Ad
Ad
Ad
Pages: [1]   Bottom of Page
Print
Author Topic: Multi-core "save" speeds. What's appropriate?  (Read 1583 times)
Michael H. Cothran
Full Member
***
Offline Offline

Posts: 136



« on: August 17, 2012, 07:29:22 PM »
ReplyReply

A couple of years ago I bought a Mac Pro 8-core 2.23 GHZ, and have been very disappointed with its processing speeds, particularly with saving PS files.

Timing my saves with PS files, here's my averages:  for a 500 MB file - 67 seconds (1 minute, 7 seconds), and for a 1 GB file - 151 seconds (2 minutes, 31 seconds). Doesn't that sound awfully slow, or am I just expecting too much? My old 1.8 GHZ G4, which this unit replaces, was almost as fast.

I don't understand a thing about multi-core processors, and whether or not it's something I personally need. Trying to find specific information as to what they do is something I've yet to find. Even the Apple Store couldn't offer me a definitive explanation. I went with 8-core simply because the sales person promised it would be fast as lightning, and run rings around my G4. Well...Not happening!

So I'd like to try something else. This time I'm thinking a Mac Pro with less cores, but a larger, faster processor. Does that make any logical sense? Am I wasting my time & money with this move?
PS is the biggest app I have, plus a few plug-ins like Nik and Photo Kit Sharpener. I have Microsoft Office, plus whatever came with the computer, and that's it. Rarely do I run anything in tangent with PS, other than iTunes.
Logged
Schewe
Sr. Member
****
Offline Offline

Posts: 5253


WWW
« Reply #1 on: August 17, 2012, 09:48:08 PM »
ReplyReply

Faster hard drives...the primary save bottleneck is drive speed.
Logged
Chris Pollock
Full Member
***
Offline Offline

Posts: 213


« Reply #2 on: August 17, 2012, 10:23:53 PM »
ReplyReply

One gigabyte in 151 seconds works out to about 6.8 megabytes per second. Just about any modern hard disk should be much faster than that. I'm guessing that the bottleneck in this case is compressing the data prior to saving.

Multiple CPU cores only help if the software is multithreaded. Some parts of Photoshop are multithreaded, but I don't know if the code for saving files is. It should be easy enough to find out - just monitor CPU usage while saving a large file. If it uses more than a full CPU core (e.g. more than 25% on a 4 core machine) there's at least some multithreading going on. I'll try this myself the next time I use Photoshop.

Another possible problem is lack of memory. If you don't have enough Photoshop may be forced to swap data out to the disk, which will drastically reduce performance. My 8GB machine suffers from this problem when I work on large panoramas. I plan to replace it with a 64GB workstation early next year.
Logged
Schewe
Sr. Member
****
Offline Offline

Posts: 5253


WWW
« Reply #3 on: August 17, 2012, 10:56:00 PM »
ReplyReply

One gigabyte in 151 seconds works out to about 6.8 megabytes per second. Just about any modern hard disk should be much faster than that. I'm guessing that the bottleneck in this case is compressing the data prior to saving.

Again, I think you need to look at HD speeds...I can save a flat 500MB 8 bit PSD in 6-7 seconds...a file of 500mbs with 1/2 dozen layers the time jumps up to 15-16 seconds (and this is in CS6 with background saving on which is a bit slower because it's a background process). The HD speed is directly tied to save times.

Now, as to my drives, it's a 6 drive RAID 0 stripped array...which is, yes, very, very fast. It tests out at 311 MBs/sec sequential reads and 275MB/sec sequential writes...

The other thing the OP hasn't bothered to mention in RAM...if the entire operation will NOT fit into RAM and a scratch disk is involved, then there will be a lot of disk thrashing when saving. Another speed bottleneck. If the scratch disk is on the same drive as the file is saving to, then it's an even bigger bottleneck.

The only time I ever really see slower saves is when saving large 16-bit multi-layered files as TIFF using zip compression because that's really CPU intensive. Unless I'm saving out an output file that needs to be FTP'ed I don't use zip compression...
Logged
pindman
Full Member
***
Offline Offline

Posts: 191


« Reply #4 on: August 18, 2012, 12:22:27 AM »
ReplyReply

Thats something we've been discussing for the last few days on a nearby thread http://www.luminous-landscape.com/forum/index.php?topic=69652.0

From my testing on three hard drive systems, drive speed makes a difference in reading or saving flat files.  But once you add multiple layers I found that that drive speed had virtually no effect as processing the file was dependant on the speed of the CPU.  Even with a dual Xeon 3.3Mhz system layered files to 3 times longer to save than a flat file of identical size.

Paul
Logged
Schewe
Sr. Member
****
Offline Offline

Posts: 5253


WWW
« Reply #5 on: August 18, 2012, 04:02:17 AM »
ReplyReply

Again, you can NOT simply eliminate hard drive speed as a major factor...it's indeed a major factor.

Yes, adding layers and channels will have an impact vs. flat files–so will compression options, but HD speed is the largest factor (IMHO).
Logged
Steve Weldon
Sr. Member
****
Offline Offline

Posts: 1441



WWW
« Reply #6 on: August 18, 2012, 04:43:03 AM »
ReplyReply

Again, you can NOT simply eliminate hard drive speed as a major factor...it's indeed a major factor.

Yes, adding layers and channels will have an impact vs. flat files–so will compression options, but HD speed is the largest factor (IMHO).

1.  Absolutely. 

2.  The current bottleneck would depend on the system.  I suppose in a system with a slow enough processor, inadequate RAM resources forcing scratch disk use, other resources funnelling off CPU/RAM resources, heat issues forcing the CPU clock speed to safe mode, etc, etc.. the bottleneck 'could' be other than the HD speed.  Unlikely with a modern well maintained system,  but I've seen more than a few systems barely limping along..   A fast SSD could 'shift' the current bottleneck.  I've seem systems with fast SSD's for OS/System and work drives.. but they've got their catalog/scratch/cache on some 60g notebook drive pulled from some system in the closet "because it had enough space.."    Granted, unlikely.. but they're out there.  And it seems many experiencing these symptoms and asking these questions own a few of them..  Roll Eyes       Not nearly enough users understand the relationship between hardware type/speeds and workflow.. and where the two shall meet.
Logged

----------------------------------------------
http://www.BangkokImages.com
Michael H. Cothran
Full Member
***
Offline Offline

Posts: 136



« Reply #7 on: August 18, 2012, 09:12:59 AM »
ReplyReply

Sorry for not posting RAM.
I have 10 GB of RAM, 1.64 TB of hard drive. Mostly unused at this point. The unit came with a 640 GB HD, and I had the store add a third party 1 TB HD. I keep my PS files on the 1TB drive. Just now I copied a 1 GB PS file from my 1 TB HD over to my original Mac 640 GB HD to test differences in save speed. Sadly, each one still took about 2.5 minutes.
I checked under "Get Info"  and "About This Mac" for any data I could find on the HD's, but could not find anything about speed. I don't know what hard drive speed is. Is it different than processor speed? BTW - just now checked - my processor is two quad cores, 2.26 GHZ.

I appreciate the responses so far, but I have to remind you that I am NOT that computer literate. I know enough to be dangerous, as they say. Which is why I'm inquiring here - I respect the knowledge that most of you have.
« Last Edit: August 18, 2012, 09:20:15 AM by Michael H. Cothran » Logged
Chris Pollock
Full Member
***
Offline Offline

Posts: 213


« Reply #8 on: August 20, 2012, 06:33:49 AM »
ReplyReply

Yes, adding layers and channels will have an impact vs. flat files–so will compression options, but HD speed is the largest factor (IMHO).
If the hard drive is the major factor, why is his net data transfer speed (amount of data divided by time required to save) at least an order of magnitude slower than the writing speed of a modern hard drive?
Logged
hjulenissen
Sr. Member
****
Offline Offline

Posts: 1615


« Reply #9 on: August 20, 2012, 06:58:21 AM »
ReplyReply

If the hard drive is the major factor, why is his net data transfer speed (amount of data divided by time required to save) at least an order of magnitude slower than the writing speed of a modern hard drive?
If the disk usage pattern is very different from that used to estimate hard drive speeds (i.e. many small fragments instead of one large chunk), then one might see significantly smaller speeds than those obtained in tests.

But I agree with you: if observed time to store an e.g. 30 MB file is an order-of-magnitude larger than that predicted from drive maximum save speed, this is a good indicator that main limiting factor is something else (cpu-bound, memory-size-bound, or something else).


Like others have said, the speedup of 8 cores over 1 core (everything else being equal) tends to be in the [1x - 8x] range. Without knowledge about the application (through experiments or inside knowledge) it can be quite hard to predict what end of that range a given application/usage mode fills.

-h
Logged
Pages: [1]   Top of Page
Print
Jump to:  

Ad
Ad
Ad