Forum Replies Created
March 18, 2022 at 5:04 am
Awesome, thank you! I’ll check it out.
November 1, 2021 at 6:02 pm
Yes, the Win10 and now Win11 platform with HDR10 option and graphics drivers (SD vs GRD) that don’t always handle HDR correctly along with all the variants in HDR specifications … and then there is the connection type HDMI, DisplayPort and the cable type required to support higher bpc at a given refresh rate … it really is a crap shoot on how something will look from PC to PC even if the same monitor is used.
I have a Samsung G9 that is supposed to be HDR with 10bit support at 60Hz. But when I enable HDR it looks more like they just increased the Gamma than actually meeting the specifications. And then I test on a 12 bpc OLED TV and it looks completely different than my Sony OLED TV.
I had someone telling me that you don’t need 10 bpc support on the monitor to use HDR10?
Not even going to talk about HDR10+ and Dolby Vision … it’s a hot mess and I was hoping Win11 would sort it out, but that doesn’t appear to be the case.
October 29, 2021 at 4:55 pm
October 10, 2021 at 7:05 pm
Welcome to Adobe Pr … this has been a major issue for many of us for years (ever since CC 2017). In most cases many recommend converting your source files into some other codec like ProRes. I found this to be too time consuming and often didn’t resolve the performance problems with Pr and used up my 80TB of storage rather quickly.
To work around the poor codec reliance that Adobe have, I do the following:
1. Convert all ingested media to Proxies (either Medium or Low res but suggest you use same H.265/H.264 or PreRes422 if you have the space )
2. Enable Proxies in Pr when working on a sequence
3. Never import Ae projects, always convert them to a video and import the video
4. Under preference, Memory, set to “Performance” (this doesn’t make much difference especially with 64GB or 256GB)
As far as your hardware, I wouldn’t consider that “sufficient”. Disk I/O performance is NOT the issue with Pr … I’ve used very fast M.2’s and SSDs and RAID no RAID … makes no difference so long as the I/O device can work at 300MB/s or better.
I have an AMD 5950X (16C/32T) and Titan RTX 24GB and Samsung EVO 980 Pro M.2’s … and Pr is still SLOW if you don’t use steps 1-4 above.
One other thing is to enable CUDA (GPU Acceleration) rather than Software … when GPU acceleration works it’s much faster but be warned that some Effects might trigger render failures with GPU Acceleration enabled (my Boris FX Continuum does). You can also get the latest CUDA drivers here: https://developer.nvidia.com/cuda-downloads … but IMHO your 1030 is far too weak a GPU to do GPU Acceleration well, so I’d turn GPU Acceleration OFF or get a better GPU.
October 5, 2021 at 5:11 pm
Yes, getting a 3090 “at retail” price is very difficult, took me 11 months to get ONE EVGA 3090 and I signed up on their waiting listing the day they became available. I refuse to pay scalper prices and even some of the pre-built system markups.
I’m currently running one Titan RTX 24GB on a AMD 5950X and it’s MUCH faster than my backup render PC (7900X with Titan X (pascal)).
If you haven’t already, be sure to get the latest CUDA support from nVidia here: https://developer.nvidia.com/cuda-downloads
October 5, 2021 at 4:23 am
Do you have an existing USB-C or Thunderbolt port? If yes, you can connect an external 10Gbps network adapter.
October 3, 2021 at 6:28 pm
Depends on how fast the “2nd PC as a render node” can render.
1 Gbps = 125 MB/s
10 Gbps = 1250 MB/s
100 Gbps = 12500 MB/s
M.2 NVMe (980 Pro local) = 7000 MB/s
Latency will be much higher in any Network environment vs. local M.2, but latency will not be relevant to the task of rendering output to file.
The bottleneck will primarily be the CPU cores hence the dependency of “how fast” is your render node at the CPU level … in my experience “render to file” (or stream) on a very fast PC can generate about 300 MB/s I/O load.
Without knowing much about your “2nd PC”, I would still suggest that 10 Gbps should be sufficient I/O performance.
You’ll need 10Gbps LAN cards, 10Gbps switch, Cat 6 or higher cable, and your cable run length is limited to about 50m (164ft).
September 15, 2021 at 6:24 pm
I ended up using Reverb, Noise, and EQ to get “close” … izotope rx (Dialog Match) would probably make it perfect, but it’s outside my current budget at $999.
September 15, 2021 at 6:20 pm
I have a QNAP also but I never link to it directly, just too slow even across 10Gbps LAN.
I keep 20TB RAID local for fast I/O, I wouldn’t recommend linking anything directly to a NAS. I know it’s done frequently but not advisable … use intermediary and then sync to NAS as needed.
Anyway, that aside, your QNAP drives will go into sleep mode and make take a few seconds to fire back up when being accessed, those few seconds could be why Ae thinks the drives are disconnected.
Do you have QTier enabled for any SSD drives you may have in your QNAP? Since SSDs don’t spool down they will always be ready for access.
With that said, if you are mapping a drive letter to the QNAP then you will need to add entry in your registry:
Add or edit entry “EnableLinkedConnections” (DWORD 32bit) and set value = 1 (Hex). This might help keep links active.
June 25, 2021 at 5:04 pm
Sadly Pr is NOT very efficient. I know a software engineers at Adobe that works on the capture features of Pr. He’s a brilliant software engineer but has warned me about encoding/decoding problems they continue to encountered.
I’ve asked him several times why Adobe doesn’t leverage more RAM usage and he indicated there is “less risk” with less RAM usage, hence why Pr is still heavily I/O reliant. Regardless, the performance issues with Adobe are all about the encoding/decoding and the types of codecs used. Even “proxies” will not save the day.
I inquired about the usage of AVX instructions and he indicated they don’t use AVX (which seems odd to me). I haven’t confirmed but you can test for AVX usage by dropping the AVX offset in your BIOS/EFI (to some much lower value like 7 — 700Mhz slower when AVX is used) … disable speedstep so the CPU frequency is constant and note your “normal” frequency (i.e. 4.0Ghz). Run HWINFO64 and monitor CPU core frequency. Now run Pr and monitor HWINFO64 CPU frequency, if you see momentary CPU drops to 3.3 Ghz then you know AVX is being used, if no drops then AVX is not being used.
The performance problems I’ve noted with Pr:
1. Software bound (CPU side) and GPU usage is minimal and only helps in very specific use cases
2. Doesn’t use RAM efficiently … there is absolutely no need to have more than 64GB RAM (I have 128GB RAM which was a complete waste even for very large projects)
3. Doesn’t leverage AVX instructions
4. Encoding/Decoding isn’t optimized (Adobe seem lost in various codecs)
The only way I’ve been able to “marginally” tackle the performance problems with Pr is to:
a. Use proxies
b. Convert all my source to a format/codec that works well on my specific PC (you’ll have to experiment because it’s NOT the same for everyone)
I’ve tried tossing hardware at Pr and it made very little difference, very high speed M.2 drives, RAM drives, more RAM, better GPUs, faster RAM, different CPUs. The single key significant performance problem is Adobe’s encoding/decoding which they seem unable or unwilling to solve.