"My Pentium 200Mhx MMX Windows machine with a good PCI graphics card gives me an acceptable game of Quake III, while my K6-2 3DNow 450 Mhx Linux box with a crappy video card is too slow to play. I suspect that a lot of money is spent on useless "upgrades" for new chips. "
I'm glad that you are happy with your p200mmx, but many many gamers are not. Your implication that a good video card is more important than a good cpu is bunk. It was common to think that way when we made the jump up from software rendering to our first 3d accelerated card (not counting Virge chips). But it just isn't true today. There is a balance between having a video card that isn't limiting the cpu and having a cpu that isn't limiting the video card.
You are also comparing different operating systems in your example, which very much biases the comparison in favor Windows, until we get the next release of Xfree86 and better drivers for video cards (non 3dfx).
PC's may have a 100 MHz bus to memory, but the pci bus is on a 1/3 divider. The PCI bus runs at 33 MHz (This is true even for 66 MHz and 133 Mhz busses, they just use 1/2 and 1/4 dividers).
The PC PCI bus has approximately 1056 megabits/second of bandwidth. 32 bits per cycle * 33 million cycles per second.
a Gigahertz ethernet card will only transfer one bit of information per clock cycle. With one billion clock cycles per second, you get 1000 megabits/second of bandwidth. So, at least in theory, a PC PCI bus can handle a gigabit card.
I'm not sure if each peripheral on the PCI bus has its own dedicated bandwidth. I would guess they have to share it. Which would mean the bus may choke off the card before it can hit One gigabit per second, by servicing other peripherals (hard disk, video card).
I wish moderators would follow the guidelines and spend more points moderating up what they like, and less moderating down. IMO, moderating down should only be for worthless posts. "Redundant" shouldn't even be an option for moderating down. Mods down should be for flamebait, and trolling. Even then, some troll posts should still probably be left alone. Let's mod up the good ones, and create a slashdot that is the intersection of all our likes, instead of one that is a union of all of our dislikes.
$19 for a cd, we must live in different parts of the country. I don't buy CDs like I used to, but I typically pay $12 to $14 for CDs. I recently bought a Metallica two disc live set (S&M - Symphony and Metallica) for around $24.
On a side note, Metallica S&M as well as Apocalyptica really show off the musicality of Metallica. I know many people probably have a knee jerk reaction to Metallica, but if you don't: get your hands on some Apocalyptica. It's a string quintet performing metallica, no vocals. Some truly amazing stuff, beautiful.
Signal 11, come on now. You don't honestly believe that any or all of the powers who try to limit or disallow digital copying don't get this. They are not stupid. Copying involving an analog step will always be around, and is unpreventable. They know this. We know this. Everyone knows it. It's not a secret.
"Methinks the editor should have done some research himself and put some of these in the story."
If you had read all the messages you would have seen that HeUnique was not only reading those links, he posted to the thread asking for more information and their (usb developers) opinion on posting this story to slashdot.
Interestingly, Alan Cox suggested waiting for a week before posting to Slashdot to give them a chance to explain and/or change back graciously before being "hit by a nuclear warhead"
Maybe Slashdot could make some sort of community watchdog section where things that seem suspicious could be posted, but in sort of a beta state before going to the main page.
how indicative of the maturity of some of slashdot's participants. this is left at zero, as all other posts not written by romco, while he gets moderated into oblivion. arguing about the merits of a post is one thing. What you are suggesting is worse for slashdot than his post.
I'm not of a strong political background or opinions, but clearly capitalism taken to an extreme (ie making a buck at all costs) without a doubt disregards ethics. That doesn't mean that all are unethical, just that there is a relationship. People are greedy and want to make a buck, oftentimes they do this at others expense. Trollking makes a good point.
I wonder how many clock cycles are actually wasted waiting on memory access. The high speed caches close to the processor help in this regard quite a bit. It would be interesting to see a simulation comparing how fast software runs on current memory systems to how fast it would run with high speed low latency memory.
There are very real reasons why you aren't going to get ram working at core speeds cheap. One big reason is that there is a huge difference in the numbers of transistors used to store a bit in the high-speed and low latency L1 and L2 caches. I believe that one transistor is used per bit in DRAM. That's pretty cheap. What is the difference in # of transistors between a P3 and a coppermine. Thats all for 256k of L2 cache. The reason that we have all these layers of cache is because it is cost effective and resource effective. Good luck making full speed ram with low latency that is inexpensive. I believe it would require a complete paradigm shift.
A nand is 4 fets, you're right. I was thinking of ttl which is 2 bipolar transistors. I remember a few years ago Intel was using a bicmos process, that i suppose could have both bipolar and mos transistors. I think they may be all mos now tho. Also AMD is mos too I believe, don't quote me tho.
"However, when the Earth's poles reverse, such technologies will cease to function properly."
I'm assuming by reverse you mean the direction of the magnetic field. I'm not familiar with with this, but why would it matter? Why would it be different than say, rotating your new magnetic chip 180 degrees?
Of course up to this point we haven't gotten that one to one ratio. Compare the Athlon and P3 (not the coppermine). The Athlon has approximately 3 times as many transistors as a P3. The athlon is fast, but nowhere near 3 times as fast. It's something like 15-20% faster.
If I understand correctly this is one of the main motivations behind Intel's Merced...err... Itanium. It's an architecture designed to be scaled up as we can make chips with more and more transistors. It's not trivial to parallelize the instruction stream.
5,500 million transistors is a mind boggling number. Think about what the last magnitude jump of a 1000 changed. What could you do with a chip that was 5,000 transistors. That would be what? 2,500 nand gates. Does anyone know how many transistors the 6502 microprocessor or 8088 processor's had?
With a thousand times more transistors, memory bandwidth problems might not be such a big deal. I think on the coppermines the 256k cache takes up around ten million transistors, give or take. That means we could have a coppermine or athlon core with approximately 138 megabytes of L2 cache.
Your budget integrated system could have everything integrated into that chip. Video, sound, I/O. Is it possible to make a chip that you could just plug your keyboard, mouse, monitor into (not the chip literally, but so there is no external logic)?
I've heard the same thing before. And while I am sure there is some truth to it, it just doesn't matter. Cyrix could have made an entire new instruction set and said: if people would just write code for our new instruction set it would be faster than Intel. Obviously they made a chip to compete with Intel. Obviously the code it was going to be running was made to run on the Intel.
I believe the cyrix chip did better when floating point operations were interwoven inbetween integer operations.
I am very skeptical that Quake could have been rewritten to perform comparably on the cyrix. It just didn't have the horsepower.
the fpu of the intel pentium mmx was pretty similar if not identical to the original pentium.
pentium mmx vs pentium: added more L1 cache, 32k vs 16k (i think, been awhile)
added mmx instructions, note that these are integer only
tweaked out some of the superduper scalar out of order supercalifragilistic part of the processor. actually I think the part is called the TLB - translation look ahead buffer.
All in all I think it was around a 10% to 15% increase in performance at a given clock speed. This put the pentium mmx ahead of the 6x86 in integer performance, and smoked it in fp performance.
The 6x86 fpu was not superior to the original pentium. I played many games of quake on my 6x86, and it was quite a bit slower than comparable Intel chips. Sure the integer was good, but the floating point was not. Now Quake isn't a true floating point test, but it is indicative. Trust me, the cyrix fpu was not faster than the pentium.
You are attributing problems to the processor, but I believe your motherboard is a more likely culprit. While the cyrix did have some issues with heat and power, the chip itself wasn't flaky. I've never heard of a K6 being flaky. I bet it's your board or system.
I bought a p150+ pretty early on. While it worked ok, it really wasn't that cheap. I think I paid 150 dollars for it, which at the time could have bought a pentium 120 chip. I wish I had gone with the Intel chip, but I didn't know what a fpu was at that point. Even though it had it's flaws, cyrix was big news when it was released. It was really hyped up, because there was nothing else at that point. AMD K5's were basically a no show. That hype may have hurt it as much as anything, because it wasn't really deserved.
Cyrix has had some problems. One big one with the 6x86 line was that the chips required more current than Intel chips. The motherboards were built to meet the demands of the Intel pentiums. The snafu came from the fact that cyrix was a budget chip, but it didn't work in the inexpensive motherboards. Cheap motherboards aren't going to throw in a good power supply for the hell of it. They have just enough to get by, in this case it was just enough for Intel chips. Up to that point, Intel was the only game in town, so it was understandable. This led to stability problems in Cyrix based systems.
The first system I built was a p150+ (60Mhz bus, 2x multiplier) with a no name HX chipset motherboard. It basically worked ok. When bought a big fan I could overclock it. When I gave it to my mom, I underclocked it to ensure stability.
It was a fine system for running linux and windows 95. It sucked for quake, but it was good for Descent I and II.
I understand now! Thanks stain ain, jabber, bonzo, ralph and dattaway. Building on what you said Ralph. If they make the "cells" very small and the transmit power low, they can use all the bandwidth of that frequency on a small number of users, thereby keeping the enduser bandwidth at a relatively high percentage of the total bandwidth of that frequency (or is it a band of frequencies?).
I read in another article (can't remember if it was here on slashdot) a prediction that wireless would overtake wired phone lines. With this push towards everything being wireless, are there shared bandwidth limitations?
If everyone is on a wireless net connection, are we sharing one big pipe of bandwidth to communicate, like coax, or do we each have a small dedicated piece of bandwidth?
I would imagine this is a problem that has been dealt with (solved?) in cell phone technology.
Mega TF is the mod that let the medics lay down syringes, wasn't it?
I was an avid Team Fortress player back in the day (basically before I started playing Starsiege Tribes). I tried out Mega TF quite a few times, but never really cared for it.
Slashdot Mirror
"My Pentium 200Mhx MMX Windows machine with a good PCI graphics card gives me an acceptable game of Quake III, while my K6-2 3DNow 450 Mhx Linux box with a crappy video card is too slow to play. I suspect that a lot of money is spent on useless "upgrades" for new chips. "
I'm glad that you are happy with your p200mmx, but many many gamers are not. Your implication that a good video card is more important than a good cpu is bunk. It was common to think that way when we made the jump up from software rendering to our first 3d accelerated card (not counting Virge chips). But it just isn't true today. There is a balance between having a video card that isn't limiting the cpu and having a cpu that isn't limiting the video card.
You are also comparing different operating systems in your example, which very much biases the comparison in favor Windows, until we get the next release of Xfree86 and better drivers for video cards (non 3dfx).
PC's may have a 100 MHz bus to memory, but the pci bus is on a 1/3 divider. The PCI bus runs at 33 MHz (This is true even for 66 MHz and 133 Mhz busses, they just use 1/2 and 1/4 dividers).
The PC PCI bus has approximately 1056 megabits/second of bandwidth. 32 bits per cycle * 33 million cycles per second.
a Gigahertz ethernet card will only transfer one bit of information per clock cycle. With one billion clock cycles per second, you get 1000 megabits/second of bandwidth. So, at least in theory, a PC PCI bus can handle a gigabit card.
I'm not sure if each peripheral on the PCI bus has its own dedicated bandwidth. I would guess they have to share it. Which would mean the bus may choke off the card before it can hit One gigabit per second, by servicing other peripherals (hard disk, video card).
I wish moderators would follow the guidelines and spend more points moderating up what they like, and less moderating down. IMO, moderating down should only be for worthless posts. "Redundant" shouldn't even be an option for moderating down. Mods down should be for flamebait, and trolling. Even then, some troll posts should still probably be left alone. Let's mod up the good ones, and create a slashdot that is the intersection of all our likes, instead of one that is a union of all of our dislikes.
$19 for a cd, we must live in different parts of the country. I don't buy CDs like I used to, but I typically pay $12 to $14 for CDs. I recently bought a Metallica two disc live set (S&M - Symphony and Metallica) for around $24.
On a side note, Metallica S&M as well as Apocalyptica really show off the musicality of Metallica. I know many people probably have a knee jerk reaction to Metallica, but if you don't: get your hands on some Apocalyptica. It's a string quintet performing metallica, no vocals. Some truly amazing stuff, beautiful.
Signal 11, come on now. You don't honestly believe that any or all of the powers who try to limit or disallow digital copying don't get this. They are not stupid. Copying involving an analog step will always be around, and is unpreventable. They know this. We know this. Everyone knows it. It's not a secret.
"Methinks the editor should have done some research himself and put some of these in the story."
If you had read all the messages you would have seen that HeUnique was not only reading those links, he posted to the thread asking for more information and their (usb developers) opinion on posting this story to slashdot.
Interestingly, Alan Cox suggested waiting for a week before posting to Slashdot to give them a chance to explain and/or change back graciously before being "hit by a nuclear warhead"
Maybe Slashdot could make some sort of community watchdog section where things that seem suspicious could be posted, but in sort of a beta state before going to the main page.
how indicative of the maturity of some of slashdot's participants. this is left at zero, as all other posts not written by romco, while he gets moderated into oblivion. arguing about the merits of a post is one thing. What you are suggesting is worse for slashdot than his post.
Open source commie?
I'm not of a strong political background or opinions, but clearly capitalism taken to an extreme (ie making a buck at all costs) without a doubt disregards ethics. That doesn't mean that all are unethical, just that there is a relationship. People are greedy and want to make a buck, oftentimes they do this at others expense.
Trollking makes a good point.
I wonder how many clock cycles are actually wasted waiting on memory access. The high speed caches close to the processor help in this regard quite a bit. It would be interesting to see a simulation comparing how fast software runs on current memory systems to how fast it would run with high speed low latency memory.
There are very real reasons why you aren't going to get ram working at core speeds cheap. One big reason is that there is a huge difference in the numbers of transistors used to store a bit in the high-speed and low latency L1 and L2 caches. I believe that one transistor is used per bit in DRAM. That's pretty cheap. What is the difference in # of transistors between a P3 and a coppermine. Thats all for 256k of L2 cache.
The reason that we have all these layers of cache is because it is cost effective and resource effective. Good luck making full speed ram with low latency that is inexpensive. I believe it would require a complete paradigm shift.
A nand is 4 fets, you're right. I was thinking of ttl which is 2 bipolar transistors. I remember a few years ago Intel was using a bicmos process, that i suppose could have both bipolar and mos transistors. I think they may be all mos now tho. Also AMD is mos too I believe, don't quote me tho.
"However, when the Earth's poles reverse, such technologies will cease to function properly."
I'm assuming by reverse you mean the direction of the magnetic field. I'm not familiar with with this, but why would it matter? Why would it be different than say, rotating your new magnetic chip 180 degrees?
Of course up to this point we haven't gotten that one to one ratio. Compare the Athlon and P3 (not the coppermine). The Athlon has approximately 3 times as many transistors as a P3. The athlon is fast, but nowhere near 3 times as fast. It's something like 15-20% faster.
If I understand correctly this is one of the main motivations behind Intel's Merced...err... Itanium. It's an architecture designed to be scaled up as we can make chips with more and more transistors. It's not trivial to parallelize the instruction stream.
5,500 million transistors is a mind boggling number. Think about what the last magnitude jump of a 1000 changed. What could you do with a chip that was 5,000 transistors. That would be what? 2,500 nand gates. Does anyone know how many transistors the 6502 microprocessor or 8088 processor's had?
With a thousand times more transistors, memory bandwidth problems might not be such a big deal. I think on the coppermines the 256k cache takes up around ten million transistors, give or take. That means we could have a coppermine or athlon core with approximately 138 megabytes of L2 cache.
Your budget integrated system could have everything integrated into that chip. Video, sound, I/O. Is it possible to make a chip that you could just plug your keyboard, mouse, monitor into (not the chip literally, but so there is no external logic)?
I've heard the same thing before. And while I am sure there is some truth to it, it just doesn't matter. Cyrix could have made an entire new instruction set and said: if people would just write code for our new instruction set it would be faster than Intel. Obviously they made a chip to compete with Intel. Obviously the code it was going to be running was made to run on the Intel.
I believe the cyrix chip did better when floating point operations were interwoven inbetween integer operations.
I am very skeptical that Quake could have been rewritten to perform comparably on the cyrix. It just didn't have the horsepower.
the fpu of the intel pentium mmx was pretty similar if not identical to the original pentium.
pentium mmx vs pentium:
added more L1 cache, 32k vs 16k (i think, been awhile)
added mmx instructions, note that these are integer only
tweaked out some of the superduper scalar out of order supercalifragilistic part of the processor. actually I think the part is called the TLB - translation look ahead buffer.
All in all I think it was around a 10% to 15% increase in performance at a given clock speed.
This put the pentium mmx ahead of the 6x86 in integer performance, and smoked it in fp performance.
In the Ars article they mention it as an estimate not a given fact.
The 6x86 fpu was not superior to the original pentium. I played many games of quake on my 6x86, and it was quite a bit slower than comparable Intel chips. Sure the integer was good, but the floating point was not. Now Quake isn't a true floating point test, but it is indicative. Trust me, the cyrix fpu was not faster than the pentium.
You are attributing problems to the processor, but I believe your motherboard is a more likely culprit. While the cyrix did have some issues with heat and power, the chip itself wasn't flaky. I've never heard of a K6 being flaky. I bet it's your board or system.
I bought a p150+ pretty early on. While it worked ok, it really wasn't that cheap. I think I paid 150 dollars for it, which at the time could have bought a pentium 120 chip. I wish I had gone with the Intel chip, but I didn't know what a fpu was at that point. Even though it had it's flaws, cyrix was big news when it was released. It was really hyped up, because there was nothing else at that point. AMD K5's were basically a no show. That hype may have hurt it as much as anything, because it wasn't really deserved.
How can a chip with 256k L2 cache (on die) and 64k L1 cache only have 2.2 million transistors?!?!
Athlons have around 20 million or so. I think my celeron is around 8 or 9 million. My pentium mmx was around 3 million.
Just doesn't seem right.
Cyrix has had some problems. One big one with the 6x86 line was that the chips required more current than Intel chips. The motherboards were built to meet the demands of the Intel pentiums. The snafu came from the fact that cyrix was a budget chip, but it didn't work in the inexpensive motherboards. Cheap motherboards aren't going to throw in a good power supply for the hell of it. They have just enough to get by, in this case it was just enough for Intel chips. Up to that point, Intel was the only game in town, so it was understandable. This led to stability problems in Cyrix based systems.
The first system I built was a p150+ (60Mhz bus, 2x multiplier) with a no name HX chipset motherboard. It basically worked ok. When bought a big fan I could overclock it. When I gave it to my mom, I underclocked it to ensure stability.
It was a fine system for running linux and windows 95. It sucked for quake, but it was good for Descent I and II.
I understand now! Thanks stain ain, jabber, bonzo, ralph and dattaway. Building on what you said Ralph. If they make the "cells" very small and the transmit power low, they can use all the bandwidth of that frequency on a small number of users, thereby keeping the enduser bandwidth at a relatively high percentage of the total bandwidth of that frequency (or is it a band of frequencies?).
All in all, a pretty slick system. Thanks guys.
I read in another article (can't remember if it was here on slashdot) a prediction that wireless would overtake wired phone lines. With this push towards everything being wireless, are there shared bandwidth limitations?
If everyone is on a wireless net connection, are we sharing one big pipe of bandwidth to communicate, like coax, or do we each have a small dedicated piece of bandwidth?
I would imagine this is a problem that has been dealt with (solved?) in cell phone technology.
Mega TF is the mod that let the medics lay down syringes, wasn't it?
I was an avid Team Fortress player back in the day (basically before I started playing Starsiege Tribes). I tried out Mega TF quite a few times, but never really cared for it.