Polaris vs Pascal for Video Editing/Adobe Premiere Pro CC

[2blzd]

It's been pretty well established for many years now that more cuda cores = better performance in Adobe Premiere for gpu accelerated video editing. The mercury playback engine that Premiere is based around was developed with CUDA in mind. Adobe eventually added OpenCL support for GPU acceleration and AMD Cards did well with this but still everyone in the community recommended top-end Nvidia cards.

LinusTechTips did a video this past spring that pretty much debunked this and showed AMD and Nvidia on par with each other with everything from a 970, R 290 to a Fury and 980ti.

https://www.youtube.com/watch?v=g7cQK8jFPzo

Getting to the point I was pretty set on getting a 1070 for the price/performance and amount of cuda cores and vram. But since I saw Linus's video, I'm starting to wonder if a RX 480 for $200ish less would be a better buy?

I'm not a heavy gamer. Thoughts?

 

[thesmokingman]

The creative arts industries are moving away from proprietary and flocking to open source compute, something to think about.

 

[kraatus77]

polaris has ocl 2.0 vs pascal's ocl 1.2/3. obviously polaris is a better choice.

 

[Stuka87]

AMD typically has better compute, and OpenCL support is MUCH better on AMD cards.

But if you want to game a lot, the 1070 could be a better choice if you dont mind spending twice as much.

 

[ShintaiDK]

polaris has ocl 2.0 vs pascal's ocl 1.2/3. obviously polaris is a better choice.

Is it? Show me the benches for the above application. Unless you recommend something without knowing the numbers. And remember the application in question got cuda support.

https://www.pugetsystems.com/labs/articles/GTX-1070-and-GTX-1080-Premiere-Pro-Performance-810/

We already know Polaris is a terrible miner card due to low compute performance.
http://cryptomining-*********/tag/rx-480-hashrate/

Note something? Yep, lack of memory bandwidth is making it hard for both.

 

[kraatus77]

Is it? Show me the benches for the above application. Unless you recommend something without knowing the numbers. And remember the application in question got cuda support.

https://www.pugetsystems.com/labs/articles/GTX-1070-and-GTX-1080-Premiere-Pro-Performance-810/

We already know Polaris is a terrible miner card due to low compute performance.
http://cryptomining-*********/tag/rx-480-hashrate/

ocl 2 > ocl 1.2. mining performance has nothing to do with ocl performance. and amd has always performed better on ocl. cuda is irrelevant because ocl is available.

your benchmark already shows almost no difference between 1080-980ti. and linus's video shwos same thing for 980ti- fury/ 970-290. so i don't see the point spending 2x amount for 10 seconds difference. lol

i'll show you benchmarks on 29th. if it performs bad he can buy 1070, it's not like he can buy rx480 now. does it ? so please stop with your personal attacks ().

 

[Bacon1]

Is it? Show me the benches for the above application. Unless you recommend something without knowing the numbers. And remember the application in question got cuda support.

https://www.pugetsystems.com/labs/articles/GTX-1070-and-GTX-1080-Premiere-Pro-Performance-810/

The GTX 1070 is about 1% faster than the GTX 980Ti

The GTX 1080 is about 4% faster than the GTX 980Ti

 

[SimianR]

Weird that 25MH's is considered "terrible mining performance" for the RX 480 considering thats the same performance of a stock 390 but using considerably less power. For reference a 1080 mines at about 25 Mh's as well.

 

[ShintaiDK]

ocl 2 > ocl 1.2. mining performance has nothing to do with ocl performance. and amd has always performed better on ocl. cuda is irrelevant because ocl is available.

Cuda is relevant since that's what the majority of the users are using for this application.

And you didn't prove anything was faster due to a possible version difference if it is even supported. Its like claiming a card that supports a newer Cuda version is automatically faster. It isn't.

You claimed Polaris is the better choice, you just didn't back it up with anything.

The GTX 1070 is about 1% faster than the GTX 980Ti

The GTX 1080 is about 4% faster than the GTX 980Ti

Exactly, and memory bandwidth is the reason.

 

[Bacon1]

Cuda is relevant since that's what the majority of the users are using for this application.

And you didn't prove anything was faster due to a possible version difference if it is even supported. Its like claiming a card that supports a newer Cuda version is automatically faster. It isn't.

You claimed Polaris is the better choice, you just didn't back it up with anything.



Exactly, and memory bandwidth is the reason.

https://www.youtube.com/watch?v=g7cQK8jFPzo

290 is already faster than 980 TI in most tests, thus would be faster than 1080 as well.

 

[Stuka87]

Is it? Show me the benches for the above application. Unless you recommend something without knowing the numbers. And remember the application in question got cuda support.

All GCN cards support OpenCL 2.0.

 

[kraatus77]

Cuda is relevant since that's what the majority of the users are using for this application.

And you didn't prove anything was faster due to a possible version difference if it is even supported. Its like claiming a card that supports a newer Cuda version is automatically faster. It isn't.

You claimed Polaris is the better choice, you just didn't back it up with anything.



Exactly, and memory bandwidth is the reason.

it's better choice based on amd's performance with ocl ( look at history lol) . watch linus's video. even r9285 matches a 980ti lol. anyway. he can't buy rx480 so doesn't really matter what i say does it ? better version only means better future capabilities. and more performance.

 

[ShintaiDK]

https://www.youtube.com/watch?v=g7cQK8jFPzo

290 is already faster than 980 TI in most tests, thus would be faster than 1080 as well.

In most?

You mean the 3 tests they made? 1 draw, one each way? Not to mention the 390X somehow lost the one that went the AMD way.

 

[ShintaiDK]

it's better choice based on amd's performance with ocl ( look at history lol) . watch linus's video. even r9285 matches a 980ti lol. anyway. he can't buy rx480 so doesn't really matter what i say does it ? better version only means better future capabilities. and more performance.

So does the GTX 970 in the video. GTX970 faster than 390X while the Fury seems to failed. That's what happens when the bottleneck is elsewhere and cherry pick

But again, if you dont have numbers. How can you recommend something?

 

[kraatus77]

so basically all perform almost same, since rx480 have lower tdp and he doesn't play games much, rx480 is a better choice for almost half price. thanks for proving my point.

 

[ShintaiDK]

Getting to the point I was pretty set on getting a 1070 for the price/performance and amount of cuda cores and vram. But since I saw Linus's video, I'm starting to wonder if a RX 480 for $200ish less would be a better buy?

I'm not a heavy gamer. Thoughts?

You should look on the video decode capabilities as well of the decode block. You may lack decode on some cards for certain formats. GTX 1060 is coming soon as well.

 

[ShintaiDK]

so basically all perform almost same, since rx480 have lower tdp and he doesn't play games much, rx480 is a better choice for almost half price. thanks for proving my point.

RX 480 and 1070 got the same TDP.

 

[kraatus77]

RX 480 and 1070 got the same TDP.

i meant actual power consumption. 480 will have lower than 150w. a gpuz ss already shows 120w something.

 

[ShintaiDK]

i meant actual power consumption. 480 will have lower than 150w. a gpuz ss already shows 120w something.

You should look on the load part. We still have to see it tested and fully loaded.

 

[kraatus77]

You should look on the load part.

sigh, it's 112w @ 94% load, so 120w @ 100% load, let's add another 20w still less than what 1070 consumes. :thumbsup:

leave it man, he can't buy it so doesn't really matter. i'm out

 

[2blzd]

Kraatus77 -

You need to educate yourself a bit better on the history of CUDA/Premiere Pro/Adobe in general and maybe perhaps read my entire OP. CUDA is very relevant to this topic, my topic, as Premiere Pro is heavily accelerated by CUDA. IF you're just stating AMD cards are better because of OCL 2.0, you're missing the point entirely. Also, this has nothing to do with Mining. General compute performance does not translate/scale perfectly across the board for every app, especially when Premiere is as heavily tuned for CUDA as it is.

By no means I am saying you're wrong, but just to dismiss something based on not having the newest spec is silly. We all have the proof and results in front of us from Linus's video but he only used 3 examples/benchmarks and they were all exporting/encoding. Most of the speedup benefit you get with GPU acceleration in Premiere Pro is from not having to pre-render everything and having a fluid and smooth scrolling timeline playback while you're applying multiple effects. I don't think these are things you can easily benchmark.

I think Linus's tests were a very limited sample of what you actually do within Adobe Premiere, which is more than just exporting. Most of the higher end visual effects are 32bit and cuda accelerated, which he completely ignores in this video. I was hoping to get more insight beyond what's out there.

 

[sirmo]

no brainier rx480

 

[Stuka87]

RX 480 and 1070 got the same TDP.

They most definitely do not the same TDP.

AMD rates the TDP on their cards as ACTUAL TDP. So if the RX 480 is a 150W card, thats peak power. nVidia rates their TDP as average gaming power consumption, not peak power consumption.

TPU shows the 1070 as having 151W peak gaming power consumption.

The RX480 is listed on TPU as a 120W part (technically could be 150W peak since it has a 6pin connection).

If they were truly the same, the 1070 would have a single 6pin, instead of a single 8 pin.

 

[Bacon1]

Kraatus77 -

You need to educate yourself a bit better or maybe perhaps read my entire OP. CUDA is very relevant to this topic, my topic, as Premiere Pro is heavily accelerated by CUDA. You're just stating AMD cards are better because of OCL 2.0, you're missing the point entirely. Also, this has nothing to do with Mining. General compute performance does not mean across the board for every app, especially when Premiere is as heavily tuned for CUDA as it is.

??

The video you linked showed AMD already on par with Nvidia's greatest. So why are you bringing up CUDA as something special for Premiere Pro?

 

[2blzd]

??

The video you linked showed AMD already on par with Nvidia's greatest. So why are you bringing up CUDA as something special for Premiere Pro?

his benchmarks were not using any cuda accelerated features in premiere. read my entire reply. There is more to Adobe Premiere than just exporting videos. You actually have to edit at some point and use FX. Said FX are GPU accelerated.

Also that was a direct response to his statement of AMD > Nvidia because OCL 2 > OCL 1.2