Computer bottlenecks

[FishTankX]

I believe that some people on this board (From what I saw in the 6GHZ thread) believe that there is 1 bottleneck in the system. It's actually a complex system of bottlenecks.
This thread is dedicated to Sohcan, who remained calm in the face of insults. You're cool. Espically you're thread on the 64 bit computing, you've done me a great service in terms of dispelling myths.

For clarity, my definition of bottleneck is the thing that is keeping the rest of the system from performing at speed. I.E. Videocard in a Videocard limited situation. The CPU could get much more work done if it wasn't waiting on the videocard. I.E. 2 CPU in RC5 the memory could supply the CPU with work much faster, but the CPU needs to finish the next key before another one can be transfered, resulting in rather large amounts of bus idletime.

The bottleneck hierarchy, goes as follows

1.The CPU is the biggest bottleneck in the system if the CPU is fed quiclky enough and the user can notice that it's not running at absolutley full speed
^
|
2.The memory (bandwidth) is the bottleneck if the CPU cannot be fed fast enough
^
|
3.The videocard is the bottleneck if the CPU spends all of it's time waiting for the videocard to finish drawing frames
^
|
4.The AGP bus is the bottleneck if your videocard is capable of fulfilling the requested operations but the AGP bus isn't capable of transfering the whole workload fast enough
^
|
5.Your memory capacity is the bottleneck if you're unable to cache all of the data needed at hand for the task in RAM.
^
|
6.The PCI bus is a bottleneck if your application is I/O limited, and your I/O devices are capable of fulfilling the load but you're limited by the bus.
^
|
7.It is established that the HDD is the biggest bottleneck in the system, *if* memory isn't able to cache all the data in a program currently running
End of story. Done. Finished.
^
|
8.The CD-ROM/Floppy is the bottleneck in data moving operations involving either or and the HDD is able to keep up

9.There is one ultimate bottleneck in system performance. The Operating system.. If the rest of the system is up to the job but you're waiting on the operating system to work stuff out, it don't matter how fast you're going. It's not going to go much faster unless you get that O/S problem worked out.

Now here are some examples of real life situations in which such things are bottlenecked.
Note:All are capable of being bottlenecked by your wallet or the thing between the keyboard and the chair.
1.Scientific simulation, or trying to install something on devilishly fast system with no DMA turned on.
2.Database situations where the Memory can't keep up with the CPU, but *can* cache the database.
3.1600X1200 UT2K3. on a P4 3.06GHZ on flyby
4.Massive CAD work/Some 3d workstation applications
5.You have hit the limit of X86/OS in terms of memory addressability. I.E. Huge corporate database's running on 2000/XP.
6. Ultra SCSI 320 RAID on a 32/33 PCI bus (A bit idiotic anyways..) Or for a more reasonable example, 3 Ultra SCSI 320 RAID arrays at max speed on a single 64/66 PCI bus.
7.Anything that there's no hope of 5 handling in this current generation, I.E. database's with 10+ Terrabytes of data on anything except the most powerful of supercomputers.
8.DVD ripping
9.I need help. Sohcan, you have any examlpes?

FishtankX hoping to settle this mad discussion.

Anybody disagree with my assesment?

 

[MrGrim]

Originally posted by: FishTankX

7.It is established that the HDD is the biggest bottleneck in the system, *if* memory isn't able to cache all the data in a program currently running

When an application is loaded some of its parts (the most vital and commonly used ones) are stored in the memory and others stay on the hard drive (swap file). That course of action is taken even if the memory is capable of storing everything at once. If that wasn't the case, imagine what would happen everytime you loaded some sizable applications; exactly what happens when you have 10 IE windows open along with Excel, Word and UT2003, it's called thrushing. My point is that faster storage mediums will not benefit a certain portion of computer users, it would benefit absolutely everyone.

I'm not exactly sure what your point is. I think that with your examples you have proved that the bottleneck in a system depends on the type of use of the system. With that in mind, I'm not sure I understand the meaning of your bottleneck hierarchy.

 

[Whitedog]

My biggest bottleneck is Funds... seems my computers are sucking money from my funds faster than I can put funds in... if you know what I mean.

 

[Woodchuck2000]

I agree with your hierachy in principle. It doesn't allow for cases where there are multiple bottlenecks however. It's quite possible to have the CD-ROM drive and Hard Disk as bottlenecks at the same time (for example, installing a game and also shunting files around at the same time) or indeed for the graphics card and memory subsystem to both cause bottlenecks simoultaneously. I agree with the list of components, just not the way you've ordered them.

Also, I don't agree that the OS is a bottleneck by your definition. Talking about bottlnecks, we should be looking at only the hardware so it's irrelevant and distracting to consider the OS...

For what it's worth...

 

[teddymines]

Maybe I mis-read your post, but I don't see how the CPU can ever be the bottleneck. It performs operations on data, which are fetched from a memory controller. Anything off the CPU core will take time to issue the request and wait for the data to come in, and also write it back out when done. Even within the chip, some memory operations require multiple cycles.

Also, the CPU is always running at "full speed". The question I'd be asking is this: out of the total number of cycles the CPU sees, how many are used specifically for my task? I would be surprised if all of 1% is used for your task, such as a photoshop rotate. I'll wager that most is idle waiting for memory, HD, or other OS-related overhead.

 

[fluxquantum]

Originally posted by: Whitedog
My biggest bottleneck is Funds... seems my computers are sucking money from my funds faster than I can put funds in... if you know what I mean.

LMAO!!! that's a great one

 

[KF]

> My point is that faster storage mediums will not benefit a certain portion of computer users, it would benefit absolutely everyone.

" Operating systems cache read and writes in a file cache, so for those workloads whose working set of the input and output fit within main memory, I/O requests will not even touch the I/O subsystem in the common case. " Sohcan


>When an application is loaded some of its parts (the most vital and commonly used ones) are stored in the memory
> and others stay on the hard drive (swap file). That course of action is taken even if the memory is capable
> of storing everything at once.

If I run a program that takes 1 megabyte, Windows keeps part of it on the HD? I don't believe so.

> If that wasn't the case, imagine what would happen everytime you loaded some sizable applications; exactly
> what happens when you have 10 IE windows open along with Excel, Word and UT2003, it's called thrashing.

10M of program will do a lot. There is no reason to cause thrashing if some one temporarily runs a little used module that sits on the HD practically all the time. Thrashing is normally a programming mistake, due to some unforseen interaction.

>My point is that faster storage mediums will not benefit a certain portion of computer users, it would benefit absolutely everyone.
It will benifit everyone, but not so they would notice, or pay a dollar extra for the negligible improvement. I spend way more time waiting in one store checkout line than I do waiting for my HD in a week. But unlike most people I am willing to pay a little extra (not a lot) for a faster HD, mostly just for knowledge that it faster.

 

[dullard]

Originally posted by: teddymines
Maybe I mis-read your post.

A good definition of a bottleneck is something that if replaced with a faster piece will have a significant impact on the speed of your program. Basically if I pull out a 1500+ Athlon XP and replace it with a 2400+, and my programs speed up significantly then the CPU was the bottleneck. If I pull out my 7200 rpm HD and replace it with a solid state hard drive and my programs speed up significantly then the hard drive was the bottleneck. If I replace SDRAM with RDRAM and my programs speed up significantly then the memory bandwidth was the bottleneck. Anytime you see a benchmark that has a significant change with a new component, then that component was a bottleneck in that program. Look at benchmarks that are everywhere on the web. Sometimes you see a huge difference in speed with various CPUs, sometimes you don't.

I agree that different programs will have different bottlenecks. Many games will not speed up with a faster CPU since the video card is the bottleneck - while many other games will speed up quite a bit with a faster CPU.

Here are what I think are some of the most common computer uses and the major bottleneck:
Internet: modem bandwidth.
Email: human input.
Word processing/Spreadsheet/Typical office work: human input.
Gaming: either CPU or video card.

That right there will cover roughly 90% of computer usage. And thus those are the major bottlenecks.

Here are some other less commonly used computer programs:
Video editing: HD bottleneck (although this can be eliminated) / CPU bottleneck.
Photo editing: CPU bottleneck for the vast majority of uses, HD bottleneck for professional applications without enough memory.
Calculations (homework, office, scientific): CPU/memory bottleneck.
Program Installation: CD drive bottleneck.
CD/DVD burning: Can be almost anything depending on what is being done (burner speed, HD, internet, memory).

From the previous thread, I am a moron from stating this: Yes HD can be the bottleneck if you don't have enough memory or don't use a solid state device, but these are rarely used cases compared to the major uses listed above. I still say the vast majority of computer users load their program and then never access the HD again for an hour or more (such as when doing office work, writing email, playign games). Thus during that hour, the HD speed is meaningless since the HD isn't being accessed).

 

[MrGrim]

Originally posted by: KF
> "Operating systems cache read and writes in a file cache, so for those workloads whose working set of the input and output fit within main memory, I/O requests will not even touch the I/O subsystem in the common case. " Sohcan

so for those workloads whose working set of the input and output fit within main memory

That's exactly what I said. Some parts are stored in the memory and others are not.

If I run a program that takes 1 megabyte, Windows keeps part of it on the HD? I don't believe so.

No, but if you load IE, MS World, Lotus Notes, Outlook or any medium size application parts will be stored on the HD. Would you like me to get started on Java applications? Are you questioning whether applications are loaded as a whole into the memory?

> If that wasn't the case, imagine what would happen everytime you loaded some sizable applications; exactly
> what happens when you have 10 IE windows open along with Excel, Word and UT2003, it's called thrashing.

10M of program will do a lot. There is no reason to cause thrashing if some one temporarily runs a little used module that sits on the HD practically all the time. Thrashing is normally a programming mistake, due to some unforseen interaction.

Oh jesus, why do you pretend to know what you are talking about when you don't have a clue? This type of comments really do piss me off you know; "Programming mistake". Virtual memory is transparent to the programmer. Let me see you run SQL Server on a quad P4 Xeon with 128MB of RAM.

If you don't know something LET IT BE. What I said was that if you loaded the whole of the program into RAM every single time, switching between applications would cause thrashing.

Thrashing occurs when a system spends more time processing page faults than executing transactions

Hence, moving a whole program from the RAM to the HD and the other way round every single time you switch applications would cause thrashing.

>My point is that faster storage mediums will not benefit a certain portion of computer users, it would benefit absolutely everyone.

It will benifit everyone, but not so they would notice, or pay a dollar extra for the negligible improvement. I spend way more time waiting in one store checkout line than I do waiting for my HD in a week. But unlike most people I am willing to pay a little extra (not a lot) for a faster HD, mostly just for knowledge that it faster.

That is not correct. You wait for your hard drive A LOT more than you think you do. I/O is the most time consuming operation in a computer system. Compared to how fast a CPU can perform operations I/O is a joke.

 

[dullard]

Originally posted by: MrGrim
That is not correct. You wait for your hard drive A LOT more than you think you do. I/O is the most time consuming operation in a computer system. Compared to how fast a CPU can perform operations I/O is a joke.

But you are still missing the main point. If you ask the vast majority of computer users to watch their HD light while running their programs, they would say it never ever came on during 99.9% of their computer use. Thus during most of the time using the computer there is NO HD ACTIVITY. Yes switching between programs will use it, but the vast majority of people can wait that 1 second to switch then use the next program for a half hour without any HD activity. Some people, very few people, need to load 10 programs and switch repeatedly - so it is possible that they will benefit. But saving the rest of us 1 second every half hour really isn't that big of a deal. Yes during that 1 second of waiting the HD is the bottleneck - what about the rest of the 29 minutes and 59 seconds?

 

[ai42]

But you are still missing the main point. If you ask the vast majority of computer users to watch their HD light while running their programs, they would say it never ever came on during 99.9% of their computer use. Thus during most of the time using the computer there is NO HD ACTIVITY. Yes switching between programs will use it, but the vast majority of people can wait that 1 second to switch then use the next program for a half hour without any HD activity. Some people, very few people, need to load 10 programs and switch repeatedly - so it is possible that they will benefit. But saving the rest of us 1 second every half hour really isn't that big of a deal. Yes during that 1 second of waiting the HD is the bottleneck - what about the rest of the 29 minutes and 59 seconds?

Thats an exaggerated point, I think for most computer usage you are talking 40-80% non-hard drive activity, depending on what you are doing.

It all comes down to what you are doing.
If you are running a server I promise your hard drive will run more than 1% of the time.
If you are gamming then you would use a fair amount of hard drive to load maps and such and even refer to classes etc would use a fair amount of HDD activity.
If you are working with large databases then your hard drive would be used a lot
The only thing I can see such little hard drive activity (still higher than 1% though) would be web surfing/word processing perhaps smaller spreadsheets.

Also depends on how much RAM you have if you ahve 4-5gb your hard drive will barely be touched but if you've worked with 64/128mb of ram on an moderm OS I promise you HDD use is very high.

Also when you compare the numbers when your RAM/CPU can mvoe around data at 8.5GB+ per second and your hard-drive maybe averages 50-60mb a second there is a huge differance.

 

[dullard]

Thats an exaggerated point, I think for most computer usage you are talking 40-80% non-hard drive activity, depending on what you are doing.

Many of my computers are on 24/7 (in constant use, not sitting idle) - to tell me that the HD is running between 10 and 19 hours each day is laughable.

"It all comes down to what you are doing."
Exactly what I said in my first post. The second post you quoted was just talking about the majority of users - think about your grandmother.

"If you are running a server I promise your hard drive will run more than 1% of the time."
Again your grandmother probably doesn't run a server. In fact I bet 99% of people don't run servers. I fully admit there are some uses where there is intense HD activity. But I truely think the majority of us are more CPU/video card/modem bandwidth/human input bottlenecked for the majority of the time. Thus the HD is only a bottleneck for a small minority of the time. If a typical user had $500 to upgrade I'd recommend a new CPU, new video card, or a cable modem (depending on use) far before I recommend SCSI raid because this will give them far more speed boost that way. I suppose you'd recommend a 500 MHz Celeron with integrated graphics and SCSI raid to the typical user when I recommend a 2400+ XP with a good video card and a fast 8 MB buffer HD. Honestly I think they'd be happier with my recommendation. But you can have your beliefs.

"If you are gamming then you would use a fair amount of hard drive to load maps and such and even refer to classes etc would use a fair amount of HDD activity."
I'm a huge gamer and generally I load a map (when yes the HD is the bottleneck unless I put the map into memory - ramdrive). But then I play for 30 minutes before I win the map and have to go to the next map. I'd much rather have 50% faster framerates during my 30 minutes of playing than to have the 2 second map loading time reduced to 1 second. Sure the HD might be occasionally used during the map, but that is almost without exception a background activity that doesn't affect the game speed (since the game is currently CPU/video card/CD drive/human input limited).

"If you are working with large databases then your hard drive would be used a lot"
Again this isn't the vast majority of users.

"The only thing I can see such little hard drive activity (still higher than 1% though) would be web surfing/word processing perhaps smaller spreadsheets."
Finally this is the vast majority of users. How much HD activity do you have while typing email? Yes there is HD activity during downloading but for most of us the modem is the bottleneck at that time - until we have modems that can download 50 MB/sec it will be this way.

"Also depends on how much RAM you have if you ahve 4-5gb your hard drive will barely be touched but if you've worked with 64/128mb of ram on an moderm OS I promise you HDD use is very high."
Correct I can make a computer that has very high HD use - like WinXP running on 64 MB memory. But that is a design flaw, not a true bottleneck in the sense this discussion is about. In that case you'd increase speed a lot faster by upgrading memory than upgrading to a faster HD. So the memory was a lot more of a bottleneck than HD speed.

"Also when you compare the numbers when your RAM/CPU can mvoe around data at 8.5GB+ per second and your hard-drive maybe averages 50-60mb a second there is a huge differance."
But you can use your ram as a ramdrive. Or you can get solid state hard drives. Thus you can with enough money get the speeds you want. But you cannot buy a 10 GHz CPU or a Geforce6, etc. So With enough money the HD bottleneck can be easily reduced or removed - but you cannot say that about other bottlenecks. So for the majority of people - if they have a HD bottleneck it is due to lack of money.

 

[teddymines]

Just about every software product I've used seems to keep "rarely used" items on the HD until they're called. For example, start editing a document in Word, then attempt to change the font. There is a noticable delay as font information is retrieved through OS calls. Powerpoint's Set Up Show menu item requires disk access. These tasks, and probably most others, call the disk on the first access, then the program or OS seem to cache it. I'm not sure if memory watermarks or time is the criteria used to un-cache.

Personally, slowdowns for than 1 second are noticable. I'd rather a program fit as much as possible into memory on program load. I'd like to see more of my free memory used for this purpose. I think the fear is that is a new program needs to be loaded, there could be significant delay as 200MB of memory is swapped out onto a slow EIDE disk.

 

[MrGrim]

Originally posted by: dullard

Originally posted by: MrGrim
That is not correct. You wait for your hard drive A LOT more than you think you do. I/O is the most time consuming operation in a computer system. Compared to how fast a CPU can perform operations I/O is a joke.

But you are still missing the main point. If you ask the vast majority of computer users to watch their HD light while running their programs, they would say it never ever came on during 99.9% of their computer use. Thus during most of the time using the computer there is NO HD ACTIVITY. Yes switching between programs will use it, but the vast majority of people can wait that 1 second to switch then use the next program for a half hour without any HD activity. Some people, very few people, need to load 10 programs and switch repeatedly - so it is possible that they will benefit. But saving the rest of us 1 second every half hour really isn't that big of a deal. Yes during that 1 second of waiting the HD is the bottleneck - what about the rest of the 29 minutes and 59 seconds?

I'm sorry, but you are wrong. You don't wait 1 second in an hour, you wait MUCH more than that. If you don't believe me do some reading up.

 

[MrGrim]

Guys, it really isn't that hard. Just think about it, how many operations can a CPU perform in 1 ms? How long does it take a hard drive to find something? An average of 9 ms? And then after it finds it, it has to retrieve it! It's as simple as that, I/O is slow!

Talking about the HD led is beyond funny. Do you think that your led blinks everytime your hard drive performs a seek? That is not the case.

I honestly don't have anything more to say. If you don't believe me just do a google search or something. I/O is the slowest process in today's computer systems.

 

[tbates757]

no offense dullard but you don't really know what you are talking about.

 

[dullard]

Originally posted by: tbates757
no offense dullard but you don't really know what you are talking about.

Just please will some of you tell me which $1000 computerwould be the faster overall for a typical user:

A) 500 MHz Celeron with integrated graphics and SCSI raid
or
B) 2.53 GHz P4 with Geforce4 Ti4400 and WD Special edition with 8 MB buffer.

Then tell me which computer is closest to what you currently have/want.

For those who'd rather have (A) and think the kid down the block playing Quake and your elderly relatives typing email and the schoolteacher calculating grades in Excel, etc would be better off with (A): give me your reasons. I really think a post like the one by tbates757 without any reasons for choosing (A) is a waste of Anandtech's HD space and bandwidth.

For tbates specifically: which computer (A) or (B) would you want to join Team Anandtech D2OL? I'm just curious since I haven't ran that disease search program - is it specifically HD bottlenecked?

I realize the HD is used frequently, but the key is how much of that time is it the bottleneck? Sure the HD is accessed when you click on the Word font menu, but even on the 5400 rpm HD I'm currently using, it is all loaded by the time I can move the mouse from the the down arrow to the fonts (meaning moving the mouse 0.5 inches was the bottleneck not the HD). And this is a one time deal on every version of Word I've used - meaning the next font change doesn't access the HD. If I really wanted to I could have put these fonts on a Ramdrive or solid state HD and then it would have been accessed in a fraction of a millisecond.

Yes I/O is slow but how many CPU calculations are done and how many I/O are done typically? The vast majority of operations are CPU calculations. So much that in most cases it is the bottleneck.

 

[MrGrim]

If you don't believe us, perhaps you will believe them

The biggest bottleneck in the computer is the hard drive
the I/O bottleneck grows ever worse without some change in the fundamentals of data storage
Beating the I/O Bottleneck
Breaking the I/O Bottleneck

Just a few links I found on google ...

 

[dullard]

Originally posted by: MrGrim
If you don't believe us, perhaps you will believe them

Just a few links I found on google ...

Hmm maybe you'll belive Anand. Look how all the programs did better with a faster CPU. If the HD was a true bottleneck increasing the speed of other components wouldn't affect the results. But it did - so during much of that benchmark the HD wasn't a true bottleneck but the CPU was instead. This CPU effect wasn't a fluke as shown by both AMD and Intel (depending on which company you like to trust).

Or maybe you'll believe Tom. Look how those benchmarks changed with a different videocard. A simple upgrade there could make those programs 10 times as fast. Hmm seems like the HD wasn't the bottleneck there either.

Darn Sharky had the same experience. Altering parts that weren't the HD increased speed. Look at all those home, game, office programs speeding up with a different CPU. If the HD was a true bottleneck the speed would depend only on the HD speed as the other parts aren't holding up the computer.

Need I go on. Yes the HD does affect speed. But no you don't need a faster HD to increase the program speed. Thus that violates the definition of a true bottleneck.

I can boot a computer, remove all drives and still run the programs. Thus you can run programs without HD access. I don't see why you can't understand this point. Why are you ignoring solid state hard drives which are much faster? Why are you ignoring all the most frequent computer uses and only focussing on the minor uses? Why haven't you answered a single question - are you just to afraid to admit you were wrong?

 

[teddymines]

Originally posted by: dullard

Originally posted by: MrGrim
If you don't believe us, perhaps you will believe them

Just a few links I found on google ...

Hmm maybe you'll belive Anand. Look how all the programs did better with a faster CPU. If the HD was a true bottleneck increasing the speed of other components wouldn't affect the results. But it did - so during much of that benchmark the HD wasn't a true bottleneck but the CPU was instead. This CPU effect wasn't a fluke as shown by both AMD and Intel (depending on which company you like to trust).

Or maybe you'll believe Tom. Look how those benchmarks changed with a different videocard. A simple upgrade there could make those programs 10 times as fast. Hmm seems like the HD wasn't the bottleneck there either.

Darn Sharky had the same experience. Altering parts that weren't the HD increased speed. Look at all those home, game, office programs speeding up with a different CPU. If the HD was a true bottleneck the speed would depend only on the HD speed as the other parts aren't holding up the computer.

Need I go on. Yes the HD does affect speed. But no you don't need a faster HD to increase the program speed. Thus that violates the definition of a true bottleneck.

I can boot a computer, remove all drives and still run the programs. Thus you can run programs without HD access. I don't see why you can't understand this point. Why are you ignoring solid state hard drives which are much faster? Why are you ignoring all the most frequent computer uses and only focussing on the minor uses? Why haven't you answered a single question - are you just to afraid to admit you were wrong?

I hesitate to enter this fray, but I will anyway. I don't think people disagree that speeding up the CPU, adding memory, or getting a better video card will increase performance. The reviews you cited, as well as others we've all read, convey that well. Maybe an analogy will help explain my viewpoint.

When you drive across state, you can go 65 mph most of the time, except for the few times you have to slow down to 30 mph for small towns. If you could increase your speed to 75, you would definitely get to your destination quicker, but you still have to slow down to 30 for the towns. I see it the same way with computers. Speeding components other than the hard drive will increase performance, but as long as operating systems rely upon the hard drive for storage, we have to wait during this access. For some, the wait seems very long.

Also, your definition of bottleneck is different than mine. You state above "If the HD was a true bottleneck the speed would depend only on the HD speed as the other parts aren't holding up the computer." I disagree. You can speed up other components so that the time between HD accesses is decreased.

 

[Mardeth]

So someone didnt actually know what "bottleneck" means? Lol must be a real hardware expert.

 

[dullard]

Originally posted by: teddymines

When you drive across state, you can go 65 mph most of the time, except for the few times you have to slow down to 30 mph for small towns. If you could increase your speed to 75, you would definitely get to your destination quicker, but you still have to slow down to 30 for the towns. I see it the same way with computers. Speeding components other than the hard drive will increase performance, but as long as operating systems rely upon the hard drive for storage, we have to wait during this access. For some, the wait seems very long.

Also, your definition of bottleneck is different than mine. You state above "If the HD was a true bottleneck the speed would depend only on the HD speed as the other parts aren't holding up the computer." I disagree. You can speed up other components so that the time between HD accesses is decreased.

Here is the definition of a bottleneck that many people here are using: a chain is only as strong as its five weakest links. Gee that just isn't quite how the saying goes. Or a process is only as fast as its slowest 3 parts. Hmmm still doesn't quite sound right. Maybe since they are using non-standard definitions.

Here is my definition of a bottleneck. A chain is only as strong as its weakest link. It doesn't matter how strong the other links are - it will fail when the weak link fails. Or maybe this one: a process is only as fast as its slowest part. I could go on and on with the phrases that are commonly used. Imagine you are on a road and there is no exit (you must go forward). Imagine there are plenty of lanes and the speed limit is really high. Imagine the police make a roadblock and let 5 vehicles through per minute. It doesn't matter how many lanes are available. It doesn't matter what the speed limit is, you cannot go any faster since your progress is bottlenecked (limited) by the police. The only way to go faster is to add more police and get more vehicles through the roadblock per minute. That is a true bottleneck. I can see what you are arguing - that after you leave the roadblock you can get home faster if there were more lanes and if the speed limit was high. There is a practical limit though. Take gaming for example - there is a definite amount of time that will be spent between loading levels (with an infinitely fast computer it still takes time for the human to make choices and play the game). I think the main part of gaming is the game and not the map loading. I'd much rather have more fps while gaming than have the next map load up 1 second faster. YES during the map loading most computer configurations are HD bottlenecked. I agree with that. But what about the rest of the time - usually the HD is the last thing that should be considered. I could care less how long I wait between maps (as long as it is reasonable - I wouldn't wait hours) - it is the 30 minutes of fun in between that are bottlenecked by CPU/video card/human/modem that are important to make sure there is no bottleneck. And in the majority of uses, this ~30 minutes far outlasts the ~2 seconds for map loading. Meaning in that example 99.9% of the game was spent with the speed limited by things other than the HD. In many games even when you think it might be the HD limiting during map loading - it often is the CD/DVD drive accessed for loading a movie/sound track not the HD. And the CD drive is much slower than the HD.

As for your 30 mph and 75 mph example. That is great for the highway. But what about those of us who take the freeway? There are no small towns to slow us down. The same thing CAN be done with drives - if you are willing to spend the money. Thus the HD is only the bottleneck due to economical spending. It can be easilly improved. You can eliminate all the 30 mph zones (HD with 350 MB/sec sustained transfer rate, at this speed the bus from the drive to the CPU is the bottleneck). I'd rather increase the 75 mph zones since the 30 mph ones are easy. Actually we are reaching the point where the speed limit isn't the problem, but the cars just cannot go much above 100 mph. No amount of money can help you go much faster (1000 mph for example). Thus there are easy methods to eliminate the HD bottleneck if you have money. The same isn't true with a CPU or videocard bottleneck. Thus in general terms the CPU/video card needs to be speed up to keep up with solid state hard drives.

 

[ai42]

I can boot a computer, remove all drives and still run the programs. Thus you can run programs without HD access. I don't see why you can't understand this point. Why are you ignoring solid state hard drives which are much faster? Why are you ignoring all the most frequent computer uses and only focussing on the minor uses? Why haven't you answered a single question - are you just to afraid to admit you were wrong?

No you can't boot a computer and take out the drives and still be working. Try it sometime. I bet it will freeze instantly once you take out the data cable, not when the next hard drive access is.

Solid state hard drives are impractial for now and a good while to come, I wish that Solid state drive thread never came up its a pipe dream, will not be practial for consumer use for years to come, and may never be practical. Also it still impractical for businesses too as you pay 80-100 times more per GB, and if your running a web host you could give your customers about 20kbs at that cost.

What is the most frequent computer uses? As by your definition web surfing and word documents ia about it. When most computers out there are used for business and I promise you they do much more than that. Maybe to consumers it is the majority of what they do, but hardware manufcatuers cannot product products to only consumers they need to hit the business market as well.

Finally even on your examples the hard drive is used more than 99.9% of the time. Just maybe if you kept your computer idle and left it on for 24/7 and used it for 5 mintues a week to play solitare MAYBE, but I doubt it!

 

[Bovinicus]

I'm surprised no one has mentioned the main area in which the HDD becomes the bottleneck: Booting up the operating system. Why do you think computers don't turn on "instantly?" It is because the entire kernel, drivers, and other OS required files must be loaded from the HDD into memory. Granted, computers nowadays boot up extremely fast compared to those of yesteryear, but nonetheless. How about when loading a map in Q3A? How about opening Adobe Photoshop. Those operations take a good 10+ seconds more than they would if all the information was in memory right from the start. I was surprised at how much faster my system ran when I upgraded from a 5400RPM drive to a 7200RPM drive. Everything just had so much more snap. I/O and HDD operations are definitely some of the main bottlenecks in any computer experience. If you say otherwise, you are just wrong. =)

 

[CrazySaint]

I want faster HDDs so I want have to spend over 4 hours waiting for scandisk to run on my 80GB HDD