, that's a given.

I've got my copy here of XP x64 it was loaded and wiped, you'd be lucky if your mouse works with it. 

Hey Bill (or anyone), if I wanted to reasonably benchmark an x86 vs. an AMD64, both new and both running SuSE 9.3, what's a reasonable benchmark to use? I've got both here on my desk, and I've been curious to see how the numbers compare. Any thoughts?

Tom, they're already out there, Spec2000(INT,FPU,etc) were you can compare an AMD64 machine to something like a SparcUltra10 or whatever.

-Your Evil Computer Scientist Network Nark

[KA1ZGC]

100+ G swap, Bill? Damn. Did you actually make (a) swap partition(s) that size (I'm assuming you did), or are you doing that in a swapfile? I'm only asking because of the size, that's a lot of disk space to be dedicating to swap space, even in this day and age. The FS overhead of a swapfile sucks, though you'll never convince MICROS~1 of that.

Tom, XP x86 runs with no swap file on the machine.

Under linux it's a 100GB swapspace on my RAID1+0, buffered reads I get on average of 110MB/Sec transfer and from cached reads its 1700MB/Sec. I don't know if I'll even need it, it's easy enough to turn on and off. The machine is part of my MQP, it's just a lab rat right at the moment. It's all imaged up waiting for evil impending sceintific doom. MUHAHAHA

FWIW, we build our machines with swap partitions that are only 2x RAM, and even running the toughest queries we've got against the largest datasets we've got, our systems barely even touch swap space. Mind you, the host really only serves as an aggregation point, but when it's time to hand back the final answer, that's all in the host, and some of the return sets are gargantuan. Someone running Oracle or some other software-based database would probably swap a lot more frequently, but we're a hardware-based database, which changes the dynamic. Still, we pound these things 24x7x365 and they rarely swap.

This is with Redhat Enterprise Linux, so your mileage may vary.

On the other hand, if you've got LVM, resizing partitions is child's play, and the perf hit is negligible. With a RAID set, it's the most maintainable way to go.

Tom, I have an AMD64 box here on my desk that our "IT" (ahem ahem) guys bought from some bulk-sales outfit. I've run Redhat Enterprise Linux 4.0 and SuSE Linux 9.3 on it and it leaves the x86 machines in the dust, at least for what few things I've done on them. Haven't tried any MICROS~1 64-bit offerings, but if it's anything like the pain they went through going from 16 to 32, I'd say wait a few years. If you don't mind the unix environment, any Linux or any BSD should do nicely.

For a 32 Bit OS there's no performance boost on an AMD64 machine. Actually if you read the benchmarks for common applications, the P4 actually beats the AMD64 most of the time. When you compare lets say Linux for pentium vs AMD64 you're comparing apples and oranges. For double presision integer or fp it mops the floor with x86, that's a given.

I've got my copy here of XP x64 it was loaded and wiped, you'd be lucky if your mouse works with it. 

Yep, I should have been more clear about that, both Linuces (RH and SuSE) were AMD64-specific kernels. I was pleasantly suprised to see that the SuSE kit installed an AMD64-specific kernel out-of-the-box, rather than a generic kernel. IIRC, I had to build an AMD64-specific kernel under RH.

Hey Bill (or anyone), if I wanted to reasonably benchmark an x86 vs. an AMD64, both new and both running SuSE 9.3, what's a reasonable benchmark to use? I've got both here on my desk, and I've been curious to see how the numbers compare. Any thoughts?

Tom, they're already out there, Spec2000(INT,FPU,etc) were you can compare an AMD64 machine to something like a SparcUltra10 or whatever.

I'll have to check that out. I've got PowerPC4, 5, Sparc, PA-RISC, Itanium², and now AMD64 all sitting here for 64-bit platforms, and I've often wondered who would win a given race. I'm sure the two Alphas (EV67) I just bought would lose, but that's because they're only 600 MHz. Too bad they're all running different OSes, it's harder to get a gut feel for actual platform performance that way.

Good luck with that hidden agenda to undermine the republic, Bill! 

--Thom
Kilimunjaro Africa One Zulu Goat Cheese

p.s. Still a cold beer waiting for you in Framingham

[KE1GF]

Good luck with that hidden agenda to undermine the republic, Bill! 

--Thom
Kilimunjaro Africa One Zulu Goat Cheese

p.s. Still a cold beer waiting for you in Framingham

Which brings me to the ultimate question, why is that drink always best served cold?

Just remember, I'm just a puddle skip away here in Hopkinton. If you can't make it back to base for whatever reason your always more than welcome here. If I'm in a state of consciousness enough to answer the land line or a radio I'll answer back, if not you can always try the old fashoned methods.

-Bill 'GF