1. M5000 is a waste of money for this purpose.
   Get the GTX960 - it has the latest hardware decoder/encoder (NVENC block) from nvidia. Or wait for the GTX1070. This NVENC hardware block (ASIC) is separated from the CUDA cores and it has a huge importance in the final encoding process. That's the Hardware Acceleration that is used in PD14.
   Note that the final product has to be encoded in either H264 or H265 to benefit from the embedded hardware encoder.
   SVRT is a different approach compared to HA , when the video content and format is left unchanged, it just "skips" the video portion that was not touched, it doesn't use the GPU at all. But you are manipulating the videos so that will not work.


2. For those tests that you saw, they didn't include any editing, just pure encoding. So it does not tell anything about how the CPU performs with the effects that you are planning to apply. The GPU (the actual CUDA cores this time) can be used for some of the effects in PD14, but not for all of them. A six core CPU will help a lot with those effects that you are planning. That Xeon is only a 4 core and has a Passmark benchmark of only 10303. Not a best buy IMO. My first generation Xeon X5670 (6 core) has a PassMark of 12747 and workstations with one or even TWO of those can be foud very cheap on eBay, compared to a new PC. Look for a Dell T7500 for example: http://www.ebay.com/itm/Dell-Precision-T7500-2x-Xeon-X5670-2-93ghz-Hex-Core-48gb-2x-1tb-Win7-Pro64-/231948228863?hash=item36013118ff:g:fhQAAOSwXshWquJR



3. SSD used for video storage are not only a waste of money but they will also die soon. The video editing and encoding process outputs data at a very slow speed (like 12MB/s). And even reads are not very stressful - a video with 100Mbps datarate equals only to 12.5MB/s read speed. Even a slow laptop HDD can easily handle that, there is no benefit to gain from moving that storage on a SSD. Or even from separating the input from the output, those events (read video - convert video - write video) are not concurent. If you really want something decent, get 3 HDD and a hardware RAID controller and install those HDD in a RAID 5 configuration - resulting capacity will be 2/3 of the total but you will have redundancy and increased speed for copy operations (outside of PD14 processing)

Quote:

1. M5000 is a waste of money for this purpose.
   Get the GTX960 - it has the latest hardware decoder/encoder (NVENC block) from nvidia. Or wait for the GTX1070. This NVENC hardware block (ASIC) is separated from the CUDA cores and it has a huge importance in the final encoding process. That's the Hardware Acceleration that is used in PD14.
   Note that the final product has to be encoded in either H264 or H265 to benefit from the embedded hardware encoder.
   SVRT is a different approach compared to HA , when the video content and format is left unchanged, it just "skips" the video portion that was not touched, it doesn't use the GPU at all. But you are manipulating the videos so that will not work.


2. For those tests that you saw, they didn't include any editing, just pure encoding. So it does not tell anything about how the CPU performs with the effects that you are planning to apply. The GPU (the actual CUDA cores this time) can be used for some of the effects in PD14, but not for all of them. A six core CPU will help a lot with those effects that you are planning. That Xeon is only a 4 core and has a Passmark benchmark of only 10303. Not a best buy IMO. My first generation Xeon X5670 (6 core) has a PassMark of 12747 and workstations with one or even TWO of those can be foud very cheap on eBay, compared to a new PC. Look for a Dell T7500 for example: http://www.ebay.com/itm/Dell-Precision-T7500-2x-Xeon-X5670-2-93ghz-Hex-Core-48gb-2x-1tb-Win7-Pro64-/231948228863?hash=item36013118ff:g:fhQAAOSwXshWquJR



3. SSD used for video storage are not only a waste of money but they will also die soon. The video editing and encoding process outputs data at a very slow speed (like 12MB/s). And even reads are not very stressful - a video with 100Mbps datarate equals only to 12.5MB/s read speed. Even a slow laptop HDD can easily handle that, there is no benefit to gain from moving that storage on a SSD. Or even from separating the input from the output, those events (read video - convert video - write video) are not concurent. If you really want something decent, get 3 HDD and a hardware RAID controller and install those HDD in a RAID 5 configuration - resulting capacity will be 2/3 of the total but you will have redundancy and increased speed for copy operations (outside of PD14 processing)

Sonic, very good information and reading, thanks



I might add when I was in my first month of PD14 I tested the same 4K Gaming video with 1-M.2 Samsung 941, 2-Samsung 840 SSD, and a 5TB Tosiba hard drive. None of these hard drives made any difference in the speed to complete the "produce" final step.

I even tried a RAM disk (that has the fastest speed) and it didn't make any difference because... the bottle neck is never in the HDD speed.

Sure, a RAID5 makes sense for data safety, but that's another story.

Quote:

1. M5000 is a waste of money for this purpose.
   Get the GTX960 - it has the latest hardware decoder/encoder (NVENC block) from nvidia. Or wait for the GTX1070. This NVENC hardware block (ASIC) is separated from the CUDA cores and it has a huge importance in the final encoding process. That's the Hardware Acceleration that is used in PD14.
   Note that the final product has to be encoded in either H264 or H265 to benefit from the embedded hardware encoder.
   SVRT is a different approach compared to HA , when the video content and format is left unchanged, it just "skips" the video portion that was not touched, it doesn't use the GPU at all. But you are manipulating the videos so that will not work.


2. For those tests that you saw, they didn't include any editing, just pure encoding. So it does not tell anything about how the CPU performs with the effects that you are planning to apply. The GPU (the actual CUDA cores this time) can be used for some of the effects in PD14, but not for all of them. A six core CPU will help a lot with those effects that you are planning. That Xeon is only a 4 core and has a Passmark benchmark of only 10303. Not a best buy IMO. My first generation Xeon X5670 (6 core) has a PassMark of 12747 and workstations with one or even TWO of those can be foud very cheap on eBay, compared to a new PC. Look for a Dell T7500 for example: http://www.ebay.com/itm/Dell-Precision-T7500-2x-Xeon-X5670-2-93ghz-Hex-Core-48gb-2x-1tb-Win7-Pro64-/231948228863?hash=item36013118ff:g:fhQAAOSwXshWquJR



3. SSD used for video storage are not only a waste of money but they will also die soon. The video editing and encoding process outputs data at a very slow speed (like 12MB/s). And even reads are not very stressful - a video with 100Mbps datarate equals only to 12.5MB/s read speed. Even a slow laptop HDD can easily handle that, there is no benefit to gain from moving that storage on a SSD. Or even from separating the input from the output, those events (read video - convert video - write video) are not concurent. If you really want something decent, get 3 HDD and a hardware RAID controller and install those HDD in a RAID 5 configuration - resulting capacity will be 2/3 of the total but you will have redundancy and increased speed for copy operations (outside of PD14 processing)

Thank you very much for the thoughtful, detailed response. I have a few explanations, comments, and questions on your points:

I'm with you on the GTX series, and I know that's what the vast majority of users run... but it doesn't seem to be available in the corporate IT box I'm working in: the approved Dell 5810 platform only offers relevant Nvidia products (or a few Firepros) in the Quadro line, so I get my pick between the K420, K620, NVS510, K1200, K2200, M2000, M4000 (8GB, 1664 CUDAs), the M5000 (8GB, 2048 CUDAs), or the K6000 / M6000 (12GB, 3072 CUDAs, not going to happen). Since the first viable option is the M4000, it's really just a choice between the M4000 or M5000.

The original videos are avi encoded… apparently very efficiently… by two custom high-powered graphics cards. In my trial-version PD 14 tests with a mid-grade Dell Precision engineering laptop (highly underpowered for this application), PD 14 only produced massive avi output sizes (4 to 11 times the MB of the original video), and Mpeg variants were the only other options I could pick. Those Mpeg's were slightly smaller to slightly larger than original avi files. So far, a 1280x720 seems to be the max rendering resolution that makes any sense, and my limited work with the PD 14 Trial suggests that Mpeg is the way to go for output. So does the NVENC (ASIC) have relevance in that case?

I did not try a side-by-side video, partly because I didn't think the hardware could handle it and tell me anything meaningful - would this have to be a PIP reduced-size approach, or can I actually position and render them side-by-side equal size?

The corporate pricing upgrade difference from the M4000 to M5000 is currently only around $150: the upgrade to a Xenon 6 core 3.5 GHz E5-1650 v3 comes in between $200 - $280 more, and so far it's been a bit of a hard sell that the 6-core has value warranting the cost. A member of the evaluation justification team uses Premiere Pro and believes the 6-core won't do anything significant for me.

Does the PD 14 image stability processing use the GPU or the CPU cores? What about PD 14 rotation of the entire video? Which effects are handled by the CPU core hyperthreading?

Any other video thoughts or suggestions?

Drive selection has been frustrating. The three-hard-drive config I would like (360 SSD, 500 SSD, 2TB 7200 rpm) is evidently not possible in the 5810 platform if I dismiss a PCIe SSD harddrive option. (The Dell configurator acts like there is a combined issue between motherboard architecture and chassis hard-drive 2.5”/3.5” form-factor.) The thought: 3 separate drives would avoid creating bandwidth slowdowns. But to get three SATA drives in the 5810, all must be either 3.5” disk-drives, or be 2.5” SSD, and the two non-OS drives must be identical drives. That forces me to compromise between needed size and reasonable price, at two 1TB SSD drives. But the overall cost of that hard-drive layout is way more than my first configuration using a 2 TB storage drive with a SATA SSD OS/Program drive and a PCIe SSD source drive. (I'm being steered away from the PCIe drives based on durability concerns.)

However, your comments (and SoNic67) have me thinking that maybe the best hardware solution is to drop back to two drives. Could I use a 500GB or 1TB SSD, with a 2TB target/storage drive? The reason for thinking a large source drive is the temporary files created from scrubbing through and editing/highlighting two source videos that can at times each exceed 2 hours and 2GB. It has been alleged/postulated that this scenario could fill a 250Gb with shadow files, and possibly push a 360GB or 500Gb drive capacity, creating performance issues. Any further thoughts on hard drives?

Lastly, does PD 14 contain a "rendering queue" concept that lets me edit multiple projects, and then start them all automatically rendering when I leave for the day? Or, do we have to manually initiate project rendering, one at a time? (Our source videos can occasionally be produced on all three shifts totaling 4 to 16 hours of videos, so that individually initiating rendering may be a challenging task to maintain timely throughput. The P.P. rendering queue is a cited reason/assumption on why we "don't need to go overboard" on hardware content - like with a 6-core CPU - because "they'll all be finished in the morning when you come back in".)

THANK YOU so very much for the input! I'm anxious to nail this down, load up PD 14 on the new computer and get started, so I can provide an additional hardware configuration to the rendering performance comparison numbers in the forum. [Nearly ordered: Xeon E5-1660 v3, 8C, 3.0/3.5GHz; Quadro M5000 8GB 2048C; 32GB 4x8 DDR4 2133MHz; 1TB SATA SSD; 4TB 5400rpm SATA]
Superior performance results are never produced by marketing slogans or financial strategies - they are carefully engineered.

Oops. Sorry, the Nvidia Quadro M4000 is 8GB, 1664 CUDA cores. I updated the original post above to correct it. [Nearly ordered: Xeon E5-1660 v3, 8C, 3.0/3.5GHz; Quadro M5000 8GB 2048C; 32GB 4x8 DDR4 2133MHz; 1TB SATA SSD; 4TB 5400rpm SATA]
Superior performance results are never produced by marketing slogans or financial strategies - they are carefully engineered.

OK, I didn't know, corporate is different.

My home PC is actually a Dell workstation and it has a six core Xeon (12 threads) and a GTX960, but the work one has a Quadro, for similar reasons like yours. I can tell you that the PD14 scales well with the number of cores, including the Hyperthreading ones. So definitely I would go for the 6 core Xenon, the Quadro M2000 (because the number of CUDA cores is somehow irrelevant in PD14 and all M2000, M4000 and M5000 share the same NVENC ASIC hardware, being Maxwell 2 chips). The trick here is to output the file in a H264 (or H265) format, that's the only way the NVENC is usable. Trial won't let you use the NVENC at all.

For the Storage part, get a SSD for the OS and programs, but get a RAID5 consisting of 3 drives for storage. Personally I have a Dell provided 3ware 9650SE-12ML - a little bit of overkill for my three drives, but it can scale so well Probably your workstation comes with some form of RAID too.

As for the batch processing... didn't try that till now, I don't need it.

The rendering queue question is really a separate topic, so I submitted it here as a separate new question:
http://forum.cyberlink.com/forum/posts/list/0/48621.page#254703 [Nearly ordered: Xeon E5-1660 v3, 8C, 3.0/3.5GHz; Quadro M5000 8GB 2048C; 32GB 4x8 DDR4 2133MHz; 1TB SATA SSD; 4TB 5400rpm SATA]
Superior performance results are never produced by marketing slogans or financial strategies - they are carefully engineered.