# GPU decoding | CUDA

**URL:** <https://casparcgforum.org/t/gpu-decoding-cuda/5424>\
**Category:** General\
**Created:** [February 13, 2023, 8:11pm UTC](https://casparcgforum.org/t/gpu-decoding-cuda/5424 "2023-02-13T20:11:38Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Koipse](https://casparcgforum.org/letter_avatar_proxy/v4/letter/k/85e7bf/32.png) [@Koipse](https://casparcgforum.org/u/Koipse)\
**Post date:** [February 13, 2023, 8:11pm UTC](https://casparcgforum.org/t/gpu-decoding-cuda/5424/1 "2023-02-13T20:11:39Z")

</div>

Hi!

First, i want to express my sincerest gratitude for that wonderful software! 🙂

About question: just want to know, would it be possible to decode the stream via gpu?

I use Quadro P1000, CUDA 11.7, Driver 515.65.01 and in Caspar configuration use next parameters in «args»:

```auto
-format mpegts 
-codec:v h264_nvenc 
-filter:v format=pix_fmts=yuv420p,fps=50,interlace -flags:v +ilme+ildct 
-mpegts_service_type advanced_codec_digital_hdtv
-mpegts_original_network_id 0x1122
-mpegts_transport_stream_id 0x3344
-mpegts_service_id 0x5566
-mpegts_pmt_start_pid 0x1500
-mpegts_start_pid 0x150
-flush_packets 0 
-pcr_period 20 -pat_period 0.10 -sdt_period 0.25 
-fflags +genpts -mpegts_flags +system_b 
-rc:v cbr_ld_hq 
-muxrate 11.5M -max_delay 2M -b:v 10.5M -minrate:v 10.5M 
-maxrate:v 10.5M -bufsize:v 2M 
-g:v 50 
-codec:a aac -b:a 192k -filter:a pan=stereo|c0=c0|c1=c1</args>

```

And in path i use next settings:

`<path>udp://some_mcast?pkt_size=1316&localaddr=some_ip&bitrate=11000000&buffer_size=3400000&fifo_size=500000&burst_bits=100000</path>`

In result, see that Casparcg use GPU but only for encoding:

enc dec  
% %  
28 0  
29 0  
29 0  
30 0

Thank you very much!

Peace 🙂

---

<div class="post-metadata">

**Author:** ![Sidon](https://casparcgforum.org/user_avatar/casparcgforum.org/sidon/32/1299_2.png) [@Sidon](https://casparcgforum.org/u/Sidon)\
**Post date:** [March 6, 2025, 3:15pm UTC](https://casparcgforum.org/t/gpu-decoding-cuda/5424/2 "2025-03-06T15:15:10Z")

</div>

I find myself asking the same question.

In my case I’m trying to decode a HEVC TCP stream at 4K and 8K, but since the codec is very GPU oriented I cannot get a smooth decoding only with CPU.

I tried adding the ‘nvenc’ flags but I didn’t see any difference in GPU use.

Shouldn’t there be an actual improvement when adding “-codec:v hevc\_nvenc” ?

I am trying very wild values for buffer and probe sizes just to see if I can get a proper stream, this string probably doesn’t make any sense, I’m still going through the documentation about all this.

> PLAY 1-10 “tcp://10.32.15.2:8888?gop-size=30buffer\_size=150M” -s 7680x4320 -r 25 -tune:v zerolatency -preset:v ultrafast -crf:v 25 -analyzeduration:v 150M -probesize:v 150M -codec:v hevc\_nvenc -format:v mpegts