What size image files do you typically work with? If you have plenty of RAM, disk performance becomes less of an issue as to how fast you can work. I would suggest that RAID 0 can be immensely frustrating if reliability is an issue.
Gary
My image file size can get up to about 500 MB but I presume could get higher with the new PS version supporting 16 bits. I plan to use 2GB RAM in the system. I am leaning against RAID because of reliability issues and the risc of data loss with drive failure in a RAID 0 system.
Assuming no RAID, but the ability to use up to 4 drives, what would the optimum configuration be? Specifically, would having the OS, PS program, and scratch disc ALL on 3 physically separate discs significantly improve performance compared to having the OS and actual PS program on one disc and the scratch disc on a physically separate disc? If it would help significantly to have 3 physically distinct discs, how much would performance dcrease if the PS program itself were on a 7200RPM HD with the other two discs being 10000RPM (as opposed to all 3 discs being 10000RPM)?
Thanks.
Howard
I worked the other way around. I am using my system for both Photoshop and Premiere.
Drive 1 (720O RPM IDE 133 drive) for PS and OS
Drive 2 (2x 120GB 7200 RPM SATA drives, combined in RAID 0) for data
Works fine so far (but the system is only runnign for a week now), but considering changing Drive 1 also to a raid 0 or 1 configuration (depending what will frustrate me more drive-speed or data-loss ;-).
Asus has a motherboard (P4C800 Deluxe) which support to sets of RAID: 1x S-ATA and 1x IDE133 (but am sure Asus is not the only one)
Just my 2 Euro-Cents
Nils
dear Howard-
obviously RAID 0 is the fastest hard drive setup (and with SCSI drives to boot) but you really won’t see huge PS performance increase with this-
PS is all about RAM -when it runs out, yer done. if you’re doing 500MB files you will definitely run out of RAM, and sadly, theres nothing you can really do about it. the application limit for 32 bit procs is 2GB, and PS won’t even give you the full 2gB to use, on my system, i have 4gb of RAM but PS’s will only actually use 1777MB.
now, i did tests to see if changing the scratch disk would help. putting scratch on a separate hard drive as your OS, any hard drive, will definitely be faster, but the actual SPEED of that drive is irrelevant -ie putting your scratch on a SCSI drive as opposed to an IDE has no real difference. I even tested putting the scratch on a ramdisk, but it wasn’t that much faster. for those big files what u really need is a 64bit version of PS AND a new 64 bit chip AND a 64 bit version of windows -then PS could use more RAM and those 500MB files would fly -perhaps in 2004 sometime?
the only thing u can really do now is to work with one large file at a time, lower history, and hope u dont use up that RAM. performance drops i’d say some 800% when PS runs out of RAM. fast drives will NOT help at all, although windows and PS will load and run better on a nice SCSI setup. RAID and SCSI are really more helpful for video.
I am planning to use 2 GB RAM, realizing that PS will not be able to use all of it. I am not sure why drive speed and RAID 0 will not help with the scratch disc. After reading the prior post I was thinking that perhaps it would be worthwhile to set up a RAID 0 array with two 10000RPM discs to serve as the scratch and also hold image files currently being worked on while having PS and the OS on a separate disc. Since when RAM runs out PS must use the scratch disc why will increasing the read/write speed to the scratch by using faster drives in a RAID 0 not improve speed. I would think that by increasing read/write by as much as 40% with faster dives and RAID 0 should improve the manipulations PS must do to the scratch to use it. Why is this not the case?
As far as 64 bit systems I have agonized over this for a while..build now or wait…of course there is always something else to wait for :>)
I think it will be some time until Microsoft has a 64 bit OS that is out of beta testing AND PS has a 64 bit version AND even then putting in gigs and gigs of RAM can get pricey!
Howard
you have a point: i too thought that "in theory" there should be a way to maximize scratch disk usage, but the gains are minimal. Chris Cox said a ramdisk wouldnt help, so i did an experiment:
tested 3 setups:
one with scratch on the same drive -C:/
one with the scratch on a different drive E:/
one with the scratch set to a RamDisk of 1.5 GB called drive R:/
now, once PS runs out of ram, it should use the scratch, and since the OS is "tricked" by the ramdisk to think of the extra ram as a separate hard disk, the speed you would think should at least APPROACH what it was when using memory -but i only noticed improvements of some 10-20% using the ramdisk -and remember, ram is like a thousand times faster than the fastest SCSI raid setup u could ever create. so, while i have not tested it, i can only assume that if a ramdisk does not help, why would RAID? on all 3 setups, times were IDENTICAL when PS had enough RAM to use.
interestingly, just putting the scratch disk on the diff drive resulted in performance gains I’d say of somewhere from 200-300%: the ramdisk only added some 10-20%. the real problem is that once PS runs out of RAM, it runs some 700-800% WORSE. for whatever reason, i dont think PS uses ram and scratch efficiently, my conservative estimates showed that PS wanted some 16 TIMES the file size for both Ram and scratch, thus, when I had ressed up a file say from 50-500MB, I had used up all the Ram (1777MB) and had a scratch temp file that was over 5 GB! this is 64 bit territory now.
didn’t write PS for sure, but for some reson, PS "uses" the scratch disk much more innefficiently than it uses RAM -even when the scratch actually "IS" RAM! seems PS is all about RAM, not aboud hard drives. they tell me that video editing: ala After Effects, and Avid, is exactly the reverse: files are so huge its all about hard drive speed, and not much about RAM.
SCSI or RAID may make sense for you anway though, if so, do some tests and let us know. my setup proved to me that PS would simply rather have a separate drive (any separate drive), even when that drive (an IDE) is twice as slow as the one you have the OS on (SCSI).
From what I’ve gleaned here from Adobe people is that RAM does cache or buffer what is written to the scratch space, and the more RAM it has to work with, the more it can wait before writing from RAM to the cache. I’ve got Windows on one drive, programs and Windows swap file on another, and Photoshop scratch on a RAID 0 partition. It works well for me. The largest images I work on start out in the half-gig range and get reduced from there. I do have to wait, but rarely enough to go to the kitchen for another cup of coffee.
Ellen,
once PS runs out of ram, it should use the scratch, <<
That’s not how Photoshop works. Under most conditions* Photoshop uses the Scratch file as its PRIME memory. It uses RAM as a Cache for the scratch file. So it is not a case "once PS runs out of ram" at all. You will ALWAYS has a scratch file, regardless of how small the image or how large your amount of available RAM.
* Some filter operations are carried out entirely in real RAM
Tim, I know there is marked benefit to having a physically seaparate scratch disc, but is there a benefit to putting Photoshop on a physically different disc than the OS (ie having a three disc setup with OS, PS program, and scratch all physically separate)??
Also, is there benefit to having the Windows swap on a physically different disc than Windows itself, as compared to, say, a separate partition on the OS disc?
Howard
Len, if that is the case regarding PS always using the scratch disc, then doesn’t it also follow that the faster your ability to read/write from the scratch (ie a faster hard drive or a RAID array) the better PS performance should be?
Or, as suggested, is there some reason that the faster scratch access doesn’t help??
It seems that the OS or the applications in use are always going to want to be going back to disk for some library or another, and that may be stored under the OS umbrella, or with the application. I’ve been curious, but I don’t have a tool to watch which disk is being accessed at any particular moment. Intuitively, it would seem that having separate disks for OS, App, OS page file, App scratch would be ideal. If you’re going to gang two of them together, it might be better to have OS and App together. Maybe someone who know more about the hardware workings will jump in.
A side question, is the memory assigned to Pshop shared between the application and the data, or is it just dataspace?
As to your question to Len, that’s why I put a RAID 0 together, for the fastest access for Pshop to scratch.
Len, if that is the case regarding PS always using the scratch disc, then doesn’t it also follow that the faster your ability to read/write from the scratch (ie a faster hard drive or a RAID array) the better PS performance should be?
Yes.
A side question, is the memory assigned to Pshop shared between the application and the data, or is it just dataspace?
I think I recall Chris saying it’s shared.
Howard,
Yes, that follows, but Photoshop will attempt to use ‘free time’ to write/read the scratch wherever possible, so faster drives won’t always show the improvement anticipated.
If all accesses can be made during ‘free time’ you wouldn’t notice any improvement in having a fast raid, whereas with very large images with multiple layers you may notice a considerable improvement.
ellen – part of what you’re missing is the OS gets involved in reading and writing to disk. And windows likes to cache things in RAM (which makes RAM disks UGLY).
Also, if you’re going to do RAID — get a dedicated RAID controller card, don’t just use the OS.
Chris, this particular motherboard is new and has on board RAID so it will be a hardware, as opposed to software RAID…so am I correct in assuming that you believe a RAID 0 scratch will speed things up?
Howard
Chris…since you are listening :>)…..is there any advantage to putting the PS program on a separate disc from the OS and scratch (3 discs) or are the gains to be had from this miniscule (particularly if the OS disc is 10000RPM and the program disc 7200)?
Thanks.
Howard
Miniscule gain.
The biggest gain is having the scratch on a fast drive, different from the OS swapfile volume.
looks like I stand corrected on a few things.
you guys got me thinking, so i did some new tests with the same action. remember, I have 3 hard drives: 1x 15K SCSI and 2 x 7200 IDE drives:
once again, all times were identical until RAM ran out, then:
SETUP #1>>>>>>>>>>>>>>>>>>>>SETUP #2>>>>>>>>>>>>>>SETUP #3>>>>>>
OS on SCSI>>>>>>>>>>>>>>>>>>OS on SCSI>>>>>>>>>>>>>OS on IDE
PS on the same SCSI>>>>>>>>>PS on diff IDE>>>>>>>>>PS on diff IDE
Scratch on a diff IDE>>>>>>>Scratch on same SCSI>>>Scratch on diff SCSI
21.2>>>>>>>>>>>>>>>>>>>>>>>>30.3>>>>>>>>>>>>>>>>>>>40
35.3>>>>>>>>>>>>>>>>>>>>>>>>40.5>>>>>>>>>>>>>>>>>>>39.6
42.3>>>>>>>>>>>>>>>>>>>>>>>>52.3>>>>>>>>>>>>>>>>>>>56.4
——————————————————— 98.8 seconds >>>>>>>>>>>>>>>123.1 seconds>>>>>>>>>>136 seconds
so maybe you guys can explain this: with my test here, the "slowest" scratch drive won! (setup #1)
it seems PS prefers both the OS AND PS7 itslef to be on the fastest drive, whether the scratch itself was on the fastest drive -seemed LESS important.
I would have thought setup 3 would be the fastest, since all 3 are on totally diff drives, and the scratch was on the fastest SCSI; but no, in my little test here, it seems ironically that the speed of the scratch disk is actually less impt than the speed of your OS or PS drive.
my rankings then for the most impt performance factors then is:
1)free RAM -under 1777MB, PS blazes away -but when this is all used up:
2) then having a separate hard drive for scratch
3) speed of your OS and PS drives is NEXT most impt
4) finally, speed of scratch disk -the least impt!
any takers on my rankings here? look, I’m certainly not going to match wits with Chris Cox here, but my results seem to contradict his last post: in my scenario, with only 1 fast drive, you should make it your OS and PS program drive, NOT your scratch.
The only thing that comes to mind is that in your particular hardware, the bottleneck is in the bus talking to the SCSI controller talking to the drive vs. bus talking to on-board controller. I suspect that the program’s scratching is going to get beyond any burst write speeds to the drives. If you want to test this, copy a large file (tens of MB at least) from one IDE to the other, and the same file from the original IDE source to the SCSI drive. What is the state of fragmentation on all of your drives?
In terms of optmizing performance is there any benefit to putting the Windows Swap file on a separate partition than the OS but on the same physical disc? Is there any advantage to putting it on a separate disc than the OS if that disc is slower than the OS disc (ie 7200RPM instead of 10000RPM)? What about putting the Swap file on a fast RAID array…but that array is the same array as used for the scratch disc…is it correct to assume that will slow things down?
I am learning a great deal from you folks on this thread…thanks.
Howard
am using win xp pro, sp3
copied a 875 MB file from drive to drive. do these numbers tell you guys anything?
from IDE #1 to IDE #2:———–27 seconds (32.3 MB/s)
from SCSI to IDE——————-15 seconds (58.2 MB/s) from IDE to SCSI——————27 seconds
from IDE to same IDE————–15 seconds
from SCSI to same SCSI————-27 seconds
have to admit this is above my head, I don’t really ‘know’ enough about buses and exactly how SCSI works -but are you suggesting that my SCSI drive is slower than it should be? even if it was, i still don’t understand why the 1st setup won… or why copying a file from within the SCSI onto the same SCSI is slower than on IDE…
re: swapfile
on my main boot OS on the SCSI (setup #1 and #2), i have turned OFF the pagefile completely. i have 4gb ram, the default always gave me a 2GB swapfile, and this seemed overkill. i know time and time again people on this forum say this is ‘bad’ -but this is the OS that runs the best! it never crashes -and in my PS action test, this is the fastest.
for setup #3, i used my backup OS installed on my western digital -and this was the poorest performer with the ‘default’ 2GB pagefile.
while i highly respect the opinions here in these forums, i sometimes find that performance on my system runs contrary to ‘popular’ opinion.
ie: the best PS performance i get is on a SCSI boot with no pagefile whatsoever, and a separate IDE scratch.
ok, my final numbers, i’ve put my work in on this one! to test if it was my SCSI that was the prob, i removed it totally from test 4:
Setup # 4———————Setup # 5——– –Setup #6
OS & PS on same IDE——–OS & PS on same IDE—-OS & PS on same SCSI Scratch on diff IDE——–Scr on SCSI———– Scr on same SCSI
44.0————————-40.6——————30.9
46.5————————-39.5——————40.4
57.0————————-56.0——————51.1
____________________________________________________________ 147.5————————136.1—————–122.4
Howard:
#6 versus #2 differed only by a second, showing that PS itself doesn’t seen to care much about being installed on a faster (or even separate) drive, so it’s probably a waste to have a sep drive just for it: its really all just about OS and scratch:
once again, the faster SCSI drive worked best AS THE OS (+50% improvement), NOT scratch, (only +10%).
#5 had the faster SCSI scratch and improved from #4, but as I’ve said, NOT AS MUCH as improving your boot drive, and just having a separate (albeit ‘slower’) scratch:
SETUP #1
OS & PS on same SCSI
Scratch on a diff IDE
21.2
35.3
42.3
——
98.8 seconds
Howard,
any benefit to putting the Windows Swap file on a separate partition than the OS but on the same physical disc?
No, in fact there may be a penalty, as the drive has to work harder, moving back and forth between the two partitions. If you can, put the Windows swapfile on a different physical drive.
… putting Windows swap file on slower disc …
There may be a penalty in performance, but not one you would notice easilty. And putting it on the same RAID array as the PS scratch file would improve the swapfile’s performance, balanced against the performance penalty/issue of the swap and scratch files on the same disk.
This is a great thread in terms of dissecting PS performance as it relates to hardware!
So it seems like after all is said and done there are a few setups contending for most likely to give the best performance:
1) OS on 10000RPM disc with Swap File on same disc, PS program on a separate partition of the same disc and a separate RAID 0 array for the scratch disc.
2) PS program and OS on a 10000RPM disc with Swap file on a separate partition on a separate disc at 7200 RPM disc (this disc will be there anyway for long term storage) and the Scratch on a separate RAID 0 array.
Which of these would be best. I guess this is tantamount to asking if putting the Swap on a separate disc but at 7200 RPM instead of 10000RPM like the OS is better than keeping the OS and Swap on the 10000RPM disc. Anyone know for sure? Can the system read/write to more than one disc simulaneously (in a non-RAID) or is the benefit of havin a separate Swap disc that it doesn’t have to search?
Also what is the optimal size for a Swap partition if put on another disc?
Howard
Howard,
I think you are getting into the "hard to measure the differences" area 🙂
I would go with option number 2, simply because the OS/application, swapfile, and Photoshop scratch file are all on different drives.
As far as the optimal size for a swap partition, Windows won’t let you create a swapfile bigger than 4096MB per partition. I would just put the swapfile in the most used partition.
I don’t have all the different drive speeds and controller you appear to have, but here is my setup.
IDE ATA133 Controller
IDE bus 1 primary – 80GB 7200rpm – contains OS and all applications, and Photoshop scratch file IDE bus 1 secondary – 40GB 7200rpm – contains swapfile and other temp stuff IDE bus 2 primary – 120GB 7200 rpm – contains datafiles
This works exceptionally well, and I have no issues with apparant performance.
The only advantage to putting Win paging on a separate partition is that if you don’t set it big enough, and it expands, it will most likely be fragmented. I’m not so fond of MS attempt to mix everything up anyway, TweakUI will let you specify default locations for programs, shared files, user files and so forth. Worth a look if you’ve got anything more than "c:\".
Regarding Ellen’s tests-
With your enormous amount of installed memory (4 Gig), using inconsistent windows page filing systems is going to heavily skew your results; having enough RAM to put the page file in rather than write is to disk is going to be considerably faster, and not using this same scheme for all the drive setup’s will certainly make some drive configurations look faster than others. To accurately test the different drive arrangements, they should all use the same method for handling the page file- otherwise you arbitrarily give advantage to one set up and the results no longer are useful for forming an informed opinion… for me, anyway.
Most motherboards available only offer 4 DDR memory slots; on such a board, 4 Gig of ram would require 1 Gig sticks- at nearly triple the cost of 512 meg sticks. While price wasn’t indicated as an issue, having that much memory available should change the way we tackle optimizing Photoshop; with only 2 Gig of memory I doubt very much that using most of it for the Windows Page File would be optimal.
