Now that P4 chips are drawing over 100 watts of power, I agree and would probably choose AMD for a casual workstation *if* I could find a motherboard that is stable. I hear of too many lock-up problems with ASUS, Kbit, ECS, etc etc etc. How come I never hear of lock-up problems with Intel motherboards?
However, I definitely choose PIII over Athlons a year ago. A PIII draws half as much power. I use a 200W power supply on my existing PIII's. They have *never* crashed.
I actually do not think it is the AMD cpu. I'm sure the AMD cpu works to spec. The RAM is Samsung DDR.
I think the problem is probably the chipset or motherboard. If AMD produced their own motherboard and chipsets, they would probably be stable as a rock. I believe that my PIII system uses an Intel chipset.
Anybody who deals with "mission-critical" machines will tell you that Intel puts out better chips.
I've noticed an alarming trend on Slashdot. The parent post was moderated as a "2". Because this post bashes AMD, it will be immediately moderated to either a "0 Flamebait", or "-1 Troll".
You are talking straight out of your ass. The K7 micro-architecture (Athlon) is no more "RISC-like" than the P6 micro-architecture (PIII). They are both out-of-order superscalar cores. They both convert the IA-32 instructions to "micro-ops". They both have branch predictors and a re-order buffer. In fact, I would consider the P4 "net burst" micro-architecture to be more "RISC-like". AMD will not be able to scale the clockrate of the K7 like Net-burst. Some of this may have to do with its enormous L1 cache. Smaller caches are faster. That's a rule of thumb. Intel uses small L1 caches so that they can deliver data in a single clock cycle at high frequencies.
When comparing a PIII vs. Athlon, the extra electricity the Athlon uses will outweigh the price/performance benefits over several years of 24/7 use. An Athlon 1GHz draws about 70 watts. A PIII 1GHz draws about 30 watts. Do the math.
Unfortunately, this doesn't hold for the P4 because it sucks much much more power.
Why do Slashdot moderators label any message describing a problem with an AMD Athlon as a "Troll"?
This website is not inpartial.
I feel the same way about Athlons and so do most businesses I work with.
I own 2 machines. One is a 900 MHz Athlon with a K7S5A Motherboard. The other is a PIII 866MHz. They both run Linux. The PIII has *never* crashed on me. The Athlon hard-freezes every month or so. I've got a quality power supply in both machines.
Furthermore, any price difference I paid between the PIII and Athlon will be made up for 3 years from the extra electricity that the Athlon draws (75 watts compared to 30 watts).
Well, I'm a geek...and I've extensively studied computer architecture in grad. school. I won't put an Athlon into an important machine. Its fine for casual desktop use, but in my experience, an Athlon is not as stable as a PIII (I have no experience with a P4). PIII's draw half as much power thus you don't need to worry about a great power supply. You don't need to worry about subtle incompatibilities such as the AGP cache coherence bug. And there are just too many lousy Athlon motherboards out there. That may not be AMD's fault, but its the truth. I know dozens of Athlon users who consistently tell me that their Athlon machine will freeze every few months or so. I've never heard of such a thing with PIII's, that is unless they are using a lousy OS such as Win9x.
Granted I do in fact use an Athlon for one of my machines and am generally happy with it. I just wouldn't trust it when things really matter (like a server situation or when doing a 10-week numerical computation).
They will bite you in the a%&. I wasn't paying attention to the quality of recent drives and purchased an IBM 60GXP based on my previous impression of IBM drives. It started giving my problems after a week of use. So I called the place I ordered the drive from, http://datadrive4all.com. They told my to run the IBM Drive & Fitness Test. I did, and it showed no errors. Then I installed my OS again, and after a few days of use, it crashed. So once again, I ran the DFT test this time stressing it for several hours. I managed to get an error, and processed the RMA through datadrive4all.com.
So I sent that place the drive, and they called me to say that they tested the drive with DFT and found no problems. I couldn't convince those idiots otherwise. So they charged me a 15% restocking fee and put the drive back on their shelves to sell to some other poor sap.
I realize all drives can fail, but further research revealed that there is a class action lawsuit against IBM for the quality of their GXP line of drives!!!
I then bought the "Caviar" of IDE disk drives (Western Digital). Its been running great and is quiet and fast.
No, I'm saying that it would be extremely difficult to implement an out-of-order superscalar execution engine when your operations are not simple.
I am saying the instruction set architectures (ISAs) don't effect the internal micro-architecture like they used to because we've figured out how to do translation for typical CISC ISAs. Of course doing this translation will cost you in silicon area.
The difference here is that there was much steeper a learning curve, and no normal engineers could actually do complex tasks - i.e. create branches etc. We had a complete groud dedicated to ClearCase.
You are kidding me, right? I haven't used Clearcase for about a year but yet I still remember "cleartool mkbranch ". If Oracle software developers can't manage making branches on their very own CM tool, then my confidence in Oracle's DBMS just plummeted!
You aren't missing anything. Mozilla is a dog. Its Javascript is also flaky. On Linux, I use Galeon which uses the rendering component of Mozilla. It is much better and faster. I really wish someone would develop a browser on Windows using only Gecko. I don't need nor care about the Mozilla "Development" platform.
CISC vs. RISC is irrelevant. Intel and AMD simply convert the CISC instructions to RISC-like instructions internally. The Intel/AMD cores are out-of-order superscalar designs just like the rest of them (well, Sparc is actually in-order but that's a whole different story).
I created my own homepage about 2 months ago. I just searched for my name on Google. The page showed up as the first hit. Then I searched for my name on AltaVista, and I scanned through several pages of hits without seeing my homepage.
us poor sobs will be stuck with the same old file concept from the 60's: A stream of bytes associated with a filename and stored in a heirarchical tree of directories.
"Film & Slide Scanner". Can that scan regular photographs?? If not, do you know of a normal flatbed scanner that has hardware dust & scratch removal? Is it less than $1000?
I originally intended to do all of my work using the Gimp. So I purchased an Epson Perfection 1240u scanner (i realize that it is amateur quality only). Then I discovered the "Dust & Scratch removal" function included with Adobe Photodeluxe which came bundled with the scanner. Because photodeluxe does not offer full-blown color correction, I either use Photoshop exclusively (at school), or I remove the dust&scratches using Photodeluxe and then reboot to Linux to use Gimp for color correction, cropping, etc.
I'm becoming more and more interested in this little photo restoration hobby I started on and would be interested in hearing about higher quality scanners that have advanced hardware features such as dust/scratch removal...
1) I am scanning pictures that are 20-40 years old. I don't believe they had 4 megapixel Nikon digital cameras back in 1969.
2) I am a student. My school lab has really fast machines with Photoshop already installed. Its worth the inconvenience to do my retouching using a Photoshop machine not in my apartment. If I wasn't a student and had a real job, it would be well worth the $600 for Photoshop. The Dust&Scratch removal is simply amazing. Like I said, doing the same thing in Gimp could take an extra hour a picture. Lets see....I have about 400 old Polaroids with scratches in them. 400 hours is well worth $600 to me. Also, who says you need to upgrade every 2 years?? A friend of mine owns Photoshop 4.0 and it has 95% of the features of Photoshop 6.0. He's had that software for 5 years now I believe.
If Gimp had "Remove Dust&Scratches", it would be good enough for me. I don't need nifty fonts or CMYK. I've tried the Gimp "Despeckle" filter but it doesn't work like I want it too. Not even close.
The little things in Photoshop make the difference. For example, Photoshop has a filter called "Remove Dust & Scratches". I'm doing photo restoration for my family pictures. That filter is absolutely amazing. Nothing compares in the Gimp. Sure, anything is possible using the Gimp, however with Photoshop, it just takes 5 seconds as compared to 30 minutes to remove scratches cleanly.
I highly recommend the ECS K7S5A. Its never ever crashed on me and I've got both the on-board Audio and LAN going on linux just fine. I only paid $65 for it!
EPIC is indeed Intel-speak for VLIW...sort of. True VLIW is where everything is statically scheduled. EPIC relaxes this. It groups independent operations into a single instruction word. The processor core is then allowed to schedule this however it wants.
No, you are wrong. The instruction set is irrelevant. IA-32 is dynamically converted to a RISC-like language. Anti-dependencies are dynamically renamed. There are dozens of physical registers for which many instances of the logical registers can be renamed if there is no true dependence.
What you are suggesting is to move to a VLIW-type instruction set. VLIW is exactly specifying what function unit does what, in the instruction itself. DSP instruction sets are VLIW-like. So is the MMX extensions (sort of). IA-64 is doing exactly this except that it is allowing some dynamic behavior instead of everything being static which is VLIW-purism.
I sort of agree with him. If an SMT processor runs 2 threads from different processes, how will it handle virtual memory? I suppose you could implement an SMT processor with separate TLB tables...that would be the only way. I wonder if Hyperthreading does this?? Otherwise running 2 threads from different processes won't work.
Oh, and from my knowledge, most UNICES flush the TLB on a context-switch. Someone please correct me if I'm wrong. I do realize that some architectures (MIPS R10000 I believe) can associate a PID with each TLB entry. This would also be a solution to the problem.
Slashdot Mirror
Thanks. I'll check out Tyan motherboards. I've never heard of them.
Now that P4 chips are drawing over 100 watts of power, I agree and would probably choose AMD for a casual workstation *if* I could find a motherboard that is stable. I hear of too many lock-up problems with ASUS, Kbit, ECS, etc etc etc. How come I never hear of lock-up problems with Intel motherboards?
However, I definitely choose PIII over Athlons a year ago. A PIII draws half as much power. I use a 200W power supply on my existing PIII's. They have *never* crashed.
I actually do not think it is the AMD cpu. I'm sure the AMD cpu works to spec. The RAM is Samsung DDR.
I think the problem is probably the chipset or motherboard. If AMD produced their own motherboard and chipsets, they would probably be stable as a rock. I believe that my PIII system uses an Intel chipset.
Anybody who deals with "mission-critical" machines will tell you that Intel puts out better chips.
I've noticed an alarming trend on Slashdot. The parent post was moderated as a "2". Because this post bashes AMD, it will be immediately moderated to either a "0 Flamebait", or "-1 Troll".
Slashdot is so impartial it makes me sick.
You are talking straight out of your ass. The K7 micro-architecture (Athlon) is no more "RISC-like" than the P6 micro-architecture (PIII). They are both out-of-order superscalar cores. They both convert the IA-32 instructions to "micro-ops". They both have branch predictors and a re-order buffer. In fact, I would consider the P4 "net burst" micro-architecture to be more "RISC-like". AMD will not be able to scale the clockrate of the K7 like Net-burst. Some of this may have to do with its enormous L1 cache. Smaller caches are faster. That's a rule of thumb. Intel uses small L1 caches so that they can deliver data in a single clock cycle at high frequencies.
When comparing a PIII vs. Athlon, the extra electricity the Athlon uses will outweigh the price/performance benefits over several years of 24/7 use. An Athlon 1GHz draws about 70 watts. A PIII 1GHz draws about 30 watts. Do the math.
Unfortunately, this doesn't hold for the P4 because it sucks much much more power.
Why do Slashdot moderators label any message describing a problem with an AMD Athlon as a "Troll"?
This website is not inpartial.
I feel the same way about Athlons and so do most businesses I work with.
I own 2 machines. One is a 900 MHz Athlon with a K7S5A Motherboard. The other is a PIII 866MHz. They both run Linux. The PIII has *never* crashed on me. The Athlon hard-freezes every month or so. I've got a quality power supply in both machines.
Furthermore, any price difference I paid between the PIII and Athlon will be made up for 3 years from the extra electricity that the Athlon draws (75 watts compared to 30 watts).
Well, I'm a geek...and I've extensively studied computer architecture in grad. school. I won't put an Athlon into an important machine. Its fine for casual desktop use, but in my experience, an Athlon is not as stable as a PIII (I have no experience with a P4). PIII's draw half as much power thus you don't need to worry about a great power supply. You don't need to worry about subtle incompatibilities such as the AGP cache coherence bug. And there are just too many lousy Athlon motherboards out there. That may not be AMD's fault, but its the truth. I know dozens of Athlon users who consistently tell me that their Athlon machine will freeze every few months or so. I've never heard of such a thing with PIII's, that is unless they are using a lousy OS such as Win9x.
Granted I do in fact use an Athlon for one of my machines and am generally happy with it. I just wouldn't trust it when things really matter (like a server situation or when doing a 10-week numerical computation).
They will bite you in the a%&. I wasn't paying attention to the quality of recent drives and purchased an IBM 60GXP based on my previous impression of IBM drives. It started giving my problems after a week of use. So I called the place I ordered the drive from, http://datadrive4all.com. They told my to run the IBM Drive & Fitness Test. I did, and it showed no errors. Then I installed my OS again, and after a few days of use, it crashed. So once again, I ran the DFT test this time stressing it for several hours. I managed to get an error, and processed the RMA through datadrive4all.com.
So I sent that place the drive, and they called me to say that they tested the drive with DFT and found no problems. I couldn't convince those idiots otherwise. So they charged me a 15% restocking fee and put the drive back on their shelves to sell to some other poor sap.
I realize all drives can fail, but further research revealed that there is a class action lawsuit against IBM for the quality of their GXP line of drives!!!
I then bought the "Caviar" of IDE disk drives (Western Digital). Its been running great and is quiet and fast.
No, I'm saying that it would be extremely difficult to implement an out-of-order superscalar execution engine when your operations are not simple.
I am saying the instruction set architectures (ISAs) don't effect the internal micro-architecture like they used to because we've figured out how to do translation for typical CISC ISAs. Of course doing this translation will cost you in silicon area.
The difference here is that there was much steeper a learning curve, and no normal engineers could actually do complex tasks - i.e. create branches etc. We had a complete groud dedicated to ClearCase.
You are kidding me, right? I haven't used Clearcase for about a year but yet I still remember "cleartool mkbranch ". If Oracle software developers can't manage making branches on their very own CM tool, then my confidence in Oracle's DBMS just plummeted!
You aren't missing anything. Mozilla is a dog. Its Javascript is also flaky. On Linux, I use Galeon which uses the rendering component of Mozilla. It is much better and faster. I really wish someone would develop a browser on Windows using only Gecko. I don't need nor care about the Mozilla "Development" platform.
CISC vs. RISC is irrelevant. Intel and AMD simply convert the CISC instructions to RISC-like instructions internally. The Intel/AMD cores are out-of-order superscalar designs just like the rest of them (well, Sparc is actually in-order but that's a whole different story).
Wyse was big back in the 80's. They were the second-largest maker of terminals for mainframes.
I created my own homepage about 2 months ago. I just searched for my name on Google. The page showed up as the first hit. Then I searched for my name on AltaVista, and I scanned through several pages of hits without seeing my homepage.
us poor sobs will be stuck with the same old file concept from the 60's: A stream of bytes associated with a filename and stored in a heirarchical tree of directories.
That's funny. Mozilla 0.98 crashes on me constantly on Linux. Mozilla 0.96 is as stable as a rock. Anyone else seeing this behavior?
"Film & Slide Scanner". Can that scan regular photographs?? If not, do you know of a normal flatbed scanner that has hardware dust & scratch removal? Is it less than $1000?
I originally intended to do all of my work using the Gimp. So I purchased an Epson Perfection 1240u scanner (i realize that it is amateur quality only). Then I discovered the "Dust & Scratch removal" function included with Adobe Photodeluxe which came bundled with the scanner. Because photodeluxe does not offer full-blown color correction, I either use Photoshop exclusively (at school), or I remove the dust&scratches using Photodeluxe and then reboot to Linux to use Gimp for color correction, cropping, etc.
I'm becoming more and more interested in this little photo restoration hobby I started on and would be interested in hearing about higher quality scanners that have advanced hardware features such as dust/scratch removal...
1) I am scanning pictures that are 20-40 years old. I don't believe they had 4 megapixel Nikon digital cameras back in 1969.
2) I am a student. My school lab has really fast machines with Photoshop already installed. Its worth the inconvenience to do my retouching using a Photoshop machine not in my apartment. If I wasn't a student and had a real job, it would be well worth the $600 for Photoshop. The Dust&Scratch removal is simply amazing. Like I said, doing the same thing in Gimp could take an extra hour a picture. Lets see....I have about 400 old Polaroids with scratches in them. 400 hours is well worth $600 to me. Also, who says you need to upgrade every 2 years?? A friend of mine owns Photoshop 4.0 and it has 95% of the features of Photoshop 6.0. He's had that software for 5 years now I believe.
If Gimp had "Remove Dust&Scratches", it would be good enough for me. I don't need nifty fonts or CMYK. I've tried the Gimp "Despeckle" filter but it doesn't work like I want it too. Not even close.
Commercial software is often worth the price.
The little things in Photoshop make the difference. For example, Photoshop has a filter called "Remove Dust & Scratches". I'm doing photo restoration for my family pictures. That filter is absolutely amazing. Nothing compares in the Gimp. Sure, anything is possible using the Gimp, however with Photoshop, it just takes 5 seconds as compared to 30 minutes to remove scratches cleanly.
I highly recommend the ECS K7S5A. Its never ever crashed on me and I've got both the on-board Audio and LAN going on linux just fine. I only paid $65 for it!
EPIC is indeed Intel-speak for VLIW...sort of. True VLIW is where everything is statically scheduled. EPIC relaxes this. It groups independent operations into a single instruction word. The processor core is then allowed to schedule this however it wants.
No, you are wrong. The instruction set is irrelevant. IA-32 is dynamically converted to a RISC-like language. Anti-dependencies are dynamically renamed. There are dozens of physical registers for which many instances of the logical registers can be renamed if there is no true dependence.
What you are suggesting is to move to a VLIW-type instruction set. VLIW is exactly specifying what function unit does what, in the instruction itself. DSP instruction sets are VLIW-like. So is the MMX extensions (sort of). IA-64 is doing exactly this except that it is allowing some dynamic behavior instead of everything being static which is VLIW-purism.
Yeah...I was being a little arrogant. I see far too many posts on Slashdot about people talking out of their asses.
My undergrad architecture course was pretty basic and didn't cover anything advanced like SMT, multiple issue, OOO cores, speculation, etc.
I looked over your prof's website and will read some of his papers when I get a chance.
I sort of agree with him. If an SMT processor runs 2 threads from different processes, how will it handle virtual memory? I suppose you could implement an SMT processor with separate TLB tables...that would be the only way. I wonder if Hyperthreading does this?? Otherwise running 2 threads from different processes won't work.
Oh, and from my knowledge, most UNICES flush the TLB on a context-switch. Someone please correct me if I'm wrong. I do realize that some architectures (MIPS R10000 I believe) can associate a PID with each TLB entry. This would also be a solution to the problem.