Mac Pro and Photoshop performance

[antonio99]

Hi,

I have a 2,26GHz 8 core cMP (MP4,1) with El Capitan.
It's pretty classic, no SSD, standard GT120, 16GB of RAM.

After seeing so many threads about upgrading the MacPro I'm thinking in some update, but first my use case.

I use mostly CameraRAW (for RAW processing) and Photoshop (CS6).

Converting the RAW files is relatively quick (it seems to uses most cores, and process more than 16 RAW files per minute). However, working with CameraRAW UI is slow, for example, changing from one file to another, the "warning" (indicating rendering is underway) sign on the top right corner stays on for about 1-2 seconds.

Using Preview.app to show 5MPix JPEG resulting from RAW conversion, and then pressing the "down" arrow to scroll the list is advancing at around 3-4 pictures per second.
When the images are smaller, it advances much faster.

Anyway, I'm wondering what sort of upgrade (software/hardware) could increase responsiveness of the UI as stated above.

Would a CPU upgrade help? From the (limited) testing above, it looks like more cores would get me faster RAW processing, but I'm not sure it would help with the UI sluggishness, whether it is CameraRAW (for RAW) or Preview.app (for JPEG).

The RAW files are around 20MB, they are probably compressed. So a workflow that used some sort of cache could be a good improvement (does Lightroom has such caching?).

I checked the speed of the (software) RAID array and it's clocking at around 400MB/sec R/W, so I doubt an SSD would help since read/write performance does not seems to be the bottleneck.

Would a GPU upgrade help? I've read many articles on CS6, Photoshop and GPU and it looks like GPU support is limited to some specific filters.
It does not look like they are decompressing/demosaicing the RAW files with the GPU.

Would a RAM upgrade help? From what I read, it looks like after 32GB there's no increase in performance of several tools.

What is your experience with Photoshop, CameraRAW and UI sluggishness?
What does one needs to be able to change from one photo to another much quicker?
Would upgrading from CS6 to CC help?

 

[antonio99]

Some information I've been finding about this:

https://forums.adobe.com/thread/1828580
http://photo.stackexchange.com/ques...pu-more-important-for-photoshop-and-lightroom
https://www.dpreview.com/forums/thread/3143728
https://forums.adobe.com/message/7533539#7533539

 

[justinkr]

I run photoshop premiere pro and after effect with my old mac pro 2,1 hdd driver 1tb external 2 tb 3x2 cpu 16gb ram gtx 750 ti never had any problem. It run smoothly. Just for a photoshop u should be fine. U don't need to upgrade. I feel like Cc is a lot better than cs6 for sure.

 

[TonyK]

My MP 3,1 has 32GB of memory and a 500GB SSD (Crucial). It runs PS CS6 fairly well. All my RAW work happens now in Lightroom 5 (not 6) and I'm happy enough with my results. The difference would be if you are professional or advance amateur (like me).

But getting more memory and an SSD in your MP 4,1 won't hurt. In fact my wife's 4,1 has exactly this configuration. She's a writer and editor so she could run 16GB if it weren't for her tendency to leave so many applications open and so many tabs in her multiple browsers.

Both sets of upgrades were completed in the last 12 to 14 months and we've had good results using Crucial drives in these systems.

 

[orph]

can you give more info on your raid?

a SSD tends to be the first and best upgrade for tasks like this (also the simplest)

 

[antonio99]

can you give more info on your raid?

a SSD tends to be the first and best upgrade for tasks like this (also the simplest)

It's 3x1TB SATA 7200rpm disks, benchmarks (AJA) state 400MB R/W.
I tested the disks individually, then 2x1TB, and then 3x1TB and could see the benchmark getting better.

In this case I'm having doubts about getting better UI performance by upgrading to SSD.
I don't know if there are other tests I can do, but it looks like RAW decompressing/demosaicing is what takes time, and it does not look like it is GPU accelerated.

The other computer I can compare this with has similar CPU frequency (Core2Duo 2,4GHz) and has similar UI sluggishness, I wish I could compare it with 3,33 or 4GHz to see if it is a matter of GHz.

Anybody using CameraRAW and using such high-speed CPUs? (even if that includes a hackingtosh)

 

[h9826790]

The main idea to use SSD for system / apps / etc is because of its low latency, not the sequential speed. No matter how many HDD you RAID together, the random 4K QD1 read performance won't improve. That's why a SSD can make a computer much more "responsive" than a HDD RAID array that has the same sequential speed.

 

[orph]

yep h9826790 is on the ball

 

[antonio99]

The main idea to use SSD for system / apps / etc is because of its low latency, not the sequential speed. No matter how many HDD you RAID together, the random 4K QD1 read performance won't improve. That's why a SSD can make a computer much more "responsive" than a HDD RAID array that has the same sequential speed.

Thanks for your insight.
Where can I see the results of the "4K QD1 read performance" on the AJA benchmark?

I did another test, loaded just 3 RAW files in CameraRAW and cycled thru them.
With 16GB of RAM, I'm pretty sure that 3 RAW files should get into the filesystem cache pretty quickly, which seems to be verified since I don't hear the drives anymore.
Yet, the UI is still taking the usual 2 seconds to change from one RAW file to the other.

I would like to improve such UI responsiveness, does anybody has experience in that regard?

While I could image RAW decompression being a serial (as opposed to parallel) operation (well, depending on compression algorithm), I could imagine demosaicing being parallelisable since it the image could be split in blocks.

I have to try FastRawViewer (some links said that is faster than Adobe tools)

 

[orph]

i do want to mention that AJA benchmark is for video editing for checking drives that have large single files ie long video clips.

you do have the drives set to not sleep ? (can RAID drives sleep?)
you can not let them sleep or you will have lockups when the drives need to wake up.

always got to say this do you have the preferences correctly set up, this was the problem last time some one had a problem (he was on CAD software one little change in preferences and it was fixed)

for most people a SSD will relay help, a PCI card which holds 2 SSD's (which you can raid) will be a fairly cheep option (or 2 pci cards with a SSD in each) for a massive speed boost past that the law's of demising returns is massive.

edit
i just looked looked at AJA system test (is that the app your using?) it's fairly overt about being a video test if it's the same app.

 

[antonio99]

i do want to mention that AJA benchmark is for video editing for checking drives that have large single files ie long video clips.

you do have the drives set to not sleep ? (can RAID drives sleep?)
you can not let them sleep or you will have lockups when the drives need to wake up.

always got to say this do you have the preferences correctly set up, this was the problem last time some one had a problem (he was on CAD software one little change in preferences and it was fixed)

for most people a SSD will relay help, a PCI card which holds 2 SSD's (which you can raid) will be a fairly cheep option (or 2 pci cards with a SSD in each) for a massive speed boost past that the law's of demising returns is massive.

edit
i just looked looked at AJA system test (is that the app your using?) it's fairly overt about being a video test if it's the same app.

I used AJA benchmark because I saw it used in other places and was readily available.
It is likely it is geared towards video, but I was interested in read speed.
You or somebody else mentioned that SSD's have less latency, but once the files are on the filesystem cache I think it does not matter.
If you know of other freely available benchmark tools, I could also run them.

The MacPro has no internet access, so I cannot use apps from the app store because they install themselves. I'd need DMG files or something similar (like in the old days, pre-jail, I mean, pre walled garden)

Anyway, other people seem to be seeing something similar, i.e.: that the time to read the file is negligible compared to the time to process the file: https://forums.macrumors.com/threads/mac-pro-cpu-upgrade.2031133/page-2

I will try to figure out more tests.

 

[orph]

make shore your preferences are correct this relay important.

you can try looking at
https://macperformanceguide.com/topics/topic-Photoshop.html
GPU acceleration (on windows) some people consider it best off tho
https://www.pugetsystems.com/labs/articles/Adobe-Photoshop-CS6-GPU-Acceleration-161/
cpu core use by Photoshop
https://www.pugetsystems.com/labs/articles/Adobe-Photoshop-CC-Multi-Core-Performance-625/

cpu upgrade to X5677

buy a imac/windows box with a 4c and high ghz speed

if your computer is ofline CC is not an option

(when you reference some one try not to use comments but references that have done extensive testings)

good luck

 

[h9826790]

Where can I see the results of the "4K QD1 read performance" on the AJA benchmark?

AJA doesn't measure that. You can go for Xbench, which is free.

http://xbench.com

This benchmarking software is not very up to date, so, don't use it for other test, but disk test actually gives you more details than most other disk benchmarking software.

You will see both sequential and random speed, and if you compare the SSD to your HDD RAID array, you will see the SSD is much faster on the random read (4K) section. You will realise this number is way lower than the Sequential read (256K speed (the manufacture marketing speed). And this is the real reason why HDD is slow on running OS and RAID 0 won't help (because OS is not a single huge file but thousands of small files).

Anyway, some example for you. (My 5,1 is working 100% on encoding, however, not much R/W on both the 840Evo and WD HDD. So, the performance figure there should be reasonably accurate. However, the RAM disk performance may be affected a lot).

A WD Red 6T HDD's 4k random read speed is ~1MB/s

A 840 Evo SSD connected to a SATA II port. ~30MB/s (lots of consumer SSD at this range), and it's ~30x faster than HDD . (This is also the reason why SATA III won't shorten the boot time. The SSD simply cannot saturate the SATA II bandwidth when loading small files. e.g. during boot)

A RAM disk (for fun), but I guess this is similar to what the PCIe SSD can do

If anyone has other SATA SSD, PCIe SSD, HDD RAID, SSD RAID, please post some benchmark here, so that OP can know more about on their performance. And decide which one fit his need.

 

[antonio99]

AJA doesn't measure that. You can go for Xbench, which is free.

http://xbench.com

This benchmarking software is not very up to date, so, don't use it for other test, but disk test actually gives you more details than most other disk benchmarking software.

You will see both sequential and random speed, and if you compare the SSD to your HDD RAID array, you will see the SSD is much faster on the random read (4K) section. You will realise this number is way lower than the Sequential read (256K speed (the manufacture marketing speed). And this is the real reason why HDD is slow on running OS and RAID 0 won't help (because OS is not a single huge file but thousands of small files).

Anyway, some example for you. (My 5,1 is working 100% on encoding, however, not much R/W on both the 840Evo and WD HDD. So, the performance figure there should be reasonably accurate. However, the RAM disk performance may be affected a lot).

A WD Red 6T HDD's 4k random read speed is ~1MB/s
View attachment 687573
A 840 Evo SSD connected to a SATA II port. ~30MB/s (lots of consumer SSD at this range), and it's ~30x faster than SSD . (This is also the reason why SATA III won't shorten the boot time. The SSD simply cannot saturate the SATA II bandwidth when loading small files. e.g. during boot)
View attachment 687574
A RAM disk (for fun), but I guess this is similar to what the PCIe SSD can do
View attachment 687572
If anyone has other SATA SSD, PCIe SSD, HDD RAID, SSD RAID, please post some benchmark here, so that OP can know more about on their performance. And decide which one fit his need.

Thanks a lot! Will try that bench.

I'm surprised by the RAM disk test though.

I mean, RAM has basically same access time to any address (hence the "Random" in RAM), thus I would have expected that Random and Sequential to be similar (DDR can have banking issues depending on how the DDR banks are enabled for access).

Also, Uncached Writing is faster than Uncached Reading?
Looks like there must be some cache, if anything thru the CPU L3/L2 cache, right?

 

[h9826790]

Thanks a lot! Will try that bench.

I'm surprised by the RAM disk test though.

I mean, RAM has basically same access time to any address (hence the "Random" in RAM), thus I would have expected that Random and Sequential to be similar (DDR can have banking issues depending on how the DDR banks are enabled for access).

Also, Uncached Writing is faster than Uncached Reading?
Looks like there must be some cache, if anything thru the CPU L3/L2 cache, right?

I don't know the technical details, but I guess writing is easier than reading, because the controller can write on any empty space, for reading, must go to the specific address. So it may require more time to locate the data, which eventually reduce the reading performance.

The RAM disk performance is not accurate anyway (because my cMP is now working hard), however, my guess would be the latency is not necessary from the RAM itself, but something like from the OS (e.g. I/O request). The RAM now work as SSD, so, it's under control of the file system, and the file system's latency will also affect the RAM disk performance. May be someone can give you better answer, I am a totally outsider in this area.