Which GPU you have for video rendering doesn't matter as long as you have one

[Master Disaster]

Video source so sorry guys, nothing to quote.

 

Essentially Jay decided to try and see if Vegas huge compute potential would affect video rendering times, he wanted to know if Vega was a good choice for YouTubers to render videos with.

 

The test

 

For his tests he compared Vega 56 to GTX 1070 and Vega 64 to a GTX 1080, to account for current pricing he also threw in a V64 vs a GTX 1080ti though he made it clear that that would be an invalid test as soon as pricing returned to normal. All cards used were reference cards, no AIB custom used at all, no GPU overclocks at all but boost was enabled at defaults where available. He was encoding the the H.264 4K YouTube preset.

 

Testing methodology

 

Jay got a sample project for the latest version of Adobe Premiere (the eagle project) and rendered it on his test bench. The test bench is running a 7900X (at 4.6Ghz) and 32GB 3200Mhz Dominator Platinum. As a control he rendered the project on the CPU alone (no GPU at all) and then he tested each cards render times measuring the time in seconds.

 

The results (in seconds so slower is better)

 

CPU - 326s

1070 - 133s

1080 - 134s

1080ti - 133s

Vega 56 - 133s

Vega 64 - 133s

 

At this point Jay got a little confused so he decided to test a few more things, CPU alone with no HT, a 1050 and a 660, he also tested the 1080ti and the 660 with HT off.

 

CPU (No HT) - 467s

1050 - 133s

1080ti (No HT) - 214s

660 - 133s

660 (No HT) - 215s

 

 

The conclusion

 

As you can clearly see, which video card you have makes zero difference as long as you have one. The results were for all intents and purposes, identical across every GPU tested. In fact the results show your CPU still has more affect on render time than your GPU does even when rendering is GPU offloaded.

 

Source

 

Very interesting results, certainly not what I was expecting at all. I guess this kinda invalidates Linus's editing suite update video, he coulda given them all 1050s and production would have remained the same.

 

Thoughts? I'd be very interested in hearing any ideas for what's going on here, there's no way a 660 is being bottlenecked by a 7900X, I wonder if its some kind of software throttling in play by Premiere?

[suicidalfranco]

133 seconds? lol, should have done it with FCP

[LePawel]

This just means that gpu rendering is bottlenecked by another component. 

He should find which and do more tests based with different classes of that bottlenecked component.

[ARikozuM]

I thought we knew this already?

 

Also, you can still quote JayzTwoCents (proper title) by transcribing his words.

[Master Disaster]

2 minutes ago, suicidalfranco said:

133 seconds? lol, should have done it with FCP

That was a YouTube 4K render, anyway the time itself isn't what's important, its the time difference between cards.

[Jurrunio]

Just now, LePawel said:

This just means that gpu rendering is bottlenecked by another component. 

He should find which and do more tests based with different classes of that bottlenecked component.

If a 7900x overclocked to 4.6GHz still bottleneck a gtx 660, then it's pointless to do more testing because people with this level of CPU power certainly has more powerful parts than a 660.

 

7 minutes ago, Master Disaster said:

Very interesting results, certainly not what I was expecting at all. I guess this kinda invalidates Linus's editing suite update video, he coulda given them all 1050s and production would have remained the same.

 

Thoughts?

Linus gets a lot of his parts from sponsors and others for project builds, so I guess he didnt spend anything for the cards in production machines anyway.

[Master Disaster]

Just now, ARikozuM said:

I thought we knew this already?

 

Also, you can still quote JayzTwoCents (proper title) by transcribing his words.

Well I didn't and obviously neither did Jay otherwise he wouldn't have wasted his time running this test.

 

I shortened the name because the thread title was too long and no, when using video sources a summery is fine. No need for direct quotes.

[Master Disaster]

14 minutes ago, LePawel said:

This just means that gpu rendering is bottlenecked by another component. 

He should find which and do more tests based with different classes of that bottlenecked component.

He did, we went as low as a 660 and still saw no change.

 

I wonder if the bottleneck is Premiere itself? I wonder if modern GPUs have reached the point where Premiere is unable to fully saturate them with data to process?

[ARikozuM]

19 minutes ago, Master Disaster said:

Well I didn't and obviously neither did Jay otherwise he wouldn't have wasted his time running this test.

My first guess would be the CPU and second would be the I/O between disks and RAM.

 

[Damascus]

23 minutes ago, Jurrunio said:

If a 7900x overclocked to 4.6GHz still bottleneck a gtx 660, then it's pointless to do more testing because people with this level of CPU power certainly has more powerful parts than a 660.

 

Linus gets a lot of his parts from sponsors and others for project builds, so I guess he didnt spend anything for the cards in production machines anyway.

A 7900X at 4.6ghz will also bottleneck a 1080 ti by that logic 

This software gives no shits about how  powerful the GPU is as long as there is a GPU, Linus established as much in 2016

 

 

Edited September 1, 2017 by Damascus
2016 not 2013, though there may be an earlier video

[Master Disaster]

1 minute ago, Damascus said:

A 7900X at 4.6ghz will also bottleneck a 1080 ti by that logic 

This software gives no shits about how  powerful the GPU is as long as there is a GPU, Linus established as much in 2013 or so

Can you link that please, it was before my time here and I'm confused why he'd give all his editors Titans if he knew that a 660 would give the same level of performance.

[Damascus]

Just now, Master Disaster said:

Can you link that please, it was before my time here and I'm confused why he'd give all his editors Titans if he knew that a 660 would give the same level of performance.

Rendering may not change much but editing sure will.  Also, he doesn't pay for those systems so he don't give a Puuuuuuug

 

Refresh the page, I updated the initial response

[LePawel]

21 minutes ago, Master Disaster said:

He did, we went as low as a 660 and still saw no change.

 

I wonder if the bottleneck is Premiere itself? I wonder if modern GPUs have reached the point where Premiere is unable to fully saturate them with data to process?

23 minutes ago, Jurrunio said:

If a 7900x overclocked to 4.6GHz still bottleneck a gtx 660, then it's pointless to do more testing because people with this level of CPU power certainly has more powerful parts than a 660.

There's more to a pc than a gpu and a cpu. Imo it'd be interesting to see how they perform on a x4 slot, see if it's a bandwidth issue. 

[Jurrunio]

Just now, LePawel said:

There's more to a pc than a gpu and a cpu. Imo it'd be interesting to see how they perform on a x4 slot, see if it's a bandwidth issue. 

using a x4 slot is actually making bandwidth a possible issue.

[Master Disaster]

1 minute ago, Damascus said:

Rendering may not change much but editing sure will.  Also, he doesn't pay for those systems so he don't give a Puuuuuuug

 

Refresh the page, I updated the initial response

Ahh I gotcha and yeah that makes sense. Its more about smooth timeline scrubbing than it is about render times.

 

Didn't Linus get those Red co processor thingies to help with timeline scrubbing though? Is that because Red 8K footage?

[Damascus]

Just now, Master Disaster said:

Didn't Linus get those Red co processor thingies to help with timeline scrubbing though? Is that because Red 8K footage?

Yup, though he only got one because they are 7 grand a peice.

 

Honestly you will see a very minor difference between gaming GPUs from high to low end but pro level certified GPU's may give a serious bump

[Master Disaster]

6 minutes ago, huilun02 said:

And when you're not rendering in Adobe?

That would be an interesting follow up for sure, Jay said at the end of the video he's going to try and do a follow up with 3D Modelling as the project instead of Video rendering.

 

Its kinda hard to test FCP though, I mean how is he supposed to change GPU in a Mac without going for a 4 year old Mac Pro?

[cj09beira]

i am pretty sure the problem come from the software used being shitely optimized 

[LePawel]

21 minutes ago, Jurrunio said:

using a x4 slot is actually making bandwidth a possible issue.

Jesus Christ, that's the point. If all gpus show same perf. drop, you're looking at possible bottleneck even at full speeds. 

You're not really familiar with research and testing are you... 

[Prox1ma]

I think the tests must be done with other software to see if it is a software optimization issue. I most probably think it is.

[Sakkura]

1 hour ago, Jurrunio said:

If a 7900x overclocked to 4.6GHz still bottleneck a gtx 660, then it's pointless to do more testing because people with this level of CPU power certainly has more powerful parts than a 660.

A GTX 660 still has more GFLOPS than an overclocked 7900X, so maybe that's the problem.

[raphidy]

IIRC, only some effects from Adobe software are utilize by the GPU computing. If you use Sony Vegas before version 13? It can be CUDA accelerated with keplers cards and newer versions can only use opencl (for processing only and not rendering) and AMD would have an edge in Vegas nowadays.

[Jurrunio]

33 minutes ago, LePawel said:

Jesus Christ, that's the point. If all gpus show same perf. drop, you're looking at possible bottleneck even at full speeds. 

You're not really familiar with research and testing are you... 

So you suspect PCIe bandwidth to be a limitation on cards using PCIe 3.0 x16 and you suggest using PCIe 3.0 x4 to make this possible bottleneck even more serious? Seriously how do you do research with logic like that.

[DocSwag]

There's a few things with this that you have to take into account 

-This was done in premier. If you used something else (FCP, Sony Vegas, etc.) you could potentially see different results

-This was ONLY an encoding test. Jay didn't test ANYTHING else. So we have no idea what it could look like for other things.

[Mihle]

What about in Davinci Resolve?