Any benefit of using Thunderbolt 3

Hey all,

My past two systems I have purposely built with Thunderbolt 3 support (though I am not sure if my Threadripper with Gigabyte Designare x399 will truly support it.. it does have a TB3 header on the m/b so I am hoping), largely to utilize some sort of plugged in super fast storage setup for working with larger projects. I never really gave it much thought as to what or how exactly the storage would best be used for video editing purposes. To be honest, I always figured I would have a few TBs of SSD storage in my machine, to handle source material, and use the TB3 setup to render to.

With that in mind, I am wondering exactly how much bandwidth is used when rendering a file to the system? Will having TB3 or even a USB 3.1 Gen 2 10Gbps connected storage improve render times? I would assume not since usually the delivery format is Youtube/vimeo sized 4K, at a few GBs or so in size, that takes several minutes or longer to render. I would assume then that I could roughly calculate the maximum bandwidth needed from render to storage medium based on the final output size and how long it takes to render. I know it isnt exactly accurate, but if it takes 5 minutes to render a 5GB file, then theoretically I could render to a USB 2 flash drive and not lose any performance.

So.. I will assume then that the real.. and possibly only benefit of a locally attached TB3 setup (assuming it could even handle the 40Gbps writing to some sort of large raided array of SSD drives given that each SATA 3 SSD drive typically can handle about 500MB/s write speed) is for working with large source files, such as RAW or DNxHR source files... is that about right? I mean I realize you could also set it up as the transcoding storage that Resolve would use to optimize media too as well.. but essentially that is the same thing.

I mostly work with 4K DNxHR SQ/LQ files, so I am thinking a RAID 0 2 drive SSD SATA 3 setup over a USB-C 3.1 Gen 2 would offer very nice performance for source files. I am wondering if it is worth it to buy larger 2TB and 4TB SSD drives internally, possibly even PCIe x4 NVMe drives, over an external 2 bay Raided USB 3.1 Gen 2 setup. Will that sort of setup give that much of an editing performance boost to be worth the added costs of internal drives? I am talking noticeably improved performance, like at least 2x if not more faster. Otherwise, it seems pointless to spend thousands on 2/4TB NVMe drives for this purpose.

The problem to me with trying to use an SSD for source files is that you're inevitably going to run into projects that need a relatively-massive number of files. It's not unusual for features (as one example) to go well beyond 8TB or 10TB, at least for final finishing. (For editing, it can go to hundreds of terabytes, even a Petabyte or two.)

I think even a RAID0 traditional drive can handle a 4K project, assuming a speed of at least 400MBps or something. Try a Blackmagic speed test and see what happens. I'm using a 64TB RAID5 at the moment and typically get about 1100MBps, and it's been very reliable.

I agree with Marc about what you'd generally need for reading, and that it should be huge.
As for your setup, its the drive/array you Cache/render to that should be your fastest possible one, you can pre-calculate its size based on your usual/expected workload and turnaround (how long are your projects, what do you export to, how long they stay 'live' etc')...

I think the real benefits of TB3 are system/project dependent and not necessarily speed related, at least not in a way that would matter to most today.
its probably more about expandability, future-proof-ness, daisy-chaining...

The problem to me with trying to use an SSD for source files is that you're inevitably going to run into projects that need a relatively-massive number of files. It's not unusual for features (as one example) to go well beyond 8TB or 10TB, at least for final finishing. (For editing, it can go to hundreds of terabytes, even a Petabyte or two.)

I just can't imagine the type of data management that a Petabyte project would require. I'm currently working on a feature which is housed on two 40Tb drives. We managed to squeeze that down to 25Tb for the color grading (could have been less but Resolve kept crashing during the Trim & Copy process, which is annoying in its own way). Copying the 25Tb footage from one source to another took the better part of two days. Being a one-man-band, that means my whole system is locked up while copying the data...not fun.

As a reference, I'm currently editing a feature with 2 internal 4TB drives on a RAID0 and getting realtime playback with 4.6k 3:1. It handles basic grading but of course as soon as you start to apply some OFX, tracking or NR, you get into trouble but that's what optimized media and proxy playback settings are for.

as soon as you start to apply some OFX, tracking or NR, you get into trouble but that's what optimized media and proxy playback settings are for.

Not sure that's drive speed related as much as it is CPU/GPU related.

Imo, thunderbolt is of limited use when it comes to external storage, unless you want to stick to consumer grade DAS devices.

If you have some technical ability, its far cheaper to re-purpose old hardware into a custom built NAS. Throw in some HBAs and a 10 gig NIC, and you have a fast and flexible storage devise. Specially when you want to move beyond 8 discs.

as soon as you start to apply some OFX, tracking or NR, you get into trouble but that's what optimized media and proxy playback settings are for.

Not sure that's drive speed related as much as it is CPU/GPU related.

It is, I'm just giving an example of how 3:1 plays with a RAID0 and then what causes the slow down. Like Marc said, if someone is doing a 4K project, a RAID0 is fine as long as it's on something faster than USB3 around 400MB/s. SSDs right now just aren't cost effective enough to run on them for large projects almost of us know.

Good stuff all. So one thing threw me for a loop.. rendering speed seems to be the slowest aspect of any project. Especially as you add more tracks, more FX, etc. Why would you need your output rendering drive to be the fastest? I would think the source footage drives where RAW and DNxHR source footage is quite large, should be the fastest drives and.. especially when rendering out to youtube/vimeo/etc, which is much smaller file sizes, that a slower drive (e.g. built in or external USB) would suffice without being a bottleneck?

Like I said I get it if you are using Resolves built in ability to cache/convert footage to a better editing format... those needing to be fast and large. But the final rendering output of an h.264 file? That doesnt make sense to me.

Another interesting reply.. having a 10gig network to a 10gig NAS.. vs a 40gig thunderbolt 3 local setup (assuming comparable space and number of drives). I would think using a USB 3.1 Gen 2 with a RAID 0 4 bay enclosure would be similar to using 10gig network to a NAS?

Mind all of you.. this is purely a hobby for me. I dont make any money doing this, nor am I a pro like most of you. I have spent thousands now on computer, storage, NAS, etc.. I am basically out of dough to expand much beyond a few more drives.

As for what my hobby entails.. I basically am shooting soccer games of my kids from a hipod/elevated tripod using Sony AX53 recorded via HDMI out to a Atomos Shogun at DNxHR SQ. Each game is about 300GB or so in size. I am saving ALL of the games (about 30 or so) to then pull clips from to assemble in to highlight reels for those interested for college purposes. As someone else eluded to, it takes a long time to copy files over the network.. in my case a 1Gb setup. Even at 10gig, not only would it be very slow to move all the files to pull clips from, but the amount of storage would be large (for me anyway). To be honest, I may not even need that.. I wrote a little software program (still working on it) that pulls clips from DNxHR files.. I provide the filename, start/stop time and output location, and it uses ffmpeg to do the work. So, at this point, my thought is to run that program against my files on the NAS over the network, storing the clips on a local SSD. In most cases, it is going to be about 30 to 60 5 to 20 second long clips. The one thing I havent figured out yet is how long ffmpeg will take to seek/grab clips over the network vs local files.. but I can at least let it run at night when I sleep... just have to feed it a list of clips to pull from multiple files.

Now that I read what I said.. my question is probably pointless for my needs. My original point was to figure out if I should look to move/store all games on to a large TB3 array, and pull clips from that, over TB3.. vs NAS storage. At the end of the season, once the clips are pulled for all players.. I will delete the games, to free up the space. Heck, even a USB 3.1 gen 2 setup with 2 12TB drives in RAID 0 would probably be a lot faster than pulling clips from the NAS over the 1gig network. Now if I could run ffmpeg ON the NAS directly.. and only have to store the clips across the network..that would be nice. I suspect that ffmpeg, like any video player, will stream the full video across the network to pull clips.. and not just figure out what byte within the file to start at, and then pull only the clip video.

I just can't imagine the type of data management that a Petabyte project would require.

Read this story on how Fotokem handled the 1PB of data for Star Wars: The Last Jedi (with all the finishing done by Walter Volpatto in Burbank):

http://postperspective.com/star-wars-la ... a-fotokem/

Imo, thunderbolt is of limited use when it comes to external storage, unless you want to stick to consumer grade DAS devices.

I'm using a 64TB G-Tech Studio XL connected via TB2, and I've been very happy with it, having done about 24 features in 2017. I have a series of backup drives that I use, all Thunderbolt, including some bare drives in two different Thunderbolt docks. So far, I haven't had to use the backups, but it's always a good idea to have a safety net. I don't have the need to go to anything bigger or faster yet, but no doubt as the price of storage drops, no doubt a 128TB Thunderbolt 3 drive will be out for less than $5K, and I think that's pretty affordable as these things go. It's always a struggle to come up with a backup strategy that will work.

Good stuff all. So one thing threw me for a loop.. rendering speed seems to be the slowest aspect of any project. Especially as you add more tracks, more FX, etc. Why would you need your output rendering drive to be the fastest?

There are projects that have to be turned around very quickly, like network TV commercials or any project that hinges on a delivery contract. I've worked on films before where literally, you could get sued if you were late in delivering the final version.

Having said that, determining the workflow for anything generally boils down to time and money. I often start my conversations with filmmakers on having them tell me what the nature of their project is, followed by when they need it finished, and based on that, we come up with a reasonable post budget.

I have worked on several 4K projects where my poor beleaguered Trashcan Mac limps along a 4fps to struggle to get through 2 hours of material. When you have to work around crashes and/or GPU glitches, this can be a 48-hour process by itself. HD is not so bad, but it's still not real-time. The new "improved" Trashcan Mac I'm using now is much improved, but I try not to push it beyond 10fps for 4K renders.

DNxHR SQ at 4K is only 73.5 MB/sec at 30p and 147 MB/sec at 60p. To saturate a 10 gig network, would require a pretty substantial machine, because you would have to be rendering at ~500 fps.


Do you have any ssd's in your machine currently m.2 would be ideal? If you do, you can run a quick test to see what your machine is capable of. Just render a 2 or 3 minute clip, and then you can use the render time to calculate how much bandwidth you actually need.

I'm using a 64TB G-Tech Studio XL connected via TB2, and I've been very happy with it, having done about 24 features in 2017.

I think you missed my point Marc. That's going for $6k on b&h right now.
https://www.bhphotovideo.com/c/product/ ... _64tb.html

If you have the mindset, you could build that for under $3K running on a 10 gig Nic. You could probably do 40gig direct connect copper for under $4k easily. Imo the higher end consumer stuff like this is a rip off.

I'm using a 64TB G-Tech Studio XL connected via TB2, and I've been very happy with it, having done about 24 features in 2017.

I think you missed my point Marc...

Perhaps you missed Marc’s point... workflow comes down to time and money. Would you rather spend your time building your own storage system, or spend it doing paying jobs? Horses for courses. Not that the two are mutually exclusive, but you have to budget your time wisely.



Sent from my iPhone using Tapatalk

OK it's 2X as fast as Thunderbolt 2 but are there any benefits of using Thunderbolt 3 40Gbps vs 40Gb Ethernet for external storage?

I think a big advantage is TB3 uses the new standard USBC cables which will connect almost any external device including storage, GPUs, monitors (8K), and even your smartphone. Devices can also be daisy-chained.


A basic intro

In terms of fast storage for editing prices look attractive with up to 2,600MBps transfer speeds
LaCie 6big Thunderbolt 3 6-Bay Desktop RAID Array with 60TB HDD (6x 10TB) $4699 and 12big 120TB $9499 (uses 10TB IronWolf Pro HDDs)
Accusys claims one petabyte (1280TB) and transfer rates up to 4,500MB/s using TB3 T-Share Expansion.

If you are talking Petabytes looks like Avid Nexus (up to 4.8PB)

New iMac Pro has 4 X TB3 ports - if you are looking at a new PC, TB3 seems like a must.

More TB3 hardware expected this week at CES2018

Perhaps you missed Marc’s point... workflow comes down to time and money. Would you rather spend your time building your own storage system, or spend it doing paying jobs? Horses for courses. Not that the two are mutually exclusive, but you have to budget your time wisely.

I mean no offence to the OP, but based on what he is shooting with he doesn't sound like a professional, so I'm willing to bet he doesn't want to just throw money out the window.

Some members here seem to forget that not everyone is a professional.

If you have the mindset, you could build that for under $3K running on a 10 gig Nic. You could probably do 40gig direct connect copper for under $4k easily. Imo the higher end consumer stuff like this is a rip off.

I don't care. I like the security of having a name brand and a company only a couple of hundred miles away I can call for support. The G-Tech has been rock-solid, and reliability is the most important thing to me -- not saving money.

BTW, I bought my G-Tech Studio XL used on eBay (3 months for one production), then replaced all the drives myself, so it actually didn't cost nearly as much as the list price of the standard 64TB model. The point of the thread was to say yes, Thunderbolt is a very efficient I/O port for small systems. You can go into much faster systems for shared storage and large facilities, but that wasn't the o.p.'s question.

Perhaps you missed Marc’s point... workflow comes down to time and money. Would you rather spend your time building your own storage system, or spend it doing paying jobs? Horses for courses. Not that the two are mutually exclusive, but you have to budget your time wisely.

Exactly. I needed something that just works, right out of the box. The G-Tech did that fine (with some tweaking on my end).

If you work just with DNxHR (read and write) then TB3 may be bit overkill (TB2 or 10Gbit will be enough). Even best mode at 60p 4K is <500MB/sec so in range of a single good SSD or 3xHDDs in RAID 0 (if you use good, even 2xHDDs).

Thunderbolt 3 is quite fast (in theory data channel is around 3.9GB/sec), so if you get decent RAID box you can use it for storing source assets or render assets (or even both). It's quite good solution when you don't need network access.
If you want something out of the box mentioned G-Tech or LACIE should be fairly good (LACIE 12big gives about 2.5GB/sec from 12 HDDs in RAID 5).

There is plenty affordable ways to get storage for 500MB/sec- starting with single SSD or simple RAID over HDDs. Well made 10gbit network will do around 1GB/sec and it's also not that expensive these days.

I can wholeheartedly second G-Tech, but had horrible experiences with LACIE – more than one of their controllers died one me and where I live I was not the only victim.

I mostly work with 4K DNxHR SQ/LQ files, so I am thinking a RAID 0 2 drive SSD SATA 3 setup over a USB-C 3.1 Gen 2 would offer very nice performance for source files. I am wondering if it is worth it to buy larger 2TB and 4TB SSD drives internally, possibly even PCIe x4 NVMe drives, over an external 2 bay Raided USB 3.1 Gen 2 setup. Will that sort of setup give that much of an editing performance boost to be worth the added costs of internal drives? I am talking noticeably improved performance, like at least 2x if not more faster. Otherwise, it seems pointless to spend thousands on 2/4TB NVMe drives for this purpose.

4K DNxHR SQ/LQ files are at most around 150MB/sec, so if you have 1 storage for read and another for write then this doesn't need anything special at all. Any 2/4TB NVMe is pure waste of money in this case. Even mentioned USB 3.1 G2 is enough, although I somehow don't fully trust USB when it comes to long terms storage access (even if it does work at the end).
One thing which SSD brings is good access time, but this is more noticeable when you work with compressed media like h264. In some apps scrubbing seems to be faster when h264 long GOP based files are on SSD and this is not about speed, but seeking time. With DNxHR being I frame code this is probably less noticeable if at all. Be realistic, don't waste money. 4K DNxHR SQ/LQ is nothing special in terms of storage needs (if you have crazy fast machine which can render at 3x faster than realtime then you still need just 500MB/sec).

I can second the advantage of better seek times in Resolve when using long-GOP formats.

The SanDisk Extreme 900 is a nice device with TB3. If I have to work with H.264, I try to break the project down into smaller chunks and keep the stuff on the 1.92 TB version I use. Very snappy!

I can wholeheartedly second G-Tech, but had horrible experiences with LACIE – more than one of their controllers died one me and where I live I was not the only victim.

I have had the same experience in the past. Not a Lacie fan here.

One thing which SSD brings is good access time, but this is more noticeable when you work with compressed media like h264. In some apps scrubbing seems to be faster when h264 long GOP based files are on SSD and this is not about speed, but seeking time. With DNxHR being I frame code this is probably less noticeable if at all.

One major issue for me with any Long-GOP format like H.264 is the lack of embedded timecode. I find that frame-by-frame cueing and audio syncing is very sloppy -- in any program -- and for this reason alone, I always advise that people transcode to an intraframe codec like DNxHD or ProRes.

I totally agree with you, but for quickly slapping together some H.264 stuff such a drive really helps.

One thing which SSD brings is good access time, but this is more noticeable when you work with compressed media like h264. In some apps scrubbing seems to be faster when h264 long GOP based files are on SSD and this is not about speed, but seeking time. With DNxHR being I frame code this is probably less noticeable if at all.

One major issue for me with any Long-GOP format like H.264 is the lack of embedded timecode. I find that frame-by-frame cueing and audio syncing is very sloppy -- in any program -- and for this reason alone, I always advise that people transcode to an intraframe codec like DNxHD or ProRes.

You are talking about totally different thing.
I just said that SSD helps with seeking for h264 long GOP files (at least in some apps). Haven't said a word about using h264 as a proxy format (or as a preferred proxy format).
Besides- h264 based files can have timecode as this rather depends on container which you use. Use MOV and Resolve will see timecode (MP4 also can store TC but this is less common- not sure if Resolve sees it). It's all down to how you transcode your files.
For example- I used h264 proxy files for subtitling needs (because o size). I just use either I frame only or IP, so I can have very fast scrubbing. It works very well. Don't forget that h264 can be I frame only (or I,P which is also fairly good for seeking). What brakes seeking a lot is B frames and very long GOPs, like default x264 250 frames ones.

Wow.. loving this tech talk!! You guys are all way ahead of me. What I would give to actually talk to some of you on how you got your careers going!! I love doing video work and editing.. but sadly.. at least for me in my mid 40s, my day job as a software engineer pays decently well and I dont think I could make the transition into some sort of part time/full time editor. Nor would I have a clue how to even get work like what you guys talk about. This has really become a passion hobby of mine, to the point that despite I dont have any paying jobs, I have been trying to build my gear the past couple years in hopes that I might somehow break in to the business even for simple jobs. As I said today my main bit is shooting my kids soccer games and hopefully getting paid a couple hundred bucks to make highlight reels for some of their team mates from all the footage. I just built a Threadripper system with 64GB ram, only one 1070 GPU but will replace with two next-gen nvidia cards when they come out soon, and using a single NVMe drive for now, but planned on picking up 2 more (my board handles 3 NVMe drives.. Gigabyte Designare x399.. nice board) when the new Samsung 970/980 come out, RAID 0 two of them to make a 2TB boot setup, and possibly a 4TB drive for the 3rd slot when those come out. I am basically thinking by end of this year, my system should be done. I cringe at the overall cost of this when its done.. and dont dare tell my wife!! But I do hope that it is worthy of good 4K editing.

I record on the Atomos Shogun on a 1TB SSD. I use the gtek USB 3.1 dock (Atomos and gtek has some sort of partnership, and gtek is good stuff, so I went with their co-opted products thinking it must be good). I plug that in to my USB 3.1 port and I assume I get 10gbps transfer speeds though I know the SSD itself does maybe 500Mbps at most. In my system now I have a 950 pro 512gb as my boot/os/app drive, and a pair of 512GB SSDs raid 0 on SATA3 for a 1TB drive, which is mostly used for my day job VMs and some games (yup.. in my mid 40s and still game).

I recently bought a dual bay USB 3.1 gen 2 external box.. and was thinking of putting 2 7200rpm drives in RAID 0 in there. Maybe a couple 4TB or 8TB..not sure. My thought was, if/when I have a large enough 4K project that might result in a few TBs of source material, copying it from wherever (1TB SSD, NAS, etc) to this external drive, then using that during editing over the USB 3.1 (or.. maybe even USB 3.0??) would be faster than trying to use my 1Gbps network. Anyway, my goal was to set up this machine and external/internal drives to provide the best possible lag-free editing experience I could without spending a movie production budget on gear. I am sure when I put in one or two high end 1180 (or whatever they will be called) with 8GB or 16GB RAM, I should see a dramatic increase in handling more nodes, FX, etc. I still dont quite understand if resolve (and to a lesser extent, fusion) use the GPUs during renders, or only when editing on the timeline. I think I grok'd that the GPUs are only used for timeline editing.. and the CPU is used for rendering.. thus the 16-core threadripper should do a much better job rendering than my old i7 quad core. Or do I have this backwards?


What complicates this for me is.. my goal is to get the upcoming GH6, or possibly if Black Magic has a new more affordable 4k micro ursa (one can dream right) that offers RAW footage, I wanted to be sure that what I am building right now can handle that too. Eventually, 8K though honestly I cant see working with 8K for many years to come.. unless I am somehow able to break in to the pro editing circle and get paying jobs!

So.. if I grok what you all are saying.. RAID 0 7200rpm drives over USB 3.1 (maybe 3.0 given that even in RAID 0 I dont think 2 x 7200rpm drives would saturate the 3.0 bandwidth since each drive does about 150MBps read speeds if that) should be plenty fast for working with 4K DNxHR source material without having to transfer it to NVMe/SSD drives. Given that SATA3 is 6gbps, and USB3 is 5Gbps.. I wouldnt see a dramatic increase in using internal SATA3 SSD drives... e.g. it would be faster but not like 2x to 3x faster assuming the USB3 can utilize most of its bandwidth. USB3.1 should be even better. Is that right? What happens in the case as I scrub through the timeline where I may have several tracks of video with 4K source.. does Resolve try to pull all those files at the same time to render them.. and if so would that then be a better reason to use internal NVMe/SSDs vs an external USB 3.0/3.1 setup? I havent done anything with more than 2 tracks of video, but just so I understand, if there were say 7 or 8 tracks all overlapping (e.g. layers. perhaps with masks, FX, etc)?


And to finally get back to my OP.. it probably isnt worth worrying about TB3 storage for my sort of hobby work at this point, because USB3.1 should be fast enough (right?). Given that USB 3.2 is finalized and should be out by end of year.. I dont even know given the time frame if I should bother with TB2. At this point, my board is the only threadripper board that has a TB3 header on it, but Gigabyte hasnt said/announced anything that indicates they will offer TB support on the board yet.. and for some odd reason you cant find PCIe x4 TB3 add in cards. I havent even found a TB2 PCIe add in card.. I dont quite understand why that is.. but with USB3.2 coming soon, that is a moot point in the coming year.

Thank you all again for all the good info. Appreciate hearing from you pros who do this for a living. Really nice that you all are part of this forum and offer your knowledge!

Resolve will read many files at the same time only if you have them stacked above each other and each has some transparency set (or some keying, PIP etc). Otherwise only top one is read (others are not visible so there is no point to read them )

Eventually, 8K though honestly I cant see working with 8K for many years to come.. unless I am somehow able to break in to the pro editing circle and get paying jobs!

I would not get too excited about going past 4K. My Threadripper 1950X rig plays back DCI 8K 9:1 RED files about 7 fps with no nodes, where it can play DCI 4K fully graded at 30. Maybe someone with access to a workstation with two massive Xeons could do better, but I certainly would not want to work on an 8K project. This is without even going into the cost of broadcast quality monitoring.

In terms of disk speed, the good old "spinning rust" is plenty fast enough if you put a few of them together. RAID0 or RAID10 depending on your risk tolerance and backup policy (please back up).

What complicates this for me is.. my goal is to get the upcoming GH6,

You have at-least another year, and more likely 2 before the GH6 comes out. Panasonic doesn't turn over the GH line very fast.

You are talking about totally different thing. I just said that SSD helps with seeking for h264 long GOP files (at least in some apps). Haven't said a word about using h264 as a proxy format (or as a preferred proxy format).

I don't care. I think Hate Dot 264 is bad for anything in post. Show me a drone camera that shoots H.264 with embedded SMPTE timecode.

I have no problems with anybody shooting drone material with a Sony, Canon, Red, or Alexa camera that can provide 10-bit Raw or ProRes files with timecode. But the Hate Dot 264 kills me, because it's just been too heavily stepped on. I don't have a problem with an intraframe codecs like the higher-end XAVC formats.

I don't care. I think Hate Dot 264 is bad for anything in post. Show me a drone camera that shoots H.264 with embedded SMPTE timecode.

Just what we need, another h.264 is {insert negative descriptor here} post..........

@Dan, given that they announced the GH5s today, you are no doubt right. Color me unimpressed... but then.. expecting cameras to advance this fast every year is a bit ridiculous anyway. I do hope BM has something either at CES or NAB. I am not a pro by any means, but I am kind a quality snob.. which is why I agree with Marc regarding h.264. I think it is fine for delivery to youtube, but otherwise I would prefer to stay in RAW/CinemaDNG/DNX range for all source video if at all possible. Despite that lack of a budget to afford big cameras and lens, I still strive for as close as I can get, which for me, shooting 4K out HDMI uncompressed to my Atomos even at the 8bit or 10bit 420 output my ax53 puts out is worth the added cost and work flow time.

Still I would love to be able to record CinameDNG RAW right from the sensor and get the better color range from a BM camera. Still trying to figure out if the micro studio camera with my Atomos recorder would be a better option than the Sony AX53 for getting good video.

You are talking about totally different thing. I just said that SSD helps with seeking for h264 long GOP files (at least in some apps). Haven't said a word about using h264 as a proxy format (or as a preferred proxy format).

I don't care. I think Hate Dot 264 is bad for anything in post. Show me a drone camera that shoots H.264 with embedded SMPTE timecode.

It has nothing to do with h264 itself, but company which makes drone and decided that there is no need for TC. As I said- timecode is mainly stored in container not codec headers, so every h264 file in MOV (or even MP4) can have timecode. Go and start petition to drone makers to add it.
Other than this- it takes about 5 seconds to add TC track to any MOV file.

@Dan, given that they announced the GH5s today, you are no doubt right. Color me unimpressed... but then.. expecting cameras to advance this fast every year is a bit ridiculous anyway.

The gh5s is actually more video oriented than the gh5 is. They removed IBIS that I know a lot of videographers don't like, and it has Dual Native ISO, that a lot love.

I do hope BM has something either at CES or NAB. I am not a pro by any means, but I am kind a quality snob.. which is why I agree with Marc regarding h.264. I think it is fine for delivery to youtube, but otherwise I would prefer to stay in RAW/CinemaDNG/DNX range for all source video.

Have you shot with a GH5 yet, if not I think you might be pleasently surprised by its internal h.264 10 bit 4:2:2.

h.264 gets a bad rap, because a lot of manufactures implement it very poorly. For example h.264 supports everything from mush, all the way up to lossless 14bit 4:4:4 ALL-I (basically raw in everything but name). Unfortunately a lot of manufacturers lean more heavily towards the mush, so consumers can use super cheap storage.

I personally don't even trust BM for h.264 creation, i find the setting/options offered in resolve far to minimal.

BM uses OS h264 capability on Mac and PC as far as I understand. This allows BM not to be forced to pay any h264 royalties, but in the same time it's almost useless as quality and control is poor.
Switching to Mainconpcet H264 SDK would be already a huge step, but this adds way more costs, so Resolve could not be free.

I don't care. I think Hate Dot 264 is bad for anything in post. Show me a drone camera that shoots H.264 with embedded SMPTE timecode.

Just what we need, another h.264 is {insert negative descriptor here} post..........

lol... what happened to this thread..

I think Its not a question of what 264 is capable of (14 bit lossless 444) but how we as Post Professionals/Colorists encounter it mostly. and in that aspect I agree with Marc.

Since we have no control on how camera manufacturers implement it on one side, and access to very good & proven work codec on the other (Cineform, Prores, DNX)... One just accepts it as a (mostly) crappy acquisition codec (excluding high end implementation), or a very good/popular web delivery codec.

BM uses OS h264 capability on Mac and PC as far as I understand. This allows BM not to be forced to pay any h264 royalties, but in the same time it's almost useless as quality and control is poor.
Switching to Mainconpcet H264 SDK would be already a huge step, but this adds way more costs, so Resolve could not be free.

Do you mean for exports/renders?
What do other systems use Andrew?

Most other companies use Mainconcept SDK for different codecs. It's not amazing, but acceptable.

I don't care. I think Hate Dot 264 is bad for anything in post. Show me a drone camera that shoots H.264 with embedded SMPTE timecode.

Just what we need, another h.264 is {insert negative descriptor here} post..........

lol... what happened to this thread..

I think Its not a question of what 264 is capable of (14 bit lossless 444) but how we as Post Professionals/Colorists encounter it mostly. and in that aspect I agree with Marc.

Since we have no control on how camera manufacturers implement it on one side, and access to very good & proven work codec on the other (Cineform, Prores, DNX)... One just accepts it as a (mostly) crappy acquisition codec (excluding high end implementation), or a very good/popular web delivery codec.

How is this a codec itself 'fault'?
Such a blame comes from not fully understanding things. h264 can be used for both cases and companies straight away jump into it. Fact that drone footage has no TC has not much to do with h264. Put Alexa on drone and you will have TC. Many cheaper cameras also support TC and they use h264. These are all shortcuts driven by ever demanding users for even cheaper things.
We all know that you will never have everything shot on Alexa or RED and it's up to you to deal with it.
5 sec and you can add TC track into every MOV without any re-encoding or even re-wrapping. It's instant. There are solutions for this. I have my own one, but when said it's not free people were not happy to pay 20$ for it. Their choice

How is this a codec itself 'fault'?

I don't think it is... and when it comes in XAVC or AVC-inta flavors I even like it.

But drones, less expensive cameras and inferior Log modes use 264 in ways that make colorists everywhere bleed from their eyes and its not because they lack TC...

Yep, but this again- has nothing to do with codec itself.
"Cheap" drones or cameras uses accordingly cheap recording "mode". Plain and simple.
If you remove this some of those cameras not going to be much worse than pro ones (at fractional cost)

You just need to work with whatever source you are given. I would love to shot my holidays with Alexa and then have all this possibilities to make it look A or totally opposite B, but this is not going to happen.

You are talking about totally different thing. I just said that SSD helps with seeking for h264 long GOP files (at least in some apps). Haven't said a word about using h264 as a proxy format (or as a preferred proxy format).

I don't care. I think Hate Dot 264 is bad for anything in post. Show me a drone camera that shoots H.264 with embedded SMPTE timecode.

I have no problems with anybody shooting drone material with a Sony, Canon, Red, or Alexa camera that can provide 10-bit Raw or ProRes files with timecode. But the Hate Dot 264 kills me, because it's just been too heavily stepped on. I don't have a problem with an intraframe codecs like the higher-end XAVC formats.

I think it is important to separate the technology from common usage. Devices that encode H.264 are typically lower quality and typically use higher compression rates than devices that use for instance raw or ProRes. XAVC, which is also H.264 being one of the exceptions.

But the technology is similar. H.264 all-intra is not very different from ProRes. Quality wise H.264 can encode all the way up to lossless.

Getting back to the ops needs, I'm a hobbyist as well, and work with the following codecs regularly.

h.264, h.265, DNxHR SQ, DNxHR HQ, DNxHR HQX, and some ProRes 422 every now and then.

My personal machine is a 6850k, and a gtx 1070. I use two internal RAID 0 drives for projects. I have back-ups of the source stored else ware till the project is done.

3 x 1TB Seagate Barracudas

E.png (28.97 KiB) Viewed 18615 times

3 x 4TB Seagate Barracudas.

F.png (27.84 KiB) Viewed 18615 times

The raw source gets dumped onto the F drive. Resolve renders the DNxHR HQ or HQX master out to the E drive, and then I use ffmpeg to transcode the deliverable back to the F drive. The benefit of this methodology, is that the drives are never doing mixed reads and writes, so they are giving maximum performance all the time.


Your motherboard has 8 sata connections, so if you can use them all in RAID, you should be able to build some raid drives that will work well and won't break the bank.

+2 with Dan:
i6850k = a good CPU => 40 PCIE lanes
and like him: always one or more discs in input and always other discs in output.

i also run dual arrays, push and pull, in my case SAS connection not Tbolt, 8x4Tb WDRedPro's in each enclosure, they run around 1200-1500 depending on space avb on any given day

So interesting.. my NVMe is insanely fast for reads/writes, though it is my boot/app drive. I had planned on.. if its possible to set up 2 NVMe drives in RAID 0 and still have my 3rd one for boot... getting 2x2TB NVMe eventually for source files. I still am not sure if each NVMe gets its own x4 lanes and thus 2 in RAID 0 would give nearly double the performance or not. I been trying to find out if each SATA3 plug gets its own 6gbps, so RAID0 for 2 drives would be 12gbps theoretical, or if they all share the 6gbps in total. Also trying to find out if USB3 (and 3.1) is the same deal.. if I were to use 2 USB3 flash drives, and software raid them, do they each get their own 5Gbps port.. or do they all share a single 5Gbps bandwidth.

So.. I would ideally want the outbound (rendering) drive to be the faster of the two? I still dont understand that. If a render is say pulling several video tracks... I would think the source footage drive being the fastest would be better for renders.

I would agree that using two different drives, especially on different channels/ports/etc would be much better than one drive both reading and writing at the same time.

So interesting.. my NVMe is insanely fast for reads/writes, though it is my boot/app drive. I had planned on.. if its possible to set up 2 NVMe drives in RAID 0 and still have my 3rd one for boot... getting 2x2TB NVMe eventually for source files. I still am not sure if each NVMe gets its own x4 lanes and thus 2 in RAID 0 would give nearly double the performance or not. I been trying to find out if each SATA3 plug gets its own 6gbps, so RAID0 for 2 drives would be 12gbps theoretical, or if they all share the 6gbps in total. Also trying to find out if USB3 (and 3.1) is the same deal.. if I were to use 2 USB3 flash drives, and software raid them, do they each get their own 5Gbps port.. or do they all share a single 5Gbps bandwidth.

Why do you need 2x NVMe in RAID0? This is only needed for work with 4K+ DPX, EXR, TIFF etc.
Single NVMe will be more than fast enough for any compressed format, including ProRes XQ 4k/6K.

It's not that easy (well- depending on the hardware) to put NVMe drives into RAID0 and double the speed. You need good motherboard for this. All your question about ports depend on implementation- with good motherboards ports are independent, but at some point they will saturate anyway. For NVMe you need CPU with many PCI-E lanes as each NVMe= 4 lanes, so quite a lot.
If you like to push NVMe a lot then read this:
https://www.extremetech.com/computing/2 ... readripper

You can also always buy one of those 4x NVMe to PCI-E x16 expansion cards and then do software RAID0 on 4 drives. This should be close to saturating x16 PCI-E, e.g. 14GB/sec. There are not that many hardware based RAID 0 solutions for NVMe. Some motherboards do it, other than this you're looking into enterprise solutions.

So interesting.. my NVMe is insanely fast for reads/writes, though it is my boot/app drive. I had planned on.. if its possible to set up 2 NVMe drives in RAID 0 and still have my 3rd one for boot... getting 2x2TB NVMe eventually for source files.

It's possible with some motherboards, and in some cases with some special adapter cards. The issue is 2 nvme drives in raid 0 doesn't have enough space for your needs. Earlier in the thread, you stated 300GB per game, and about 30 games, Thus you would need 9TB of storage space. The largest Nvme drive on the market is the 2TB Samsung 960 pro, at $1260 each. To have enough space you would need at least 5 for a grand total of $6300. If you have that kind of money to burn on storage, don't waste it on an nvme raid. Buy or build a NAS or DAS, you will get a much better return on your investment.

I still am not sure if each NVMe gets its own x4 lanes and thus 2 in RAID 0 would give nearly double the performance or not.

You need to check the manual of your motherboard for that. A lot of the boards with multiple m.2 mounts share pcie-e lanes between the secondary m.2 mounts and the sata ports.

I been trying to find out if each SATA3 plug gets its own 6gbps, so RAID0 for 2 drives would be 12gbps theoretical, or if they all share the 6gbps in total.

This is another check your manual thing. 99% of the time each ports gets a full 6gbps of batdwidth, but some boards will share bandwidth between ports or other devices.

Keep in mind that no sata ssd or regular hard-rive will saturate a sata connection. 6gbps is 750MB/sec and ssds top out at around 550, and hardrives at 250.

So.. I would ideally want the outbound (rendering) drive to be the faster of the two? I still dont understand that. If a render is say pulling several video tracks... I would think the source footage drive being the fastest would be better for renders.

who said the outbound drive always needs to be faster?

You should be basing your drive speeds and capacities based on the codecs and the amount of footage you will be pushing to and from them.

It's possible with some motherboards, and in some cases with some special adapter cards. The issue is 2 nvme drives in raid 0 doesn't have enough space for your needs. Earlier in the thread, you stated 300GB per game, and about 30 games, Thus you would need 9TB of storage space. The largest Nvme drive on the market is the 2TB Samsung 960 pro, at $1260 each. To have enough space you would need at least 5 for a grand total of $6300. If you have that kind of money to burn on storage, don't waste it on an nvme raid. Buy or build a NAS or DAS, you will get a much better return on your investment.

I have a Gigabyte Designare Ex X399 board. It has 3 x NVMe slots, with RAID capability on those. As well it has separate SATA3 with RAID.

I do NOT need to load all games at once. In fact, what I have the kids do is go through each game, which I have added burn-in of timecode to the video. They get me a list of each video, and the start/stop times of clips they want. I have asked them to put that in a specific file format. I then wrote a little bit of code that loads that format, and uses ffmpeg to pull the clips from the files across the NAS system. I do have a 5 drive 40TB (32TB realized) Synology NAS in RAID.. hmm.. the Synology equivalent of RAID6. I have the clips put out on one of my SSD drives (havent fully finished setting this box up.. but ideally the "source" drive.. most likely my 2x2TB SATA3 RAID0 SSD drive. 4TB should be WAY more than enough for a bunch of clips.

You need to check the manual of your motherboard for that. A lot of the boards with multiple m.2 mounts share pcie-e lanes between the secondary m.2 mounts and the sata ports.

Yup.. still a brand new board.. very little info on it yet. It has a TB3 Header too yet no info on whether or not it will eventually support TB3.

This is another check your manual thing. 99% of the time each ports gets a full 6gbps of batdwidth, but some boards will share bandwidth between ports or other devices.

Keep in mind that no sata ssd or regular hard-rive will saturate a sata connection. 6gbps is 750MB/sec and ssds top out at around 550, and hardrives at 250.

So.. I would ideally want the outbound (rendering) drive to be the faster of the two? I still dont understand that. If a render is say pulling several video tracks... I would think the source footage drive being the fastest would be better for renders.

I was hoping that would be the case. However, with 8 ports, I was a little worried with a Threadripper system and already having 64 PCI lanes, plus the 3x x4 NVMe lanes (not sure if those are CPU bound or part of the total 64 lanes Threadripper makes available), I was curious if 8 x 6Gbps was also available, or would be shared. Really dont know.

who said the outbound drive always needs to be faster?
You should be basing your drive speeds and capacities based on the codecs and the amount of footage you will be pushing to and from them.

Someone in this thread unless I misunderstood.. said that the render drive should be the fastest drive. I cant imagine that is true vs source drive where multiple 4K DNxHR clips reside and used on the timeline. But I wasnt sure.

Agreed on basing drives.. if I am fortunate enough where 2 of my 3 NVMe slots can be RAID0, and the other one can still be a separate non-raid boot/os drive, then I would like to add 2 x 1TB or 2x2TB NVMe next-get Samsung 970/980 drives to those slots, and use the combined RAID0 setup for my source footage drive. I would then use the 2x4TB SATA3 RAID0 SSD setup to render output to. I would think that would be about near ultimate performance you could hope for today sort of TB3 connected to an 8bay SSD setup with SATA3 drives in RAID0 to churn out 40Gbps over TB3.