M380 M390 M395 M395X Thread

[joema2]

Fantastic. I'll run this when I get home.

OK I did some tests and we will need to use the 4k version of that file -- the HD version runs too quickly to get an accurate time. The file is 138 MB. Go the Vimeo link on this page, select the Download button and download the 4k version:

For most effects rendering should run in the 40-60 sec range which is long enough for accuracy but short enough for convenience.

Please remember to delete the effect from the Inspector pane and delete the render files after each test: with Library selected, do File>Delete Generated Library Files, ensure Delete Render Files and "All" is selected. That keeps FCP X from re-using any render files between tests.

Also when you import it to FCP X use the "leave files in place" option and turn off all the other import options, esp. turn off proxy and optimized media. We just need the same procedure for us and anybody else that wants to run it.

 

[joema2]

Fantastic. I'll run this when I get home.

OK I re-ran all the FCP X benchmarks on the new 4k file from Vimeo. My updated numbers are below. I also added an additional test using the Photon Pro video noise reduction plugin. It is very GPU intensive. It's about 2.2x faster than my 2013 iMac 27 with GTX-780m.

https://joema.smugmug.com/photos/i-gRvVr9h/0/X2/i-gRvVr9h-X2.jpg

 

[huffy15]

quite the assumption! Pray, tell us how this is calculated!

I'll try tomorrow with my new m395x, but from istat my current imac with the m380 is running at 1.1gb with just with macrumors running on chrome. Hopefully it's just because I'm using the m380.

 

[twilexia]

Ok. Running the tests now. I'm documenting everything so in case I messed up somewhere you can tell me and I can rerun.

Open Final Cut Pro Trial
New Event
Import — leave files in place - downloaded 4k movie (138mb)
Open in Timeline


For some reason there seem to be already 3 default effects applied - not sure if this is same for you (Transform, Trim, Distort). I ended up deleting these before running the effects below. Will be running them multiple times to be sure. Also I'm not sure if clicking around/moving around tabs will affect the time of the render.
Open Effects view

Drag effect onto clip - ctrl R and timer at same time

Sharpen - 1:07.9 (this was with the three above effects still applied), 52.49
delete effect from view (ctrl+v)
delete generated effect (all)
Water Pane - 52:02, 53:30
Underwater - 56:42, 1:01.18, 53.66
Stabilize - there was no effect in library named stabilize... I did see something in Inspector but when I checked it, it started the process "transcoding and analyzing clip." with the other effects this didn't happen, so I'm thinking I messed up somehow
Gaussian - 53.37, 53.27
Aged Film - 52.87, 56.01
Add Noise - 59.22, 54.96

Seems like I'm getting quite a bit of variance in re-runs, do you see the same?

 

[joema2]

...
For some reason there seem to be already 3 default effects applied - not sure if this is same for you (Transform, Trim, Distort). I ended up deleting these before running the effects below.

I forgot about those. I just left all the default effects in place.

Just remember to delete each new effect you apply before going to the next one, then delete the render files before applying the next effect:

- Click on library in the left-hand pane
- File>Delete Generated Library Files
- Checkbox Delete Render Files
- Press the All button

- there was no effect in library named stabilize... I did see something in Inspector but when I checked it, it started the process "transcoding and analyzing clip."

Yes stabilize is always in the Inspector pane. If the clip is clicked (yellow outline in the timeline), then if you click stabilize it is automatically applied without dragging anything.

Stabilize has two passes, analyzing then when complete the orange render bar on the timeline shows it's not yet applied. From that point it's just like the other effects -- you do CTRL-R to render and apply it. That is what I measured: from CTRL-R until it finished.

...Seems like I'm getting quite a bit of variance in re-runs, do you see the same?

I got maybe 1 sec variation. I'm not sure what is causing that on your system.

I would suggest turning off Time Machine and Spotlight indexing which can cause some variation.

To turn off Time Machine, click on that in the menu bar (clock icon with backward arrow), then Open Time Machine Preferences and turn it off.

To disable Spotlight indexing, go to System Preferences>Spotlight, click the Privacy Tab, then the + button at bottom left, and select each of your hard drives. That will disable it. Just remember to reverse these steps when you're done.

Things like this are never easy.

 

[jerwin]

I'll try tomorrow with my new m395x, but from istat my current imac with the m380 is running at 1.1gb with just with macrumors running on chrome. Hopefully it's just because I'm using the m380.

Ah. That. One of those istat menu items I've learned to ignore.
Figure the screen buffer is 60 Megabytes, and... since OSX treats each window as a texture (Quartz extreme)-- i suppose you could get up into the gigabytes, given enough overlapping windows..

but maybe it's just a case of the window server exploiting resources that aren't needed elsewhere,

 

[twilexia]

I forgot about those. I just left all the default effects in place.


Yes stabilize is always in the Inspector pane. If the clip is clicked (yellow outline in the timeline), then if you click stabilize it is automatically applied without dragging anything.

Stabilize has two passes, analyzing then when complete the orange render bar on the timeline shows it's not yet applied. From that point it's just like the other effects -- you do CTRL-R to render and apply it. That is what I measured: from CTRL-R until it finished.


I got maybe 1 sec variation. I'm not sure what is causing that on your system.

I would suggest turning off Time Machine and Spotlight indexing which can cause some variation.

To turn off Time Machine, click on that in the menu bar (clock icon with backward arrow), then Open Time Machine Preferences and turn it off.

To disable Spotlight indexing, go to System Preferences>Spotlight, click the Privacy Tab, then the + button at bottom left, and select each of your hard drives. That will disable it. Just remember to reverse these steps when you're done.

Things like this are never easy.

Alright, thanks. Time machine was already off but I turned off spotlight and will re-run the FCPX effects now - this time leaving the 3 default effects in....

Sharpen: 1:03.86
Water Pane: 1:00.59
Underwater: 59.54
Stabilize: 53.89
Gaussian: 39.57, 39.45
Aged Film: 1:04.30
Add Noise: 54.09

 

[joema2]

...
Sharpen: 1:03.86
Water Pane: 1:00.59
Underwater: 59.54

You are on a 2015 iMac 27 with 4Ghz i7 and a 2GB M395? Please re-state your system specs.

Maybe also reboot OS X and re-run a couple of tests.

Mine is a 2015 iMac 27, 4Ghz i7, 32GB (Apple RAM), 1TB SSD, 4GB M395X
OS 10.11.1, FCP X 10.2.2.

I don't think the disk should make any difference in this case.

 

[twilexia]

You are on a 2015 iMac 27 with 4Ghz i7 and a 2GB M395? Please re-state your system specs.

Maybe also reboot OS X and re-run a couple of tests.

Mine is a 2015 iMac 27, 4Ghz i7, 32GB (Apple RAM), 1TB SSD, 4GB M395X
OS 10.11.1, FCP X 10.2.2.

I don't think the disk should make any difference in this case.

I'm on 10.11, shoot. Will have to update. FCPX is 10.2.2. 16GB ram (1600 Mhz). Everything else is the same besides the graphics card.

 

[joema2]

I'm on 10.11, shoot. Will have to update. FCPX is 10.2.2. 16GB ram (1600 Mhz). Everything else is the same besides the graphics card...

I don't understand why the performance difference (thus far) between your system and mine. Of course we expect the M395X to be a little faster but not that much difference. Also these FCP X effects are not generally 100% GPU bound. There is a lot of CPU involved. It must be due to something else. 10.11 vs 10.11.1 shouldn't make that much difference either. However I suggest upgrading the OS and trying again.

 

[twilexia]

I don't understand why the performance difference (thus far) between your system and mine. Of course we expect the M395X to be a little faster but not that much difference. Also these FCP X effects are not generally 100% GPU bound. There is a lot of CPU involved. It must be due to something else. 10.11 vs 10.11.1 shouldn't make that much difference either. However I suggest upgrading the OS and trying again.

I am thinking it is because I only had 16GB RAM installed. I'm putting in another module now to make 24GB RAM (1600 mhz when using the two default apple modules). Unfortunately I do not have another 16GB RAM to make a full 32 GB set, which is what I assume you have. I'm also guessing your ram speed is 2133 mhz?

Running the tests again now officially on OSX10.11.1 and Final Cut pro 10.2.2 (sorry folks, basically hijacked this thread...?)

Sharpen: 41.05
Water Pane: 42.28
Underwater: 40.62
Stabilize: 40.32
Gaussian: 27.19 (still not sure why this one is so fast)
Aged Film: 52.16
Add Noise: 42.52

I think the RAM increase made a big increase. With another 8 GB I suspect even more increases but unfortunately unless you are willing to test on your end with the setup of 2 default 4gb apple modules + two 8 gb ram modules, we won't be comparing apples to apples (literally)

As you mentioned these effects are mainly CPU bound, what effects would you say are GPU bound? Maybe we can try the following - how about this - try exporting with the following settings - Clip with default effects +Add Noise effect, export with the following settings and open with quicktime - I got 14.69 seconds.

 

[joema2]

I am thinking it is because I only had 16GB RAM installed. I'm putting in another module now to make 24GB RAM (1600 mhz when using the two default apple modules). Unfortunately I do not have another 16GB RAM to make a full 32 GB set, which is what I assume you have. I'm also guessing your ram speed is 2133 mhz?

Running the tests again now officially on OSX10.11.1 and Final Cut pro 10.2.2 (sorry folks, basically hijacked this thread...?)

Sharpen: 41.05
Water Pane: 42.28
Underwater: 40.62
Stabilize: 40.32
Gaussian: 27.19 (still not sure why this one is so fast)
Aged Film: 52.16
Add Noise: 42.52

I think the RAM increase made a big increase. With another 8 GB I suspect even more increases but unfortunately unless you are willing to test on your end with the setup of 2 default 4gb apple modules + two 8 gb ram modules, we won't be comparing apples to apples (literally)

As you mentioned these effects are mainly CPU bound, what effects would you say are GPU bound? Maybe we can try the following - how about this - try exporting with the following settings - Clip with default effects +Add Noise effect, export with the following settings and open with quicktime - I got 14.69 seconds.

 

[joema2]

Here is the spreadsheet results comparing your benchmarks to mine. This is a lot closer to what I was expecting. Thanks for doing the tests. I will do the other test you mentioned tomorrow.

If you want to run some more GPU-specific tests, you can run the ones from here: http://www.geeks3d.com/gputest/

My results for those are already posted in my previous results link.

 

[twilexia]

Here is the spreadsheet results comparing your benchmarks to mine. This is a lot closer to what I was expecting. Thanks for doing the tests. I will do the other test you mentioned tomorrow.

If you want to run some more GPU-specific tests, you can run the ones from here: http://www.geeks3d.com/gputest/

My results for those are already posted in my previous results link.

Great. Thanks a lot for your help as well. I'll now run the other GPU tests that you have on your list.

Edit: What resolution are you running these Gpu tests at (the ones from geeks3d)? And are you running Stress Test or Benchmark? Thanks.

I also placed my order for 2 more modules of 8 GB 2133mhz RAM. I think if we are to compare apples to apples it will have to be with the same # and speed of RAM. (if anyone cares the RAM I got is http://www.newegg.com/Product/Product.aspx?Item=N82E16820231760 - recommended by another user here on macrumors)

 

[looking4anotebo]

I should have waited another day to place my order.... does not look like the 395x is worth the price premium.

 

[magbarn]

Here is the spreadsheet results comparing your benchmarks to mine. This is a lot closer to what I was expecting. Thanks for doing the tests. I will do the other test you mentioned tomorrow.

If you want to run some more GPU-specific tests, you can run the ones from here: http://www.geeks3d.com/gputest/

My results for those are already posted in my previous results link.

Now that's more like it. The difference here is about the same as what's seen on Barefeats' testing. In other words $300 more for a 10% or so improvement. The i7 upgrade is a much better value if you're going to do cpu intensive stuff and for video editing as the higher frequency + hyperthreading add up to over 30% improvement in many tests.

 

[twilexia]

I should have waited another day to place my order.... does not look like the 395x is worth the price premium.

I don't think we have tested the VRAM capabilities yet, it seems so far the FCPX tests we ran were more reliant on RAM than anything else. This is something I'm very curious about too so I greatly appreciate joema2 willing to take the time out to test this. No one has really done an apples to apples comparison on exactly what VRAM does (and i keep stressing apples to apples because a lot of times people will test two graphics cards with different VRAMs but also dramatically different processing units/shaders)

If the apple status order is still at "processing" you can hit cancel and wait until more tests comes out. If it's at "preparing to ship" or "shipping" then the only thing you can do is return it after it's delivered.

Personally I still feel the 395x is worth it at least in terms of futureproofing. Maybe right now it's hard to tell due to a lot of software not leveraging the VRAM to its full capabilities yet. But perhaps in the future there will be.

Does anyone know of a good tracker to show you how much VRAM a GPU is utilizing?

 

[joema2]

As you mentioned these effects are mainly CPU bound, what effects would you say are GPU bound? Maybe we can try the following - how about this - try exporting with the following settings - Clip with default effects +Add Noise effect, export with the following settings and open with quicktime - I got 14.69 seconds.

What FCP X effects are GPU-intensive and to what degree is hard to determine. You must render the effect while observing GPU processor activity in iStat Menus or similar utilities. Even though we were running these FCP X effects as GPU tests, they vary in how GPU-focused they are. I can tell from iStat Menus that many of them have more CPU than GPU activity which lessens the relevance.

The export test you mentioned is testing more encoding (inherently CPU bound) than rendering. I can tell from your execution time it was pre-rendered. However I ran the test both ways (pre-rendered and not) on my 2015 iMac 27 with M395X and 2013 iMac 27 with 780m (both 32GB):

2015 iMac 27 M395X, ProRes export of 4k test clip with Add Noise effect, no pre-render: 36.6 sec
Same export after pre-render was 13.7 sec (vs your 14.69 sec)

The 2013 iMac 27 780m did 45.3 sec and 25.3 sec (vs your 14.69 sec)

What resolution are you running these Gpu tests at (the ones from geeks3d)? And are you running Stress Test or Benchmark?

1920x1080, windowed (not full screen), and Benchmark not Stress Test.

 

[OSB]

Now that's more like it. The difference here is about the same as what's seen on Barefeats' testing. In other words $300 more for a 10% or so improvement. The i7 upgrade is a much better value if you're going to do cpu intensive stuff and for video editing as the higher frequency + hyperthreading add up to over 30% improvement in many tests.

It’s down to workloads and how you frame the analysis.
For the applications I use, which tend to make fair-to-good use of the GPU, and for games, a 10-12% increase in performance is commensurate with the ~7% premium on the all-in costs of my machine. It’s not a slam-dunk always-buy upgrade for every workload, but within the scope of my own use case it’s an entirely adequate value.

 

[joema2]

what effects would you say are GPU bound?

I just tested a bunch of stock FCP X effects and Texture and Film Noir were both GPU intensive, as indicated by iStat Menus. Interestingly the M395X was not much faster than the 780m in these, whereas on other GPU-intensive effects the M395X was a lot faster. This just shows how complex benchmarking is. My numbers for our standard test clip were:

Texture on 2015 iMac 27 with M395X: 46.4 sec
Texture on 2013 iMac 27 with 780m: 45.0 sec
Film Noir on 2015 iMac 27 with M395X: 45.3 sec
Film Noir on 2013 iMac 27 with 780m: 43.1 sec

 

[twilexia]

Running GPU benchmarks now (system is i7, 16gb 2133mhz ram, 1tb ssd, m395)
1920x1080 windowed
Furmark no AA: 3216 (fps 53)
Furmark 8x AA: 761 (fps 12)
Tessmark8 no AA: 7206 (fps 120)
Tessmark16 no AA: 7195 (fps 119)
Tessmark32 no AA: 7198 (fps 119)
Tessmark64 no AA: 5607 (fps 93)
Tessmark64 x4 AA: 4238 (fps 70)
GiMark no AA: 3944 (fps 65)
GiMark 4xAA: 2868 (fps 47)
JuliaFP64 no AA: 3994 (fps 66)
JuliaFP64 4xAA: 3967 (fps 66)
JuliaFP32 no AA: 7197 (fps 119)
No AA:
Pixmark Piano: 731 (fps 12)
Pixmark Volplosion: 1695 (fps 28)
Plot3d: 7198 (fps 119) - ran this again - 7196 (fps 119)
Triangle: 7198 (fps 119) (exactly the same..... for real)

Joema, would you deign to give us some insight into these results and the real world performance significance of such?

Edit: running FCPX tests now on the 4k clip we used yesterday (this is with 16GB 2133 mhz ram as I removed the stock Apple modules in preparation for my two more 2133 mhz ram coming in soon):
Textures: 52.96
Film Noir: 49.84

 

[joema2]

..

Joema, would you deign to give us some insight into these results and the real world performance significance of such?

Attached is the spreadsheet with the combined GPU results. For some reason Tessmark seems slower on yours. I will re-run mine tomorrow to double-check. In our real-world FCP X tests there wasn't a huge difference in general. The M395X was a little faster but it wasn't huge.

Some of my results were slower than the 780m. You never know what little feature of the GPU or driver the test is exercising. A single test could give a much lower or higher value yet not be broadly relevant. That's why you runs lots of tests.

I am happy with my iMac with M395X; in general it is usefully faster than the 2013 iMac with 780m. Yours seems pretty good too.

 

[twilexia]

Attached is the spreadsheet with the combined GPU results. For some reason Tessmark seems slower on yours. I will re-run mine tomorrow to double-check. In our real-world FCP X tests there wasn't a huge difference in general. The M395X was a little faster but it wasn't huge.

Some of my results were slower than the 780m. You never know what little feature of the GPU or driver the test is exercising. A single test could give a much lower or higher value yet not be broadly relevant. That's why you runs lots of tests.

I am happy with my iMac with M395X; in general it is usefully faster than the 2013 iMac with 780m. Yours seems pretty good too.

Don't get me wrong, I'm not trying to show one way or the other that the M395 or M395X is better value. In fact I am leaning towards returning mine and getting the M395X. I'm truly curious about how VRAM affects performance in FCPX (and maybe Premiere Pro if you want to test that?)

For the GPU test, is the first number the time it takes to run the test, in which case shorter = better?

 

[jerwin]

It's the "score". Higher is better. How much vram does the GTX-660 have? The m395X shows some clear advantages in a great many of those tests.

 

[jerwin]

Those 119 fps figures for the m395 look suspicious.

Here are the results for my 290x, a significantly less powerful card with 2 GB ram



Looks like a bug.