The issue is not the GPU unless you don't have the correct drivers in. CC no longer requires an approved card ie GPU for the GPU acceleration. Any Nvidia card with 1GB of vram will work with the GPU acceleration. The approved list is simply what Adobe has tested. Those Xeons are to low clock speed for realtime playback. You want CPU's that are 2.6GHz or higher to maintain realtime playback since CPU's with lower clock speed have to high processing latency to keep up with the realtime playback requirements. Besides this GPU acceleration performance is significantly effected by clock speed of the CPU's once you have enough cores and threads to ideally decode/encode the media.
I suppose that could be the issue, though it's hard to understand then why my 9 year old Xeon 5150's seemed to be able to handle real-time playback of the same HD footage (though admittedly only a single stream). It's not like I'm working with 4K here. I would think that dual 2650's could handle it. Perhaps that was presumptuous of me.
The clock speed of the Xeon 5150's is 2.6GHz. No matter how many cores you have the time it takes to process data through the CPU pipeline is decided by the clock speed. Realtime applications have a limited time ie 1 sec for 30 frames for example to process that data. When the clock speed is to low then it cannot process that data in that time period.
....the spec for your CPUs shows a "turbo speed" of 3.0 Ghz, which means under a full load you SHOULD be seeing thar 3Ghz frequency on all cores, You can monitor the CPU's behavior real time with task manager.
I HAVE read on this forum in the recent past that PPro has had TROUBLE running with twenty physical cores.......there was some bug that limited the use of PPro to 16 cores, ( i think ), before PPro started having problems.
Even though that clock speed is not ideal, with 1080P DNxHD, that footage should run like butter, as it does on my current laptop and as it did on my OLD laptop from 2010 !!! I'm talking about multipletracks with effects !!
I am surprised that Eric didn't mention ,or, remember the "20 core" problem others reported a while ago.....maybe PPro fixed it....i do not know. Eric definitely knows the most about this stuff, as he BUILDS custom video editing machines at his company ADK.
My 2010 laptop had a 4 core, 8 thread i7 CPU that had a base frequency of 1.73 Ghz and it would "turbo" up to 2.3 Ghz. I only had 14 GB of system memory and an NVidia GPU with only 1.5 GB of video memory. I would convert the "pesky" canon .MOV footage into DNxHD in order to edit on that laptop.....which worked well.
My newer laptop from 2014 has 24 GB system memory, a 2 GB NVidia GPU, and an i7 Haswell that has 4 cores and 8 threads running at 2.4 base frequency, but, "turbos" on all cores under load to 3.4 Ghz.........not TOO far from your 3.0 frequency.
With my newer laptop that footage you described runs EASILY, plus I no longer even need to "transcode" as the CPU handles even the highly compressed .MOV Canon footage easily.
I am thinking that you MAY be running into that "20 core" problem that was discussed on this forum in the past.
I have several T7910 Dells in various configurations and they playback with multiple layers of RED 4.5k with zero lost frames at 100% resolution. There are some differences, I'll highlight them below:
My workstations are:
Dell 7910 Precision Tower - same
Dual Xeon E5-2650 v3 CPU's (20 physical cores) - I have dual e5-2687wv3 (faster than yours and same core count) and dual e5-2620v3 (slower than yours and less cores)
64GB's DDR4 ECC 2133Mhz memory - I'm using 128GB now, but 64GB worked just fine too for Premiere Pro --> make sure you only allocate about half the RAM to Premiere Pro when you have this much RAM - Windows caching does help and Premiere Pro doesn't seem to benefit much at all from anything over 32GB
SSD Boot Drive - same for boot
local storage -- you don't seem to put anything local -- I assign Premiere Pro to use local 4x RAID 0 SSD array for media cache and media cache db files
24TB's of Fibre Attached Shared Storage (Editstore EO) which typically tests at 500+MB/s throughput. - I tested with all files (project, media, scratch, etc. - all files except media cache and media cache DB files) on a local SSD RAID 0 array (less than 1TB) and also on a server RAID 5 array (40TB connected w/ dual 10 Gb copper using Dell Intel x540 NIC)
EVGA GTX980ti 6GB Display card (Driver 361.43) - I've used dual GTX Titan, dual GTX 980 Ti, and am currently using a single GTX 980 Ti with driver 364.51 (on CPU1, slot #2 or #4)
Black Magic Studio 4k IO - never used one, but I think I'd like to
Running Windows 10 Pro 64bit - Windows 8.1 64-bit -- I don't think there's much difference here
Premiere Pro CC 2015.2 - assuming this is the latest Adobe CC, same
My results for playback of 4 layer RED 4.5k media - zero dropped frames at full resolution (playback using Displayport 60Hz or HDMI 30 Hz) to LG 4k monitor.
So, what does this mean? What is holding your system/workflow back?
I don't have "the" answer, but offer the following thoughts...
- GTX 980 Ti is awesome, I doubt that has anything to do with dropped frames
- CPUs are fine, I doubt that there are any issues with those; Premiere Pro issues have been reported with dual CPU setups with more cores that you or I have
- RAM - as noted above, with 64GB available, I'd suggest allocating 32GB for Premiere Pro and the rest can be used for file caching
- media - when you playback and scrub, are your CPUs all being utilized? (it is easy to view CPU utilization using Task Manager / Performance); If you are dropping frames and your CPUs are not close to being max'd out, then possibly your media is not compatible with the multi-core capabilities of Premiere Pro.
- filters - do you have any 3rd party filters (i.e. Red Giant, etc.) that are single threaded on your timeline - that's one way to really tank your performance in Premiere Pro
- drive(s) - have you tried putting all your files on the boot SSD? You say that your Editstore has decent throughput, but there is more to drive performance than simply throughput; in any case, for a remote drive workflow I would suggest having a fast local drive (or SSD RAID 0 array) for media cache, media cache DB, and possibly your scratch files too
- Black Magic - are all the drivers, etc. up to date? Does Premiere Pro drop frames without using the Blackmagic card for monitoring? Have you tried their tech. support?
The max Turbo value is not used when all Cores are used unless the board allows it which the Dual Xeon boards normally don't. That value is used when less cores are utilized. We have not seen or had any reports with any Core/threading limitation with Premiere in general. Now there may be some codecs that don't thread beyond 16 cores currently but last I checked R3D does without issue so atleast 1 codec does not have this problem.
BTW Jim R3D media is handled far differently in Premiere than many other codecs both with threading and GPU acceleration with the GPU accelerated debayering. Try XAVC for example with both Xeons systems and see how the playback is compared to each other. Also keep in mind GPU acceleration is a pipeline using many components such as system ram video cards and drives or in this case networking and storage. The longer it takes to process the data through the CPU pipeline the less time the realtime app has tolerance for higher latency ie slower performance elsewhere such as networking. This means to resolve the issue you have to improve the latency somewhere like taking media locally versus networking with slower clocked CPU's. None of this is cut and dry as People think it is because so many variables are involved especially at the codec and application level.
Thanks Eric and all. I appreciate all your feedback.
Our CPU's are not being maxed...not even close. Rarely are they hitting over 20 percent in my tests. This is pretty basic footage 1080p or 1080i without any additional filters etc. I'll get a few more dropped frames during sections where overlay graphics (uncomressed AVI) are present, but I would somewhat expect that when not rendered..
Based on the insight provided, and with some additional digging, I've toyed around with several settings. Most had little appreciable affect, but some have helped, as it now seems to be pretty stable with minimal (through still present) dropped frames.
- Changed Audio Hardware from MME to ASIO.
- Limited Premiere's RAM allocation to 32Gb
- In task manager, set affinity to utilize only 16 cores for PP (unfortunately, this setting does not hold on re-launch)
- Turned Hyperthreading off in BIOS.
Taking our footage local isn't really an option based on our departmental workflow. Most projects are shared regularly and handed off from producer to editor to gfx artist to post audio. Whichever our solution, we will need to utilize our shared storage. Even with the dropped frames, the process is workable right now, I just expected better performance based on the system configuration. I guess I should have splurged for the 2687w's!
Well do you atleast have a media cache drive locally or are you putting that on the shared storage? Lowering the cores and threads essentially will cause Premiere to cache less frames just like limiting the ram used by the application to 32GB for example. This can improve performance over networked storage because Adobe is caching less frame data at a time so the wiping the memory buffers and refilling them takes less time as it's done since less data is requested. Unfortunately the CPU load you list is expected with the media or FX are not using all the cores and the processing latency is taking to long. This is one of the reasons higher clocked speed CPU's are better since many codecs don't thread out to all the cores available on the higher core slower clocked Xeons. When you switch the Audio from MME which is WDM ie Windows audio to Asio this changes the audio driver to bypass layers in the OS since Asio is your low latency protocol and that is how they achieve that low latency. WDM audio allows device sharing and the OS to synchronize audio easier with applications but the trade off is higher latency for the audio since it has to thread through OS layers. Asio also allows multichannel configurations and mapping where as WDM does not for the most part. Everything is pointing to the processing latency as the issue especially since DNxHD/HR and ProRes are Iframe codecs where each frame is presented data wise that are very efficient and normally thread better because of that. Keep in mind power management profiles lowering the clock speed of the CPU's or not allowing them to turbo correctly will show this. The 20% CPU load or less might be causing those CPU's to lower their clock speed down prematurely or keep it down. See if the bios has a Intel Speedstep setting and if so disable it. You might have to disable C1E as well but that can disable turbo on many boards. Many dual Xeon boards have power profile settings in the bios that have to be changed to performance. They are often default set to higher power management.
We have tried placing the PP mediacache on shared storage spaces as well as local. Currently, my mediacache is set to a local secondary hybrid drive.
When poking around the BIOS, I had noticed the Intel Speedstep setting but had left it enabled. I will now disable it and report back any performance gains.
Thanks again for your assistance!
Keeping the media cache locally is important especially with your network storage for media. The audio conform files are in the media cache as are other cache files. You might want to also check for power profile options built into the bios. Normally those are under advanced settings in an ACPI area.