Question Polaris vs Vega HEVC encoding

[Zepp]

I have been using hardware encoding (AMD VCE x265) on handbrake lately to quickly size down large 1080p VODs saved from Twitch to put on a 720p tablet for a long plane ride.

My WX4100 is a 14nm polaris chip with 1024 cores
I also have a 5600U laptop with Vega 7 448 cores

for whatever reason I thought to compare the encode speed between them and was shocked that the little vega is more than twice as fast.

searching around on google I never found anything to suggest there was such a large improvement in vega encoding over polaris

is this entirely expected or is there something I'm missing.

 

[Tech Junky]

I don't recall which is newer or more efficient but I'm using an a380 card that flies through converting things. Depending on the input it hits over 1000 FPS typical 1080 though usually is in the 550 range. The advantage is using Intel's QSV codec.

 

[Hitman928]

I have been using hardware encoding (AMD VCE x265) on handbrake lately to quickly size down large 1080p VODs saved from Twitch to put on a 720p tablet for a long plane ride.

My WX4100 is a 14nm polaris chip with 1024 cores
I also have a 5600U laptop with Vega 7 448 cores

for whatever reason I thought to compare the encode speed between them and was shocked that the little vega is more than twice as fast.

searching around on google I never found anything to suggest there was such a large improvement in vega encoding over polaris

is this entirely expected or is there something I'm missing.

The video encoding/decoding is handled by a dedicated block outside the GPU and so the encoding quality and speed has nothing to do with how big the rest of the GPU is. The encoder block inside the 5600u is multiple years newer than in the WX4100, so it is not surprising that it is so much faster and probably higher quality output as well.

 

[biostud]

The video encoding/decoding is handled by a dedicated block outside the GPU and so the encoding quality and speed has nothing to do with how big the rest of the GPU is. The encoder block inside the 5600u is multiple years newer than in the WX4100, so it is not surprising that it is so much faster and probably higher quality output as well.

This. Look at the multimedia engine.

 

[Zepp]

I see, I did not realize the GPU cores themselves had nothing to do with encoding.

Still surprises me the Vega is that much faster

Also the Polaris card would show 70-80% utilization while the little vega was something like 30%

 

[Tech Junky]

Well, as a comparison...

7900X CPU = 8-10 cores
A380 unoptimized ~8 cores
A380 optimized ~2-3 cores

The cores on the CPU engaged while converting media files adds up in electricity costs and fans blowing faster while in use. Since I'm using Linux there's no quick way to see the GPU % but, I expect it to be about 100% which factors into picking a low tdp GPU at ~35w. If we take the 170w tdp of the CPU 170/12 ~14W/core

CPU transcode / 112-140W
A380 - 112W
A380 QSV - 35W

Just rough estimate of W's used by the CPU. Taking the load off the CPU and pushing it to the capped GPU makes a difference but, not only the savings in W's per file the time is reduced to 1/8th of the CPU only. The $100 GPU pays for itself pretty easily in terms of the monthly bill but also time saved in waiting for files to be processed.

Each generation of GPU chip though exponentially increases the efficiency of de/encoding and rendering. However, the trick with the Arc's isn't impacted if you go with the $100 A380 or the $350 A750 as they all process video in the same manner. The only benefits of higher VRAM is better game play but not video processing.

 

[Mopetar]

The Polaris based card uses VCE 3.4, which wasn't all that great. There are even some old threads here talking about the lackluster performance. The encoder on Polaris was really just a tweaked version that had shipped back with some of the 200-series cards. It did add HEVC though, which makes it better than prior cards.

Meanwhile Vega had a newer VCE 4.0, which was AMD's second cut at HEVC hardware encoding, so it's hardly surprising that they made a lot of improvements over their initial inclusion.

I'm sure that the software you're using tries to make use of the GPU cores to speed up the encoding. The extent to which that is possible probably depends on the capabilities of the hardware encoder. It's not surprising that there would be more work for GPU cores to do on earlier versions of the hardware encoder.

 

[Zepp]

The Polaris based card uses VCE 3.4, which wasn't all that great. There are even some old threads here talking about the lackluster performance. The encoder on Polaris was really just a tweaked version that had shipped back with some of the 200-series cards. It did add HEVC though, which makes it better than prior cards.

Meanwhile Vega had a newer VCE 4.0, which was AMD's second cut at HEVC hardware encoding, so it's hardly surprising that they made a lot of improvements over their initial inclusion.

I'm sure that the software you're using tries to make use of the GPU cores to speed up the encoding. The extent to which that is possible probably depends on the capabilities of the hardware encoder. It's not surprising that there would be more work for GPU cores to do on earlier versions of the hardware encoder.

what about 7nm Vega vs 14nm Vega? should they perform similarly or does the clock speed bump on 7nm factor in to encoding?

 

[Leeea]

look at biostud's post,

it is not even using the Vega encoder, it is using a different one all together.


The 5600u is released in 2021, the encoder is entirely separate from the GPU. They did that for a reason. Vega GPU release 2017.


That 5600u is not using a 2017 encoder.


It is not apples and oranges, it is apples and legos.


If the 5600u was a traditional system, it would have a GPU video card, and a separate encoder/decoder card that just did encoding/decoding. Apples and legos. But instead of the PCI communication it is Infinity Fabric communication (vastly faster lower latency). Apples and legos.

 

[burninatortech4]

I actually have a RX 550 (E9176), an RX 560 (E9260), and a Vega 64. I also have a Vega 6 (Ryzen 3 5400U), Vega 7 (Ryzen 5 5600G), Vega 8 (Ryzen 7 5700G), Vega 11 (Ryzen 5 2400GE) and a 7800XT.

I think it would be cool to do a post where I compare them with the same res/bit rate/file type via handbrake encodes. Would anyone be interested in me running that?