Wowza Community

[Performance benchmark] Which should be chosen: Quadro K4000 or Tesla C2070

Hi guys,

I am a little bit of confused to choose between K4000 or Tesla C2070. I saw the benchmark of Wowza here in which Tesla is the most powerful Graphics card, but it was long time ago. So I have some questions, hope you can give me some advice:

  1. Do both of them support NVENC, I know that surely K4000 supports NVENC, but I don’t know about Tesla branch.

  2. Do you have a performance test (if it has the same with Wowza benchmark, it would be great) with K4000 card. Because I’m not really sure about the performance between K4000 and Tesla. My objective is to receive about 40 4CIF incoming feeds, transcode them into 4CIF and 1CIF (with watermark) streams for each feed.

  3. For transcoding, which one is better to use, NVENC or CUDA ?

Thanks for your help guys, I appreciate all your comments,

Regards,

Anh-Dzung

Hi,

  1. The K4000 supports both NVENC and CUDA for accelerated transcoding.

The Tesla C2070 supports only CUDA.

  1. We currently do not have any transcoding performance benchmarks for the NVENC cards.

We feel that the performance of NVENC will be better than CUDA, but not a huge improvement.

So the CUDA based benchmarks can be used as a rough comparison when considering NVENC.

However, 40 incoming streams for transcoding is very high, and decoding of the video

is still handled by the CPU, and not the cards. You would likely need several transcode servers to

handle that load.

  1. NVENC is the newer technology and provides improved performance over CUDA.

NVENC can also be run under Linux as well as Windows. Currently CUDA is windows only

for Wowza transcoding.

Daren

Hi,

For transcoding purposes at least, you should keep the CPU utilization at or below 60%

Once the utilization starts to go high frames start to drop (skip) and the transcended streams become

less watchable.

Daren

Hi Daren,

Thanks for the informative response. That makes sense a lot.

AD

Hi Just need some more info in this regard. I want to know if Tesla K40 support nevenc technology or not. Is there a better card available than Tesla K40.

thanks

Hi,

  1. The K4000 supports both NVENC and CUDA for accelerated transcoding.

The Tesla C2070 supports only CUDA.

  1. We currently do not have any transcoding performance benchmarks for the NVENC cards.

We feel that the performance of NVENC will be better than CUDA, but not a huge improvement.

So the CUDA based benchmarks can be used as a rough comparison when considering NVENC.

However, 40 incoming streams for transcoding is very high, and decoding of the video

is still handled by the CPU, and not the cards. You would likely need several transcode servers to

handle that load.

  1. NVENC is the newer technology and provides improved performance over CUDA.

NVENC can also be run under Linux as well as Windows. Currently CUDA is windows only

for Wowza transcoding.

Daren

Hi Daren, many thanks for your response,

Yeah 40 incoming streams is a large number, but it’s mandatory requirements. I don’t know if it is possible to use more CPU load for transcoding more incoming streams, if my server is only installed Wowza streaming engine for receiving streams, transcode them and stream to clients needing them. Because my suspicious point was that streaming part is not really a CPU-consuming work. Is it possible to let 80-90% of CPU workload for just transcoding ?

Anyway thank you and have a nice week-end.

Anh-Dzung

Hi,

For transcoding purposes at least, you should keep the CPU utilization at or below 60%

Once the utilization starts to go high frames start to drop (skip) and the transcended streams become

less watchable.

Daren

Hi Daren,

Thanks for the informative response. That makes sense a lot.

AD