Exporting benchmarks v2

General discussion regarding photography practices and Capture One workflow as well as integration with third party applications.
Please DO NOT post to this thread regarding OS specific concerns or questions.
Forum rules
For the sake of being thorough, please remember to note if you are using Mac or Windows.
Hopefully this will keep any confusion to a minimum.

Re: Exporting benchmarks v2

Postby gnwooding » Mon Jan 29, 2018 9:56 pm

In light of what everyone else has posted on TIFF results being more accurate for testing GPU's I decided to run a couple of additional tests.
I left my i7 5820k OC at 4.4GHz and I used a 12GB RAM disk to do the tests on (to ensure no disk bottleneck).

Single GTX 1080ti : 24s
2 x GTX 1080ti : 16.5s

What I observed is that when doing TIFF files at least my CPU usage is 100% across all cores so I am definitely CPU limited.
With a single GTX1080ti GPU usage peaked at 75%.
When using both cards GPU usage only peaked at about 45% on each card and it was less constant. I also notice that the cards only bother running at 1.5GHz since they only see a light load (in games and other GPU computational applications the GPU core runs at 2GHz on my cards).

So clearly even when exporting TIFF files you need a very powerful CPU to take advantage of a powerful GPU.
gnwooding
 
Posts: 21
Joined: Sat Aug 30, 2014 1:54 am

Re: Exporting benchmarks v2

Postby WPNL » Mon Jan 29, 2018 10:23 pm

garrison wrote:What do you think about adding 'format type' column - tiff or jpeg (see my post above)?

Done! (I hope I didn't mess it up, of to sleep now!)
WPNL
 
Posts: 343
Joined: Tue Dec 02, 2014 11:11 am

Re: Exporting benchmarks v2

Postby Chad Dahlquist » Tue Jan 30, 2018 3:49 am

maybe should have put out for a 1080ti over my 1080 :) but interesting to see the gains in the tiff times

gnwooding wrote:In light of what everyone else has posted on TIFF results being more accurate for testing GPU's I decided to run a couple of additional tests.
I left my i7 5820k OC at 4.4GHz and I used a 12GB RAM disk to do the tests on (to ensure no disk bottleneck).

Single GTX 1080ti : 24s
2 x GTX 1080ti : 16.5s

What I observed is that when doing TIFF files at least my CPU usage is 100% across all cores so I am definitely CPU limited.
With a single GTX1080ti GPU usage peaked at 75%.
When using both cards GPU usage only peaked at about 45% on each card and it was less constant. I also notice that the cards only bother running at 1.5GHz since they only see a light load (in games and other GPU computational applications the GPU core runs at 2GHz on my cards).

So clearly even when exporting TIFF files you need a very powerful CPU to take advantage of a powerful GPU.
Chad Dahlquist
 
Posts: 43
Joined: Thu Dec 02, 2010 12:58 am

Re: Exporting benchmarks v2

Postby Christian Gruner » Tue Jan 30, 2018 7:59 am

gnwooding wrote:So clearly even when exporting TIFF files you need a very powerful CPU to take advantage of a powerful GPU.

Well, that depends. It this case with a lot fairly small files, the CPU is very busy indeed feeding the GPU(s). If processing say 100 mp IQ3 files, you would not see the same load on CPU, as the number of files per second are less. The bottleneck also moves around depending on MP, you could say.
Test-Manager, Capture One, Phase One A/S
Christian Gruner
Crew
Crew
 
Posts: 1828
Joined: Fri Nov 30, 2012 9:48 pm
Location: Copenhagen, Denmark

Re: Exporting benchmarks v2

Postby NNN636355020530937144 » Tue Jan 30, 2018 10:33 pm

i9-7980xe overclocked to 4,2ghz
amd vega fe
c1 11.0.1
win10 1703

JPEG
45sec GPU
82sec CPU

TIFF
19sec GPU
64sec CPU

...it was interesting to see that there is very small difference compared to i9-7900x.

Storage is NVME Optane 900p and MegaRaid 9460-16i with SATA SSD and RAID10 SATA. Results were same with Optane 900p and SATA SSD.

update: if i overclock mesh from 24 to 30, tiff and jpeg cpu time drops to 77sec and 56sec.
NNN636355020530937144
 
Posts: 11
Joined: Thu Jul 13, 2017 1:20 am

Re: Exporting benchmarks v2

Postby Tom-D » Wed Jan 31, 2018 3:14 am

NNN636355020530937144 wrote:i9-7980xe overclocked to 4,2ghz
amd vega fe
c1 11.0.1
win10 1703

JPEG
45sec GPU
82sec CPU

TIFF
19sec GPU
64sec CPU

...it was interesting to see that there is very small difference compared to i9-7900x.

Storage is NVME Optane 900p and MegaRaid 9460-16i with SATA SSD and RAID10 SATA. Results were same with Optane 900p and SATA SSD.

update: if i overclock mesh from 24 to 30, tiff and jpeg cpu time drops to 77sec and 56sec.


Nvme and Sata SSD are the same? Was thinking of getting a 1TB samsung 960 Pro for my Hot Files cause of higher performance, using a 960GB SSD right now.

I'm testing PrimoCache right now for my HDDs that i use now and then, with a SSD as a cache drive.

Was also using a Ram Disk for Read and Write but the Performance does only go up by 1%.

Will post my System Today, with a P5000 Quadro GPU, < as fast as 1080. (ECC enabled).
Win 10 x64, Intel 7820X, 64 GB Ram 2666mhz, Quadro RTX 5000, M.2 2x 1 TB PCI-E.
Tom-D
 
Posts: 77
Joined: Wed Jan 13, 2010 1:44 pm

Re: Exporting benchmarks v2

Postby WPNL » Wed Jan 31, 2018 7:38 am

Tom, did you consider adding an identical SSD in RAID0?
You'll get twice the speed and storage space.
(I'd backup the important files to another location in case of a failure, which I never had but just in case)
WPNL
 
Posts: 343
Joined: Tue Dec 02, 2014 11:11 am

Re: Exporting benchmarks v2

Postby Tom-D » Wed Jan 31, 2018 3:00 pm

WPNL wrote:Tom, did you consider adding an identical SSD in RAID0?
You'll get twice the speed and storage space.
(I'd backup the important files to another location in case of a failure, which I never had but just in case)


For me, SSD raids are outdated, a Samsung Nvme M.2 SSD in a PCI-E slot gives me up to 3500 MB Read and 2100 MB Write.

Two Sandisk 960GB will give me 1000 MB Read an Write.

And Nvme's SSDs are made for Paralell Workload, Sata SSDs are not.
Win 10 x64, Intel 7820X, 64 GB Ram 2666mhz, Quadro RTX 5000, M.2 2x 1 TB PCI-E.
Tom-D
 
Posts: 77
Joined: Wed Jan 13, 2010 1:44 pm

Re: Exporting benchmarks v2

Postby Tom-D » Wed Jan 31, 2018 3:37 pm

____

Win 10, 64Bit, 1709

Capture One 11.0.1

CPU: 7820X / GPU: Quadro P5000

CPU+GPU: 43 Sec. (Jpeg) (GPU ECC on)
CPU: 78 sec. (1:18 Min)(Jpeg)

CPU+GPU: 24 Sec. (Tiff 8 bit) (GPU ECC on)
CPU: 60 Sec. (Tiff 8 bit)
CPU: 63 sec. (Tiff 16 bit)

Time Measured with Adobe Bride, Input Time last image minus Input time first image.

SSD Sandisk Extreme Pro 960GB (Input & Output)
Mainboard MSI Gaming Pro Carbon
64GB Ram, Gskill 2666 (15.15.15.35-Latency)
Samsung 960 Pro 512GB, OS PCI-E NVMe SSD.
I7 7820X, 4.6/4.3/4.3/4.3/4.3/4.3/4.3/4.6 (Ghz Cores Boost)
(Max 150 Watt Max Pull alowed (Short and Long Duration)(2.5 Ghz Mesh)
Max Temp during Bench (72°C , Core 7 <- Hottest/ 55°C, Core 0 <-Coldest)
Cooler: Prolimatech Genesis (Liquid Ultra, as Thermal Compound)

GPU Max 54°C, Max Load 82%, did hit Performance Limit.

Measured with HWinfo.

Switched From Vega FE to P5000 last week, the Vega was the most Buggy GPU i've ever had, did have Problems also with my Two Firepro W8100, after 4 Years of Trying Radeon i will never go back to it.

The P5000 is Rock Stable and i did not hear it.
Win 10 x64, Intel 7820X, 64 GB Ram 2666mhz, Quadro RTX 5000, M.2 2x 1 TB PCI-E.
Tom-D
 
Posts: 77
Joined: Wed Jan 13, 2010 1:44 pm

Re: Exporting benchmarks v2

Postby CraigJohn » Wed Jan 31, 2018 4:21 pm

NNN636355020530937144 wrote:i9-7980xe overclocked to 4,2ghz
amd vega fe
c1 11.0.1
win10 1703

JPEG
45sec GPU
82sec CPU

TIFF
19sec GPU
64sec CPU

...it was interesting to see that there is very small difference compared to i9-7900x.

Storage is NVME Optane 900p and MegaRaid 9460-16i with SATA SSD and RAID10 SATA. Results were same with Optane 900p and SATA SSD.

update: if i overclock mesh from 24 to 30, tiff and jpeg cpu time drops to 77sec and 56sec.



Definitely seeing a line of diminishing returns here. Wonder if it's at 10 or 12 cores. The kid spending on a 16 core Threadripper for C1P11 might be a little disappointed. Still interested in seeing what happens with his build.
CraigJohn
 
Posts: 55
Joined: Thu Jun 09, 2011 5:26 pm

Re: Exporting benchmarks v2

Postby CraigJohn » Wed Jan 31, 2018 4:25 pm

7820x
8GB Sapphire Radeon RX 580 Nitro+

Tiff CPU+GPU = 25.8s
CraigJohn
 
Posts: 55
Joined: Thu Jun 09, 2011 5:26 pm

Re: Exporting benchmarks v2

Postby garrison » Wed Jan 31, 2018 5:43 pm

Windows PC (Win7)
C1 version - 11.0
CPU - Intel Core i7 4770 (4-core, 8-threads, 3.40 GHz)
GPU - NVidia GTX 660

JPEG
CPU+GPU - 1:24 (84s)
CPU only - 2:33 (153s)
The GPU speed up factor is about 1.82x.

TIFF
CPU+GPU - 1:05 (65s)
CPU only - 1:56 (116s)
The GPU speed up factor is about 1.78x.
garrison
 
Posts: 51
Joined: Sat Sep 20, 2008 12:56 am

Re: Exporting benchmarks v2

Postby NNN636355020530937144 » Wed Jan 31, 2018 9:23 pm

CraigJohn wrote:Definitely seeing a line of diminishing returns here. Wonder if it's at 10 or 12 cores. The kid spending on a 16 core Threadripper for C1P11 might be a little disappointed. Still interested in seeing what happens with his build.


It seems that currently, i9-7900x with 1080ti is best combination for C1. It's interesting that even all cores are at 100% still the difference between 7900x and 7980xe is so small (in CPU only benchmarks). Maybe with CPU+GPU tests multiple GPU's on 7980xe will give better results...
NNN636355020530937144
 
Posts: 11
Joined: Thu Jul 13, 2017 1:20 am

Re: Exporting benchmarks v2

Postby NNN636355020530937144 » Wed Jan 31, 2018 9:26 pm

Tom-D wrote:Switched From Vega FE to P5000 last week, the Vega was the most Buggy GPU i've ever had, did have Problems also with my Two Firepro W8100, after 4 Years of Trying Radeon i will never go back to it.

The P5000 is Rock Stable and i did not hear it.


I did have exactly opposite experience, moving from Quadro (last 10-15 years on Quadro) to Vega FE (with past FirePro cards I did also have lot of problems). What kind of problems you had?
NNN636355020530937144
 
Posts: 11
Joined: Thu Jul 13, 2017 1:20 am

Re: Exporting benchmarks v2

Postby WPNL » Wed Jan 31, 2018 10:26 pm

Tom-D wrote:For me, SSD raids are outdated, a Samsung Nvme M.2 SSD in a PCI-E slot gives me up to 3500 MB Read and 2100 MB Write.

Two Sandisk 960GB will give me 1000 MB Read an Write.

And Nvme's SSDs are made for Paralell Workload, Sata SSDs are not.

I see, thanks for the explanation! :-)
WPNL
 
Posts: 343
Joined: Tue Dec 02, 2014 11:11 am

PreviousNext

Return to Workflow and Common Photography Exploration



Who is online

Users browsing this forum: C-M-B, NNN636924912861020152 and 1 guest