Personal computing discussed

Moderators: Captain Ned, emkubed

 
TheEmrys
Minister of Gerbil Affairs
Topic Author
Posts: 2458
Joined: Wed May 29, 2002 8:22 pm
Location: Northern Colorado
Contact:

Idea for further GPU testing

Fri Oct 12, 2018 11:50 am

There is a lot of talk on photography/video boards about how much of a difference CUDA/OpenCL make while editing photographs as well as effects in various video editing programs. However, there is not real uniform testing. TR could become the place to go for authortative information on the subject. Moreover, adding a photography build that emphasizes monitors with higher bit depth, and wider color gamut could be quite useful, in addition to identifying where photographers are overspending (namely, video cards). I have seen quite a few builds done by photographers who simply buy the highest level Nvidia card claiming it will provide the highest impact on their photo editing. There is no real information that quantifies this. My suspicion is that there is a negligible difference between the performance of a 1050 and a 1080ti. There are a few sites out there that have done a bit of testing, Puget Systems for example. They even have a publicly available download for their benchmark tool. How good it is, I have no idea. However, I haven't found a good site the unifies both video and photography reviews opencl, cuda, and gpgpu in general.
Sony a7II 55/1.8 Minolta 100/2, 17-35D, Tamron 28-75/2.8
 
DPete27
Silver subscriber
Grand Gerbil Poohbah
Posts: 3558
Joined: Wed Jan 26, 2011 12:50 pm
Location: Wisconsin, USA

Re: Idea for further GPU testing

Fri Oct 12, 2018 2:20 pm

Since (I presume) Jeff does all the GPU tests and it seems like he's into photography, you may have a winning idea here.

I know Tomshardware does a couple Photoshop tests in their CPU benchmarks.
I will say I'm generally in the dark about GPU requirements for photo/video editing builds. That may be because I'm more focused on gaming benchmarks generally. My [potentially incorrect] assumption is that once you've got an entry level GPU just to provide GPU-acceleration for rendering the view on screen, the differences are small in photo and video editing because most of the workload is still being done on the CPU.

I think you hit the nail on the head by saying that because photo/video editing doesn't get a lot of benchmarking attention, buyers are forced to "overspend" in hopes that they end up with a system that can handle what they're doing. And that applies to all components (CPU, GPU, RAM, storage). It's hard enough with these "productivity" applications to determine how your usage complexity compares to the benchmark.

One example: My friend is a graphic designer, but not inherently savvy about PC tech. He does content creation and photo editing mostly in Photoshop, and uses a paltry desktop at work (A8-6800K on IGP, 16GB RAM, hdd) that his company provides. Up until June of this year, he's also been doing the same type of work on his personal laptop (i5-2430M, Nvidia GT540M, 6GB RAM, 90GB SSD + 640GB hdd) with satisfactory results. I helped him pick out a new laptop (i5-8250U, GTX1050, 8GB RAM, 256GB SSD + 1TB hdd) this year, but it was difficult to show/tell him how much improvement he should expect to see for his $600 spent or how much more improvement he could expect to see in spending...$300 more than that, $500 more than that.

Since I frequent the SBA forum, I can say from memory that the occasional "photoshop build" that pops up generally diverges into recommendations of 8+ CPU threads, 32-64GB RAM, >$300 GPU, and 1TB+ SSDs (aka ~$2,000 just for the tower). I wouldn't doubt that covers the majority of power users, but how much power do YOU need for your specific usage?
Main: i5-3570K, ASRock Z77 Pro4-M, MSI RX480 8G, 500GB Crucial BX100, 2 TB Samsung EcoGreen F4, 16GB 1600MHz G.Skill @1.25V, EVGA 550-G2, Silverstone PS07B
HTPC: A8-5600K, MSI FM2-A75IA-E53, 4TB Seagate SSHD, 8GB 1866MHz G.Skill, Crosley D-25 Case Mod
 
TheEmrys
Minister of Gerbil Affairs
Topic Author
Posts: 2458
Joined: Wed May 29, 2002 8:22 pm
Location: Northern Colorado
Contact:

Re: Idea for further GPU testing

Fri Oct 12, 2018 5:31 pm

Totally agree. There is no solid way to quantify how much to spend on a build for photography or videography. TR could stand in the gap there.
Sony a7II 55/1.8 Minolta 100/2, 17-35D, Tamron 28-75/2.8
 
Voldenuit
Minister of Gerbil Affairs
Posts: 2806
Joined: Sat Sep 03, 2005 11:10 pm

Re: Idea for further GPU testing

Fri Oct 12, 2018 5:58 pm

TheEmrys wrote:
Totally agree. There is no solid way to quantify how much to spend on a build for photography or videography. TR could stand in the gap there.


The problem is it's hard to establish a baseline load. A given Photoshop user's workflow can vary greatly depending on the size of the images they work on, how many images they process at once, the number of layers, the use of brush tools (some of which use CPU power to simulate physical behavior), stacking, HDR, noise reduction, post processing tools, the use of third party plugins and filters.

A system that is 'overkill' for one user may very well be 'inadequate' for another. And they may even experience different bottlenecks in their workflow, so that an 'overkill' system might still end up 'inadequate' to other users that have different workflows.

I'd think that any decent modern system today with a 4+ core CPU, 8 GB+ RAM, a SSD, and any discrete GPU will probably be 'overkill' for average DSLR/MILC users just processing RAWs, but probably inadequate for professional or enthusiast content creators. The hard part will be quantifying the level of performance and correlating it to specific hardware bottlenecks if possible.
Wind, Sand and Stars.
 
TheEmrys
Minister of Gerbil Affairs
Topic Author
Posts: 2458
Joined: Wed May 29, 2002 8:22 pm
Location: Northern Colorado
Contact:

Re: Idea for further GPU testing

Fri Oct 12, 2018 6:46 pm

Take a lot at Puget Systems. They actually have a pretty nice baseline established.

https://www.pugetsystems.com/labs/articles/Photoshop-CC-2018-NVIDIA-Quadro-vs-AMD-Radeon-Pro-1202/
Sony a7II 55/1.8 Minolta 100/2, 17-35D, Tamron 28-75/2.8
 
Airmantharp
Emperor Gerbilius I
Posts: 6166
Joined: Fri Oct 15, 2004 10:41 pm

Re: Idea for further GPU testing

Fri Oct 12, 2018 6:53 pm

Voldenuit wrote:
TheEmrys wrote:
Totally agree. There is no solid way to quantify how much to spend on a build for photography or videography. TR could stand in the gap there.


The problem is it's hard to establish a baseline load.


Came to say this, and basically everything you put after.


Voldenuit wrote:
I'd think that any decent modern system today with a 4+ core CPU, 8 GB+ RAM, a SSD, and any discrete GPU will probably be 'overkill' for average DSLR/MILC users just processing RAWs, but probably inadequate for professional or enthusiast content creators. The hard part will be quantifying the level of performance and correlating it to specific hardware bottlenecks if possible.


That last line nails it.

I will share one bit of personal experience. I upgraded from an already excellent XPS13, call it the 2017 model, that has a 7500U which is a dual-core CPU. The laptop otherwise has an excellent screen, keyboard, touchpad, NVMe drive, and 16GB of RAM.

And I had three complaints with it. First two were that it wasn't a flipper (2-in-1), which is useful for retouching, that the screen while nice was a non-integerly scaled panel that required scaling to use and non-integer scaling is imperfect in Windows.

The third was that the 7500U was just plain slow when it came to doing multi-layer processing, especially stuff like HDR used to expand dynamic range*, and Intel had just released their quad-core ultrabook line. I replaced it with an 8550U ASUS 2-in-1 which otherwise has the same basic specs, but swaps SATA for NVMe, and is otherwise inferior in every way, just not enough to keep me from using it, or to downgrade (quite literally) to Dell's XPS13 2-in-1.

Going from two hyper-threaded cores to four makes a pretty big difference here.

Now, going from a quad-core hyper-threaded CPU in the i7-6700K in my desktop to the i7-8700K didn't make too much difference, but in both cases, this is workload dependent situationally dependent. On the desktop, I have four screens going and I don't mind waiting as much, and don't have to wait as much; on the laptop, I have one screen, and I'm waiting because there's a thermal budget involved.


And that's just CPUs. The issue with GPUs as noted for content creation is that their utility is very highly workload dependent and using modern content creation routines for GPU benchmarking is a good if complex issue. I'd like to see a 'content creation' build that includes baseline benchmarks that can then be updated both with new hardware and new software.

To add, a large part of the deficiency of GPUs when talking about utility for content creation is the lack of implementation of hardware acceleration in the software itself- meaning that benchmarks would likely need to evolve a bit quicker than they do on the gaming side of things.


*[when talking about HDR, there's slinging the sliders to the stops, and there's using multiple exposures to keep highlights from blowing and shadow detail from being lost to noise- the choice is personal, of course, but mine is to use the technology to produce cleaner final images]
Canon 6D|24-105/4L |70-300/4-5.6 IS USM|50/1.4 USM|50/2.5 CM|50/1.8 STM|85/1.8 USM|Samyang/Bower 14/2.8|24/2.8 IS|Sigma 150-600 | C
Canon EOS-M|11-22 IS|22/2|EF-M 18-55
 
Voldenuit
Minister of Gerbil Affairs
Posts: 2806
Joined: Sat Sep 03, 2005 11:10 pm

Re: Idea for further GPU testing

Fri Oct 12, 2018 8:35 pm

TheEmrys wrote:
Take a lot at Puget Systems. They actually have a pretty nice baseline established.

https://www.pugetsystems.com/labs/articles/Photoshop-CC-2018-NVIDIA-Quadro-vs-AMD-Radeon-Pro-1202/


That's a great article, but it's still a fairly specific workload. I don't see any use of layers and masks outside of the Photomerge tasks. I've seen 2D artists with photoshop files that literally have over a hundred layers, some of which are reference images (scans, photos, hand drawings) that is used to create a their artwork from, as well as backgrounds, sprites, animations, effects, lighting masks, etc.. Even with 8-core CPUs and fast SSDs, some of those files can take a minute or more to load. And I don't know what the rate limiting step is in those cases - is it SSD speed? scratch disk speed? RAM? Cache? CPU? Is the slowest step the loading, compositing, or calculation of visibility or transparency?

My $0.02 is that Photoshop is a very powerful tool that can be used and asked to perform a really wide variety of tasks. I'd say that most users would never go into the corner cases, but the corner cases might be where performance differences become important (such as pugetsystems' discovery that the adaptive wide angle filter was abnormally slow on the Radeon systems).
Wind, Sand and Stars.
 
TheEmrys
Minister of Gerbil Affairs
Topic Author
Posts: 2458
Joined: Wed May 29, 2002 8:22 pm
Location: Northern Colorado
Contact:

Re: Idea for further GPU testing

Sat Oct 13, 2018 4:25 am

Voldenuit wrote:
TheEmrys wrote:
Take a lot at Puget Systems. They actually have a pretty nice baseline established.

https://www.pugetsystems.com/labs/articles/Photoshop-CC-2018-NVIDIA-Quadro-vs-AMD-Radeon-Pro-1202/


That's a great article, but it's still a fairly specific workload. I don't see any use of layers and masks outside of the Photomerge tasks. I've seen 2D artists with photoshop files that literally have over a hundred layers, some of which are reference images (scans, photos, hand drawings) that is used to create a their artwork from, as well as backgrounds, sprites, animations, effects, lighting masks, etc.. Even with 8-core CPUs and fast SSDs, some of those files can take a minute or more to load. And I don't know what the rate limiting step is in those cases - is it SSD speed? scratch disk speed? RAM? Cache? CPU? Is the slowest step the loading, compositing, or calculation of visibility or transparency?


For such workloads, it is easy enough to test to see if it is the GPU or not, which is the subject at hand. Use IGP first, then try with various video cards. If a different video card makes an impact, we would know it is a worth testing. If all other factors are near top-of-the-line (cpu, m.2 nvme, ram, scratch disk, etc), then it is simply the nature of the beast. But if the GPU does have an impact, it can be measured. Their free tool is a great way to measure that, particularly as it itemizes those things that are GPU dependant. My one issue I have is that there is no results that are labeled as CUDA vs OpenCL, which in itself would be of interest.

Moreover, using something like Premiere Pro CC with some solid effects put in would be easily measurable, and of interest to the growing number of photographers who are now working as videographers as well.
Sony a7II 55/1.8 Minolta 100/2, 17-35D, Tamron 28-75/2.8
 
Voldenuit
Minister of Gerbil Affairs
Posts: 2806
Joined: Sat Sep 03, 2005 11:10 pm

Re: Idea for further GPU testing

Sun Oct 14, 2018 3:15 am

TheEmrys wrote:
Moreover, using something like Premiere Pro CC with some solid effects put in would be easily measurable, and of interest to the growing number of photographers who are now working as videographers as well.


I'd recommend checking out DaVinci Resolve, I've only started playing with it, but it seems much more powerful than APPCC, and what's more, it's free! (From the makers of the blackmagic cinema cameras). Was using Vegas 14 before, but it had issues with accurate trimming, and was looking for something with better color correction tools as well.

Although perhaps not great as a multiplatform benchmark, as it accelerates CUDA but not OpenCL afaik.
Wind, Sand and Stars.

Who is online

Users browsing this forum: No registered users and 1 guest