REDROCKET ....dead...

[sergio arguello]

I find it a bit odd that these threads never get the attention of RED employees (nick grow), and therefore we never know what they think, and much less where they stand on this increasingly obvious flaw in the way we are forced to deal with the current workflow. I myself refuse to do any upgrades to the card (RR) because ive been thru 4 cards in 3 years, and the culprit every time was the upgrade. I also keep my card in the pro
and never ever go in there. Its kept in a cool location and still had cards go out.

 

[Jeff Kilgroe]

Yep. Every other company has opened its RAW workflow up to the GPU. Sony is very similar (RED might think to similar) and it cuts like butter on even the most lightly modified GPU system. Proprietary workflows that require specific cards are not a way forward.

With RED, RAW is not the problem, it's the compression. That is where the bottleneck in the workflow happens. RED isn't about to open up their compression now that it has gone proprietary. Even with that, they still wouldn't open up their debayer and demosaic processing. Arri and Sony have not opened up theirs either. GPU acceleration on what happens after the frame is decoded is perfectly do-able. In the past GPU acceleration for the wavelet compression was indeed a problem. I'm not so sure that still applies with the latest GPUs and MIC compute cards rolling off the production lines these days. I'm hoping to see some big things at NAB next week in terms of workflow.

Also dumping 5 grand into a Pcie 2.0 card when 3.0 is the standard on higher-end PC's, and the new MacPro's will (hopefully) be using it, seems like utter and consummate folly

RED Rocket is actually PCIe v1.1, it's not even PCIe v2.0!!! It should have been PCIe v2.0. At least then we could have run it at full speed in one of the Mac Pro's X4 slots. ...Just sayin'

Was wondering if Dr Kilgroe had any info/advice on this

http://www.macrumors.com/2013/03/19...d-7950-mac-edition-graphics-card-for-mac-pro/

Nice card, but what would you do with it exactly? If you don't need CUDA, it's awesome. It would be one kick-ass card for FCPX, Photoshop, Maya, games, etc.. Not worth a crap for Premiere or Resolve or other things that use CUDA...

Absolutely agree with this. My guess is that RED anticipated high end CPUs would have cracked the real time barrier by now and saw the Rocket as a stopgap measure. FWIW I'd rather spend 5 grand on blazing fast CPUs that will make everything faster, not just R3D workflows. Hopefully that day is close, and as noted above, it will be a PCI-e 3.0 world soon as the throughput needed for "big data" is dependent on the weakest link.

I think many of us made that assumption about CPUs. But CPU compute power hasn't progressed by huge leaps and bounds over the past 5 years. Sure, systems keep gaining about 20% overall performance every 12 to 18 months, but most of those innovations are coming in the form of increased memory performance, faster I/O transactions and interconnects, larger and faster cache buffers. CPU makers (Intel) are focusing more on trimming the fat, reducing power/ TDP while also increasing the number of cores. For most every application and use out there CPUs are at a point where they are actually plenty powerful. We either need more of them or we need to find a way to make them a LOT faster at just doing what they already do. Faster isn't going to happen for the most part. Shrinking transistor path sizes close to 20nm, which is happening now, will offer some boost in speed with lower power consumption, but it's a smaller evolutionary jump. We're not too far off the limits of what the current manufacturing process and materials can provide. Unfortunately, most of the technology that could or should replace what we have now is still largely theoretical. When we hit the hypothetical process barrier at about 15nm, then we're stuck. At that point transistors will need to be supplanted with optical and/ or quantum gates and metallic circuitry will have to give way to electron or photon streams racing along "tracks" etched into carbon composites.

The other answer is for RED to support GPU decodes on nVidia/AMD/Intel. The downside is that GPU processing of image data typically has less precision than CPU decodes. It is my hope that RED is pursuing a solution using hybrid processors that will be revealed at NAB - but that's pure conjecture.

Precision is no longer a problem. Consistency was always a problem across different GPU families, models and brands, but that has largely been addressed via the latest CUDA and OpenCL releases. Actually, the image processing portion (debayer, etc..) are not the bottleneck and current GPUs can tear through that sort of stuff like it's nothing. The real bottleneck is the wavelet compression. I've always upheld the stance that GPUs don't lend themselves well to decoding wavelet data and in the past that's been mostly correct. I'm not so sure now. In fact, I'm willing to bet an Intel MIC (Xeon Phi) can do a decent job. I'm betting a nvidia Kepler GK110 (Tesla K20X, GTX Titan, Quadro K6000) can handle it too. And even if those solutions aren't perfect, I still think they need to be explored. I can put three GTX Titans in one workstation and take my crew out to a surf and turf dinner all for the price of one RED Rocket...

 

[Gavin Greenwalt]

With RED, RAW is not the problem, it's the compression. That is where the bottleneck in the workflow happens.

Seeing as RED has had the good sense to abandon pushing the high compressions (14:1 etc) for the most part... I think it's time to rethink wavelets. I would be perfectly happy with multi-threaded scanline ZIP at this point. 2:1 compression is fine by me. Leave 8:1 redcode in place as JP2k but at the lower compression ratios I would take lower compression efficiency for better system performance.

 

[JanneJansson]

I run a HP z820 2x8 core, and a 2008 mac pro w rr. The HP is silly much faster w playback and decode then the old mac w rocket.

 

[NateWeaver]

Just so Red / others know what the rest of the world is up to, F55 raw plays RT, full debayer on a 3.4 680MX iMac.

I don't think the Red Rocket only thing is going to fly for a lot longer.

 

[luigivaltulini]

I run a HP z820 2x8 core, and a 2008 mac pro w rr. The HP is silly much faster w playback and decode then the old mac w rocket.

hey Janne, you have hp z820 in osx?

 

[JanneJansson]

hey Janne, you have hp z820 in osx?

No win7

 

[luigivaltulini]

No win7

HP Z820 8 core 2ghz? whit nvida?

 

[Will Keir]

Dr. Kilgroe, I like that. hahahahaha

 

[Peter Matthes]

This seems to be the standard symptom of dead RR...

Hearing this from you ... and looking at my status window ... makes me think that mine can be added to to RIP list.

Totally messed up.

 

[Peter Matthes]

From the number of issues I've seen posted, perhaps 50-100, suggests that maybe 1%-3% of rockets fail in the first couple of years.

So I wouldn't portray it as an epidemic of poor quality.

Jim,

Mine was very lightly used ... despite being housed in a Mobile Rocket, it did not travel ... and this job was the only time it had been used for more than a few hours in a row.

This is the error message parade that I now get just trying to use the RR card to open up a clip in RedCine-X.

"Taps" is played on a distant bugle.

Very disappointing.

 

[Manfred Lopez]

We have a Red Rocket X on order for when they are released. I hope this problem doesn't carry over into the next generation.

 

[Peter Matthes]

We have a Red Rocket X on order for when they are released. I hope this problem doesn't carry over into the next generation.

You and me, both.

If there have been some improvements made in the new Rocket X card to prevent this sort of common failure, it would be great to hear them from RED.

Does anyone know the longest extended warranty period you can get for the new Rocket X card?

 

[Robert Hart]

Desperate measures only. - If your card is already out of warranty and a new buy is the only alternative, maybe try giving it a good spray with canned air with the can turned upside-down to eject liquid gas. Leave it in the computer when you do this. Use the little plastic tube which comes with the can and allow the liquid gas to flow on the board both sides if you can reach. The board will be chilled by this treatment and will attract moisture as dew. Use the can in the upright position to drive this dew off. Leave the card to dry out for about four hours. Check if any dampness or dew remains. Then if dry, try operating the computer again. Don't operate the computer with anything inside it in a damp state. If it works again, you might get another twenty hours out of it, if the video card in my computer is any indicator. - Desperate measures only. - Please take notice of ANY advice given against this desperate idea.

 

[Leroy Jackson]

I have been officially scared into NOT picking up one of these cards smh.

 

[Johnny Friday]

I'm adding myself to the 1% - 3% of dead cards....bought a battle tested card...and it NEVER worked from day one. However, i never got around to sending it back to red until after the 3 month warranty....so i'm in an interesting predicament. That said, my original RR card---bought new & one of the first production line cards......is working without a glitch after MANY years of service....

 

[Ed David]

I'm taking my card very soon to the oven - I will let you guys know how it goes.

 

[Tony Lorentzen]

Reviving this grand old thread. So did anyone end up baking their Rocket?

 

[James Hogarth]

I'm having the exact same problem did anyone ever get a solution?

What is the steps for the 'oven' idea?

All this happened after I upgraded to build 30 and I've tried rolling back but no good. I've been in contact with RED but the service I'm getting is pretty poor to say the least.

Given that these situations only arise when you are on a job (in other words must be fixed fast) I'm worried I'm going to have to jump in and buy a battle tested card to replace it. But given the poor warranty you get what is the point.

How can something so un reliable be so expensive and why is no one at RED Commenting on this?

 

[Shiblon Wixom]

I'm on rocket number #4, they typically last about a year or so, two of the four were covered under warranty. At this point the GPU in my rack system transcodes 5k at around 16 fps, so I save the rocket for when the client pays for expedited transcodes. The red rocket seems to work better than ever on RedcineX build 30. I've found once the rocket starts acting weird, it typically has only a few days of life remaining.

The old rocket only has a limited number of hours of operation, typically 18 months or less, due it's technology and older design, it's still pci-e gen 1. The new rocket x is more in line with modern GPU's and should last years under heavy load.