Cycle counts died with super-scaler architectures. This is true for both CISC and RISC machines.
Its now all about latencies and throughputs. Let me explain.
The x86 line of processors has an "add reg, reg" instruction. On an AMD64 in the naive world of cycle counts, this instruction takes 3 clock cycles. In the specs you will see that its "latency" is 3 cycles, and that exactly means that from the time this instructions operands are ready, that the result will be ready exactly 3 cycles later.
But an AMD64 is not limited to 1 addition every 3 clock cycles. The throughput for this instruction is actualy 3. The AMD64 can execute 3 additions per clock cycle.
This is explained based on two factors.
First, realize that the addition itself only takes 1 clock cycle.. the other two cycles are in setting up and retiring the execution of it. Both setup and retirement are independent steps, such that immediately after the integer execution unit performs its 1-cycle addition it can perform another one. The execution unit can handle 1 addition every clock cycle, even though the overall latency of addition is 3 cycles
Second, realize that the AMD64 has 3 integer execution units. That each of these execution units can handle 1 addition per clock cycle.
Thus, as the tech specs will tell you (or Agner Fog's fine docs), this addition operation has a 3 cycle latency but a 3-additions-cycle throughput.
Now back to CISC vs RISC. It seems "obvious" that both should be able to benefit from super-scaler architecture, and that is true.
The problem for RISC is that the execution time of a piece of code is the sum of the latencies of the instructions in the longest dependency chain. (a dependency chain is a sequence of instructions where each depends on the result of the previous instruction.) RISC architectures were able to crank up the MHZ by simply making the latency of an instruction longer. So while a typical 200mhz Pentium 2 latency was 3 cycles, it was 6 for the 400mhz Alpha.
In real world scenarios, there is an awfull lot of dead time where only a single dependency chain is left, and thats where most of the benefits of RISC evaporates. Both the 200mhz CISC and the 400mhz RISC machine each took the same real-world time to execute that longest dependency chain, because they both used the same sized transistors to perform the same sequence of operations.
This isn't to say that Intel isnt guilty of cranking up the MHZ by extending latencies.. the pentium 4 was basically cranked-up in the same way that RISC machines were cranked up.. Its instruction latencies were much longer than AMD64's, so even though Intel had the MHZ edge, the AMD64 actualy performed on par, and in many ways better, with a lower clock speed.
There are real advantages to RISC (manufacturing cost and power efficiency).. but in real world scenarios, performance was never one of them.
In assembly you had to fetch from memory, operate, write to memory. The number of cycles to do this was greater than on say, a 68K.
liar liar pants on fire.
What you are claiming is that on Intel, you have to do this:
mov eax, [foo]
add eax, ebx
mov [foo], eax
Yeah... you CAN do that... but you can also do:
add [foo], ebx
omg where did your bullshit arguement go? drink the punch much there, fanboi?
The CISC vs RISC debate was never about what you imagine. The reason RISC was so attractive was because it *didn't* have the option of that second method.. that second method is "complex".. its a read/modify/write instruction and risc machines don't do that because RISC machines were all about MHZ.
RISC became popular precisely because they could crank up those MHZ.. the Alpha being the star of the show.
You just declared exactly the opposite of what was going on back then. The PPC failed precisely because it was an underpowered piece of crap. It was RISC without the performance advantage of RISC. The alpha was performance king when it clocked (MHZ) at 1.5x to 2.0x faster than Intel's offerings. The PPC couldnt boast that. The only folks who didnt notice were Apple and their fanbois
Master Charge owns the debts of others
United States owes debts to others
See the difference? Its hard for the government to write off bad debt when its the one thats upside down.
..and whats this about a tax deduction? The solution to the problem of not enough tax revenue, you believe, is to reduce tax revenue further? Did you even think about it before you typed it?
..and whats this about right wingers? Did someone convinced you that this was a political idiology problem (with the 'right' being the 'enemy') rather than a fiscal numbers problem?
The national debt is now over 11 trillion dollars and there hasn't been a real budget surplus since 1969.
The federal government has borrowed money and can not pay the prior lenders back, so they borrow more money in order to do so. This has been going on for so long that only a few people still recognize it as an active ponzi scheme, as is evident by the ignorant people modding my post as a troll.
The interest payments on the debt will crest half a trillion dollars this year, and will be over a trillion/year by the time Obama's 1st term is up. The government will likely borrow more from social security (why not? its doomed anyways) in order to pay off prior lenders as it becomes more and more difficult to find new lenders.
There are other areas where the government is actively ponzi as well, such as medicare (medicare is a bigger problem than social security.)
It wont be possible to take down the pyramids through force (taxation at gunpoint) for much longer. If ponzi could have gone out and robbed some banks, he could have paid off all of his lenders. But imagine the point where even if he robbed every bank in the country he couldn't pay them off. America is pretty much there, right now, at the point of no return.
The problem with testing for 2% error rates on "usability" problems, is that there is a real chance that at least one of your testers is throwing a wrench into the gearts on purpose.
Pick 150 random people from a "representative voter sample" of a population, like say 150 people selected for jury duty in new york city ("you people vote red.. the rest vote blue") and you are going to have more than 2% assholes that want the system to fail..
..and more than 2% drooling retards.. (remember, 50% of the people are below average at whatever it is you are sampling)
The ultimate question is, is the system reasonably foolproof. That is not to say that 100% of fools are accounted for, but rather that a reasonable number of them are.
"Some men you just can't reach. So you get what we had here last week..."
Obviously the problem is way outside of your grasp since you can only fathom using whatever is already bundled with your distro. So much for freedom.
Off the top of your head, what would the name of the best linux statistics package be? Give up yet?
Its called "R", name based on the original "S"
You probably want a GUI for it, so you have to find something like "PMG" or "RKWARD" (probably awkward to use, right?)
Maybe you want an IDE for it, so maybe "Bluefish", "Vim", or "Geany"
Another thing that Linux is seriously in need of is DESCRIPTIVE SOFTWARE NAMES.
GIMP - This is WORSE than non-descriptive. Who wants to try out software thats been gimped? Sounds like crippleware to me.
VLC - I want a media player.. this thing has something to do with networking.
Sage - I need a math program, not a green spice.
This can go on and on..
Still waiting for one called LINGAA (Linux Is Not Good At Acronyms)
You can run one wire to each town, to just one location in town, and wirelessly connect the whole town..
...should be significantly cheaper than running wire to every home.
A decent wireless broadband service will eventualy come.. likely to be from the cell services, since they already have sites well located and contracted: they just need the proper equipment and the incentive to do it.
This all reminds me of a crazy SSD video, which unfortunately I cannot link to (bad memory), where the chief engineer of one of these SSD companies demonstrates the I/O performance of some sort of super-raid0 SSD board with like 16 or 32 flash modules on it...
With this evil-genius grin on his face, after showing the benchmarks in some popular benchmarking program (bad memory again), he says something like "thats ten thousand IOPS per second", then he raises his eyebrows and then repeats it "ten thousands IOPS per second"
I got excited and spooked out all at the same time.
They have already gotten literally millions of free player-created "creatures." The cost of creating all that content would have easily been hundreds of millions of dollars if they hired modelers/artists to make it.
You can bet anything that in the next year or two, E.A. will release games that are not anything like Spore, but will leverage the player created art they got for free from Spore players.
"Thats my Ososaplopicus!! What the hell is it doing in StupidGame 2010, and why did they rename it HaHaHaSucker?"
Will sacrificed his original goal, that of something similar to evolution, for a toy completely unrelated to evolution.
Remember all of Will's talks about the high cost of content and how the game industry needed to change?
He had lead everyone to believe that "procedural content" was the goal, when the real goal was apparently to get millions of people to create content for free: all your creature designs are owned by E.A and are automatically uploaded to their servers.
And that explain the evolution sacrifice nicely, as well as the "free" creature creator. They wanted no limits to creativity in order to maximize the diversity and quantity of the free content they were going to aquire.
The binaries executed via my emulator are just source code too.
In fact, the modern x86 itself is simply an interpreter (based in hardware) which translates programs into a RISC language which can be natively executed.
Let me be clear. Your idea of "compiled programs" is call machine code, also known as machine-readable source code.
There is no way for anyone starting EVE today to ever catch up to those who started a year ago, and those that started a year ago will never catch up to those who started two years ago,.. and so on.
I played eve for nearly 6 months. When a big content patch came out that was essentialy ships I wouldn't be able to fly for at least another 6 months, yet I had to compete directly against one (freighters), I decided the system sucked more than I had already suspected.
I was a very successfull high security hauler and trader who had his market taken away by superships he could afford to buy, but could not fly... I had to train another 10 or so skills (one of which taking almost two months to train all by itself)
Basically, I had to pay them $15 x 6months = $90 in order to continue my trading career, and thats assuming that a new update wasn't going to again push the bar even further away from me.
EVE, mostly a great game, but the leveling/training system needs to go.
Sure, and Apple had the same sort of school contracts here in America at that same time.
Motorola was far and away the king of desktops back then.
Apple Macintish, Atari ST, and Amiga were all using Motorola 68000 processors.
Second-chairs were Zilog and MOS processors, used in coleco consoles, atari consoles, commodores prior to the amiga, as well as tandy's and the apple's prior to the macintosh.
Slashdot Mirror
Cycle counts died with super-scaler architectures. This is true for both CISC and RISC machines.
.. the pentium 4 was basically cranked-up in the same way that RISC machines were cranked up .. Its instruction latencies were much longer than AMD64's, so even though Intel had the MHZ edge, the AMD64 actualy performed on par, and in many ways better, with a lower clock speed.
.. but in real world scenarios, performance was never one of them.
Its now all about latencies and throughputs. Let me explain.
The x86 line of processors has an "add reg, reg" instruction. On an AMD64 in the naive world of cycle counts, this instruction takes 3 clock cycles. In the specs you will see that its "latency" is 3 cycles, and that exactly means that from the time this instructions operands are ready, that the result will be ready exactly 3 cycles later.
But an AMD64 is not limited to 1 addition every 3 clock cycles. The throughput for this instruction is actualy 3. The AMD64 can execute 3 additions per clock cycle.
This is explained based on two factors.
First, realize that the addition itself only takes 1 clock cycle.. the other two cycles are in setting up and retiring the execution of it. Both setup and retirement are independent steps, such that immediately after the integer execution unit performs its 1-cycle addition it can perform another one. The execution unit can handle 1 addition every clock cycle, even though the overall latency of addition is 3 cycles
Second, realize that the AMD64 has 3 integer execution units. That each of these execution units can handle 1 addition per clock cycle.
Thus, as the tech specs will tell you (or Agner Fog's fine docs), this addition operation has a 3 cycle latency but a 3-additions-cycle throughput.
Now back to CISC vs RISC. It seems "obvious" that both should be able to benefit from super-scaler architecture, and that is true.
The problem for RISC is that the execution time of a piece of code is the sum of the latencies of the instructions in the longest dependency chain. (a dependency chain is a sequence of instructions where each depends on the result of the previous instruction.) RISC architectures were able to crank up the MHZ by simply making the latency of an instruction longer. So while a typical 200mhz Pentium 2 latency was 3 cycles, it was 6 for the 400mhz Alpha.
In real world scenarios, there is an awfull lot of dead time where only a single dependency chain is left, and thats where most of the benefits of RISC evaporates. Both the 200mhz CISC and the 400mhz RISC machine each took the same real-world time to execute that longest dependency chain, because they both used the same sized transistors to perform the same sequence of operations.
This isn't to say that Intel isnt guilty of cranking up the MHZ by extending latencies
There are real advantages to RISC (manufacturing cost and power efficiency)
In assembly you had to fetch from memory, operate, write to memory. The number of cycles to do this was greater than on say, a 68K.
liar liar pants on fire.
.. its a read/modify/write instruction and risc machines don't do that because RISC machines were all about MHZ.
.. the Alpha being the star of the show.
What you are claiming is that on Intel, you have to do this:
mov eax, [foo]
add eax, ebx
mov [foo], eax
Yeah... you CAN do that... but you can also do:
add [foo], ebx
omg where did your bullshit arguement go? drink the punch much there, fanboi?
The CISC vs RISC debate was never about what you imagine. The reason RISC was so attractive was because it *didn't* have the option of that second method.. that second method is "complex"
RISC became popular precisely because they could crank up those MHZ
You just declared exactly the opposite of what was going on back then. The PPC failed precisely because it was an underpowered piece of crap. It was RISC without the performance advantage of RISC. The alpha was performance king when it clocked (MHZ) at 1.5x to 2.0x faster than Intel's offerings. The PPC couldnt boast that. The only folks who didnt notice were Apple and their fanbois
..because its linux.
Its supposed to be better than windows.. and doesnt it have a windows emulator?
This is not your grandmothers OS (yet.)
You got it backwards.
..and whats this about a tax deduction? The solution to the problem of not enough tax revenue, you believe, is to reduce tax revenue further? Did you even think about it before you typed it?
..and whats this about right wingers? Did someone convinced you that this was a political idiology problem (with the 'right' being the 'enemy') rather than a fiscal numbers problem?
Master Charge owns the debts of others
United States owes debts to others
See the difference? Its hard for the government to write off bad debt when its the one thats upside down.
It isnt just social security...
The national debt is now over 11 trillion dollars and there hasn't been a real budget surplus since 1969.
The federal government has borrowed money and can not pay the prior lenders back, so they borrow more money in order to do so. This has been going on for so long that only a few people still recognize it as an active ponzi scheme, as is evident by the ignorant people modding my post as a troll.
The interest payments on the debt will crest half a trillion dollars this year, and will be over a trillion/year by the time Obama's 1st term is up. The government will likely borrow more from social security (why not? its doomed anyways) in order to pay off prior lenders as it becomes more and more difficult to find new lenders.
There are other areas where the government is actively ponzi as well, such as medicare (medicare is a bigger problem than social security.)
Check out (U.S. Government Accounting Office) http://www.gao.gov/cghome/d08446cg.pdf to see how grim the pyramid really is.
It wont be possible to take down the pyramids through force (taxation at gunpoint) for much longer. If ponzi could have gone out and robbed some banks, he could have paid off all of his lenders. But imagine the point where even if he robbed every bank in the country he couldn't pay them off. America is pretty much there, right now, at the point of no return.
Just as the government learned it by watching Ponzi, these youtubers learned it by watching the government.
It doesnt look to me like Tesla focused on 'flash'
It looks to me like the focused on filling a demand
If 'demand' = 'flash' then you've got some explaining to do as to why you think that thats a bad thing.
What are all the 'affordable' electric vehicles that you are speaking about? Golf carts?
The indy game devs are kicking some ass.
Portal is one example, another is World of Goo.
Left for Dead isn't exactly cookie cutter either. Sure there have been other survival horror games, but L4D really made it blossom.
Of course, some of my favorite games (like Mindrover) were never a big hit.. so maybe my opinion isnt exactly representative of the market.
The problem with testing for 2% error rates on "usability" problems, is that there is a real chance that at least one of your testers is throwing a wrench into the gearts on purpose.
..and more than 2% drooling retards.. (remember, 50% of the people are below average at whatever it is you are sampling)
Pick 150 random people from a "representative voter sample" of a population, like say 150 people selected for jury duty in new york city ("you people vote red.. the rest vote blue") and you are going to have more than 2% assholes that want the system to fail..
The ultimate question is, is the system reasonably foolproof. That is not to say that 100% of fools are accounted for, but rather that a reasonable number of them are.
"Some men you just can't reach. So you get what we had here last week..."
Is english your third language?
..you forgot...
"..while reading J.K. Rowlings Harry Potter and the Philosophers Stone, published by Bloomsbury, on my Amazon Kindle"
"We" were never talking about application friendly menus.
..and if you don't know what R is, then maybe its because it has a bad name. You know what Mathematica is, right?
I am not sure why, but you talked about them in direct reply to me.
Was I supposed to know thad you were out in left field talking to yourself?
Obviously the problem is way outside of your grasp since you can only fathom using whatever is already bundled with your distro. So much for freedom.
Off the top of your head, what would the name of the best linux statistics package be? Give up yet?
Its called "R", name based on the original "S"
You probably want a GUI for it, so you have to find something like "PMG" or "RKWARD" (probably awkward to use, right?)
Maybe you want an IDE for it, so maybe "Bluefish", "Vim", or "Geany"
Another thing that Linux is seriously in need of is DESCRIPTIVE SOFTWARE NAMES.
GIMP - This is WORSE than non-descriptive. Who wants to try out software thats been gimped? Sounds like crippleware to me.
VLC - I want a media player.. this thing has something to do with networking.
Sage - I need a math program, not a green spice.
This can go on and on..
Still waiting for one called LINGAA (Linux Is Not Good At Acronyms)
You can run one wire to each town, to just one location in town, and wirelessly connect the whole town..
...should be significantly cheaper than running wire to every home.
A decent wireless broadband service will eventualy come.. likely to be from the cell services, since they already have sites well located and contracted: they just need the proper equipment and the incentive to do it.
They (google) don't have to do anything other than take your stuff down. They have no duty to rehost it just because you countered.
They may still submit a DMCA takedown notice againt it, and google will still take it down...
..and now what do you do? Get a laywer? Cry? yeah...
Linux, all about freedom, right?
Yet here you are, saying "but we don't want that."
This all reminds me of a crazy SSD video, which unfortunately I cannot link to (bad memory), where the chief engineer of one of these SSD companies demonstrates the I/O performance of some sort of super-raid0 SSD board with like 16 or 32 flash modules on it...
With this evil-genius grin on his face, after showing the benchmarks in some popular benchmarking program (bad memory again), he says something like "thats ten thousand IOPS per second", then he raises his eyebrows and then repeats it "ten thousands IOPS per second"
I got excited and spooked out all at the same time.
Of course, this really doesnt matter.
They have already gotten literally millions of free player-created "creatures." The cost of creating all that content would have easily been hundreds of millions of dollars if they hired modelers/artists to make it.
You can bet anything that in the next year or two, E.A. will release games that are not anything like Spore, but will leverage the player created art they got for free from Spore players.
"Thats my Ososaplopicus!! What the hell is it doing in StupidGame 2010, and why did they rename it HaHaHaSucker?"
Will sacrificed his original goal, that of something similar to evolution, for a toy completely unrelated to evolution.
Remember all of Will's talks about the high cost of content and how the game industry needed to change?
He had lead everyone to believe that "procedural content" was the goal, when the real goal was apparently to get millions of people to create content for free: all your creature designs are owned by E.A and are automatically uploaded to their servers.
And that explain the evolution sacrifice nicely, as well as the "free" creature creator. They wanted no limits to creativity in order to maximize the diversity and quantity of the free content they were going to aquire.
The binaries executed via my emulator are just source code too.
In fact, the modern x86 itself is simply an interpreter (based in hardware) which translates programs into a RISC language which can be natively executed.
Let me be clear. Your idea of "compiled programs" is call machine code, also known as machine-readable source code.
Why aren't they programs?
Be very careful now. You may accidentally define scripts as non-programs, pcode such as java, and maybe even binaries run under an emulator.
My beef with EVE is the leveling system.
.. and so on.
There is no way for anyone starting EVE today to ever catch up to those who started a year ago, and those that started a year ago will never catch up to those who started two years ago,
I played eve for nearly 6 months. When a big content patch came out that was essentialy ships I wouldn't be able to fly for at least another 6 months, yet I had to compete directly against one (freighters), I decided the system sucked more than I had already suspected.
I was a very successfull high security hauler and trader who had his market taken away by superships he could afford to buy, but could not fly... I had to train another 10 or so skills (one of which taking almost two months to train all by itself)
Basically, I had to pay them $15 x 6months = $90 in order to continue my trading career, and thats assuming that a new update wasn't going to again push the bar even further away from me.
EVE, mostly a great game, but the leveling/training system needs to go.
Sure, and Apple had the same sort of school contracts here in America at that same time.
Motorola was far and away the king of desktops back then.
Apple Macintish, Atari ST, and Amiga were all using Motorola 68000 processors.
Second-chairs were Zilog and MOS processors, used in coleco consoles, atari consoles, commodores prior to the amiga, as well as tandy's and the apple's prior to the macintosh.