I am using my PC at work, and speed is important. That includes I/O speed, so now I am using SCSI disks. However, it seems quite difficult to find quiet ones, and even within a manufacturer's line, they vary tremendously. Are there any SCSI disks are are recommendable for a quiet system, or is it hopeless?
(Although I've seen some "benchmarks" claiming ATA is as fast as SCSI, may experience has been otherwise. With ATA, I/O slows down the whole system, and if one program is doing I/O, all other I/O is delayed.)
We are ordering some new PCs and I'd like them to be both powerful and quiet.
Quiet SCSI a contradiction?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
I've been for quite some time a pure scsi user.
Strange to go back to IDE for my desktop, but the price/space was just too much for SCSI.
However, for quite some time, I had 2 7200 and 1 10k rpm scsi hdd in my desktop pc, all that with a pretty quiet PSU and CPU HSF. And didn't find it too noisy.
Since that, I reached the (in)famous noise threshold from which I headed toward the horizon of a silent PC, thanks to a screamer 4cm fan on a video card and a lighter case.
Bought a sonata and a seagate V HDD and played with my then 2 10k RPM IBM HDD to transfer the that and found that having them isolated, using the standard HDD mount of the sonata, helped quite a lot so that they were usable to me (however, look at the disclaimer below).
These 2 drives were IBMs and, if you look at storagereview.com perf. database for HDD, there're the loudest ones
But I cannot believe it was 55 dB(A).
Anyway, if money/space is not that much a problem, I think that 15k RPM seagate SCSI HDD should be a good speed/quiet combination. For example, the Cheetah X15-36LP.
Of course, you'll have to isolate them from the case.
Disclaimer and other notes :
- I like sound seeking noise of HDD (as much as I like 'clicky' keyboards).
- For the people regarding use of screaming SCSI with disdain, look at the speed/perf benchmark, especially when you move from the 'quietest' possible choice to a more perf-oriented one.
- For example, the Cheetah are (about) twice the speed of a Barracuda
Strange to go back to IDE for my desktop, but the price/space was just too much for SCSI.
However, for quite some time, I had 2 7200 and 1 10k rpm scsi hdd in my desktop pc, all that with a pretty quiet PSU and CPU HSF. And didn't find it too noisy.
Since that, I reached the (in)famous noise threshold from which I headed toward the horizon of a silent PC, thanks to a screamer 4cm fan on a video card and a lighter case.
Bought a sonata and a seagate V HDD and played with my then 2 10k RPM IBM HDD to transfer the that and found that having them isolated, using the standard HDD mount of the sonata, helped quite a lot so that they were usable to me (however, look at the disclaimer below).
These 2 drives were IBMs and, if you look at storagereview.com perf. database for HDD, there're the loudest ones
But I cannot believe it was 55 dB(A).Anyway, if money/space is not that much a problem, I think that 15k RPM seagate SCSI HDD should be a good speed/quiet combination. For example, the Cheetah X15-36LP.
Of course, you'll have to isolate them from the case.
Disclaimer and other notes :
- I like sound seeking noise of HDD (as much as I like 'clicky' keyboards).
- For the people regarding use of screaming SCSI with disdain, look at the speed/perf benchmark, especially when you move from the 'quietest' possible choice to a more perf-oriented one.
- For example, the Cheetah are (about) twice the speed of a Barracuda
That's it.dago wrote:UPDATE (after reading rman answer) :
- my main problem was a 1-2 Hz noise due to 2 drives not in perfect sync.
- most of the high-pitched noise went away when moved the sonata. Probably due to combination of grommets and better damping.
I mounted my Cheetah 15k.3 on foam, similar to the article on this site, and it is now dead silent.
Yes, and they are made to resist such high temps (!= design, higher quality stuff, you get what you paid for).dont they get really hot too?:\
According to the specs, they can take up to 10°C more than their IDE counterparts.
In my server, they (the 10k) are between 40-45°C, while my 'cuda is in the low 30°C. And I fear most a failure for the later one than from the SCSI
-
Katana Man
- Patron of SPCR
- Posts: 350
- Joined: Mon Mar 10, 2003 11:14 am
I know this is a very old thread, but please let me resurrect it for a simple-minded question.
From what I understand (just off the top of my head), isn't maximum IDE throughput 133mb/s? Also, isn't the theoretical maximum throughput for SCSI 160mb/s? Finally, isn't the theoretical maximum throughput for a PCI card 133mb/s? IF all of the above are true (and remembered correctly), then how does SCSI perform better than EIDE (because SCSI generally goes through a PCI adapter)? And also better than SATA (transfer rate of 150mb/s), for that matter?
That question's been bothering me for awhile now, I hope someone can provide some insight here for me...
-PS
From what I understand (just off the top of my head), isn't maximum IDE throughput 133mb/s? Also, isn't the theoretical maximum throughput for SCSI 160mb/s? Finally, isn't the theoretical maximum throughput for a PCI card 133mb/s? IF all of the above are true (and remembered correctly), then how does SCSI perform better than EIDE (because SCSI generally goes through a PCI adapter)? And also better than SATA (transfer rate of 150mb/s), for that matter?
That question's been bothering me for awhile now, I hope someone can provide some insight here for me...
-PS
PS,
SCSI discs are often physically faster (lower access / seek times, faster burst, better cache algorithms, better sustained transfer and tagged command queuing that really works, unlike SATA currently).
Single drives (ata or scsi) don't usually get anywhere near the bottleneck of the interface of SCSI, U-PATA, SATA or PCI. Or if they do, it's a temporary burst limitation.
According to what I've read, the theoretical maximum SCSI standards is 320 MB/s for Ultra320 SCSI. Also, some SCSI controllers are (just like some SATA controllers) directly integrated on the motherboard and using not the PCI, but another memory bus for transfers.
I'd really like SATA to start performing myself. However, in practise I've noticed that even my 3 year old Ultra SCSI 80 Mb/s LVD set up is noticeably faster in use than any Ultra-Ata disk I've ever used. Yes, it's also noisier and hotter, but I have much less system stalls and choppy performance than on a very good ide setup.
I'm really looking forward to SAS (serial attached SCSI as well), although in 2004-2005 it'll probably still be prohibitively expensive for us ordinary users.
IMHO, YMMV.
regards,
Halcyon
SCSI discs are often physically faster (lower access / seek times, faster burst, better cache algorithms, better sustained transfer and tagged command queuing that really works, unlike SATA currently).
Single drives (ata or scsi) don't usually get anywhere near the bottleneck of the interface of SCSI, U-PATA, SATA or PCI. Or if they do, it's a temporary burst limitation.
According to what I've read, the theoretical maximum SCSI standards is 320 MB/s for Ultra320 SCSI. Also, some SCSI controllers are (just like some SATA controllers) directly integrated on the motherboard and using not the PCI, but another memory bus for transfers.
I'd really like SATA to start performing myself. However, in practise I've noticed that even my 3 year old Ultra SCSI 80 Mb/s LVD set up is noticeably faster in use than any Ultra-Ata disk I've ever used. Yes, it's also noisier and hotter, but I have much less system stalls and choppy performance than on a very good ide setup.
I'm really looking forward to SAS (serial attached SCSI as well), although in 2004-2005 it'll probably still be prohibitively expensive for us ordinary users.
IMHO, YMMV.
regards,
Halcyon
ok, hope folks don't mind me ressurceting this thread to throw my $.02
reinforcing what halcyon mentioned of why scsi performs better --
mainly because of the protocol involved.
now, if you measure performance while say, just copying a single file, or
any single threaded i/o, then there's practically NO difference between scsi
and ide with harddrives of the same rpm.
however, throw other i/o in the mix and the difference is huge!
the authors of O'Reilly's "PC hardware in a Nutshell" book did a test
between a 7200 rpm scsi vs ide doing: gigabit ftp transfer and file copy
(about 1 GB in size) and tried to use the pc while doing such i/o such as
meansuring how long it takes to launch Ms word, etc. With the ide drive,
everything took magnitudes longer.
One advatange of SATA over PATA however, is its independent channel
(i.e. star/host bus topology) allowing concurrent access to devices.
because with pata, you cannot concurrently access a slave and master on
the same channel.
p.s. good article on scsi:
http://arstechnica.com/paedia/s/scsi-1.html
reinforcing what halcyon mentioned of why scsi performs better --
mainly because of the protocol involved.
now, if you measure performance while say, just copying a single file, or
any single threaded i/o, then there's practically NO difference between scsi
and ide with harddrives of the same rpm.
however, throw other i/o in the mix and the difference is huge!
the authors of O'Reilly's "PC hardware in a Nutshell" book did a test
between a 7200 rpm scsi vs ide doing: gigabit ftp transfer and file copy
(about 1 GB in size) and tried to use the pc while doing such i/o such as
meansuring how long it takes to launch Ms word, etc. With the ide drive,
everything took magnitudes longer.
One advatange of SATA over PATA however, is its independent channel
(i.e. star/host bus topology) allowing concurrent access to devices.
because with pata, you cannot concurrently access a slave and master on
the same channel.
p.s. good article on scsi:
http://arstechnica.com/paedia/s/scsi-1.html
Looking at the database at storagereview.com, from most quiet SCSI drive to loudest:
1. Seagate Barracuda 36ES2 -- 39.3 dB
2. Seagate Cheetah 36ES -- 44.6
3. Seagate Cheetah 15K.3 -- 45.1
4. Fujitsu MAP3147 -- 47.0
5. Maxtor Atlas 15K -- 47.6
... and others
Have a look at the site for more information; it is a good one. And don't forget drive performance.
FWIW, I got a 15K.3. The Barracuda has pretty lousy performance; the 15K.3 is faster than the 36ES (the half dB difference is negligable), and is roughly the same, but a bit lower, performance than the next two (and two to three dB matters to me).
DrJ
1. Seagate Barracuda 36ES2 -- 39.3 dB
2. Seagate Cheetah 36ES -- 44.6
3. Seagate Cheetah 15K.3 -- 45.1
4. Fujitsu MAP3147 -- 47.0
5. Maxtor Atlas 15K -- 47.6
... and others
Have a look at the site for more information; it is a good one. And don't forget drive performance.
FWIW, I got a 15K.3. The Barracuda has pretty lousy performance; the 15K.3 is faster than the 36ES (the half dB difference is negligable), and is roughly the same, but a bit lower, performance than the next two (and two to three dB matters to me).
DrJ
We in my previous work once replaced UW scsi Barracuda 7200RPM 4G drive with 30G 5600RPM lousy IDE crap 
As SCSI lover (i has both my Amigas on SCSI completely, damn you , IDE s*it!) i feared that the performace drop will be drastical
However much to my surprise, its othervise. The whole machine boot faster, Photoshop work with big files (swaping) faster (measuret little over 20% speed-up) and maybe only programs did not starting faster!
So, i come to conclusion that it must be the damn winblows or wheather, witch slow-down the 1Ghz Atlon with the UW SCSI
Into the tread, there are some interesting remars, however the truth is, that IDE and SCSI drives are almost same. The "better working some special commands" are true, however the performace gain it minimal, unless you do heavy multi-drive operations, witch windows large file cache did not handle
And there are something highly against scsi, witch no-one mention. And thats the sad fact, that 99% of SCSI are as external PCI card. And now the hard part. Even the teoretical PCI bandwitch are big, it never ever can reach or even tough it into reality. So, the IDE winning there in terms of the maximum bandwitch peak pretty easily. 133Mhz for defragmented file read is possible on all modern drives over 160G with decently fast (3G+) CPUs
But there is just where we begin! The PCI card share the bus with many other cards and devices and all this humping one innocent interrupt (IRQ sharing Win2k and higher MS OS). So, in reality, everything like huge network activity and sound + gfx, for example, WILL slow-down the SCSI notably.
On ATA crappy word is standard, that devices using DMA - they writing the data what CPU asked them for directly to memory W/O the CPU even assist Thats nice and cause the 0-1% CPU load during copying. However SCSI on PCI card lack of these abilities and ofcourse need to also transfer the data trought the bus - witch is shared with another cards (at least NIC and soundcard) and it will need CPU assistance in order to push the data to memory.
And last word - i learned that simple PC do job better. Just either the winblows support is terrible, or support for IDE is exceptional, while SCSI/RAIDs sufers? I don't know, but i know for sure i never install any more RAIDs into my machine, even i got kick-ass disk benchmarks 
As SCSI lover (i has both my Amigas on SCSI completely, damn you , IDE s*it!) i feared that the performace drop will be drastical
However much to my surprise, its othervise. The whole machine boot faster, Photoshop work with big files (swaping) faster (measuret little over 20% speed-up) and maybe only programs did not starting faster!
So, i come to conclusion that it must be the damn winblows or wheather, witch slow-down the 1Ghz Atlon with the UW SCSI
Into the tread, there are some interesting remars, however the truth is, that IDE and SCSI drives are almost same. The "better working some special commands" are true, however the performace gain it minimal, unless you do heavy multi-drive operations, witch windows large file cache did not handle
And there are something highly against scsi, witch no-one mention. And thats the sad fact, that 99% of SCSI are as external PCI card. And now the hard part. Even the teoretical PCI bandwitch are big, it never ever can reach or even tough it into reality. So, the IDE winning there in terms of the maximum bandwitch peak pretty easily. 133Mhz for defragmented file read is possible on all modern drives over 160G with decently fast (3G+) CPUs
But there is just where we begin! The PCI card share the bus with many other cards and devices and all this humping one innocent interrupt (IRQ sharing Win2k and higher MS OS). So, in reality, everything like huge network activity and sound + gfx, for example, WILL slow-down the SCSI notably.
On ATA crappy word is standard, that devices using DMA - they writing the data what CPU asked them for directly to memory W/O the CPU even assist
Thats nice and cause the 0-1% CPU load during copying. However SCSI on PCI card lack of these abilities and ofcourse need to also transfer the data trought the bus - witch is shared with another cards (at least NIC and soundcard) and it will need CPU assistance in order to push the data to memory.And last word - i learned that simple PC do job better. Just either the winblows support is terrible, or support for IDE is exceptional, while SCSI/RAIDs sufers?
I don't know, but i know for sure i never install any more RAIDs into my machine, even i got kick-ass disk benchmarks Well, windows does have some issues with their SCSI drivers -- this is well known (see the forums on www.2cpu.com, for example). I wouldn't know; I only use SCSI drives with FreeBSD; the windows machines all have IDEs.
SCSI controllers certainly have DMA capabilities, and have for a very long time. If yours does not, then it or the operating system is probably misconfigured.
For desktop applications, it is true that IDE and SCSI drives often give comparable performance. Partially this is due to controller algorithms -- those for SCSI tend to perform best in server-like environments (with heavy I/O); IDEs are tuned more for the desktop.
I would also guess that most SCSI drives are connected to a 64 bit bus, which are very common in server boards. Mine is, though I only needed a 33 MHz card for the U160 interface. That can easily handle two SCSIs delivering data at their maximum rate, and would probably only lose a bit on bursts with more drives attached. There are quite a few U320 controllers on the market if greater throughput is required.
Finally, the connection between the north and south bridges varies a lot. Mine has separate connections for the 64 bit and 32 bit channels, so activity on the 32 bit side of things does not affect SCSI performance at all. Most 32-bit cards don't consume a lot of bandwidth anyway.
I'm not trying to win converts here. SCSI drives are expensive, generate a lot of heat, noise and vibration. They also tend to be made much better and last longer. But when you need them, you really need them, and an IDE just won't do the job.
DrJ
SCSI controllers certainly have DMA capabilities, and have for a very long time. If yours does not, then it or the operating system is probably misconfigured.
For desktop applications, it is true that IDE and SCSI drives often give comparable performance. Partially this is due to controller algorithms -- those for SCSI tend to perform best in server-like environments (with heavy I/O); IDEs are tuned more for the desktop.
I would also guess that most SCSI drives are connected to a 64 bit bus, which are very common in server boards. Mine is, though I only needed a 33 MHz card for the U160 interface. That can easily handle two SCSIs delivering data at their maximum rate, and would probably only lose a bit on bursts with more drives attached. There are quite a few U320 controllers on the market if greater throughput is required.
Finally, the connection between the north and south bridges varies a lot. Mine has separate connections for the 64 bit and 32 bit channels, so activity on the 32 bit side of things does not affect SCSI performance at all. Most 32-bit cards don't consume a lot of bandwidth anyway.
I'm not trying to win converts here. SCSI drives are expensive, generate a lot of heat, noise and vibration. They also tend to be made much better and last longer. But when you need them, you really need them, and an IDE just won't do the job.
DrJ