Nimble
Member
So running dual-channel or double dual-channel is actually creating a bottleneck somewhere else, even though 3D usage is reported as 50% lower. When I go to pull the feed from OBS Virtual Camera @ 4K60, OBS and FFmpeg are dropping 10% off all frames, if the ram is in quad-channel I don't drop any frames.
This entire time I wasn't using XMP in any of the RAM configurations, once enabling it which clocked my RAM from 2133mhz to 3000mhz, I'm observing less overall 3D usage in OBS while in a quad-channel RAM configuration. Previously I was at 89% in the worst case scenario, where I was running a 4K60 feed of OBS Virtual camera with a mosaic of my 3x4K60 capture cards. Now with the XMP profile enabled I'm at 60% usage doing the same test, which also explains why my usage was lower when testing with my other set of 4x16GB sticks of RAM, as their base clock without XMP was higher.
Obviously whatever the core "problem" is is way over my head, and I'm not sure if this is limited to the X399 platform. I could test on my Intel system, but I'd have to move a ton of stuff around to get my capture cards slotted in. Whatever the case, it would appear I get the best performance in quad-channel at higher clock speeds, which makes so much sense.
In this new configuration (quad-channel XMP @ 3000mhz), I'm looking at about 75-80% 3D usage while pulling 3 instances of 4K60 Virtual cam into 3 separate instances of FFmpeg, which is what I've been trying to do this entire time. 75-80% usage really doesn't leave me much headroom to do anything else, but maybe I could further lower this usage by upgrading to a kit of quad-channel memory with higher clock speeds.
Now if you could excuse me while I let the rest of my brain leak out my ears I'd appreciate it lmao
This entire time I wasn't using XMP in any of the RAM configurations, once enabling it which clocked my RAM from 2133mhz to 3000mhz, I'm observing less overall 3D usage in OBS while in a quad-channel RAM configuration. Previously I was at 89% in the worst case scenario, where I was running a 4K60 feed of OBS Virtual camera with a mosaic of my 3x4K60 capture cards. Now with the XMP profile enabled I'm at 60% usage doing the same test, which also explains why my usage was lower when testing with my other set of 4x16GB sticks of RAM, as their base clock without XMP was higher.
Obviously whatever the core "problem" is is way over my head, and I'm not sure if this is limited to the X399 platform. I could test on my Intel system, but I'd have to move a ton of stuff around to get my capture cards slotted in. Whatever the case, it would appear I get the best performance in quad-channel at higher clock speeds, which makes so much sense.
In this new configuration (quad-channel XMP @ 3000mhz), I'm looking at about 75-80% 3D usage while pulling 3 instances of 4K60 Virtual cam into 3 separate instances of FFmpeg, which is what I've been trying to do this entire time. 75-80% usage really doesn't leave me much headroom to do anything else, but maybe I could further lower this usage by upgrading to a kit of quad-channel memory with higher clock speeds.
Now if you could excuse me while I let the rest of my brain leak out my ears I'd appreciate it lmao