CPU Speed - can someone explain to a numpty?
Discussion
I know a little about how CPUs work, and can appreciate that different designs of CPU get more done in each beat of the clock, but can someone explain this more thoroughly to me please, or reccomend a good website that explains it all?
What prompted my question was that I've just got a new PC that runs an Intel Duo Core processor at 2.40Ghz. I also use a friend's 2Ghz Intel laptop for recording music. When we've finished a song, we have to 'bounce it down' to disk, which is quite a singuarly processor intensive task. The 2Ghz laptop chugs away for an hour, progress bar steadily advancing, before it usually blue screens! I think it once finished after about an hour and a half. My new PC takes 90 seconds to complete the entire task. Now, obviously, the actual clock speed isn't that different, but as you can see, it is at least 50-100 times quicker. I should add that the RAM on my new PC is significantly quicker as well (800Mhz dual channel).
What prompted my question was that I've just got a new PC that runs an Intel Duo Core processor at 2.40Ghz. I also use a friend's 2Ghz Intel laptop for recording music. When we've finished a song, we have to 'bounce it down' to disk, which is quite a singuarly processor intensive task. The 2Ghz laptop chugs away for an hour, progress bar steadily advancing, before it usually blue screens! I think it once finished after about an hour and a half. My new PC takes 90 seconds to complete the entire task. Now, obviously, the actual clock speed isn't that different, but as you can see, it is at least 50-100 times quicker. I should add that the RAM on my new PC is significantly quicker as well (800Mhz dual channel).
ThatPhilBrettGuy said:
Is the laptop hardisk light on all the time? If so it may be constantly swapping to to lack of RAM. If not sounds like the laptop is overheating and the CPU is clocking itself right down to not melt. Certainly something wrong with it.
I can't remember to be honest, I'll have a look. The file produced from this activity is about 300Mb, and the laptop has 1Gb of RAM and is running Windows XP. Obviously, it may still be using HDD swap space, so that may be it. To be honest, the laptop, with its tiny screen any puny processing power, isn't really suitable for music processing. My desktop, in contrast, is a joy to use. The difference in CPU power just interested me.
Also, what is almost universally accepeped as the fastest consumer CPU on the market (Intel Duo Core 6800) has a clock speed of 2.93Ghz. Interesting, as I'm sure we got up above that in the previous generation of Intels. Can anyone explain? I assumed that without special prallel processing code (is this called 'Hyper Threading' nowadays?), most applications just used one processor.
Certainly sound like you have a problem with the laptop as that is an excessive amount of time even for a 300MB file. When you say 'Bounce it down to disc' what exactly do you mean? I assume you are recording the music directly to the PC, but then are you converting it to a different format? Either way should only take a few minutes even with such a large file.
Quite simply CPU's complete 1 instruction per clock cycle so depending on how many clock cycles per second there are (GHZ) then theoretically the more information can be processed. This isnt the total sum of it as of course you can get 32bit & 64bit processors as well as different instruction sets, not to mention the different levels of cache and now of course dual core processors. However, as a general rule of thumb you wont be a million miles away.
Quite simply CPU's complete 1 instruction per clock cycle so depending on how many clock cycles per second there are (GHZ) then theoretically the more information can be processed. This isnt the total sum of it as of course you can get 32bit & 64bit processors as well as different instruction sets, not to mention the different levels of cache and now of course dual core processors. However, as a general rule of thumb you wont be a million miles away.
As has been said, the laptop is almost certainly running out of memory. Either because it doesn't have enough in the first place or there's loads of other stuff running in the background and using it all up. 1Gb is adequate for Win XP under normal use, you should get away with it if the system is clean.
Having to swap stuff to disk is the performance killer on modern PCs, processors and memory are all so fast these days and disks are so slow (relatively). I'd almost always choose more RAM over a faster CPU or bigger harddisk when speccing a PC as it hurts performance SOOOO much if you run out.
If you look in Task Manager at the performance tab, check the total physical memory against the 'commit charge' total value while the machine is running slowly (or just check the peak value after its finished). If the commit charge is significantly bigger, you're losing huge amounts of performance from running out of RAM. Buy more or run less stuff
Having to swap stuff to disk is the performance killer on modern PCs, processors and memory are all so fast these days and disks are so slow (relatively). I'd almost always choose more RAM over a faster CPU or bigger harddisk when speccing a PC as it hurts performance SOOOO much if you run out.
If you look in Task Manager at the performance tab, check the total physical memory against the 'commit charge' total value while the machine is running slowly (or just check the peak value after its finished). If the commit charge is significantly bigger, you're losing huge amounts of performance from running out of RAM. Buy more or run less stuff
Ok.
The interesting thing you're encounting shows the different between laptop and desktop machines. Even if both the desktop and laptop had the same generation of CPU in them (Core Duo / Core 2 Duo / P4-M / Whatever) the desktop will generally be faster than the laptop. More so in something disk intensive.
Laptop drives, simply put are designed to be survivable and low power. Speed whilst nice is not the highest priority.
The interesting thing you're encounting shows the different between laptop and desktop machines. Even if both the desktop and laptop had the same generation of CPU in them (Core Duo / Core 2 Duo / P4-M / Whatever) the desktop will generally be faster than the laptop. More so in something disk intensive.
Laptop drives, simply put are designed to be survivable and low power. Speed whilst nice is not the highest priority.
road2ruin said:
Certainly sound like you have a problem with the laptop as that is an excessive amount of time even for a 300MB file. When you say 'Bounce it down to disc' what exactly do you mean? I assume you are recording the music directly to the PC, but then are you converting it to a different format? Either way should only take a few minutes even with such a large file.
Quite simply CPU's complete 1 instruction per clock cycle so depending on how many clock cycles per second there are (GHZ) then theoretically the more information can be processed. This isnt the total sum of it as of course you can get 32bit & 64bit processors as well as different instruction sets, not to mention the different levels of cache and now of course dual core processors. However, as a general rule of thumb you wont be a million miles away.
Quite simply CPU's complete 1 instruction per clock cycle so depending on how many clock cycles per second there are (GHZ) then theoretically the more information can be processed. This isnt the total sum of it as of course you can get 32bit & 64bit processors as well as different instruction sets, not to mention the different levels of cache and now of course dual core processors. However, as a general rule of thumb you wont be a million miles away.
The bouncing to disk takes all your components WAV files (for our songs, typically around 10 to 15), applies any post processing (levels, effects, etc) and creates one, single, large wav file. It is well known for being very processor intensive, and my book on the software says that over half an hour is common for bouncing a file in the highest quality.
Clock speed is only one part of the overall speed of a computer. Clock speed is only useful when comparing two otherwise identical CPUs. A Core2Duo CPU is very, very different to an older Netburst (Pentium IV) or even non CoreDue based laptop CPU. It can acheive much more per clock cycle. In real terms a 2Ghz Core2Duo is over twice as fast as a 3.6Ghz Pentium IV.
In addition the Core2Duo is dual core (basically 2 whole CPUs). It sounds like the laptop CPU is probably only single core.
As noted above desktop drives are much faster than laptop one, your desktop probably has much faster RAM (and may well have more)...
In addition the Core2Duo is dual core (basically 2 whole CPUs). It sounds like the laptop CPU is probably only single core.
As noted above desktop drives are much faster than laptop one, your desktop probably has much faster RAM (and may well have more)...
robbieduncan said:
Clock speed is only one part of the overall speed of a computer. Clock speed is only useful when comparing two otherwise identical CPUs. A Core2Duo CPU is very, very different to an older Netburst (Pentium IV) or even non CoreDue based laptop CPU. It can acheive much more per clock cycle. In real terms a 2Ghz Core2Duo is over twice as fast as a 3.6Ghz Pentium IV.
Now we're talking. What exactly is it that makes the Core2Duo so much faster than a P4, even when running in single processor mode?
RobM77 said:
robbieduncan said:
Clock speed is only one part of the overall speed of a computer. Clock speed is only useful when comparing two otherwise identical CPUs. A Core2Duo CPU is very, very different to an older Netburst (Pentium IV) or even non CoreDue based laptop CPU. It can acheive much more per clock cycle. In real terms a 2Ghz Core2Duo is over twice as fast as a 3.6Ghz Pentium IV.
Now we're talking. What exactly is it that makes the Core2Duo so much faster than a P4, even when running in single processor mode?
Advancments in CPU design (tweaking cache size / location on die, etc.) and refinments in instruction sets as well as better refinment of the manufacturing processes used.
tank slapper said:
I have just put together a machine using a Core 2 Duo, and am well impressed with how fast it is. Much better then I expected it to be.
Not played to much but the laptop side of things (Core Duo & Core Solo) but apparently the chips are an amazing leap ahead of the old P4-M stuff. Apparently the Core 2 Duo's are equally impressive.
I wonder how AMD's X2 systems perform head too head as Intel have really stolen the limelight with their dual core chips.
thepassenger said:
tank slapper said:
I have just put together a machine using a Core 2 Duo, and am well impressed with how fast it is. Much better then I expected it to be.
Not played to much but the laptop side of things (Core Duo & Core Solo) but apparently the chips are an amazing leap ahead of the old P4-M stuff. Apparently the Core 2 Duo's are equally impressive.
I wonder how AMD's X2 systems perform head too head as Intel have really stolen the limelight with their dual core chips.
Well, apparently they don't. Intel's new dual core chips are firmly ahead of AMD's best. I have no allegiances either way and Intel was an easy choice.
The Core Duos are not hugely more advanced than the Pentium M, as they are in fact based on them. The Core 2 Duos have there heretage in the mobile Pentium M, which was designed to be effecent at low clock speeds thus consuming little power.
The Key thing which effects CPU effecency is the length of the pipeline, i.e. how many stages the processor splits each operation into. A long pipeline means each invidual operation is broken down into more stages and takes more CPU cycles to complete. However the idea is that you can stack up operations like baseball players on bases, so when you have all bases full the throughput is high. While on the surface having a CPU with a long pipeline seems like a very bad idea as the effecency is initally low. Extending the pipeline allows for much higher clock speeds.
The P4 was designed to scale much faster than Intel eventually managed, i've read 8-10GHz a number of times as the original end point. Another problem with the P4's long pipeline or NetBurst architechture is that in some situations you need to know the result of the previous instruction/s before you can know what the next instruction is. Obviously if you are still computing it (P4's pipeline has something like 20 stages
) you only have two was to deal with it; Wait or Guess. The P4 guess's and due to it's extremely cleaver branch prediction it is amazingly accuate. However if it's wrong it gets it very wrong and can have to recalculate huge amounts of instructions.
Obviously this is very off the top of my head and i *may* have got things the wrong way round here n there
also i think i've interchangably used instructions and operations, sorry if this confusses.
If you want to read more most of the hardware sites explain this kinda stuff when they review new CPUs
anandtech
xbitlabs
penstar system
tomshardware
hexus
techreport
they all cover the same things to a greater or less extent.
The Key thing which effects CPU effecency is the length of the pipeline, i.e. how many stages the processor splits each operation into. A long pipeline means each invidual operation is broken down into more stages and takes more CPU cycles to complete. However the idea is that you can stack up operations like baseball players on bases, so when you have all bases full the throughput is high. While on the surface having a CPU with a long pipeline seems like a very bad idea as the effecency is initally low. Extending the pipeline allows for much higher clock speeds.
The P4 was designed to scale much faster than Intel eventually managed, i've read 8-10GHz a number of times as the original end point. Another problem with the P4's long pipeline or NetBurst architechture is that in some situations you need to know the result of the previous instruction/s before you can know what the next instruction is. Obviously if you are still computing it (P4's pipeline has something like 20 stages
) you only have two was to deal with it; Wait or Guess. The P4 guess's and due to it's extremely cleaver branch prediction it is amazingly accuate. However if it's wrong it gets it very wrong and can have to recalculate huge amounts of instructions. Obviously this is very off the top of my head and i *may* have got things the wrong way round here n there
also i think i've interchangably used instructions and operations, sorry if this confusses.
If you want to read more most of the hardware sites explain this kinda stuff when they review new CPUs
anandtech
xbitlabs
penstar system
tomshardware
hexus
techreport
they all cover the same things to a greater or less extent.
mr_yogi said:
Pipeline length
Quite so. The length of the instruction pipeline was the reason why the Althon seemed to perform so well against the early P4's and why when intel stepped on the accelerator the P4 would nudge ahead.
I belive the P4's instruction cache (for want of a term) was 20 long and the Athlon 12 back in those days? So not much of a difference numerically but...
I also seem to recal that the original P4-M development was done by not the 'main dev teams' but by a small group of engineers who thought the P3 had more life in it. End result a hybrid of the P3 and P4 chip dubbed Mobile Pentium 4. Although this hybrid bit could well be a urban myth.
The p4-M was done by Intels Isreal team, was a non mainstream projectthat ended up better than the old P4 core & became the core2duo.
The P4's longer pipeline did help it acheive higher clock rates though at a cost, less work typicaly done per clock cycle, the branch prediction done just that, predicted what parts of the code could be loaded simultaniusly (rather than predict results). Initialy it was well styfled because Intel couldnt afford the silicon real estate for the cache nessecary to make it as quick as it could be, and the memory bandwidth it needed just wasnt there (unlike the Athlon which was far less bandwidth restricted).
The new Core2duo's are cracking CPU's by all accounts and the first in an age (apart from the p4-m) that I'd consider using.
As to the Op's laptop as many have suggested, it sounds like a swapping issue, the CPU's arnt 100 times different in performance, thats the sort of performance dropdown when you go from main memory to hard drive.
The P4's longer pipeline did help it acheive higher clock rates though at a cost, less work typicaly done per clock cycle, the branch prediction done just that, predicted what parts of the code could be loaded simultaniusly (rather than predict results). Initialy it was well styfled because Intel couldnt afford the silicon real estate for the cache nessecary to make it as quick as it could be, and the memory bandwidth it needed just wasnt there (unlike the Athlon which was far less bandwidth restricted).
The new Core2duo's are cracking CPU's by all accounts and the first in an age (apart from the p4-m) that I'd consider using.
As to the Op's laptop as many have suggested, it sounds like a swapping issue, the CPU's arnt 100 times different in performance, thats the sort of performance dropdown when you go from main memory to hard drive.
RobM77 said:
robbieduncan said:
Clock speed is only one part of the overall speed of a computer. Clock speed is only useful when comparing two otherwise identical CPUs. A Core2Duo CPU is very, very different to an older Netburst (Pentium IV) or even non CoreDue based laptop CPU. It can acheive much more per clock cycle. In real terms a 2Ghz Core2Duo is over twice as fast as a 3.6Ghz Pentium IV.
Now we're talking. What exactly is it that makes the Core2Duo so much faster than a P4, even when running in single processor mode?
Complex question. You can read about it here: http://arstechnica.com/articles/paedi
The mobile P4 and Pentium M are two TOTALLY different designs, the mobile P4 is a lower power version of the P4 where as the Pentium M is in fact based off the Petium 3. The P4 was designed by Intels Main CPU design centre (or one of), whereas the Pentium M was designed by Intel's Israeli team. So it's a bit of luck that they had the Pentium M when the P4 hit the wall.
Somethng that hasnt been mentioned, the actual reason for having pipelines in the first place is cause relative to the speed of the core of the processor it takes frikkin ages to get instructions in from ram, during which time the chip would be sat there doing jack for hundreds of cycles inbetween instructions. So you queue things up ahead of time to keep it fed. But it all gets very hard when you're trying to guess ahead at whats coming next to keep the pipes full, since whats coming next could depend on the outcome of something that hasnt happened yet.. 
of course thats just one small part of the whole thing. The ARS technica link is good, you can read quite a bit around the subject on that site if it interests you.
of course thats just one small part of the whole thing. The ARS technica link is good, you can read quite a bit around the subject on that site if it interests you.
Gassing Station | Computers, Gadgets & Stuff | Top of Page | What's New | My Stuff