i7 3930k vs. i7 5960x Render Speed Comparison

doublehamm wrote on 2/9/2016, 3:25 PM

EDIT: I have SCS Benchmark Project results below after my initial test!

EDIT #2: Here are my current best results as shown lower in the thread:

Preliminary tests are in. TLDR: lightly overclocked i7 5960x @ 3.978GHz (64GB RAM @ 2400MHz on ASUS Rapage V Extreme MOBO) had significant improvements in render times over moderately overclocked i7 3930k @ 4.4GHz (32GB RAM @ 1600 MHz on ASUS Rampage IV Extreme MOBO).

Originally I was able to get the 5960x to run at 4.5GHz for most of the day. EVERYTHING worked wonderfully on the PC. The only issue I had was when I did a longer render in Vegas Pro 12 and Vegas crashed 20-40 minutes into each render. I tried at 4.5GHz, 4.4GHz, 4.3GHz, and 4.1GHz with the same result. The rest of the PC seemed very stable, so I am unsure what was going on there. Once I went back to the Bios and used a dummy "EZ" tuner, I was able to complete an 8+ hour render overnight with no issue.

Eventually I would love to get the 5960x where most seem to get it running, but for now with the current settings, I am happy with the improvement. If I can figure it out I will do another test!

First, I used TeamViewer to control both PCs on the 5960x PC. This made it pretty easy to compare side by side what was going on, as well as take simultaneous screenshots of the 2 PCs. To be the most fair of tests, I used the exact same project, exact same 15 second section of the project to render. Both PCs had the project files on identical SSD C: drives (Samsung Evo 850 500GB), and I rendered directly to the Desktop on the C: Drives. Usually I would render off of and on to separate drives, but as of today I did not have that luxury on the 3930k PC.

Here are the effects applied to the clip. Some basic CC effects and NeatVid to give it some beef:

I rendered them both to Sony XAVC S MP4 files. Here are my render settings:

This is a screen shot of both PCs when the 5960x hit 50%, the 3930k was at 40%

This is a screen shot when the 5960x hit 100%, while the 3930k PC was still at 78%:

This is a screen shot when the 3930k PC hit 100% and you can compare both render times:

i7 5960x = 3:45 render time. i7 3930k = 4:51 render time.

Lastly, I have a screenshot of HWiNFO 64 running on Both PCs to show the settings.

The i7 3930k has a 4 year old Corsair H60 liquid CPU cooler on it with a single 7 year old case fan pulling air IN to the case from the back side. I was originally not so sure how this cooler would perform, but it is seems to do just fine at these settings (4.4GHz). RAM was running at 1600MHz all manually overclocked.

The i7 5960x has a brand new Corsair H100i dual 120mm fan radiator at the top of the case pulling air IN to the case. I do not have fans for a push pull setup as they are just shy of actually fitting inside the case as the ATX 8 pin power connector gets in the way. CPU was running at 3.978GHz as set by the "EZ" tuner in the bios. RAM running at 2400MHz as set by the “EZ” tuner in the bios.

For temperatures, The 3930k had 1 core top out at 69C, the rest were 66C and cooler. The 5960x had 2 cores top out at 62C, but most of the cores topped out under 60C.

Both of these PCs were running in the same 12'x13' office with a closed door, so if I left the door open it may have even been a bit cooler as it did heat the room up in a hurry.

Both of these tests were started within a half of a second of each other - as fast as I could move between the 2 monitors.

What I did notice is the 5960x PC seems to have a slight delay to start the render (about 2 seconds longer than the 3930k) so in the first few seconds it appeared as though the 3930k might be ahead, but that quickly changed as the render went on.

That’s all I have for now! If I can get this PC overclocked any further successfully, I would gladly share the results again.

Comments

OldSmoke wrote on 2/9/2016, 3:38 PM

It would have been better if you would have used the SCS Benchmark project as this will give us a better relation in terms of render times and improvement.

doublehamm wrote on 2/9/2016, 3:40 PM

I think I remember looking for this before, do you have a link to it? I can also do longer render tests as well.

doublehamm wrote on 2/9/2016, 3:55 PM

I found a thread where you had a link to it but it says file not found.

John222 wrote on 2/9/2016, 4:29 PM

http://download.sonycreativesoftware.com/whitepapers/vp11_benchmark.zip

OldSmoke wrote on 2/9/2016, 4:34 PM

John was faster.

You should also try and convert the SCS benchmark project to 4K. Its actually quite simple. Render all media to XAVC-Intra 29.97p, change your project settings to 4K, replace the 1080 media with the 4K media and change all generated media to 4K by selecting them all and running the "Resize Generated Media" script.

doublehamm wrote on 2/9/2016, 5:19 PM

Working on it! Are these all essentially the same footage? Can I just render out one then replace all with the 4k file? I will do it separately until I hear otherwise, just trying to shave a few minutes.

OldSmoke wrote on 2/9/2016, 5:29 PM

There are 3 different media inside which you render separate to XAVC-I 4K.

doublehamm wrote on 2/9/2016, 5:39 PM

Doing it on 2 PCs. What I was asking is if the footage was the same, and I am rendering them all to 4K, they will all be the same anyhow with different names.

OldSmoke wrote on 2/9/2016, 5:46 PM

I still don't quite get what you mean. The original project contains three different media in different codecs but you render them all out to one XAVC-I codec. That will give you three XAVC-I files (.mxf) with the original file name. Then you use the replace media function to get the new 4K files into the 4K project.
It is important that you resize the generated media too.

doublehamm wrote on 2/9/2016, 6:06 PM

No worries, I got them. How do you want me to render? XAVC S or XAVC-I?

EDIT: Nevermind that, I got both versions to share, working on compiling info now!

doublehamm wrote on 2/9/2016, 7:32 PM

4K SCS Benchmark Tests are in!

The first series of pics are from the XAVC-I render. This was pretty impressive! I still showed where the 3930k was at when the 5960x was at 50% and 100%, although I understand that different parts of the project were easier to render. The end results were still pretty good!

XAVC-I Total Render Times:

3930k: 5:54
5960x: 3:17

5960x was 79.7% faster!

First the project media that was converted:

And the results - Again I was able to use simultaneous screen shots by using TeamViewer (No I do not get $ for saying that, just stating how I was able to do both PCs at one time)

Next up was the XAVC-S Render

Total Render Times:

3930k: 3:23
5960x: 2:38

5960x was 28.5% faster.

I also reset the HWiNFO session before this render. This render was a touch warmer, but I think it is more due to this being both in a small closed room and the first render heated up the room quite a bit.

OldSmoke wrote on 2/9/2016, 8:34 PM

Ok, I am a bit confused. I just rendered my SCS Benchmark 4K project on my 3930K @4.3GHz and a R9 Fury X in 3:19.
Did you enable your GPU and set your preview ram to the default 200?
What is your "OpenCL Memory Size Filter" setting on the "Internal" settings tab? Mine is 1024.
Also my clips are larger files, about 1GB larger then yours.
Rendering to XAVC-S took 3:11

Edit: I just read that your are reading and rendering to and from the same drive? And that is your C: drive? I am reading from a RAID10 and I render to a SSD which is still slower then the other way around.

doublehamm wrote on 2/9/2016, 8:47 PM

Are you using Vegas Pro 12 or 13? This is Vegas Pro 12 if that helps. Also your 3:19 render, was that XAVC-I?

I am changing my Dynamic RAM preview to 200 as you suggest. GPU acceleration of video processing is set to AMD (Hawaii) I have a pitcairn card in there as well for a 3rd monitor.

Where is the Internal Settings tab?

I am only reading and writing from the same drive for fair comparison. My 3930k PC only has one drive in it at the moment.

OldSmoke wrote on 2/9/2016, 8:50 PM

Yes, the 3:19 is XAVC-I and VP13. I will try VP12 too.
You can access the "Internal" tab by holding down SHIFT when opening "Preferences"; there will be a new tab called "Internal". On the bottom of the tab you type in "opencl mem" and you will see it. Change value, apply and restart Vegas.
Also render the original project to XDCAM EX 1080 29.97p... that should be really fast, below 20sec.

Edit:
VP12 results: XAVC-S=3:15; XAVC-I=4:10
It seems there is a difference how XAVC-I is implemented in VP13.

doublehamm wrote on 2/9/2016, 9:06 PM

I am downloading trial of VP 13 to see. Also, FWIW the 3930k PC has a nVidia GTX 680. The 5960x has R9 390x, but not sure how much those impct the actual rendering. As far as I knew, these renders were CPU only and I do not see a GPU option for the XAVC profiles.

OldSmoke wrote on 2/9/2016, 9:10 PM

There is no GPU option for XAVC but any effect will be accelerated via OpenCL in the same way as it would be for timeline preview. The R9 390X will do a much better job with the OpenCL then the GTX680 which may not help at all.

doublehamm wrote on 2/9/2016, 10:03 PM

Here are my results in VP13. It is faster.

I am also wondering if your larger XAVC-I files were because they came from VP 13, and if they might make a difference on the render. If I feel ambitious tomorrow I will try, but I am behind on some other editing. So far at least the XAVCS seems to be a significant improvement.

I also noticed whenever the VP13 watermark popped up, the render seemed to slow just a bit, so that might actually be slowing this down slightly as well.

Lastly I did the HDCAM EX in 17 seconds.

doublehamm wrote on 2/10/2016, 12:01 PM

Well, I got off my behind and purchased VP 13. I suspected the watermarks were slowing down the renders and I think I was correct! Here are my results of XAVC-I, XAVC-S and XDCAM HD renders with VP13.

EDIT: I also re rendered all the footage though VP 13 and they were a bit larger than the VP 12 counterparts I rendered yesterday.

astar wrote on 2/10/2016, 7:59 PM

The 4K Sony test file, keeping in line with the original, should probubly be XAVC-Intra, XAVC-L, and XAVC-S formats combined. Also since XAVC is 10-bit, tests should be run with the project setting of 32-bit FP Video levels. Just a thought.

An EXR, DXP, TIFF combo test might be good as well in ACES mode.

I have always thought this about the Anandtech reviews stating there is a limit to GPU acceleration benefits. When in reality their tests are all being run in 8-bit mode, when they should be testing 10-bit codecs in 32-bit mode for a degree of difficultly. Tests in 32-bit full mode would show a huge difference in card performance.

A new 4K/HD test that would be good is where the most popular formats are represented: EXR, DXP, TIFF, DNG, PNG sequences, XAVC-i-L&S, HDCAM, XDCAM, AVCHD, and Cineform are represented. All run in 32-bit full project mode. This would be more of an uphill slope test for Vegas, and really show hardware differences in compute abilities.

doublehamm wrote on 2/10/2016, 8:16 PM

Newest tests at 4.5GHz. Thank you for the help OldSmoke. Now fingers crossed this is 100% stable over the next couple days.

I will run the 32 bit tests, but is there anything to compare it to?

OldSmoke wrote on 2/10/2016, 8:23 PM

Much better results!

I can do it on my 3930K system too, that would give us 3 data points already. I like the idea of using all three XAVC versions, one for each media clip, just let me know how did it.

Edit:
Try one more thing. Nothing to do with overclocking but disable resample on all events, it's all progressive anyways.

doublehamm wrote on 2/10/2016, 9:01 PM

OldSmoke I did not see your disable resample message until now, but this is what I got. I have not tried 32 bit mode since Vegas 10 was the current (I have no real use for it), I am blown away.

XAVC-I 32 bit render = 2:34
XAVC-L 32 bit render = 2:47
XAVC-S 32 bit render = 2:26

OldSmoke wrote on 2/10/2016, 9:14 PM

These are great results!

I think what astar meant was that instead of having all three clips as XAVC-Intra in the 4K project, we should use one XAVC-Intra, one XAVC-Long and one XAVC-S. Then switch the project settings to min. 32bit levels only and then give it a swirl. Also try with 32bit full as that would preserve 10bit content. All that should show is heavy load on the GPU.

doublehamm wrote on 2/10/2016, 9:19 PM

OldSmoke

Well, I have procrastinated on my other house work all week, but I will be doing this test again when I get a chance with all 3 files you mentioned.

On a side note, I noticed that when rendering XAVC-I, changing Dynamic RAM preview from 200 to 1024 helps a TON! I have no idea what that option really is, but it greatly speeds up the XAVC-I renders. For XAVC-S I have to move it down to your suggested 200 for the best performance.