I don't get it. If compiler optimization makes so little difference, why do some people see substantial benefits of using Intel's compiler (vs. gcc) on AMDs?
The move away from ia32 would be a positive one, on the condition that AMD didn't do retarded things in x86-64. It is my understanding that it is a b!tch to design and validate a high-performance ia32 chip because of all the complexities. IIRC, several years ago, at a time when Intel was pending billions in R&D for the next chip, the R&D budget on all the big iron RISC chips didn't add up to Intel's ia32 development, many of them beating Intel. AMD has to spend the big bucks too, although I imagine they are spending a little less.
I think it might help compiler design and software validation as well.
They claimed unlimited internet, now they claim that it was the always-on connection, not the number of bits. OK, now is that really so hard for them to state in the first place? They should be clear about the number of bits per day / week / month they'll haul on an account too and give users tools to keep track of that.
Copyrighted works can be parodied in a product, but it might be slightly different for trademarks, and how big a single trademark "parody" is in the business model.
IIRC, Weird Al gets permission from the original song owners for most of his songs anyway.
I did buy an ATI card on eBay, installed it and decide it wouldn't work for what I wanted to do with it, sold it a month later for about $50 more. All I added was a couple better pictures and a spare $5 DVI to VGA dongle that I bought to use it.
Apart from proper spelling, good photos and a reputation for accurate descriptions, another thing that seems to help is to be willing to ship overseas, especially for the retailer exclusive items that overseas bidders couldn't get. It is a bit of a hassle but I think it helps the prices, and usually I sell to domestic bidders anyway.
The Crown Vics had a higher than normal tendency to have similar problems with its rear gas tank, at least enough to know it is at least somewhat statistically significant, although I don't know by how much. You'd think Ford would know better because that is asking for massive PR trouble.
Skysurfer is one, I think through SkyVision company. It allows the use of an old 6ft c-band satellite dish. It is still one-way, requiring a modem and phone line for the uplink but the costs are less. Last I checked, they were a lot more up-front about the fair access policy too.
The problem is that it is a very uphill battle. Even far superior products fall by the wayside because of established user base. Saving maybe 10% more space & bandwidth and having a slightly better UI isn't enough for most people.
RAR is used so (relatively) rarely that the download tine and bytes saved by RAR is wasted because I have to find the decomressor, even if I have a local copy.
I'm not sure if it would save much on the server side because the webmaster would have to deal with complaints that the download is incompatible with the standard program they have. I've only seen RAR on websites that try to be the esoteric of esoteric.
Actually, Anand's has shown that the single memory channel isn't much of a hindrance on AMD's chips, or rather, dual channel isn't very beneficial yet, at least on the 64 bit chips.
I think the current biggest benefit of dual channel on AMD is being able to run more memory.
It's too bad that laptop hardware support for AMD, well, sucks. That I can tell, no company is willing to make a solid performing mobile AMD based laptop. Sure, there are some desknotes, but those are kludges in the first place, I wouldn't buy any desknote, regardless of whose chip is in it.
Maybe I'm crazy but the systems I run that have ECC are incredibly stable even when using Windows. My Alpha got 100+ days uptime with daily use on Windows NT4. I have an old Xeon that easily did 40 days, and I shut it down by mistake.
One thing to keep in mind that you'll want a backup that is routinely disconnected from the computer. Heck, some people had reported problems with upgrading to MacOS 10.3 where certain Firewire controllers went nuts and blanked several user's Firewire hard drives.
I have read through most of the pdf. It looks to be mostly a compilation of data from a great many sources that aren't from organizations on the level of NIST. I thought they were doing some testing, I had hoped to read about accelerated aging & testing methodologies as well as see some actual testing data and how they derive life expectancies.
Unfortunately, the author used "Amaray" as a general description of plastic DVD keepcases and not distinguished it from the "Amaray" brand DVD keepcase.
Indeed, hard drives often don't last as long as removable media. If the hard drive fails, it isn't as if you can remove the media and just put the platters in another drive. Well, it is possible but I hear it is pretty expensive to do this and risky too.
The key I guess is to keep a spare drive in a second system and backup routinely and replace the drive as a routine. That may not allow for rolling back to a particular date (virus, discovery of otherwise corrupt files, etc), like the standard tape backup cycle often allows.
IMO, the AMD Athon 64 3000+ plus a basic decent A64 board aren't that cost prohibitive compared to a Barton. I don't think you'll have to spend $500 to upgrade to that.
Do 64 bit binaries really use more code space? Really?
64 bit mode is the width of the register, not the length of the instruction word. Alpha's register width is 64 bit, but its words are 32 bit, IIRC, like a great many other 64 bit architectures. While the machine code may take a little more space (IIRC, never more than 50%) than ia32 for the same code functionality, that is easily accounted for in the RISC-CISC differences between the two architectures.
This might be an additional thing to nail people with for cracking the net nanny software.
The DMCA covered this but IIRC, the Library of Congress specifically exempted the nanny software from the DMCA. So now, you can crack the database, but you can't data post much data from the list.
I am fairly sure that you aren't the only one to come across this. I don't think it is the deep pipelines...
IIRC, the P4 has only one full standard FP unit (vs Athlon's & Opteron's 3 FP units?), the majority of the floating point was expected to be done by SSE2 optimized code but it kind of backfires a little if the number crunching isn't done by SSE2 code.
A girl who buys a guy an electronic gift is a total babe.
Assuming they know enough. I wouldn't expect most guys to know what kind of shoes to buy, and I wouldn't expect most women to know what kind of hardware to buy. As much as a lot of guys would think all the shoes are the same, a lot of women that I know would think all the computer processors are the same.
I'm still not finding anything that shows I am wrong.
The Nyquist theorem doesn't state that reproduction is perfect at 2x sampling rate of the highest frequency, it only states that it is the minimum sampling rate needs to be 2x the highest frequency in order to capture the signal. Higher sampling rates do capture more accurate detail particularly on the phase accuracy.
http://members.aol.com/ajaynejr/nyquist.htm
In fact, if sampling a 1kHz sine wave at 2kHz, one has to sample at the peaks or get a magnitude error. If the samples happen at the zero crossing (out of phase sample), one gets no signal samples at all:
Nyquist's theorem proves (with mathematics and I've done the derivation myself) that sampling at (slightly more than) 2x the highest frequency can EXACTLY reproduce the original wave. Not approximate. Exact.
This assumes that the samples are at the peaks, if the sample 180 degrees out of phase of the signal, sampling the valleys, then you have no signal. Not getting a singal when one should expect one is a pretty large error to me.
If an input sine wave is near 1/3rd the sampling rate, you can easily get a bunch of nasty phase and magnitude modulations as part of your output signal. I suppose with spline interpolation a lot of that is fixable with interpolations but that may still result in errors.
I am not an audiophile but I will note these things:
The Nyquist theory is an absolute best-case, and assumed that you sampled at the peaks.
Even with four samples per wavelength you can get pretty weird looking sample data. IIRC, EEs try to get at least eight samples per shortest wavelength to get a decent waveform representations, less than that and you can get some noticable potential frequency and phase shifting errors. On CD audio, that makes it a little over 5kHz.
Slashdot Mirror
I don't get it. If compiler optimization makes so little difference, why do some people see substantial benefits of using Intel's compiler (vs. gcc) on AMDs?
The move away from ia32 would be a positive one, on the condition that AMD didn't do retarded things in x86-64. It is my understanding that it is a b!tch to design and validate a high-performance ia32 chip because of all the complexities. IIRC, several years ago, at a time when Intel was pending billions in R&D for the next chip, the R&D budget on all the big iron RISC chips didn't add up to Intel's ia32 development, many of them beating Intel. AMD has to spend the big bucks too, although I imagine they are spending a little less.
I think it might help compiler design and software validation as well.
They claimed unlimited internet, now they claim that it was the always-on connection, not the number of bits. OK, now is that really so hard for them to state in the first place? They should be clear about the number of bits per day / week / month they'll haul on an account too and give users tools to keep track of that.
Copyrighted works can be parodied in a product, but it might be slightly different for trademarks, and how big a single trademark "parody" is in the business model.
IIRC, Weird Al gets permission from the original song owners for most of his songs anyway.
I did buy an ATI card on eBay, installed it and decide it wouldn't work for what I wanted to do with it, sold it a month later for about $50 more. All I added was a couple better pictures and a spare $5 DVI to VGA dongle that I bought to use it.
Apart from proper spelling, good photos and a reputation for accurate descriptions, another thing that seems to help is to be willing to ship overseas, especially for the retailer exclusive items that overseas bidders couldn't get. It is a bit of a hassle but I think it helps the prices, and usually I sell to domestic bidders anyway.
The Crown Vics had a higher than normal tendency to have similar problems with its rear gas tank, at least enough to know it is at least somewhat statistically significant, although I don't know by how much. You'd think Ford would know better because that is asking for massive PR trouble.
If ugly is the only strike against it then I don't see how it fits the list.
If it were an unreliable piece of trash, then it would.
There are alternatives to DirecWay and Starband.
Skysurfer is one, I think through SkyVision company. It allows the use of an old 6ft c-band satellite dish. It is still one-way, requiring a modem and phone line for the uplink but the costs are less. Last I checked, they were a lot more up-front about the fair access policy too.
Details are at www.cband.net
For $250 / month and into the thousands to install, wouldn't it be cheaper / better to rent personal office space and subscribe to business DSL?
The problem is that it is a very uphill battle. Even far superior products fall by the wayside because of established user base. Saving maybe 10% more space & bandwidth and having a slightly better UI isn't enough for most people.
RAR is used so (relatively) rarely that the download tine and bytes saved by RAR is wasted because I have to find the decomressor, even if I have a local copy.
I'm not sure if it would save much on the server side because the webmaster would have to deal with complaints that the download is incompatible with the standard program they have. I've only seen RAR on websites that try to be the esoteric of esoteric.
Actually, Anand's has shown that the single memory channel isn't much of a hindrance on AMD's chips, or rather, dual channel isn't very beneficial yet, at least on the 64 bit chips.
I think the current biggest benefit of dual channel on AMD is being able to run more memory.
It's too bad that laptop hardware support for AMD, well, sucks. That I can tell, no company is willing to make a solid performing mobile AMD based laptop. Sure, there are some desknotes, but those are kludges in the first place, I wouldn't buy any desknote, regardless of whose chip is in it.
How about just simple ECC?
Maybe I'm crazy but the systems I run that have ECC are incredibly stable even when using Windows. My Alpha got 100+ days uptime with daily use on Windows NT4. I have an old Xeon that easily did 40 days, and I shut it down by mistake.
One thing to keep in mind that you'll want a backup that is routinely disconnected from the computer. Heck, some people had reported problems with upgrading to MacOS 10.3 where certain Firewire controllers went nuts and blanked several user's Firewire hard drives.
I have read through most of the pdf. It looks to be mostly a compilation of data from a great many sources that aren't from organizations on the level of NIST. I thought they were doing some testing, I had hoped to read about accelerated aging & testing methodologies as well as see some actual testing data and how they derive life expectancies.
Unfortunately, the author used "Amaray" as a general description of plastic DVD keepcases and not distinguished it from the "Amaray" brand DVD keepcase.
Indeed, hard drives often don't last as long as removable media. If the hard drive fails, it isn't as if you can remove the media and just put the platters in another drive. Well, it is possible but I hear it is pretty expensive to do this and risky too.
The key I guess is to keep a spare drive in a second system and backup routinely and replace the drive as a routine. That may not allow for rolling back to a particular date (virus, discovery of otherwise corrupt files, etc), like the standard tape backup cycle often allows.
IMO, the AMD Athon 64 3000+ plus a basic decent A64 board aren't that cost prohibitive compared to a Barton. I don't think you'll have to spend $500 to upgrade to that.
Do 64 bit binaries really use more code space? Really?
64 bit mode is the width of the register, not the length of the instruction word. Alpha's register width is 64 bit, but its words are 32 bit, IIRC, like a great many other 64 bit architectures. While the machine code may take a little more space (IIRC, never more than 50%) than ia32 for the same code functionality, that is easily accounted for in the RISC-CISC differences between the two architectures.
This might be an additional thing to nail people with for cracking the net nanny software.
The DMCA covered this but IIRC, the Library of Congress specifically exempted the nanny software from the DMCA. So now, you can crack the database, but you can't data post much data from the list.
I am fairly sure that you aren't the only one to come across this. I don't think it is the deep pipelines...
IIRC, the P4 has only one full standard FP unit (vs Athlon's & Opteron's 3 FP units?), the majority of the floating point was expected to be done by SSE2 optimized code but it kind of backfires a little if the number crunching isn't done by SSE2 code.
A girl who buys a guy an electronic gift is a total babe.
Assuming they know enough. I wouldn't expect most guys to know what kind of shoes to buy, and I wouldn't expect most women to know what kind of hardware to buy. As much as a lot of guys would think all the shoes are the same, a lot of women that I know would think all the computer processors are the same.
I'm still not finding anything that shows I am wrong.
e fo rms/nyquist.html
The Nyquist theorem doesn't state that reproduction is perfect at 2x sampling rate of the highest frequency, it only states that it is the minimum sampling rate needs to be 2x the highest frequency in order to capture the signal. Higher sampling rates do capture more accurate detail particularly on the phase accuracy.
http://members.aol.com/ajaynejr/nyquist.htm
In fact, if sampling a 1kHz sine wave at 2kHz, one has to sample at the peaks or get a magnitude error. If the samples happen at the zero crossing (out of phase sample), one gets no signal samples at all:
http://www.cs.ust.hk/faculty/layers/comp342/wav
If you would explain, I would be happy to read it.
I realize I did mistate it a little bit, but I don't see where my thought was wrong and if you aren't willing to correct me, then I can't learn.
Attacking the least relevant part of my post really looks like a strawman argument anyway.
Nyquist's theorem proves (with mathematics and I've done the derivation myself) that sampling at (slightly more than) 2x the highest frequency can EXACTLY reproduce the original wave. Not approximate. Exact.
This assumes that the samples are at the peaks, if the sample 180 degrees out of phase of the signal, sampling the valleys, then you have no signal. Not getting a singal when one should expect one is a pretty large error to me.
If an input sine wave is near 1/3rd the sampling rate, you can easily get a bunch of nasty phase and magnitude modulations as part of your output signal. I suppose with spline interpolation a lot of that is fixable with interpolations but that may still result in errors.
do you care to elaborate?
I am not an audiophile but I will note these things:
The Nyquist theory is an absolute best-case, and assumed that you sampled at the peaks.
Even with four samples per wavelength you can get pretty weird looking sample data. IIRC, EEs try to get at least eight samples per shortest wavelength to get a decent waveform representations, less than that and you can get some noticable potential frequency and phase shifting errors. On CD audio, that makes it a little over 5kHz.