Rocket-X was working now it Stopped & vanished from PCI list

[Blair S. Paulsen]

Really looking forward to UHD/4K video/monitor out from the RR-X. My Quadro 4000 can't do better than 2560x1600 :-(

It remains to be seen how much performance we'll get out of the RR-X in a TB2 expansion chassis. In the short term, an X16 internal physical slot is a much better host.

Cheers - #19

 

[Mark Andersen]

if TB2 is 4X then I don't see how this will change, unless you can use two busses and get 8X, but still only half the speed of internal slots soldered to the board.

 

[Mark Andersen]

Really looking forward to UHD/4K video/monitor out from the RR-X. My Quadro 4000 can't do better than 2560x1600 :-(

It remains to be seen how much performance we'll get out of the RR-X in a TB2 expansion chassis. In the short term, an X16 internal physical slot is a much better host.

Cheers - #19

I get this now by using the Blackmagic Studio 4K, RCX uses this for monitoring, the Studio 4K is connected via HDMI to a sony 4K monitor and a Flanders monitor (in HD) simultaniously.

 

[Medavoym]

This is troubling.

Any reason we should go for the new Mac Pro + RRX in a Thunderbolt 2 chassis?

Will things improve in the future?

 

[Blair S. Paulsen]

Don't Panic

Don't Panic

There is every reason to believe there will be huge improvements in performance over the next few months. TB2 chipsets and drivers are still getting dialed in by the vendors, still some conflicts to resolve in how it all knits together, etc. Apple has chosen to put more device drivers/firmware updates into OSX Mavericks itself vs everyone downloading 3rd party code that may wreak unforeseen havoc. Philosophically I agree with the strategy; Apple engineers vet the drivers so you don't have to. In practice… time will tell.

Last I heard RED engineers were re-writing some code to address initial issues with the RR-X in TB attached enclosures. Once everyone is playing nicely together we can assess how critical of a bottleneck the 4x PCI header might be. In theory, if the RR-X mostly contributes compute power, and has moderate throughput requirements for most/all tasks, then it might be a great wing-card for the Neo MP - just not right away.

Cheers - #19

 

[Mark Andersen]

At the moment my nMP with FirePro 700 transcodes 6K footage at the exact same speed as my Rocket x (with Open CL disabled) which had me thinking that the RRX was doing nothing, it wasn't doing nothing, just the same job as the Mac Pro. With my MBP however, the RRX doubles the transcode speed. BTW I'm getting 10 frames a sec. with RRX in a TB2 enclosure on both my MBP and nMP, and 10 frames Ps with open CL on the nMP.

 

[Nick Fury]

At the moment my nMP with FirePro 700 transcodes 6K footage at the exact same speed as my Rocket x (with Open CL disabled) which had me thinking that the RRX was doing nothing, it wasn't doing nothing, just the same job as the Mac Pro. With my MBP however, the RRX doubles the transcode speed. BTW I'm getting 10 frames a sec. with RRX in a TB2 enclosure on both my MBP and nMP, and 10 frames Ps with open CL on the nMP.

Hi Mark,

I am getting 4X transaode times with my RRX in Tb2 MBR, vs. no card, using alpha firmware, so the future is bright. Now, YMMV, depending, of course, on the computer in use. The advantage to the RRX in transcode times, even in Beta firmware is still real, with MBR, vs. no RRX. I have rented the RRX , in the Tb2 enclosure, with very happy users.

Thank you very much

Fury

 

[Zeb B]

Really looking forward to UHD/4K video/monitor out from the RR-X. My Quadro 4000 can't do better than 2560x1600 :-(

It remains to be seen how much performance we'll get out of the RR-X in a TB2 expansion chassis. In the short term, an X16 internal physical slot is a much better host.

Cheers - #19

Doesn't this blow up the case for purchasing a dedicated nMP for exclusive use with 6K RCX, DVR and FCPX? Maybe my 2010 MP tower still has legs or I get one of those badass Ted Machines:

http://www.devildemonstrategy.com

 

[Zeb B]

First Grant and Chris from Red replaced my card and now it works flawlessly. However I do have some insights. The RRX is much more powerful than a standard Rocket. It is built for 16 lane PCIe bus. ALL Thunderbolt (1 and 2) enclosures are limited to 4 PCIe lanes. At this time I am not sure how much of the PCIe bandwidth that RRX is using but according to post it is only use 1 of 3 cores that built into the card. I have read report here that they have the card running at more than 60fps in FF6k. I had a new 12 core mac pro with D700s and 64Gb ram literally going to be shipped this last week.... after a two month delay. I cancelled it the day before it shipped. I ordered a used 2012 mac pro, upgraded it with Twelve 3.46 Ghz Westemere processors, 128 Gb of ram, an internal 3 Tb SSD eSATA Areca Raid all for about the same amount as the new mac pro. However this didn't solve my GPU issue. I wanted something faster than the 2013 MacPro (Dual FirePro D700s), so I added a Cubix PCI expander. This is the real deal, no flimsy Thunderbolt cable, this connects to the PCIe host controller via 4 cables that look like they could power your house. I then put Nvidia K5000 Mac Quadro in the the double wide PCIe slot (slot number 1) then used the Slot 2 (the only remaining 16 lane slot in a MacPro) to connect the Cubix box. In the Cubix box I mounted the RRX in the first slot, and then put THREE (3) Radeon R9 280x GPUs. Now for the best part, not only is each card running at the full 16 lane speed, I know have 4 GPUs, each of which is more powerful than the GPUs in the new MacPro.

The result. According to Barefeats.com, the new MacPro Runs the the LuxMark OpenCL bench mark at 3483 (Ksamples/sec). My machine runs the same bench mark at 8634!!!!

So what does this mean. It means I can now render FF6k faster than real time, since FCPX doesn't have a frame counter per se, it will render an out an hour of video at FF6k in about 40 minutes.... The only downside is the room where all this happening gets pretty warm and I have had to add additional airflow to keep it below sauna temperatures.

I have spent a ton of time on this and would be happy to share the things I have learned with anyone who is interested.

BTW, the reason that I used the K5000 Mac Quadro is I couldn't get any other card to drive a 4k monitor in OS 10.9.2. The machine would be faster with all R9 280x. Also the R9 290x uses a different architecture and while it will function as a GPU you can not use it for the GUI or as your boot card.

Peace

Are you considering making a side biz out of this? Kinda like rehousing Kowa lenses. I would be interested in a turnkey Mac 6K RRX realtime solution

 

[Blair S. Paulsen]

IMO the jury is still out on what a NeoMP with a TB2 expansion chassis hosting an RR-X will do once things are better sorted.

If you want to start ripping through Dragon 6KFF footy ASAP and prefer Macs, then Michael's build out makes sense. My biggest hang up is the lack of TB I/O on the towers. I installed a CalDigit FASTA card that has 2 eSATA and 2 USB3 ports to tide me over but I'm jonesin' to take my entire ecosystem over to TB2.

Currently running the RR-X in a 2009 4,1 single CPU MP tower with a Quadro 4000 in slot 1 which plays most files in real time. Adding RAM and swapping in a K5000 might give me real time playback on virtually anything, but those are sunk costs into an older machine.

Perhaps the key issue is whether you'd rather put several grand into rehabbing a tower that gets you jamming sooner (and could potentially fill secondary roles for years) - OR - get on the NeoMP train and endure the climb up the hill (optimization) before you get to speed across the valley.

Since Zeb is so patient, he'll probably just wait for Apple to sort things out… hehe

Cheers - #19

 

[Zeb B]

Right . . . I'm the guy who yells at the microwave "FASTER!!!"