Welcome to our community

Be a part of something great, join today!

  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

best way to run 2 rockets

Matthew Love

Well-known member
Joined
Dec 22, 2010
Messages
292
Reaction score
0
Points
0
Age
42
Location
Los Angeles, CA
Hey guys.. lately i've been doing some jobs where we run 40/90fps at 5k/4k at the lowest compression ratios (40fps/5kFF/5:1 example) and eat through 128SSDs in roughly 11 minutes.

Typically for a job like this I'm having enough trouble keeping up with the offloads (mac pro tower with sonnet 4EP Esata card running to RAIDs, so my offload is about 20 minutes for a full 128GB SSD to two places, a master and a backup)

NOW, when production wants to walk with the transocdes that night typically it takes me 2 hours to even get them into their hands, which I feel is normal but if I could speed it up I'd love to (Nothing like having a producer sitting with you as the stage gets cleaned up and you're still transcoding).

For jobs like this i'd like to bring on a 2nd rocket but not sure what the best setup would be.. Currently the rocket lives in my mac pro tower in the 8X slot, above it is an empty 4X slot. Could I use that? would having a rocket in an 8X and 4X lane cause issues? would I notice any quicker of a render?

THanks!
 
What is the generation and specs of your Mac Pro? How are you doing transcodes now? Transcoding to which formats?

I'm assuming your Rocket is probably in an X16 slot, or second one up from the bottom (slot-2), with your GPU below it in slot-1. The only Mac Pros to ever have an X8 slot was the '06/'07 model where you could assign PCIe lane configurations. The 2008 thru current all have two X16 slots and two X4 slots. Rocket works OK in an X4 slot, but if your system is fast enough to keep up with it, you can definitely run into a bottleneck there. IMO, we're at the tipping point where you may stand to gain more by building or buying a second computer to aid in transcodes vs. another Rocket, given the same amount of money to spend.

If you do go with the second Rocket, both Premiere and Redcine-X will use it. But before you spend that money, we should identify where your bottlenecks are in the transcode workflow. If the bottleneck is CPU power and / or storage speed, you can stack all the Rocket cards in the system you want, it won't make it go any faster. The Rocket mostly just accelerates the wavelet decompression, which is the bulk of the process in decoding and displaying an R3D frame. If your bottleneck is with your output codec, and it most likely is, adding another Rocket won't help.
 
I'm getting ready for a 2 Rocket system on my 2008 3,1 8 Core Mac Pro but with the Kona and RAID there aren't anymore slots.

Its the copy time that needs to be addressed IMO. So I think the anwser is to make a copy to a SSD at +750MB/s w Thunderbolt or Through PCIe. This "Turbo" copy to your SSD becomes your Source to render from. Get your render going, then go worry about making Master and Back up copies for the producer.


I'm looking to put together an expansion box of 2 Red Rockets and a PCIe SSD. or 1 RED rocket in the machine and 1 in the expansion Box with the PCIeSSD. What ever works best and fastest.


I want to spec out is this gear and get copy and render results.

Let me know what you guys think...?

Dynapower USA NA211A-GPU Netstor TurboBox PCI Express Expansion Box for Desktop
OR
THE MAGMA THUNDERBOLT 3T SLOT EXPANSION http://www.magma.com/magma-announces-three-slot-expansion-chassis-thunderbolt-technology
with
2 RED Rockets
and
1 TB OWC Mercury Accellsior SSD.


http://www.bhphotovideo.com/c/produ...3_SLOT_PCIe_EXPANSION_ENC_PCIe_6_DESKTOP.html

http://eshop.macsales.com/shop/SSD/PCIe/OWC/Mercury_Accelsior/RAID
 
thanks you guys.

Jeff, I'm running a 2009 mac pro, quad core 2.66. running 32gb of ram, and booted of an SSD. I'm typically transcoding out via esata to a G RAID drive.

My setup is the rocket in slot 2, nothing above that (just the sdi out for the rocket) and above that a sonet 4ep esata card.

typically we shoot 5k or 4k high speed, lowest compression possible and I render out to prores422 LT, 1920.. no cropping done.
 
IS there an R3D file that people could use for benchmarking?

I have a dual red rocket setup inside a cyclone box connected to a nice mac pro. Anyone care to share a specific r3d file so we can benchmark each of our systems?

-F
 
What are you using to do the transcodes? Redcine-X or Adobe CS6 or ???

Other questions I still have are about what drives are you reading the R3D files from? If you're reading from a single eSATA drive, you're bottlenecking yourself. Same with the output. The way the R3D input via the RED SDK works, it really likes / favors fast source drives, or a good sized and somewhat intelligent read-ahead cache.

We also have to address the limitations of the Mac Pro you're using. It's a quad-core unit, meaning you're restricted to the 4 DIMM sockets for RAM and that memory is running at 1066MHz. The ProRes codec isn't optimized to scale beyond two threads. The 2.66GHz "Nehalem" core Xeon is a fine CPU, but dated as it lacks the memory bandwidth of the newer CPUs that have succeeded it. The current model iMac, which is actually about to be replaced with new offerings, can outperform your Mac Pro in many tasks.

Based on what I know and a few assumptions, I think you can get a bit of a speed boost with some faster storage. A second Rocket could be beneficial here, but let's see if we can't maximize what you have to work with first. Even a battle-tested Rocket is $3K, if you can get lucky enough to snag one. You can build a system that's 70% more powerful than your Mac Pro for about the same money. Of course, that leads you into PC land and makes ProRes encoding a difficult prospect. There's always the Hackintosh route, but I would never recommend that to anyone wanting to use the system in a real professional setting or critical environment.
 
I concur with Jeff. your disk subsystem seems like the most obvious bottleneck culprit. Especially, GRAID over eSATA. Not to say that there aren't others, but that is the most obvious one. Also it might be worth mentioning that if you are using that eSATA GRAID for both read and write, you are slowing things ever more. While eSATA is much fater than USB or FW, it still struggles to keep up with the datarates required for streaming R3Ds. Without going into the mechanics of a spinning disk, the disk can either be reading or writing at once, not both. Given that GRAIDs only have 2 drives inside them as a RAID0, they are still not optimized for performing simultaneous read/writes. Thus, having separate source and write disks will help here. In a proper hardware RAID, there are usually enough physical disks present that are distributing the read/write load, so as long as the buffers in the controller are not reaching saturation this doesn't pose an issue.
 
Ha Dan, I think that would be so fast my head would explode..

i'm thinking of doing an 8TB SAS raid to transcode to and then going form that to productions g-raid drives.
 
Wish there was no need for transcoding. I get frustrated letters from the directors and producers i work for often saying that the transcoding step is time consuming, buggy and costly and is discouraging them from shooting RED.

RED would blow the competition away if the camera could generate a ready to edit 1080 file in Pro Res or similar. it's not more resolution or a sensor with a catchy name we need - it's a quick turn around solution for 1080 TV work. (high speed footage, too. digital recorder works fine for sync sound frame rates.)
 
I do agree with you Mike, but typically with my setup if we shoot 1/2 epics at 5k at 23.98 for the day i can have offloads and prores transcodes done an hour after camera wrap. I just stay on my camera guys about offloads in the last quarter of the day.
 
i think you're gonna need a pci expander to gain the full use of the rockets. in a mac tower you have 2- 4x slots and 2 16x slots. one of the 16x slots will be for gpu automatically. not sure how the system is gonna run with one rocket in a 16x and another in a 4x. or u could put both of them in the 4x slots. but again don't know if RCX will even initialize the two rockets together if it's seen that they aren't running at full capacity.


as for director's producers not liking the hassel of transcoding. I think there is just so much mis information out there. transcoding always happens. it's part of the process. if you edit at home and go adobe premiere, yes you can skip all of that. but any post house big or boutique never edit with native master footage. they'll either transcode in house or on set, no matter what camera you are shooting. alexa direct to edit? well it records log c to the pro res, you really think an editor is gonna cut with flat footage and try to show that with clients? or spend the money on expensive plugins to put a lut on it? i transcode for alexa pro res on set all the time, baking in looks for dailies, and down to a lower quality codec like pro res proxy or dnxhd 36 so they can have a smooth editing process without needing the most powerhouse system.

avid AMA can do native stuff but it falls apart on keeping files online iin the project in something long form. most post houses don't even run the newest version of avid, so thus again, they either transcode to avid mxf in house, or on set. i just had a shoot the other day where i had dnxhd quicktimes, and avid fast import would have rewrapped the footage like lightning. post house rejected them as dailies because their version of avid didn't support fast import. so i was rewrapping them in avid 6 on set so they could have their mxf files.

transcoding is part of the deal, and i don't think that's gonna change for a long time. a offline online workflow is really important for post houses. saving on storage requirements on their sans, not needing to spend all the money on powerhouse computers for every edit bay, faster to generate out rough cuts, etc. etc.

the sooner people accept that, the easier things will get because it will be more about finding the right solutions for the job rather than going against the current.
 
Back
Top