Benchmarking

Comments

Former user wrote on 7/22/2019, 4:30 PM

@Damian Is updated now Damian.

petecarney wrote on 7/22/2019, 4:31 PM

User name  :PeteCarney
Machine   :OB
VP version  :16 (B424)
CPU   :TR1950X
Cores   :16
GPU   :Radeon VII
OC   :No
Render frame size :UHD
Encode mode  :Magix UHD 25p VCE
Render time  :1:49
FPS   :12-21

Former user wrote on 7/22/2019, 4:46 PM

@petecarney Is updated now Peter.

Former user wrote on 7/22/2019, 6:14 PM

@Howard-Vigorita I’ve just double checked my spreadsheet entries with what you supplied earlier today and I can see no discrepancies.

I had thought earlier that your 45s CPU only time was a bit off, but I got bogged down with BruceUSA'S issue. When you entered Encode mode as Mainconcept, with nothing else, I took it that you meant Cpu only.

Can I make a suggestion, rather than attempt to integrate your new info here with your previous info. just modify all of your previous post, or do a new third post with everything in it, whichever is easier for you.

I prefer to enter data from just one post, whether its the earlier one today modified or a complete new one, thanks.

Its sufficient for encode mode to only enter ... CPU, QSV, NVENC, CUDA VCE.

I think you want to add these values without HW-Acceleration, looking forward to your single complete list.

Its 00:27 my time so I’ll pick this up again tomorrow.

I welcome additions, or highlighting any data entry error I may make, so if anyone else sees such do let me know.

If its a major update, rewrite like this then please supply a complete new post with all of your testing.

 

 

fr0sty wrote on 7/22/2019, 11:31 PM

The GPU is used for timeline processing as well, so that is likely why you see usage during Mainconcept encodes.

Systems:

Desktop

AMD Ryzen 7 1800x 8 core 16 thread at stock speed

64GB 3000mhz DDR4

Geforce RTX 3090

Windows 10

Laptop:

ASUS Zenbook Pro Duo 32GB (9980HK CPU, RTX 2060 GPU, dual 4K touch screens, main one OLED HDR)

Kinvermark wrote on 7/23/2019, 12:03 AM

Briefly, I think we need to keep the number of render options ("name tags") to a minimum or the table will become unreadable. Perhaps NVENC, VCE, and "Native" will suffice. ( With HD and UHD that's already 6 categories.)

Even "native" clearly has some GPU involvement and even CPU ASIC decoding. That's OK IMHO.

Oddly, I don't see a QSV option for AVC on my Vegas, just for HEVC - which we are not testing.

Former user wrote on 7/23/2019, 4:14 AM

@Howard-Vigorita @Kinvermark “My timings are correct as posted. I think the disconnect comes from the fact that when I listed the Mainconcept encoder it got changed to read CPU in the spreadsheet.”

My understanding is that historically we used to call the render options (Kinvermarks name tags) as Cpu, Nvenc, Qsv, Vce, Cuda. I think we should stay with that. “Mainconcept AVC” does pop up as encoder option but I have been using CPU in place of, as I understand thats what it is, then either with or without HW Acc.

I'm pretty sure that to satisfy @Howard-Vigorita's needs and maybe others who want to test likewise, we need, not an extra render name i.e. MC, but an extra column named HW Acceleration. These would be say ... Nvidia, Intel, None.

I knew about this when initially laying out the suggested user data input text file. What I assumed was that for most users, they would be most interested in the best HW Acc. combination with the best render options to display the best render times and playback rates. After all I felt the focus was to see how well a particular set of hardware performed. Given we all know that wouldn’t necessarily always produce the very best render quality. But the test was never set up to check render quality.

I was also minded to keep the number of columns as few as possible, simply to fit width wise, and use the “comments” column as a catch all. The other consideration I had was that if the users data items became too onerous they might be less inclined to participate KISS. So a balance has to be struck.

For example quite a few users also add in, I put it in comments, if not already designated, the iGPU details.

Its also reasonable, and maybe what users want, to add a column for iGPU. Frankly I would add any extra columns in this order of preference .. iGPU, HW Acc. This will mean that less of the comments column will be visible in the 2 screen jpg's I post after each user update, or the view size becomes very small.

If we do now add these two columns, no problem for me, the HW Acc. one won’t be very well populated unless users go back and do more testing, or repost with updated details. I can easily populate the iGPU column by using what info is already given plus looking up CPU specifications.

In the meantime let me know what your thoughts are on all of this.

 

 

 

Former user wrote on 7/23/2019, 4:34 AM

@Howard-Vigorita Mainconcept MC is as I understand it the name of a third party company whose render codecs were bought in by Sony previously, and now used by Magix. When users used it without HW Acc. we just reported it mostly as Cpu. So MC or Cpu entries in encode mode mean the same.

This is a typical render benchmarking post, from @TheRhino in this case ...

https://www.vegascreativesoftware.info/us/forum/threadripper-1950x-rtx-2080-poor-render-time-on-vegas-pro-plugins--115499/?page=4#ca716778

Results from RED CAR Test after $1,350 USD of upgrades to an aging Xeon...

($500 9900K currently @ 4.9ghz, $200 open-box Asus Z390 WS, $350 refurbished, but like-new liquid cooled Vega64, $160 32gb DDR4, $140 H150i cooler)

100% QUIET. CPU & GPU both have radiators w/quiet fans...
INTEL 9900K = modest OC to 4.9 all cores. It does 5.0 but radiator fans spin-up...
AMD VEGA 64 Liquid Cooled @ Stock Speeds. Can't hear radiator fan...

Render with VCE = 0:33s …...............................Hevc VCE = 0:34s

Render with QSV = 0:17s ………………………..Hevc QSV = 0:23s
(6/13/2019 = 14s @ 5.0 ghz, 13s @ 5.1 ghz)

Render with Cpu =  0:52s ......………………...Hevc HDR10 = 0:45s

Interestingly, before I installed the Vega 64, I ran the CPU & QSV test & only got 0:44. According to Task Manager the AMD GPU is still doing some work when QSV is selected... Entire system pulls 300W during QSV, 260W during VCE, and 360W during CPU-only with CPU pegged at 100%. Task Manger shows 80% CPU & 28% GPU useage during QSV, 42% CPU & 52% GPU useage during VCE, and 100% during CPU...

Former user wrote on 7/23/2019, 6:23 AM

This from Peter_P ... https://www.vegascreativesoftware.info/us/forum/intel-i9-9900k-processor-review-using-vegas-pro--113429/?page=2

Red Car 1080p

HW Acceleration = Nvidia

Render with Cpu only ...  0:59s ......... [4790K ... 2:11s]

HW Acceleration = AMD R7 on i7-8700k Vp15 B416

Render with Cpu only ...  0:59s ......... [4790K ... 2:11s] /  [8700k … 1:15 / 88W]
So the i7-8700k takes 27% longer with CPU only 1080p30 AVC rendering.

Here the power consuption (~88W) and CPU ( 55°C) / core #1 (~66°C)

Former user wrote on 7/23/2019, 7:54 AM

Ok quicker doing it than talking about it. I've added 3 extra columns, iGPU, HW Acc. and split the VP version colum into two columns, VP version number and build number. If anyone wants to include hevc rendering let me know and i’ll add it in also, always better to say earlier than later. Ok, did it anyway🤪.

I've added in the iGPU's as best I know, let me know if I got any wrong.

If you want to post again, say just the HW Acc. you used while testing then do so, but modify your original post and insert the HW Acc. item, make a new post separately so that I pick it up, without data, i’ll go back and find your modified post. If you don’t want to go back updating HW Acc. etc then that's fine also.

If you want to post a lot more info then ideally modify your first user data post only, thanks.

Former user wrote on 7/23/2019, 8:28 AM

If any users have substantial changes to add, do it the way i’ve mentioned above. Two reasons why I only want to refer to one data post per user ... 1) it’s easier for me. 2) Its easier for everyone else to refer to one post if theres a query, say the ss data was incorrectly entered, thanks.

If your 1st post is a bit of a mess and you need to make substantial changes then remove it all, delete it and just repost everything again.

Another suggestion is to mark in bold and underline any new additions, modifications since your 1st. user data post.

petecarney wrote on 7/23/2019, 9:27 AM

Where are you guys getting an option to render mainconcept? I only see that option as mpeg-1 and mpeg-2, none of which allow for 4K of any kind. My VP16 doesn't even use gpu acceleration for MC MPEG-2 much less VCE on my Radeon VII. Just curious as the where that is coming from.

Cheers, Pete

Former user wrote on 7/23/2019, 9:34 AM

@petecarney Hi Pete, see screenshot ...

Toggling the Encode mode will give NV Encoder or Mainconcept AVC.

petecarney wrote on 7/23/2019, 10:15 AM

Thanks @Former user, hadn't seen that. Quite a bit slower at 2:16 vs 1:49 with AMD VCE

Former user wrote on 7/23/2019, 11:03 AM

Thanks @Former user, hadn't seen that. Quite a bit slower at 2:16 vs 1:49 with AMD VCE

@petecarney Yes, not that I claim to know much of it, but apparently the HW encodes use an “asic” chip on the GPU when encoding, so really fast all right. Why don’t you do a FHD render also using Nvenc, now that should be a lot faster, thanks for participating.

Former user wrote on 7/23/2019, 11:06 AM

@Howard-Vigorita  “I'll let you know when I've edited my 2 previous posts to add your new columns and remove now redundant comment column info.”

Looking forward to that Howard. Do point me to only one post to update from though, see my previous comments.

fr0sty wrote on 7/23/2019, 1:32 PM
My understanding is that historically we used to call the render options (Kinvermarks name tags) as Cpu, Nvenc, Qsv, Vce, Cuda. I think we should stay with that. “Mainconcept AVC” does pop up as encoder option but I have been using CPU in place of, as I understand thats what it is, then either with or without HW Acc.

That explains the confusion because historically, the Mainconcept encoder was only available in the Mainconcept "Render As" template groups and none of them were gpu enabled. But that's no longer the case with templates in the Magix AVC implementation. Every one of the encoder methods seem to be gpu enabled. Assuming you have the required gpu and it's enabled. Good move to put in more columns and match them to template terminology. I think its best to do the CPU-only timings with the Mainconcept encoder method because the other methods seem to disappear from the menu if you don't have or disable the associated gpu in video settings. Not sure if the "Use legacy gpu" general setting also impacts Magix AVC gpu usage... I keep it checked because I still use the Sony AVC for YouTube but might switch to Magix AVC/MainConcept based on what I've learned running your benchmark. I'll let you know when I've edited my 2 previous posts to add your new columns and remove now redundant comment column info.

Again, this is not true. Not all of the templates use the GPU to ENCODE. All of them use the GPU for timeline rendering unless you have turned off GPU acceleration in your video preferences, and that is why you will see GPU usage when encoding to a CPU only format. The GPU is helping with tasks like applying effects on the timeline, but when it comes time to take that rendered frame and encode it to a format, that is being done solely by the CPU on those formats.

Last changed by fr0sty on 7/23/2019, 1:34 PM, changed a total of 1 times.

Systems:

Desktop

AMD Ryzen 7 1800x 8 core 16 thread at stock speed

64GB 3000mhz DDR4

Geforce RTX 3090

Windows 10

Laptop:

ASUS Zenbook Pro Duo 32GB (9980HK CPU, RTX 2060 GPU, dual 4K touch screens, main one OLED HDR)

fr0sty wrote on 7/23/2019, 5:32 PM

You just said exactly what I said... Rendering uses the GPU if you have timeline GPU acceleration enabled. Once the frame is rendered, the encoder then uses the CPU, if you have not selected to use a NVENC/Quicksync/VCE encoder.

Systems:

Desktop

AMD Ryzen 7 1800x 8 core 16 thread at stock speed

64GB 3000mhz DDR4

Geforce RTX 3090

Windows 10

Laptop:

ASUS Zenbook Pro Duo 32GB (9980HK CPU, RTX 2060 GPU, dual 4K touch screens, main one OLED HDR)

Former user wrote on 7/23/2019, 6:23 PM

@Howard-Vigorita No hurry at all. That was a seriously easy way for me to get your data in my ss, easy peasy😀.

The only change I see that I need to do is change a few, 6, of the HW Acc. values from CPU to Off.

Former user wrote on 7/23/2019, 6:44 PM

@Howard-Vigorita I’ve updated all. But I just noticed that theres an “Off” value, 8, for GPU. I’ll change that to the Radeon 7, for i9 and to RX-580+ for the xeon.

There needs to be some value in that field, since you do physically have a GPU. It’s another matter whether it is in use or not, decided by the HW Acc. entry value.

I'll post you again when updated with changes shortly.

 

Former user wrote on 7/23/2019, 7:04 PM

@Howard-Vigorita Is updated now. Using the SS made my life a lot easier for that!

Grazie wrote on 7/24/2019, 3:06 AM

@Howard-Vigorita - Hi there! Is Turbo Boost a form of Over-Clocking? If it is, then why am I reading “NO” under OC for your top three?

@Howard-Vigorita, I’m always willing to learn and even try adding some TurboBoost to my rig! 😏

Former user wrote on 7/24/2019, 3:49 AM

@Grazie Not what is generally regarded as overclocking, as I understand it.

Former user wrote on 7/24/2019, 4:49 AM

If anyone wishes to do test renders to hevc also, in the interests of keeping things as simple as possible i’d suggest using the existing templates on my first post in this thread and use the same data rates etc, just change the “Render As” to Magix Hevc while selecting the render template.

When submitting user data enter Magix Hevc instead of Magix Avc for the “Render As” field.

I know that hevc needs less data rate for the same quality but these data rates are already low, but my main reason for doing it this way is to avoid users having too many complications with two more templates.

I've added two new columns, Cpu Manufacturer and Gpu Manufacturer, no need to add these to the user data input information as these are auto generated within the spreadsheet.