FPS is used as an example of a running program. Once the program is running, HD doesn't matter. It matters when swapping and loading software. You are not loading software nor you are swapping all the time.
Originally posted by: Gonad the Barbarian
Some of you 'tards need to realize that frames per second is not the only performance measurement of a computer. Will a slow hard drive affect fps? No. Will it affect nearly ebvery other aspect of system performance? Yes.
Pink0, please reread my previous post. You are naming valid I/O intensive applications that depend on hard drive access times and throughput, but this does not make all workloads universally I/O bound.
How many tasks actually require constant full speed hard drive access? Video and audio work yes... but for games, for example, 100mb/sec is in no way a bottleneck.Originally posted by: MaxDSP
Originally posted by: Jellomancer
Originally posted by: AMDfreak
Why would Harddrives be a bottleneck?
In today's technology, the hard drive is the utter definition of "bottleneck". Serial ATA only provides a burst rate of 150MB/s, and likely sustained transfer rates will not likely reach above 50MB/s, even with an 8MB cache. The next slowest item is the RAM, which even if you're using old EDO SIMMS will transfer data faster than a hard drive.
Being slow compared to memory etc doesn't make the hard drive a bottleneck. A car's seats don't tilt at 6500 rpm.. does that mean they are a bottleneck?
No, but then again your comparison is not relevant to the term "bottleneck". What does moving a car's seats have to do with the actual performance? If you're talking about the battery that the seat uses for power, then in reality performance would actually go down because it would take more energy to move the seat at "6500 RPM" (whatever that means) as opposed to 3500 RPM.
A more relevant comparison would be the gas tank or the fuel nozzles being the bottleneck in the car, since their performance directly affects the performance of the car. (I dunno much about cars so that may be a bit off)
The term bottleneck pretty much means that the system can only go as fast as the slowest component (which is the hard drive)
*sigh*...this is NOT at all what I am saying. Apparently you have not read my posts. Merely stating that the hard drive is the slowest component of a system does not make it *THE* bottleneck...this neglects main memory bandwidth and latency; cache bandwidth, latency, block size, and associativity; fetch, decode, issue, and retire width; branch target buffer size, associativity and width; branch predictor size and layout, number of renaming registers, number of reservation stations, fetch and issue queue sizes, reorder buffer size and commit width, number of execution units, instruction latency....would you like me to continue? There are a large number of potential "bottlenecks" in any system, each of which may manifest themselves as one of the "primary" bottlenecks in a variety of circumstances. For example, Tullsen's et al. study on instruction fetch and issue on SMT processors found that improving branch prediction from 95% to 100% in SPEC CPU improved performance by 25%, lending to the well-known fact that control prediction and speculation is a major performance issue in modern dynamically scheduled superscalar.Originally posted by: Pink0
Pink0, please reread my previous post. You are naming valid I/O intensive applications that depend on hard drive access times and throughput, but this does not make all workloads universally I/O bound.
A lot of you...umm...what gonad said are basically saying that because the hard drive doesn't slow down ALL applications it's not the system's bottleneck. That's pretty dumb...No, the hard drive is not used all the time in all applications but it is the slowest major computer component and it is the system's bottleneck.
Of course going to the extreme by eliminating the main memory layer of the memory hierarchy will make just about any serious workload I/O bound. This does nothing to show that the hard drive is the only important bottleneck in a system; when working sets for CPU bound workloads fit within main memory (ie, when a vast majority of pages accessed can be fit within main memory), page fault rates quickly become negligible and I/O latency is not nearly as important. 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.Has it occured to anyone that they only reason we have RAM Is because hard drives are such a bottleneck? If we had lightning fast non-volitile storage, we would not need ram. I'd like to see you use a system with no ram (pretend it works) and see if the hard drive is a bottleneck or not.
Merely stating that the hard drive is the slowest component of a system does not make it *THE* bottleneck
Originally posted by: Pink0
Fishtank, when you talk about the various methods that you can get around the slow boot times for PVRs and other living room appliances which is where the home PC is going what you are basically doing is supporting my point. By saying that you can use flash cards or whatever to boot up and then the hard drive from then on what you're describing is a way to bypass the bottleneck and thus admitting that the hard drive is the bottleneck in this situation.
Originally posted by: Pink0
Merely stating that the hard drive is the slowest component of a system does not make it *THE* bottleneck
Actually, that's the definition of bottlneck: the slowest part. The smallest part of the bottle. The neck of the bottle.
Yes it is. The sky is also not brown. I really don't know how people can argue with this. Bottleneck is the slowest component of the computer. The hard drive is the slowest component of the computer. What's the problem here? RAM is just a device which was invented to decrease the effect of this bottleneck. The CD rom and floppy are not crucial parts of the computer. You can run one without either. They don't count. Full fledged PCs NEED a hard drive. You can load programs through means other than floppies and CDROMs. The user is not a part of the computer. I don't care if 90% of what they do is officework and internet. Those are not applications which challenge the system and that is all that we're talking about. You can completely ignore that 90% and look only at the 10% which is challenging the system.The HDD is the biggest bottleneck in systems today. and that IS a fact!!!!
This is precisely why the term "bottleneck," in your definition, is an inaccurate term to represent performance in a system. The location of the neck in a glass bottle is a static property, the inhibitor in performance in a system is not. Performance is very dynamic in computer systems, and naming the component with the absolute slowest latency as the universal "bottleneck" is not accurate when under many workloads it goes unused; this is the basic concept that I've been trying in vain to explain to you, but astoundingly you continue to ignore the contents of my posts. I really have nothing left to say unless you actually reread my posts, or unless you would like to present an argument that CPU bound workloads or the polarity between CPU and I/O bound workloads do not exist.
Originally posted by: BD231
Originally posted by: Howard
Sometimes, but please don't make generalizations like this.SCSI is to loud and to hot.
IDE is too loud and too hot, too.
Very true, I've used SCSI before though and the spindle noise made at even just 10,000rpm is unbareable if your computer is sitting right next to you(I had an 18gig maxtor, latest modle). I love SCSI performance but it's just not a good option for normal home computer use.