Anyone running ThreadRipper with Centos VMs running FFMPEG?

My main am3+ box had a failed corsair aol pump cooler and roasted the cpu. Replaced the cooler, but the machine still blue Screens when I prime 95 it on cores 7 and 8. Whoops. I just disabled those cores in the bios for now and its stable.

Got enough pennies to build a new box.. Looking at this CPU at the moment:

https://www.newegg.com/amd-ryzen-threadripper-2950x/p/N82E16819113542

At the moment, im running windows with Vm Player. Are you running centos, and ffmpeg in VMs on this or any other thread ripper cpu? If so, are you rocking the default ffmpeg from the centos repos, or did you have to recompile things?

I did search, but it was mostly people talking about benchmarks.. I dont care about benchmarks.. my am3 box was fast enough for what I was doing, so I know this one will be fine.. Im more concerned about having to recompile my libs, vs just rocking whats in the repos.

Well, I can't answer your question regardless, but the version of centos you're using is relevant here. The library versions will be different (at least for core libraries, but possibly program libraries too, and it may affect performance regardless) on centos 7 from those on centos 8, for instance (both still receive updates, they just target different stable library versions).

Nobu said:

Well, I can't answer your question regardless, but the version of centos you're using is relevant here. The library versions will be different (at least for core libraries, but possibly program libraries too, and it may affect performance regardless) on centos 7 from those on centos 8, for instance (both still receive updates, they just target different stable library versions).

Click to expand...

Thank you. I had a sneaking feeling I was going to end up in compile hell. I have an automated work flow that I wrote that will recompile ffmpeg for me from source and poop out what i need into separate dirs. so its not super painful. I was just hoping to pick up my libs and bins and just move them over and call it a day. You can see what I'm doing in my production environment with FFMPEG if you clink the link in my sig.

Have you done any ffmpeg stuff in containers?

kdh said:

Thank you. I had a sneaking feeling I was going to end up in compile hell. I have an automated work flow that I wrote that will recompile ffmpeg for me from source and poop out what i need into separate dirs. so its not super painful. I was just hoping to pick up my libs and bins and just move them over and call it a day. You can see what I'm doing in my production environment with FFMPEG if you clink the link in my sig.

Have you done any ffmpeg stuff in containers?

Click to expand...

Nah, only used ffmpeg on a personal compy, and I'm just familiar with Linux as a hobby—why I can't really answer your question. If performance isn't a concern, and the version in the repo has all the features you need, I see no reason why not to use it. As far as compiling, I can't really help you there. I always ended up fubar on rpm distros.

I had this discussion with someone that works onsite movie shoots years ago.

My argument was to write a scheduler to ship workflow to AWS and use Spot for anything that wasn't needed in real time.

3-cents an hour per instance, and scale out to whatever is available. Ironic but where on Azure now.

In the OPs case I'd use KVM and benchmark my workflow to see what the optimal core to RAM to io ratio would be. That's been a standard when sizing datacenter compute forever. K8's can be an issue when dealing with state, we run 10million to 100million objects in flight.

Do you exceed 2 nvme drives, 64gb ram, say 12 cores as a base starting build?

What resource do you need vs scaling the workload?

Does your final output need to go anywhere besides your local machine? That was the biggie with my buddy, bc their output was viewed off-site.

Super awesome information. I'm not doing post process, but real time encoding for live broadcasting. I've been able to crush my live streaming vms on my AM3+ box down to 2gbs of ram, 2cpus, and run a 1280x720 display, with a 512k v bit rate, and a 256k a bit rate. Disk performance isn't a major issue(yet) because my VMs are not storing anything on disk, but receiving my live feeds, doing "something" with them, and more or less kinda rerouting them to the right place in real time. My ultimate goal for scaling is to move to a cloud provider so I only pay for what I need when I use it. At the moment, I'm still trying to work out my work flow for various bits of what I'm doing on my own hardware as a proof of concept. Once i think I have a working model down.. cloud baby.. cloud. I actually don't care which one I go to.. I'll figure that out once I have a working proof of concept in my house. My step 1 current production environment is still using physical servers rented in a colo.. I'm working towards going 100% cloud. Using the CPU above, and the 2 other am3+ boxes I have in my house will give me enough resources in my house(I think/hope) to give me enough cloud like resources to build a super rough cloud application. once I settle on my cloud provider or choice, I can bend what I've done to work w that cloud. My thought process anyway.