Welcome to our community

Be a part of something great, join today!

Log in
  • 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 :)

Will RR (original) benefit from TB2 Enclosure?

B

Brent Altomare

Active member
Joined
Jul 30, 2008
Messages
29
Reaction score
0
Points
0
Age
53
Location
San Diego
Website
www.1post.film
Hey All -

Just pulled the trigger on my RedRocket-X (now if I could just get my Dragon!) and I'm going to pick up a Thunderbolt2 enclosure for it. It got me thinking - would the original RR benefit from a TB2 enclosure? Anyone run a side-by-side comparison?
 
If your question is whether an original RR in a TB2 chassis will outperform one in a TB1 chassis the answer is yes, but the difference may be marginal for most operations.

Not sure how much it matters though - why would you get a TB1 external enclosure vs TB2 in 2014? Even if the original RR can't take advantage of the greater bandwidth, other things you are likely to want to put in there easily could.

Cheers - #19
 
After a weekend of testing, I didn't see any major difference between a TB1 or TB2 chassis, or even a native PCI-e x4 or x8 lane slot inside a full tower, in each case the Rocket-X connected (500 MB/s) at either x4 or x8 lanes but transcoded at pretty much the same speed in every test.
 
Hey Blair -

We have our original RR in a TB1 enclosure. However, I'm purchasing a TB2 enclosure for our new RR-X. Just wondering if I should splurge and purchase 2 new enclosures.
 
The original Red Rocket is a gen1 pci-e card and only uses the equivalent of x4 gen2 pci-e lanes, you won't see any major improvement over TB1 (x4) or TB2 (x4/x4) or native pci-e (x8) lanes. The older Red Rocket caps out at slightly better than real time performance with 4k material and slightly less than real time performance with 5k material. The Rocket-X seems to perform around 40 fps whether using TB1, TB2, or native pci-e x8 lanes (the current rocket-x configuration seems limited to 500 MB/s bandwidth), it didn't seem to matter if I was using 6k dragon / 5k epic / or 4k red-one material.

I haven't tested the Rocket-X in a full x16 lane pci-e slot just yet and obviously the old Red Rocket can't transcode anything from the Red Dragon camera.
 
A couple of quick questions:

- does the RRX still performs at about 1/3 of its total processing power? Is RED still fine-tuning it over time until it gets to 100% capacity?

- I think RRX was advertised as a card which is several times better than the original RR (five times better?- if I remember correctly). And RED charged for it accordingly, to be fair (it is very, very expensive). Do people generally find this statement true? Is the RRX several times better than RR?

Thanks!
 
Only 1/3 LED lights were active during all my testing so far, but the Rocket-X connects at 500 MB/s and transcodes at around 40 fps, however I haven't tested the Rocket-X in a full x16 lane PCI-e slot just yet and I suspect that is a requirement to get full use of the card.
 
Hi all,

How much of a benefit you see going from Thunderbolt 1 to Thunderbolt 2 with a Rocket(-X) (assuming you can feed it R3D input fast enough!) depends on the playback resolution (full, half, etc.) or the export crop/scale you are asking it to do.

At full resolution (4K+ mostly) you will hit the Thunderbolt bandwidth limit before you hit a Rocket(-X) performance limit, so to get the best performance you will want to make sure you are at half res or lower unless you shot at something like 2K, of course.

We have seen a small benefit using Thunderbolt 2 with the original Rocket if the situation is right (output res + framerate requires more bandwidth than what Thunderbolt 1 can provide), but that may not happen in some / most cases if you're outputting at lower resolutions.

For Rocket-X there's a big difference going to Thunderbolt 2 as it can process clips way faster than realtime and thus needs much more bandwidth than the original Rocket! This includes input (and potentially output) I/O speeds!

As an example: if the Rocket-X is doing 6K at 50 fps it could require 500 MB/s (that's megabytes, not bits!) of *read* I/O performance of the R3D's to keep up that speed! This all depends on REDCODE level chosen and scene complexity, of course. This does not even take write performance in to account when exporting (which you may not want to do to the same raid). The new Mac Pro is great for this as it's somewhere around 1 GB/s of read performance and very high writes as well.

Lastly when exporting CPU performance is also something to keep an eye on when doing high-speed transcoding with the Rocket-X (we have seen performance go down when ProRes hit a CPU limit for example).

Rocket-X is not at 100% performance yet, but it's definitely way higher than 1/3 Medavoym! Do make sure you are on the 2.1.x driver (and accompanying firmware) to get great performance though!

To wrap up, for Rocket-X you definitely want to use a x16 slot in a PC or use Thunderbolt 2.

Rob
 
I don't know how much bearing the 3 LED lights have on activity or performance with the Rocket-X, but I did not see any different between full res debayer of 6k dragon, 5k epic, or 4k red one material; over thunderbolt 1 or thunderbolt 2 or even native motherboard PCI-e slots, everything connected at 500 MB/s and transcoded to/from a SAS + HBA RAID, around 38-42 fps.
 
Shiblon,

Something is off then, you should see big speed difference between all of that. Thunderbolt 1 (and 2) will not be able to full res 6K debayer at 38+ fps. Math will easily proof that. Sounds like you are scaling on the Rocket-X and exporting, so your export codec could be the limit (encode speed limit). Won't know until you provide more details.

Rob
 
I seem to remember that thunderbolt 1 can greatly exceed 500 MB/s of connectivity, which seems to be the bandwidth cap of the current Rocket-X configurations, and my current system transcodes 6k HD dragon material at around 10 fps without a Rocket-X, so a realistic performance of 30-40 fps transcodes with a Rocket-X is apparently not feasible according to Red?

I'm going to run the same series of transcode tests again, I guess 1 min clips of 6k HD dragon / 5k HD epic / 4k HD red one isn't sufficient, but if Red is really indicating the Rocket-X is unable to provide faster than realtime 6k HD transcodes, what's the point?
 
Brent Altomare said:
Hey Blair - We have our original RR in a TB1 enclosure. However, I'm purchasing a TB2 enclosure for our new RR-X. Just wondering if I should splurge and purchase 2 new enclosures.
Click to expand...

As Rob noted, bus bottlenecking is not typically a factor for the original RR. I'd also expect the cost of TB2 enclosures to drop a bit once all the players are fully in that game, right now they're still optimizing the first generation of TB2 chipsets, etc. I also think the second generation might bring dual TB2 I/O to 3 slot boxes which will be great if you have one card (RAID, RR-X, GPU) capable of saturating (or nearly) a single link.

Cheers - #19
 
Rob Lohman said:
Sounds like you are scaling on the Rocket-X and exporting
Rob
Click to expand...

I think lots of folks do not understand that you can:

a) scale on the Rocket or Rocket-X via hardware

vs.

b) full 6K/5K frame moves to your frame buffer from the Rocket in full res - and you scale in your software app - which does allow you to choose scaling algorithm in many apps like Scratch, etc, - but it's slower than forcing the Rocket/Rocket-X to scale on the hardware for you

For dailies I always scale on the card - for finish I always scale in the software. There is a performance difference.
 
I suppose the old rocket at (250 MB/s) transcoding 5k HD epic material at 25 fps and 4k HD material at around 35 fps, is nearly as capable or more so (in its own limited, crippled, and deprecated ways) as a new Rocket-X is with the same exact material, and neither fully utilizes a TB1 connection.
 
Shiblon Wixom said:
I suppose the old rocket at (250 MB/s) transcoding 5k HD epic material at 25 fps and 4k HD material at around 35 fps, is nearly as capable or more so (in its own limited, crippled, and deprecated ways) as a new Rocket-X is with the same exact material, and neither fully utilizes a TB1 connection.
Click to expand...

From a practical perspective that's essentially true as far as I've seen. There really are a lot of variables in the mix though: RedCode compression ratio, complexity of scaling, read/write speeds on storage, etc.

BTW, Mark noted that scaling can be done in software and I've done quite a bit of that in Assimilate's Scratch DI tool. I bring it up because I have been able to use my scaling filter choice (Mitchell, Lanczos, Gaussian, etc) to hit the right amount of perceived sharpness in the finished image without a separate sharpening pass. Not only is it faster (in most cases), but the sharpening effect is more natural looking to me, plus it keeps me away from a "blunt force" sharpening node that could cause ringing, aliasing, etc. YMMV.

Cheers - #19
 
Rocket-X is a lot faster with 5K & 4K HD material as long as you can feed it properly and get it out again.
 
Blair S. Paulsen said:
BTW, Mark noted that scaling can be done in software and I've done quite a bit of that in Assimilate's Scratch DI tool. I bring it up because I have been able to use my scaling filter choice (Mitchell, Lanczos, Gaussian, etc) to hit the right amount of perceived sharpness in the finished image without a separate sharpening pass. Not only is it faster (in most cases), but the sharpening effect is more natural looking to me, plus it keeps me away from a "blunt force" sharpening node that could cause ringing, aliasing, etc. YMMV.

Cheers - #19
Click to expand...

In my experience - scaling on the Rocket is faster than software scale - which is why I always use hardware scale on dailies. I can run some more tests on a tricked out machine - but I have never encountered faster scale via software vs. hardware - so ... I agree on the ability to tweek the sharpness and scale algorithm - Mitchell better for UP - Lanczos better for down, etc. - but speed .... I have not seen software scale beat the hardware scale in terms of speed.
 
You must log in or register to reply here.
Back
Top