Luminous Landscape Forum
Equipment & Techniques => Computers & Peripherals => Topic started by: pindman on August 14, 2012, 12:46:48 am
-
I just set up a HP Z820 Windows 7 Pro workstation:
64GB Ram
Dual Xeon 3.3 GHz processors
Drives and my benchmarks:
300GB Intel SSD system drive (256/204 MB/s sequential Read/Write time)
Raid 5 Array, 6Gbits/s, of 4x4TB drives ((490/96 (no typo) sequential Read/Write time))
OWC Accelsior 960 GB PCIe drive (531/496 sequential Read/Write time)
I copied tiff files of approximately 4GB and a PSD of 700MB onto each of the drives. Then opened them in Photoshop CS6. There was essentially no difference in the time required to open the files between the drives. Then I saved them in non-compressed TIF (for those opened as a TIF) or non-compressed PSD for the file originally in PSD. Again, no measureable difference in time to save file.
On the other hand, Lloyd Chambers shows a marked improvement using the Accelsior PCIe drive: http://macperformanceguide.com/blog/2012/20120530_1-OWC-Mercury-Accelsior-960GB.html (which is why I ordered the Accelsior).
I had originally planned to use the Accelsior to work on large files and then store them on the Raid drives. Given the fast dual Xeons and 64GB of memory, I would think that drive speed would be a limiting factor in opening and saving files, but if it isn't then there's no point in keeping the Accelsior. But I'm wondering why there was such a marked improvement on a MacPro but not my machine.
Any thoughts?
Thanks!
Paul
-
It looks like all the drives you are comparing are quite fast for read speed. I believe that Photoshop uses just one single CPU thread for opening/saving files, so having an 8 core CPU isn't going to help you. Perhaps you've crossed the point where the hard drives are all faster than the CPU is processing the data? If that's the case, then they would all open up the image in the same amount of time since they are CPU limited.
FWIW, I compared opening a 3.8GB Tiff file on my machine. With the file on a conventional hard drive (2TB WD Caviar Black) it takes about 23 seconds to open the image. With the file on my primary SSD (Crucial M4 128GB SSD) the image opens up in about 4 seconds. This is with a 4.4GHz i7 3770k processor.
-
If you have any virus and/or spyware detecting software try disabling those for the sake of testing. You might disable scanning of, for example, .psd .tiff and so on. If possible disable that kind of thing for active monitoring, but enable it for doing weekly full system scans, or as you do.
Also check to see how much RAM PS is configured to use.
In Photoshop, make sure that the scratch drive is on a different location than the computer’s system drive (usually the C drive). Also while it’s typically a good idea to make sure that the windows paging file is on a separate drive from the computer’s system drive, with a SSD drive this may not be an issue.
I thought I read somewhere that cs6 does background saves, as a new feature of cs6. If this is the case, the save will probably be a lower priority process than it was previously. I don’t know if this feature is something that can be turned on or off.
If you have the Windows indexing service turned on, disable it.
In Windows, open task manager, click on the performance tab and open the resource manager. Do a single large file read and watch for bottle necks. Also do a single large file save and do the same. The bottle necks will show you where the performance issues are.
-
@Justan... All virus software disabled. PS configured to use most of 64GB RAM. In resource monitor (reading 4GB file) it appears that the CPU is at max when reading from the Accelsior SSD.
@Sheldon N... It looks like your machine is overclocked, and beating the pants off a dual Xeon 3.3GHz machine. And task manager shows all 8 cores being used during opening and saving files.
But the CPU in the MacPro used in Lloyd Chambers tests (Mac Performance Guide) isn't faster than mine. So why is the CPU a bottleneck for me and not the MacPro? Any other suggestions?
Thanks.
Paul
-
PS configured to use most of 64GB RAM.
Hi Paul,
Have you tried assigning less RAM to Photoshop?
Cheers,
Bart
-
Hi Paul,
what are you using to monitor the speed?
I too run a dual XEON 2.66 @ 3Ghz
96GB of ram
4x Intel ssd RAID-0 for OS
8x Intel ssd RAID-0 for temp
8x seagate SV35 RAID-0 for data
let me know and I will run the test, just copying within windows and looking at the copy window with add more detail, I see speed over 1GB/sec, but I have to be quick to see it, when copy and pasting a 4GB TIFF file
Henrik
-
@ BertvanderWolf... tried reducing RAM to photoshop... no difference
@ tived... Wow—what a system!!! I used a several programs to benchmark drive speeds. The issue is not in copying files, which is clearly faster with the Accelsior, but in the time Photoshop takes to open and write files. It takes PS6 26 seconds to open and 30 seconds to write (with no compression) a 3.9 GB layered TIFF file on the Accelsior PCIe card drive. I'd love to know how long it takes to open a comparable size TIFF on your machine.
Thanks.
Paul
-
Paul,
can you give me the pixel dimention and bit dept of the file you are testing with and i will create one and run a test later today
thanks
Henrik
-
Much appreciated, Henrik! I just did a little experiment, which I think answers the question of why the save time was the same for all drives. I was using a TIF file with multiple adjustment layers for the original tests. I flattened the the image, and then upsized so the file size was the same. This shortened saving to all drives, and saving was MUCH faster on the Accelsior PCIe hard drive as compared to the standard drives.
This indicates that the most of the time needed for saving large multi-layered files is in processing the file, and that writing to disk speed is of secondary importance.
I really appreciate your offering to make a file to test, but it seems that unless the layers were the same we wouldn't get comparable results. If you are willing, I can downsize the file to 2GB and send it by dropbox, or you can send me a multilayered file to compare time required to open and save it. I'll send you a PM with my email.
Again, thanks!
Paul
-
Paul,
just give me the pixel size and bit dept, and i can recreate a file and test with, no need to send a file ;-)
concept it the same
Thanks
Henrik
-
7760x1038, 360 pixels/inch. 16 bit RGB. 15 adjustment layers... mostly curves, one B&W.
PS shows 46/GM/1.48G, but saved TIF is 4.188Gb.
File with all the layer takes about 29 sec to load and about the same to save. If I flatten the file, and then upsize so it is about 4Gb PS will open and save it in about 6 seconds.
In real life I'm saving files with layers. It is of little advantage to save files without adjustments more quickly. If there the SSD will save time in the every day workflow I'll keep it; if not I'd rather spend the money on a trip to Australia :)
Your thoughts....?
Much appreciated!
Paul
-
As I've been looking recently for some SSDs I notice that your Intel SSD read/write could be better. It's almost as if they are on SATA 3GB/s; many of the newer SSD have above 500 MB/s. Just a thought.
See here:
http://www.tomshardware.com/reviews/buy-ssd-recommendation,3255.html
http://blog.laptopmag.com/whats-the-best-ssd-5-drives-tested
-
For sure the SSD boot drive is not stellar. In buying a workstation, rather than building one, there is a limited choice of components. I asked about a different SSD and HP's comment was that their supplied drives were tested and certified to work with the system. Since the drive is only used for booting and loading applications, it doesn't affect performance once a program is being used. And applications are very quick to load.
If you look at my last post, you'll see that flat files load and save much quicker on the very fast PCIe SSD, but there is no advantage in time to load large layered files. I agree with a note Sheldon sent me: use the PCIe SSD card for lightroom catalags and Photoshop scratch files, but store the photo files on the mechanical raid drives.
Best,
Paul
-
All modern OS's use 'unused' RAM for buffering disk reads. You have 64 gigs and the files are puny 4 gigs so I assume you're holding 58G+ unused RAM. So it might be the case if you didn't reboot your machine between the tests the photoshop is actually reading the image from the RAM and not from the disk(s).
If that's the case your file opening speed is limited by the speed of your RAM and the speed of your CPU. Might have been a better idea to go with 16-24 gigs of really low-latency RAM coupled with nicely overclocked FSB.
-
Coming late to the party here, but I suggest you save any layered files as PSD, not TIFF. If nothing else, you will have significantly smaller files.
-
HI Sty,
The file may be able to be held in ram, but you still have to write it back to disk when saving, the bottleneck is still the disk (disk be a single drive or an array)
Paul,
sorry for my absence, but I am rebuilding and waiting for a new RAID controller card Areca 1882xi 24 channel with 4GB cache and Battery Backup Unit, that should give the system a little boost ;-)
Henrik
All modern OS's use 'unused' RAM for buffering disk reads. You have 64 gigs and the files are puny 4 gigs so I assume you're holding 58G+ unused RAM. So it might be the case if you didn't reboot your machine between the tests the photoshop is actually reading the image from the RAM and not from the disk(s).
If that's the case your file opening speed is limited by the speed of your RAM and the speed of your CPU. Might have been a better idea to go with 16-24 gigs of really low-latency RAM coupled with nicely overclocked FSB.
-
What you are looking for is clock speed, which is very limited using the Xeon platform. Multicores and virtual cores certainly help sometimes, but simply having them run faster greatly improves other things - as does a faster bus speed and ram speed, other important parts of overclocking. Xeons are the only choice for huge amounts of ram, but you can not match the pure speed of a Sandybridge or Ivy Bridge based overclocked system for certain operations. It's good that you are figuring out where your particular bottlenecks are for your workflow though.
-
Hi kaelaria,
The Xeon's are slowly getting dated, but with my particular mainboard, the evga SR-2, i can still push the CPU's to past 4+GHz and with 96GB of ram and various ssd/disk arrays - i would say its still a relevant setup today and the next few years.
but you are quite right its about clock speed, but its also about how fast you can get the data to and from disk, and this is the bottleneck in all systems to day
Henrik
What you are looking for is clock speed, which is very limited using the Xeon platform. Multicores and virtual cores certainly help sometimes, but simply having them run faster greatly improves other things - as does a faster bus speed and ram speed, other important parts of overclocking. Xeons are the only choice for huge amounts of ram, but you can not match the pure speed of a Sandybridge or Ivy Bridge based overclocked system for certain operations. It's good that you are figuring out where your particular bottlenecks are for your workflow though.
-
The file may be able to be held in ram, but you still have to write it back to disk when saving, the bottleneck is still the disk (disk be a single drive or an array)
When saving the file, it is cached in system memory first, and then from cache it is written to disk. So if you have e.g. 10GB free RAM, when you save what will become a 4GB PSD, that whole PSD is "written" to cache. From the perspective of the application, the save operation will be incredibly fast, perhaps merely a few seconds.
Because you have no control over the caching algorithm, or how it purges the cache, you cannot reliably benchmark drives from Photoshop. You need a benchmarking tool that disables the system read/write cache.
-
I'm not saying it's slow or irrelevant - but I'm saying the server/Xeon based platform is limited compared to traditional ones for certain operations. For example you're limited to 1333 ram, I'm running nearly double that speed and don't need to use ECC sticks for even better speed. My CPU can go above 5GHz, again, better for some operations. Of course I'm limited to a single socket, so multicore operations suffer in comparison to your dual. Everything has a pro and a con. Personally I find PS not at all CPU bound, and unless I'm doing video work never peak my cores out for more than a couple seconds. PS ha a lot of operations that just like speed over cores. Bus, ram and CPU speed.
-
Moving slightly away from topic, but the idea with building a computer is to minimize the bottleneck.
Most computers off the shelf have a huge bottleneck in their storage sub-system, be it Mac or PC. Compared to CPU, GPU and RAM, the storage is far behind, we are negating this by either using PCIe direct storage, or making arrays through RAID controllers.
Personally I am not looking at if mine is faster or slower then the next person's machine, but that what I do add to it, will improve the tasks I do. Unfortunately its not always so. I recently added another 48GB of ram to it, which resulted in that could not overclock it, so now I have to replace this and find another brand that will let me do so. In my particular case, I can use both ECC and NONE-ECC (EVGA SR-2), and it supports 1600Mhz ram.
There are some trade offs with every system, so far there isn't a perfect one - at least I haven't found it.
I started this system back in the end of 2010, and expecting it to work till 2015, with modifications done to it as we go along.
i do alot of pano's some of them in 10's of gigabytes, but also work with multi-layered files all the time( As i am sure most of us do ;-) ), to me the biggest bottleneck is actually Photoshop inability to use multicores to save with, CS6 has helped a bit, that it can save in the background, but still its not that fast - as in I want it to be faster ;-)
with your newer systems, you can only add so much ram, before you start seeing that your clock rates starts to drop, but also you will not be able to take full advantage off it when it comes to deliver the data to the disk, it will simply just cue up.
Henrik
PS: Please share you system config
EVGA SR-2 mainboard
2x Intel X5650 @ 4.3Ghz
48GB 1333 RAM ECC (soon to be 96GB again)
12x Intel SSD 520
8x Seagate SV-35 2TB 7200rpm
2x IBM M1015 RAID HBA controller - to be replaced with Areca 1882xi 24 4GB cache + BBU
ASUS GTX-580
Antec 1200w PSU
Lian Li v2120 black case
Noctua 14 heatsinks and fans in the system, almost silent ;-)
-
I know almost nothingc ompared to guys like you and as such it's easy to be deceived by various claims. I'm in the market for a new machine and this caught my eye. any opinions, from the experts?
http://www.anandtech.com/show/5667/ibuypower-erebus-gt-custom-cooling-for-less
The RAM looks a little meagre but that can be taken care of easily. Thanks
-
Andy,
it looks like a nice machine, but its more of an allround machine
have a look around here and on sites related to photo editing and see what they use - also it very much depends on your budget and if you plan to add to the system later
all the best
Henrik
I know almost nothingc ompared to guys like you and as such it's easy to be deceived by various claims. I'm in the market for a new machine and this caught my eye. any opinions, from the experts?
http://www.anandtech.com/show/5667/ibuypower-erebus-gt-custom-cooling-for-less
The RAM looks a little meagre but that can be taken care of easily. Thanks