Announcement: Our new CyberLink Feedback Forum has arrived! Please transfer to our new forum to provide your feedback or to start a new discussion. The content on this CyberLink Community forum is now read only, but will continue to be available as a user resource. Thanks!
CyberLink Community Forum
where the experts meet
| Advanced Search >
Speed rendering tests: NVidia and Skylake GPU
Notonotonurb [Avatar]
Newbie Joined: Oct 20, 2015 15:16 Messages: 20 Offline
[Post New]
Setup:

--------

System: i7 6700 Skylake, 32 GB DDR4 ram, no overcloking, Nvidia GTX960, driver 359.06

System drive: SSD Samsung XP941, media drive: Seagate SSHD 4TB

OS: Windows 10 treshold 2

PD 14 with patch v2302

Task: rendering a short (about 4 min) UHD video in HEVC. (https://www.youtube.com/watch?v=SwfKwO-3KIo)



Both the intel GPU and NVIDIA GPU support HEVC encoding





Results

--------

First, the NVIDIA card was removed:

Hadrware optimization OFF:

TIme: 23:16 , CPU usage during transitions: 100% (92w) , CPU usage without transitions: 100% (92w)

Hadrware optimization ON -Intel GPU

TIme: 9:46 , CPU usage during transitions: 100% (92w) , CPU usage without transitions: about 70-80% (86 w)



Second , the NVIDIA card is placed in the PCIe slot:

Hadrware optimization ON

TIme: 7:00 , CPU usage during transitions: 100% (110w) , CPU usage without transitions: about 46% (NA w)



Third , the NVIDIA card is placed in the PCIe slot and the on board GPU is enabled in the BIOS

TIme: 7:04 , CPU usage during transitions: 100% (110w) , CPU usage without transitions: about 46% (NA w)



Conclusions

--------------



For this project, the Nvidia card provides the fastest rendering (speedup factor = 3.3). The Skylake GPU comes second (speedup factor = .2.4).



The combination of both graphics cards did not bring further improvements and gave the same results as the NVIDIA card.

The problem that I don't know how to enable multi GPU acceleration. Or may be the internal GPU is always enabled/used? Asus Z170, i7 6700, 32GB DDR4, GTX1070
Win 10 pro
Samsung Galaxy Note 3
Samsung NX1
optodata
Senior Contributor Location: California, USA Joined: Sep 16, 2011 16:04 Messages: 8630 Offline
[Post New]
Quote: ... The problem that I don't know how to enable multi GPU acceleration. Or may be the internal GPU is always enabled/used?


If both GPUs show up in Device Manager, then Multi-GPGPU should be active. There may also be a setting in your system UEFI/BIOS. If you want to use only your GTX960, you can disable the Intel GPU in Device Manager

YouTube/optodata


DS365 | Win11 Pro | Ryzen 9 3950X | RTX 4070 Ti | 32GB RAM | 10TB SSDs | 5K+4K HDR monitors

Canon Vixia GX10 (4K 60p) | HF G30 (HD 60p) | Yi Action+ 4K | 360Fly 4K 360°
Notonotonurb [Avatar]
Newbie Joined: Oct 20, 2015 15:16 Messages: 20 Offline
[Post New]
Quote:

If both GPUs show up in Device Manager, then Multi-GPGPU should be active. There may also be a setting in your system UEFI/BIOS. If you want to use only your GTX960, you can disable the Intel GPU in Device Manager


Thanks for the info. I connected two monitors to my PC, one to the HDMI output of the motherboard and one to the DVI output of the GTX960. With this configuration, only Intel video encoding is available. No speed improvement. Asus Z170, i7 6700, 32GB DDR4, GTX1070
Win 10 pro
Samsung Galaxy Note 3
Samsung NX1
optodata
Senior Contributor Location: California, USA Joined: Sep 16, 2011 16:04 Messages: 8630 Offline
[Post New]
I do the same thing to get Intel QuickSync to show up when producing. If I run both off my GTX780Ti, I get the nVidia HW option in PD but it's in MultiGPGPU mode.

You can run 2 copies of GPUz to monitor the activity of each GPU with different monitor hookups.

YouTube/optodata


DS365 | Win11 Pro | Ryzen 9 3950X | RTX 4070 Ti | 32GB RAM | 10TB SSDs | 5K+4K HDR monitors

Canon Vixia GX10 (4K 60p) | HF G30 (HD 60p) | Yi Action+ 4K | 360Fly 4K 360°
Notonotonurb [Avatar]
Newbie Joined: Oct 20, 2015 15:16 Messages: 20 Offline
[Post New]
Quote: I do the same thing to get Intel QuickSync to show up when producing. If I run both off my GTX780Ti, I get the nVidia HW option in PD but it's in MultiGPGPU mode.

You can run 2 copies of GPUz to monitor the activity of each GPU with different monitor hookups.


Thanks for the tip about GPUz. If I enable the onboard GPU in the BIOS, I can see both GPU in GPUz.

The HEVC rendering time with only the NVIDIA graphics card for a new project is 03:21. When both the NVIDIA and the Intel GPU are enabled (confirmed by GPUz), the rendering time is 03:23. Without GPU acceleration, the rendering time is 12:27.

Conclusion: having two GPU's does not improve rendering time. At least not in the system I have. Asus Z170, i7 6700, 32GB DDR4, GTX1070
Win 10 pro
Samsung Galaxy Note 3
Samsung NX1
[Post New]
That's normal. Each of the above GPU has a dedicated hardware pice for accelerating encoding of h264 and h265.
The NLE software (PD) can choose one hardware to pipe the encoding process trough... but it would have to "split" the file in two in order to use simultaneous the two hardware encoders. That's not likely to be implemented because of the differences between the two drivers and even the quality settings.
Windows 10 can use Multi GPU for tasks that involve the "cores" of the GPU, but the encoding hardware is something else, is not part of the "GPU cores".

This message was edited 1 time. Last update was at Dec 12. 2015 16:35

JL_JL [Avatar]
Senior Contributor Location: Arizona, USA Joined: Oct 01, 2006 20:01 Messages: 6091 Offline
[Post New]
Quote: Conclusion: having two GPU's does not improve rendering time. At least not in the system I have.

From what I've seen, if configured correctly, both GPU and iGPU will be utilized but only for timeline content that utilized OpenCL accelerated effects, so it has limited applicability in my view. This was true on all other iGPU's so I can't believe the Skylake any different, although I've not tested for verification.

Jeff
Julien Pierre [Avatar]
Contributor Joined: Apr 14, 2011 01:34 Messages: 476 Offline
[Post New]
FYI, I now have 2 Intel machines - one i5-6600k (OC at 4.5 GHz) with GTX 750Ti + iGPU, and one i7-5820k (OC at 4.3 GHz) with a GTX 960.

In my tests, the i7 Haswell-E box trounces the i5 Skylake box by 70% to 100% in rendering time. This is true regardless of using QuickSync or multi GPGPU on the Skylake machine, or just the nVidia GPU. It is also true for pure software encodes without acceleration. I have tested both H.264 and H.265 and this holds true as well.

What is nice about the Skylake is that the QuickSync supports H.265 hardware acceleration, so I can keep the old 750Ti GPU and still be able to encode H.265 with acceleration. But the QuickSync H.265 is still much slower than a GTX 960. Now, if Youtube would start supporting H.265 uploads...

I will post more detailed data about my tests data later.

But to anyone who is shopping for a CPU/GPU, I would recommend the X99 platform with quad channel RAM, and a discrete nVidia GPU, vs a Skylake with only dual-channel RAM and the iGPU. Of course, the X99 platform will cost more in general. There were some good sales around black friday, though. I would have bought a Skylake i7-6700k rather than the i5-6600k I got, but just couldn't find a chip in stock at a reasonable price. The cost of the 6700k has skyrocketed from the $300 range to $400 - $500, and there seem to be shortages. Whereas the 5820k goes on sale more regularly for under $300. I paid $276 for my 5820k, and $229 for my 6600k.

This message was edited 2 times. Last update was at Dec 17. 2015 20:39

MSI X99A Raider
Intel i7-5820k @ 4.4 GHz
32GB DDR4 RAM
Gigabyte nVidia GTX 960 4GB
480 GB Patriot Ignite SSD (boot)
2 x 480 GB Sandisk Ultra II SSD (striped)
6 x 1 TB Samsung 860 SSD (striped)

2 x LG 32UD59-B 32" 4K
Asus PB238 23" HD (portrait)
Julien Pierre [Avatar]
Contributor Joined: Apr 14, 2011 01:34 Messages: 476 Offline
[Post New]
FYI, here is my test data on Skylake i5-6600k & Haswell i7-5820k, with GTX 750TI and GTX960 respectively, and also the iGPU on the Skylake.

Both machines have 32GB RAM. But the Skylake has 2x16GB DDR4-3000, ie. dual channel, whereas the Haswell-e has 4x8GB DDR4-2666 (RAM is overclocked at 2800). Both CPUs are overclocked.

The project is 11:48 with all HD 1080i H.264 source material with a few titles, but otherwise no effects.

On the i5 box, 2 monitors are running off the GTX 750 Ti, and one monitor off the iGPU, Intel Graphics 530.

When I tested QuickSync on lines 8, 9 and 20, I had both GPUs enabled in the device manager, and all 3 monitors connected.

On the other hand, when I tested the GTX 750Ti, on lines 12, 13 and 22, I had the Intel Graphics 530 disabled in the device manager.

So, perhaps I did not get to test the Multi GPGPU mode at all. Would that require leaving both GPUs enabled in device manager, but disconnected the monitor from the Intel Graphics 530 ? It makes little sense that the rendering mode in PD should depend on which monitor is attached or not.

For H.264, the GTX 750Ti is faster than the Intel Graphics 530, yet there seems to be no way to force PD to use the faster GPU, short of disabling the Intel GPU in device manager. But that also means I lose my 3rd monitor. I still have one output left on the 750Ti, and could run that monitor off the GTX 750Ti too. But then, would Quicksync be available at all, for H.265, where I actually want it ?

For me, the optimal performance on the Skylake i5 system is obtained by using the GTX 750Ti for H.264 encoding, and using the Intel Graphics 530 for H.265 encoding (since the GTX 750Ti does not support H.265 encoding at all). But there is no way inside PD to make the GPU selection for the rendering. Enabling/disabling a GPU in device Manager seems to be the only way. This is really not very practical.

That said, the combination of the i7-5820k and the nVidia GTX 960 clearly trounces the Skylake i5 with either the GTX 750Ti, or the Intel Graphics 530.

Software rendering is also much faster on the i7-5820k, no doubt due to the 6 cores, as well as larger cache, and quad channel memory.

I am not happy at all with the stability of the Z170 chipset also, even at stock settings (without any overclock). The Gigabyte Z170XP-SLI motherboard I have has a very problematic BIOS. I may return it and exchange it for another model of motherboard.

For the i7 5820k Haswell-E, I'm using an MSI X99-A Raider motherboard, and that is a much more polished platform along with a much better BIOS.

I hope this data helps. I will also test the project listed in the original post. EDIT: can't do that since it's only a Youtube link, no project data to try.

This message was edited 1 time. Last update was at Dec 18. 2015 02:15

MSI X99A Raider
Intel i7-5820k @ 4.4 GHz
32GB DDR4 RAM
Gigabyte nVidia GTX 960 4GB
480 GB Patriot Ignite SSD (boot)
2 x 480 GB Sandisk Ultra II SSD (striped)
6 x 1 TB Samsung 860 SSD (striped)

2 x LG 32UD59-B 32" 4K
Asus PB238 23" HD (portrait)
Notonotonurb [Avatar]
Newbie Joined: Oct 20, 2015 15:16 Messages: 20 Offline
[Post New]
Thank you for posting your test results. It certainly required some work. It is a nice table.

My Skylake motherboard is an Asus ROG Maximus VIII Gene. It is very stable wiht the latest firmware update. I am using an Intel I7 6700 (wihtout a K, I could not get a 6700K). Asus Z170, i7 6700, 32GB DDR4, GTX1070
Win 10 pro
Samsung Galaxy Note 3
Samsung NX1
optodata
Senior Contributor Location: California, USA Joined: Sep 16, 2011 16:04 Messages: 8630 Offline
[Post New]
Quote: ... For me, the optimal performance on the Skylake i5 system is obtained by using the GTX 750Ti for H.264 encoding, and using the Intel Graphics 530 for H.265 encoding (since the GTX 750Ti does not support H.265 encoding at all). But there is no way inside PD to make the GPU selection for the rendering. Enabling/disabling a GPU in device Manager seems to be the only way. This is really not very practical.


You should be able to get PD to show the QS option as long as one monitor is connected to you motherboard's video connecter, and you should see the HWE option when both monitors are connected to your nVidia card. You only need to close PD before switching monitors, no need to reboot. I finally added an HDMI switch so I don't have to pull out my computer and manually switch the cable every time wink

As long as both devices are enabled in Device Manager you'll also have MultiGPGPU mode active, which you can confirm by running 2 copies of GPU-Z, and set one to monitor each GPU.

YouTube/optodata


DS365 | Win11 Pro | Ryzen 9 3950X | RTX 4070 Ti | 32GB RAM | 10TB SSDs | 5K+4K HDR monitors

Canon Vixia GX10 (4K 60p) | HF G30 (HD 60p) | Yi Action+ 4K | 360Fly 4K 360°
Notonotonurb [Avatar]
Newbie Joined: Oct 20, 2015 15:16 Messages: 20 Offline
[Post New]
I have prepared a PD14 test that we can all use to compare the rendering performance of our computers:

https://docs.google.com/uc?id=0B13xfYZIRuNVTEF4SUU2Y0Y0NDQ&export=download

Warning: this is a 1.2 GB file. It contains two short UHD clips. The first one is color corrected.



I have attached my results as a picture to this post. I used the same colum labels and naming convention as posted by Julien Pierre.



Bruno
[Thumb - pd14_test.png]
 Filename
pd14_test.png
[Disk]
 Description
 Filesize
6 Kbytes
 Downloaded:
162 time(s)
Asus Z170, i7 6700, 32GB DDR4, GTX1070
Win 10 pro
Samsung Galaxy Note 3
Samsung NX1
Alstar250 [Avatar]
Newbie Location: San Diego Joined: Jan 04, 2016 15:03 Messages: 3 Offline
[Post New]
Awesome post and thank you for the information. I have a question as I am looking at the GTX 750 Ti vs the GTX 950 and GTX 960. I have just purchased PD 14 Ultimate Suite to edit my GoPro footage. My computer is a i7 3770 (3.4ghz 4-core) with HD 4000 integrated graphics, 32gb 1600mhz DDR3 and a 1tb Samsung EVO SSD. In working with h.264 content would I experience a significant gain from the GTX 960 vs the GTX 950 or even a 750 Ti. I know the 960 had the h.265 support but I can only shoot 4K @ 15fps so I rarely use it. Id like to stay under the $150 mark if possible but I would spring for the 960 if the benefits are significant. Thanks again!h

This message was edited 1 time. Last update was at Jan 04. 2016 15:59

[Post New]
GTX 950 has the same nvenc hardware encoder like GTX960 - and it is better then the other member of the 9xx family. Definitely better than 750 one. It has several improvements in encoding quality - even for h264.

https://en.wikipedia.org/wiki/Nvidia_NVENC
I think that for strictly video editing, 950 provides very similar performance to 960. Some GPU accelerated effects might go faster on 960, but I really doubt it.
Personally I have a 960 because... 950 wasn't available yet at that time and 960 was on sale

This message was edited 2 times. Last update was at Jan 04. 2016 20:37

Alstar250 [Avatar]
Newbie Location: San Diego Joined: Jan 04, 2016 15:03 Messages: 3 Offline
[Post New]
Thanks SoNic67! I've seen the 960's down to within $20 of the 950's in some places so I can go either way. I guess my only question now would be in my not-so-knowledgeable grasp of CUDA vs Opencl, would PD14 work better with one or the other? Reason being is I've been recommended by others the likes of the R9 380 or the R7 370 in lieu of the 960/950's from Nvidia. Keep in mind I don't play any games at all and am solely looking for a card to help with HD video production. Even though I said I use almost solely h.264, the fact that I will have h.265 abilities in the two Nvidia cards is now a must for me.
JL_JL [Avatar]
Senior Contributor Location: Arizona, USA Joined: Oct 01, 2006 20:01 Messages: 6091 Offline
[Post New]
Quote: Reason being is I've been recommended by others the likes of the R9 380 or the R7 370 in lieu of the 960/950's from Nvidia.

Might be of interest, http://forum.cyberlink.com/forum/posts/list/45474.page#234931 http://forum.cyberlink.com/forum/posts/list/46690.page#242239

A little confused, if you state this,
Quote: Even though I said I use almost solely h.264, the fact that I will have h.265 abilities in the two Nvidia cards is now a must for me.
then a R9 370/380 can't be an option as no H.265 encode support.

Keep in mind, the major benefit will primarily be "Produce" of video files and only when hardware encoding is utilized, very little real editing benefit with current PD14 release.

Jeff

This message was edited 1 time. Last update was at Jan 05. 2016 22:20

Alstar250 [Avatar]
Newbie Location: San Diego Joined: Jan 04, 2016 15:03 Messages: 3 Offline
[Post New]
Thanks Jeff.... Man I get so much info from you all!

Yeah I mentioned the h.264 initially as I don't plan on shooting anything in 4K until I get a better camera. And that's when those AMD cards were recommended to me. BUT... The fact that I do plan on upgrading to a 4K capable camera I'd like to get a card that does have h.265 support to kind of future proof myself.

Seeing as these files are so demanding as it is, any help I can get to cut down on production time in any aspect is welcome. And I don't have a video card yet so might as well go with what others are having good luck with. I'm sort of handicapped in that my case doesn't fit very large cards. So I'm stuck with the single fan versions. Just ordered the MSI GTX 950 from newegg for $139 shipped. Thanks again!

This message was edited 2 times. Last update was at Jan 06. 2016 04:30

[Post New]
Good choice! AMD cards not only that are bigger, but they require much power supplies (and they generate more heat) than the 950/960.
Smaller PC cases usually have smaller power supplies, not sure what you really have there.

This message was edited 1 time. Last update was at Jan 06. 2016 05:53

Powered by JForum 2.1.8 © JForum Team