This post has been de-listed
It is no longer included in search results and normal feeds (front page, hot posts, subreddit posts, etc). It remains visible only via the author's post history.
Hi guys! I’m looking for a decent build to serve as a PMS. I did a lot of research lately, like really a lot. So I know everything about the IGPU and QS etc. Because I couldn’t find a decent answer/benchmark online I decided to find it out for myself. I have a (old) Lenovo laptop which has a I7-8550U and a UHD 620 chip. I did a little stress test on this thing because I’m aiming for a iris plus 655 chip or a UHD 630 (8,9 or 10generation) The results;
https://imgur.com/gallery/9O4lUk7
I managed to HW transcode 9 streams to 1080 10mbs. While the CPU was around 10-20% with 60c degrees. When i tried to play the 10th stream I noticed that my RAM was spiking around 90% (from the 8GB in total) ((yes I transcode with RAM)) So when the 10th stream started he was completely buffering. All the other streams still managed to work properly. The HDD where all the files where located had a usage of 8%. The files are located in a different fractal design file server system (own build) so not directly with the UHD620. I think when I upgrade the ram to 16GB or even 32GB the UHD620 can manage 15 1080 10mbs easy!
I do have a questions regarding this test; - are 265x more difficult to transcode than my original test? Because maybe in the future im going to convert everything to 265x depends on what technology will do.. right now I don’t see directly the benefits of it, and it’s not like I really need extra space (TB) I have enough room to shuck some WD.
Any thoughts about this ‘test’? It was my first time so, be easy on me! If someone can recommend me a great PMS build which has better results than this, please be my guess. I need at least 10 transcodes at the same time (got around 16-18 users at my DS918 , poor J3445..)
Cheeeeers!
Subreddit
Post Details
- Posted
- 4 years ago
- Reddit URL
- View post on reddit.com
- External URL
- reddit.com/r/PleX/commen...