Workstation recommendation please
Hello.
There is not a lot of support from P1 when it comes to hardware recommendations, unfortunately. There is not even a statistic published on which computers C1 runs at the customer base, and how many, although a lot of information is gathered by the software, if I remember an older thread correctly.
There is also not a lot to find in the forums search. Maybe you can share your experience here...
Q1: Which workstations
have been proven performance-wise for rendering / edit on screen, and process exports?
Models please...😊
Anyone experience with an older HP Z600?
Q2:Parallel processing capabilities (CPU) of C1:
Can C1 use TWO quad core or TWO six core processors?
Q3: Benchmarks
Which famous performance benchmarks would come close to simulate C1 requirements?
Q4: Graphics card
I do not feel comfortable to put my money into an overly expensive graphics card / GPU power because whenever I read someone complaining about weird screen output, or jpg output, the ONLY recommendation from Phase One seeems to be to switch hardware acceleration (OpenCL) OFF. And I am not sure if there are other advantages of (overly) expensive graphics cards.
Can you use hardware acceleration without problems, and which card do you use?
Thanks for sharing
BeO
There is not a lot of support from P1 when it comes to hardware recommendations, unfortunately. There is not even a statistic published on which computers C1 runs at the customer base, and how many, although a lot of information is gathered by the software, if I remember an older thread correctly.
There is also not a lot to find in the forums search. Maybe you can share your experience here...
Q1: Which workstations
have been proven performance-wise for rendering / edit on screen, and process exports?
Models please...😊
Anyone experience with an older HP Z600?
Q2:Parallel processing capabilities (CPU) of C1:
Can C1 use TWO quad core or TWO six core processors?
Q3: Benchmarks
Which famous performance benchmarks would come close to simulate C1 requirements?
Q4: Graphics card
I do not feel comfortable to put my money into an overly expensive graphics card / GPU power because whenever I read someone complaining about weird screen output, or jpg output, the ONLY recommendation from Phase One seeems to be to switch hardware acceleration (OpenCL) OFF. And I am not sure if there are other advantages of (overly) expensive graphics cards.
Can you use hardware acceleration without problems, and which card do you use?
Thanks for sharing
BeO
0
-
[quote="BeO" wrote:
Yes Grant, I intend to report.[quote="Wesley" wrote:
You can turn off all that send data to Microsoft so not a big deal.
Hm, I wouldn't be too sure about that.
Also, many of the MS updates are no longer documented according to this article:
and
Windows 10 November Update Resets Settings to Re-Enable Data Collection
However, privacy in WIndows 7 and 8.1 might also be "compromised" by certain updates:
Where are we heading to, if I can't even trust those who I pay for...
Oh lord.
I haven't touched Win 10 since building the rig so looks like a lot have changed.
Guess it's time to switch to Mac OS guys 😉0 -
[quote="BeO" wrote:
i7-5820K
MSI X99A SLI PLUS, incl.USB 3.1
M.2 (PCIe) SSD (samsung 950 pro, 512GB)
Sata3 SSD (samsung 850 evo, 1 TB)
32 GB RAM (2666)
Geforce 960
Not yet arrived...
cheers
BeO
BeO
Which memory are you installing?
Alain0 -
[quote="Alain" wrote:
Which memory are you installing?
Alain
4 x 8GB DDR4 - 2666 Kingston HyperX Fury
Recommended by the PC shop. They favored Kingston a little over Crucial.
The CPU is not going to be overclocked (which they don't do anyway), but RAM with 2400 should not be a problem, and one could run higher specced RAM also with lower frequencies.
I then favored 2666 for a potential future upgrade of the CPU, the next Haswell-E is supposed to support 2400 (I've read somewhere) so 2666 would be just a little over it. Minimal price difference. If it is worthwhile I don't know.
I am not an expert with PCs running above manufacturers specs though, I hope the shop's staff are... 😊
cheers
BeO
EDIT: For a quad channel CPU you should install 4 modules at least.0 -
[quote="BeO" wrote:
[quote="Alain" wrote:
Which memory are you installing?
Alain
4 x 8GB DDR4 - 2666 Kingston HyperX Fury
Recommended by the PC shop. They favored Kingston a little over Crucial.
The CPU is not going to be overclocked (which they don't do anyway), but RAM with 2400 should not be a problem, and one could run higher specced RAM also with lower frequencies.
I then favored 2666 for a potential future upgrade of the CPU, the next Haswell-E is supposed to support 2400 (I've read somewhere) so 2666 would be just a little over it. Minimal price difference. If it is worthwhile I don't know.
I am not an expert with PCs running above manufacturers specs though, I hope the shop's staff are... 😊
cheers
BeO
EDIT: For a quad channel CPU you should install 4 modules at least.
Thank, memory is one of those area's where a shop has far more experience regarding possible problems. It was also one off the possibilities on my list.0 -
Kingston or Crucial...you can't go wrong with either brand.
I'm using Crucial on my Mac. Bought their RAM two different times, no DOA 😄0 -
[quote="BeO" wrote:
i7-5820K
MSI X99A SLI PLUS, incl.USB 3.1
M.2 (PCIe) SSD (samsung 950 pro, 512GB)
Sata3 SSD (samsung 850 evo, 1 TB)
32 GB RAM (2666)
Geforce 960
Not yet arrived...
cheers
BeO
Hi BeO
Has you're workstation already arrived?
If yes, what are you're first impressions?
Alain0 -
Hi Alain,
Well, yes, it arrived, and I mainly made myself familiar with the computer before setting up my C1 environment.
Computer:
MSI Geforce GTX 960 4GD5T OC
At day 2, the computer was very loud even idle. It turned out (after hours without opending the case, which I should have done earlier) that one of the 2 Geforce fans stuck and the other was at 100%, so the stuck one needed some manual help to run. Since then the fans ran fine and quiet (mostly idle).
SSD Samsung 950 (PCIe)
It is considerably faster than the Samsung 850 Evo (Sata III), however not as fast as I thought it would be. Random read access (measured with Samsung Magican) was 160k iops instead of expected 250k-300k. The 850 Evo has around 90k.
C1 has a lot of small files, so I assume random access is important, beside sequential access.
I've read in the web that the 950 was limited due to higher RAM speed via enabled XMP changing the bus speed affecting SSD performance. So I want to try out reducing the RAM speed from 2666 to normal 2133.
Unfortunately my BIOS crashed after a week and even with the second BIOS the system doesn't boot up to Windows anymore, the SSD 950 is no longer recognized. Whether it's the mainbord or the SSD is currently under investigation at the PC shop/doctor.
Capture One Performance:
- The GTX 960 had a OPENCL benchmark around 0.25.
- i7-5820K on 3.3 GHz, Turbo Boost disabled
- RAM on 2666 MHz
- Windows and catalog residing on Samsung 950
- 4k raw files on Samsung 850 Evo, mainly Sony 24MP .ARW files
- Preview size (long edge) 1920 px
- Monitor res: 1920 x 1200
I plan to do reproducible (and more) performance tests when the PC is back, so I have only a few (and rather less precise) observations for now:
- Load 4k "All images" (until every thumbnail appeared): around or under a minute
- Loading 1 preview image: under a second.
- Loading 4 preview images: around a second.
- Double click on 1 preview (to view 100%): under a second
On first sight, OPENCL enabled or disabled didn't make a noticable difference.
No importing, processing or brush work with many layers performed yet.
It is a big improvement over my current 5 years old notebook. However, catalog performance it not a racing car.
Cheers
BeO0 -
[quote="BeO" wrote:
...
Capture One Performance:
- The GTX 960 had a OPENCL benchmark around 0.25.
....
On first sight, OPENCL enabled or disabled didn't make a noticable difference.
I'm not sure but I think anything is wrong with you PC.
My old nVidia should be faster?2014-07-31 18:09:41> Logging is now active.
OpenCL initialization...
OpenCL : found platform Apple, OpenCL Version : OpenCL 1.2 (Apr 25 2014 22:04:25)
OpenCL Device : GeForce GTX 770
OpenCL Driver Version : 8.26.26 310.40.45f01
OpenCL Compute Units : 8
OpenCL : Loading kernels
OpenCL : Loading kernels finished
OpenCL : Benchmarking
OpenCL : Initialization completed
OpenCL benchMark : 0.225264
And there is a big difference to work with C1 with and without OpenCL. Do you have latest nVidia drivers installed?0 -
Thanks RobiWan Kenobi.
Yes there is something wrong with my PC ... 😊 but I hope I'll get it back from the PC doctor soon.. 😊
No, seriously, 0.22 and 0.25 should be quite similar I think, and if you feel a big difference but I do not, that could indicate that my CPU is faster than yours, approximately as fast as your gtx 770.
But I cannot tell for sure if I have a noticable difference, I only tried a few images before I had to say goodbye...
Thanks again,
BeO0 -
Is the OpenCL benchmark done somewhere in Capture One or a third-party program? 0 -
[quote="Wesley" wrote:
Is the OpenCL benchmark done somewhere in Capture One or a third-party program?
C1. It is the benchmark figure from the ImgCore.log, checked at C1 startup.
The figure represents one important step in the image processing, don't know which, it is meant for P1 internal usage, it is not necessarily an "official" benchmark figure, however it is often referenced here in the forum to compare gfx cards performance.
Lower is better.
Gfx card must have at least 2 GB VRAM to be used and the benchmark below 2.
The benchmark is said to be somewhat linear (0.1 twice as fast as 0.2)0 -
Where is ImgCore.log or is this only on Windows? 0 -
It is on the Mac too, don't know where. Maybe it is a plain file in a log folder as it is in Windows, but if I remember right someone wrote there is a menu entry (maybe in the help menu) which allows to create a package with all the other logs, to easily send it to support, but I am not sure... 0 -
[quote="Wesley" wrote:
Where is ImgCore.log or is this only on Windows?
On Mac is it Called CaptureOneIC.log and is available through the Console under ~/Library/Logs/0 -
[quote="BeO" wrote:
Gfx card must have at least 2 GB VRAM to be used and the benchmark below 2.
Actually, must-have 1 gb, but 2 gb or more is absolutely recommended.0 -
[quote="Christian Gruner" wrote:
[quote="Wesley" wrote:
Where is ImgCore.log or is this only on Windows?
On Mac is it Called CaptureOneIC.log and is available through the Console under ~/Library/Logs/
I find mine under CaptureOneICOCL.log
I'm going to assume CO doesn't bother doing the benchmark for video cards with less than 1GB since I don't see a score.
So having it off or auto would make no difference for me.0 -
[quote="Wesley" wrote:
[quote="Christian Gruner" wrote:
[quote="Wesley" wrote:
Where is ImgCore.log or is this only on Windows?
On Mac is it Called CaptureOneIC.log and is available through the Console under ~/Library/Logs/
I find mine under CaptureOneICOCL.log
I'm going to assume CO doesn't bother doing the benchmark for video cards with less than 1GB since I don't see a score.
So having it off or auto would make no difference for me.
That is a different log. You have to look in either CaptureOneIC.log or CaptureOneICP.log
If the card has less than 1 gb of ram it will not be initiated.0 -
Performance update
Hardware and C1 setupCPU: i7-5820K, 6 cores, 3.3 GHz - Turbo Boost enabled
Board: MSI X99A SLI PLUS
SSD1: M.2 (PCIe) (Samsung 950 pro, 512 GB) - Win7 and catalog
SSD2: Sata3 (Samsung 850 vo, 1 TB) - Raw files
RAM: 32 GB (Kingston 4x8GB, 2666Mhz) - currently 2133 Mhz
GPU: MSI Geforce GTX 960 4GD5T OC - OPENCL benchmark 0.247
The SSD1 is noticably faster than the SSD2 (tested with Samsung Magician) but random read (155k IOPS) and random write is far below my expection(read >250k IOPS).
More of C1 setup
- 4.1K raw files, mainly Sony 24MP .ARW
- 400 images have layers, .comask file sizes range from 5 Kb to 90 MB (total 1.32 GB)
- Preview size (long edge) 1920 px
- Monitor res: 1920 x 1200
Catalog loading
- Load 4.1k "All images" (until every thumbnail appeared): almost a minute
- Without the layer masks (I closed C1, renamed the Adjustments folder between , and restarted C1 again to test that) loading is around 20 seconds faster !
EDIT 1: More tests showed that this performance advantage mostly is less.
EDIT 2: I created a new catalog in a (different folder) and imported the original catalog, and regenerated previews. The database is cleaned up around 30 images and .comask files less. Loading all images is now 35 secs.
Viewer
Working with the viewer is quite smoothly
- Loading 1 preview image: under a second.
- Loading 4 preview images: around a second.
- Double click on 1 preview (to view 100%): under a second
- Zooming 9 images (simultanously with <Shift>) which are already loaded in viewer : around 2 secs
That's more or less the same with or without OPENCL
Editing
- Brushing on an image which has already around 6 layers is quite fast and smoothly. OPENCL enabled or disabled doesn't make a noticable difference. In both cases CPU is 100% (??)
Preview generating
- 4.1k previews & thumbnails in about 55 minutes with OPENCL enabled
Processing
(Output folder on SSD1)
- 20 images (24MP Sony.ARW) all with a few layers, export TIFF 16 bit Full size
- CPU: 55 secs
- GPU: 35 secs
Output to JPEG 90% quality, 2048 long edge is a little faster, similar proportion CPU vs. GPU
Summary
No importing performed yet.
I haven't really worked with the new machine and C1 yet, but at first sight it definitely is much more fun than with my 5 years old notebook 😊
However, catalog performance it not a racing car, imo (especially considering that only 10% of the images are having layers).
Questions
Why is the CPU at 100% while brushing with OPENCL enabled. Is the brush always computed by the CPU?0 -
[quote="BeO" wrote:
Performance update
Hardware and C1 setupCPU: i7-5820K, 6 cores, 3.3 GHz - Turbo Boost enabled
Board: MSI X99A SLI PLUS
SSD1: M.2 (PCIe) (Samsung 950 pro, 512 GB) - Win7 and catalog
SSD2: Sata3 (Samsung 850 vo, 1 TB) - Raw files
RAM: 32 GB (Kingston 4x8GB, 2666Mhz) - currently 2133 Mhz
GPU: MSI Geforce GTX 960 4GD5T OC - OPENCL benchmark 0.247
...
Thanks
I use sessions and I see only a couple of very small files types, so I don't think that 4k IO is typical for C1, which is good for the PCIe drive. I wouldn't be surprised that the CPU is still the bottleneck for some tests. I have the impression that C1 startup is not using all cores about 30% (probably one core is the bottleneck in my 4core setup).
Editing : The 100% CPU usage is also what I see on my i5-2500K.
Alain0 -
Performance update
I created a new catalog in a (different folder) and imported the original catalog, and previews were regenerated.
The database is cleaned up around 30 images and .comask files less. Loading all images is now 35 secs.
4.1k previews in about 55 minutes with OPENCL enabled. CPU with low-medium utilization, mainly one core (two threads) were around 50% (Sony raw files are decoded on a single core, unlike P1, Canon or Nikon files, afaik).0 -
Hi Beo
Thanks for the update.
I'm also interested with the speed difference between all files (catalog and RAW) on SSD1 and all files on SSD2.
SSD1: M.2 (PCIe) (Samsung 950 pro, 512 GB)
SSD2: Sata3 (Samsung 850 vo, 1 TB)
Does the difference matter in reality?0 -
Hi Alain,
SSD1 vs SSD2
All files (catalog (incl. adjustment masks) and RAWs) is nearly identical on SSD1 and SSD2.
- Tested for (1) opening catalog and preview generation and (2) output to jpg 80% Full size.
Both types of "cross-drive operation" (i.e. catalog on SSD1 and RAWs on SSD2 as well as the other way round) resulted also in the same performance
- Tested for jpg output only
I repeated the tests several times, both SSD1, SSD2 and cross-drive tests are not very consistent i.e. range from 1:55 to 2:10 for jpg processing of 50 images.
Navigating in browser and loading in viewer was undistinguishable too.
Conditions:
50 images (24 MP Sony .arw files) many of them are cropped, each with several layers.
The CPU is always very busy. OPENCL is set to "auto" (both options).
Cheers,
BeO0 -
Jpeg aren't a good choice for disk comparisons. This is because the compression the image has to go through, which is done on CPU (Intel standard jpeg library), and the resulting limited file-size.
Instead try with uncompressed tiff in 16 bit. This will move the bottleneck, as the disk is used a lot more intensively.0 -
Thanks Christian.
The same 50 images are before, output to 16 bit tiff full size (but incl. crops) uncompressed.
2 tests each:
SSD1: 1:40, 1:45
SSD2: 1:55, 2:04
1 test:
cross-drive (catalog & raws on SSD2, output on SSD1): 1:55
The CPU profile looks very similar, quite busy. Maybe due to Sony files demosaicing not multithreaded and a handful of layer masks each.
Cheers
BeO0 -
[quote="BeO" wrote:
Maybe due to Sony files demosaicing not multithreaded
If your GPU and disk is suffuicently fast, then yep, the single core decoding could have an effect.
As raw B file is loaded while raw A is being processed, the loading of B has to be longer than the processing of A for the decoding to be a prolonging factor for the total time.0 -
In a chain there is always one weak link. It seems that both SSDs, Sata III and PCIe, are good enough for my specific processor (i.e. unlike former times where HDDs mostly were a bottleneck).
There might be other functions in C1 or other camera file types where the PCIe drive is more advantageous than for out processing of 24MP Sony files.
Cheers,
Beo0 -
Thanks BeO
I will try to replicate the test on my older machine after I install my second sata SSD (is normally in a usb enclosure for fast mobile storage).
I will test with sony A850 files and with nikon D800 files.0 -
Hi BeO
It took me longer than wanted to add the SSD to my system.
tested on my system :
i5-2500K, GPU AMD 7870, 2x sata III SSD
between 30 and 50 RAWfiles where tested.
40MB RAW's --> 150MB (average) tiff 16 bit
RAW's in a session on a SSD
Observations :
First used 16-bit tiffs with zip compression --> zip compression happens on one core, max. throughput 20MB/s. But it doesn't makes much difference in file size.
All further tests where without zip compression 😉
RAW files with only white balance, some levels and a curve: fastest and the GPU was used for about 70%, CPU had some margin and all cores where used.
added 6 layers without a mask --> the GPU usage dropped to about 50% and the CPU became the bottleneck, all cores used at 85-100%.
cropped the images and every image got at least two mask (all made with a soft brush) --> GPU usage dropped almost to 30% and the CPU was clearly the bottleneck. The same observation that I have when brushing masks
Conclusions :
The diskspeed doesn't matter for writing, I got about 30 MB/sec max. Writing to the same SSD, another SSD or a HDD didn't made any significant difference. Even a computer that's 4x times faster won't give a problem for a recent HDD that's defragged from time to time.
When using layers it seems that the CPU is used much more. My CPU is clearly the bottleneck, but all 4 cores are used.
zip compression for 16-bit tiff's is slow and single core and doesn't safe much disk space.0 -
Hi Alain, interesting. [quote="Alain" wrote:
cropped the images and every image got at least two mask (all made with a soft brush) --> GPU usage dropped almost to 30% and the CPU was clearly the bottleneck. The same observation that I have when brushing masks
Do you notice a difference in actual speed when turning OPENCL on/off? It is said a fast gfx card speeds up C1 significantly, but given your observation this might be more true or less true depending on the usage of masks...0 -
[quote="BeO" wrote:
Hi Alain, interesting.[quote="Alain" wrote:
cropped the images and every image got at least two mask (all made with a soft brush) --> GPU usage dropped almost to 30% and the CPU was clearly the bottleneck. The same observation that I have when brushing masks
Do you notice a difference in actual speed when turning OPENCL on/off? It is said a fast gfx card speeds up C1 significantly, but given your observation this might be more true or less true depending on the usage of masks...
Oh yes, the GPU is still much faster.
I did a double test run yesterday and the difference is a factor 2.65 for a test with processing 142 high ISO images with lot's of layers to 16-bit tiff.
When only using the CPU it's constant 100% and the systems is less snappy for browsing etc. When using the GPU the CPU does change between 80 and 100% and the system stays rather snappy when browsing.
BTW. When drawing masks the CPU is unfortunately at 100%.0
Post is closed for comments.
Comments
112 comments