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 :)

Is a 12 core Mac Pro strong enough for 1/2 res good 5K realtime CPU debayering?

S

Simon Jaquemet

Member
Joined
Feb 1, 2008
Messages
13
Reaction score
0
Points
1
Hello

We are in the process of building a grading suite.
We have two 8 core Mac Pros that could be upgraded to 12 cores (with a bit of hacking)

I particularly have in mind to upgrade to 2x 2.66GHz 6 core Xeons because those sell fairly cheap.

What are your experiences? Anyone with a 12core MP running Davinci? How is CPU debayering performance?
All we need is 1/2 res good playback.

Cheers
 
I've got the 2.4 12 core 2012 with 48gb ram and I can do 4k r3d 1/2 good 24fps in resolve (also running 3 gtx680 in cubix, and an 8 drive sas raid5). Epic 5k, not really, need to drop to 1/4. Dragon is going to suck big time. As it seems all the red projects I get tend to be low budget, there is no return on investment in getting a red rocket card (what, 7 grand for a turbo? No thanks), so I'll likely turn those jobs away...
What year are your 8 core? The older ones might not have high enough bus speed.
Resolve 10 will have the option to display full debayer on pause and drop down on playback like apple color used to do, so that will help.
 
I have a 12 core 64 GB RAM, 2 512 SSD and 2 2TB Mirrored Drives with 2 2GB Graphics cards and i am still having to do 1/2 or 1/4 depends on the project and what plugins I am running (that is what really kills you) I am def. getting the new RED Rocket Card for sure!!! Trust me I have a buddy with one and I use him all the time for trans-coding.
 
Even on a monster Z820 16-core 3.1GHz with 128GB RAM and GTX Titan GPU, 1/2 res is a stretch for doing serious work. For general playback in Resolve with a few easy nodes, no problem, but to do anything substantial you will be at 1/4 for 5K material and then the system isn't straining at all. And a decent 12-core Mac Pro will also handle that 1/4 just fine. Half-res 5K is pretty demanding. I'm betting the next round of Xeons due here shortly which give 1866MHz memory support as well as a bit of a speed boost on the CPU side with the shift to Ivy Bridge core design will handle 1/2 res just fine because the current Sandy Bridge Xeons can do it 65% or more of the time and are right at the cusp of making it happen. 2010-2012 Mac Pro towers are using the prior generation Westmere core Xeons where the RAM tops out at 1333Mhz and chips are 10% slower per clock per core, vs. Sandy Bridge Xeon EP.
 
Jeff Kilgroe said:
Even on a monster Z820 16-core 3.1GHz with 128GB RAM and GTX Titan GPU, 1/2 res is a stretch for doing serious work. For general playback in Resolve with a few easy nodes, no problem, but to do anything substantial you will be at 1/4 for 5K material and then the system isn't straining at all. And a decent 12-core Mac Pro will also handle that 1/4 just fine. Half-res 5K is pretty demanding. I'm betting the next round of Xeons due here shortly which give 1866MHz memory support as well as a bit of a speed boost on the CPU side with the shift to Ivy Bridge core design will handle 1/2 res just fine because the current Sandy Bridge Xeons can do it 65% or more of the time and are right at the cusp of making it happen. 2010-2012 Mac Pro towers are using the prior generation Westmere core Xeons where the RAM tops out at 1333Mhz and chips are 10% slower per clock per core, vs. Sandy Bridge Xeon EP.
Click to expand...

I'm not counting on any serious improvements, at least to the point of being able to use these files reliably at 1/2 res or better, unless/until Red revises their methodology, and thus the SDK, to achieve such performance on commodity hardware. Arri has gone the uncompressed route for now, but Sony uses wavelet based compression on the F55 and F65 and has no problem getting those files to play back at 1/2 resolution on any modern PC with sufficient GPU power (a GTX680 is more than enough). It can be done, and likely will, eventually. But for the moment, Red seems content to continue the reliance on proprietary hardware assistance, so I wouldn't count on your scenario happening any time soon.
 
M Most said:
I'm not counting on any serious improvements, at least to the point of being able to use these files reliably at 1/2 res or better, unless/until Red revises their methodology, and thus the SDK, to achieve such performance on commodity hardware. Arri has gone the uncompressed route for now, but Sony uses wavelet based compression on the F55 and F65 and has no problem getting those files to play back at 1/2 resolution on any modern PC with sufficient GPU power (a GTX680 is more than enough). It can be done, and likely will, eventually. But for the moment, Red seems content to continue the reliance on proprietary hardware assistance, so I wouldn't count on your scenario happening any time soon.
Click to expand...

That's somewhat of a dire outlook there for the hardware. Current Sandy Bridge EP Xeons can do 1/2 res 5K in Resolve to some extent. Actually some people consider it to be working fine, but I don't like the occasional hiccups or stutters that come with it. Ivy Bridge EP brings us from 8 cores up to 12 cores on a chip and on top of that we gain a 10% increase in per core performance at the same clock rate and a 15% clock up of memory speed and throughput. To lay it out there, if the new Xeon platform can't handle 5K EPIC MX in realtime in Resolve 9, then something is wrong. Furthermore, I'm hearing Resolve 10 is more efficient and should do better here, but I haven't tried it. Perhaps I should...

As for RED's reliance on proprietary hardware, I'm a bit puzzled here myself. 5K and now 6K are demanding. I can totally understand the need for hardware acceleration, but I'm not sure I understand the continuation of an expensive and proprietary accelerator card. I still think the Intel MIC platform could be better explored here. Even commodity GPUs. Comparing with Sony's wavelet implementation isn't really the same thing, RED has adopted a far more complex layered wavelet system with inherent encryption. It has some real advantages and those come at the expense of performance. But knowing what the current crop of GPU hardware is now capable of, I think I speak for many when I say that I'd be more interested in installing a couple GTX780's or GTX Titans in a system than one proprietary RED Rocket card.

We also have to take a look at software implementation as well. There is a huge disparity in the software world right now when it comes to R3D performance and how all of the various apps are integrating R3D support based on the same SDK tools. SGO Mistika handles 5K R3D with ease and on relatively common hardware. BMD/DaVinci Resolve and Adobe Premiere do alright, but is still a good bit behind what Mistika can do. Then there are a number of others that still can't seem to integrate native support. I honestly don't know what else RED can do here on the SDK front... It's pretty darn easy to work with and it's more complete than the SDK tools provided by ARRI and Sony. That is not the problem. RED needs to find other ways to get software companies motivated. I do know that one huge problem in the past was the continuous air of beta status about the whole RED program. Software developers were unwilling to commit when the system would change every couple weeks. RED has drastically slowed their SDK updates. They now feel like they are holding it back intentionally... I'm sure they are. Trying to get developers more comfortable and saving the small updates to coincide with the big ones.
 
In computer world performance is capped by the weakest link. Moreover, as Jeff notes, most companies don't want to spend money on coders if they expect significant changes in the short term. If my job was to support R3Ds on a specific platform, part of doing that efficiently would include reading the "tea leaves" and cherry picking specific points in time to devote resources. Post Dragon release, SDK rev 2 for example...

Stepping back and looking at the development curve of the digital cinema infrastructure I do think we are still in the accelerator card era. Annoying and expensive though they may be, one could also look at RedRocket cards as a powerful leverage point that allows us to work with compressed and encrypted high resolution media years before commodity hardware makes it trivial. If the upcoming RedRocket-X chews 6K R3Ds for breakfast, offers IQ similar to software decode, can operate at close to full speed in an external TB2 chassis and comes with a proper no questions asked 2 year warrantee it should be an excellent investment.

Key point - IF the RedRocket-X can live in it's own external casing that allows it to be a viable short term rental piece (or just easily reallocated to the point of current need) that doesn't require physical installation into a PCIe slot while still having sufficient throughput (TB2, miniSAS, PCIe direct cable) to decode any flavor of R3Ds in real time...

IMO the crux of the value proposition is the warrantee which is now 2 years ($395 to add a 3rd) and the redesign of the Rocket itself to better protect it from static and physical damage.

FWIW I'd pay another $500 on top of that to have a next day replacement service contract where if I'm having any issues I can call RED and have them FedEx me another card immediately. In this scenario I send them back the wonky card, they test/repair/QC it and put it in the queue for the next caller. Other companies have versions of a "no down time" coverage option for a variety of mission critical items and if administered properly the fee for the service should offset the increased overhead.

Even better, particularly for facilities, might be a combination purchase and "no down time" service contract billed at $300/month with a minimum term. If, for example the minimum time period was 2 years, your $7,200 commitment would be enough to make it good business for RED and allow you to pay monthly rather than scrape up $6,750 or put it on a credit card. Add in the knowledge that if you have any issues a fully tested card will be in your hands the as soon as the next FedEx delivery day... You can sign my ass up right now.

Cheers - #19
 
NO. You will need a RR card for 1/2 res.

You'll be okay with 1/4 though. 1/2 will be a little choppy.

That's my experience with Premiere Pro CS6 and CC.

2009 Mac Pro. 48 GB Ram 8 Core 2.9gh.

Simon Jaquemet said:
Hello

We are in the process of building a grading suite.
We have two 8 core Mac Pros that could be upgraded to 12 cores (with a bit of hacking)

I particularly have in mind to upgrade to 2x 2.66GHz 6 core Xeons because those sell fairly cheap.

What are your experiences? Anyone with a 12core MP running Davinci? How is CPU debayering performance?
All we need is 1/2 res good playback.

Cheers
Click to expand...
 
I should also add to my notes above... 1/2 res is totally do-able on that Z820 box, and even the 12-core Mac Pro, if enough RAM is available and the sequence on which you're working can be fully cached.

The Rocket is a necessity, IMO. These systems will perform without it for 1/2 res and lower for editing/ grading, but when it comes time to do full-quality renders for finishing or any intermediate full quality work, such as for compositing, the Rocket is a must-have. Just no way around it.

And, while I'm still not a fan of the expensive and proprietary accelerator card, I totally agree with everything Blair just said. I look at the current Rocket card, of which I own two. I have never had any troubles with either of mine <knock wood>. And the first one I received was out of RED's first shipment. Those cards have paid for themselves time and time again. At over 3 years in lifespan and a price, due to MX sensor upgrade bundle and buying my second card used, I have less than $7000 invested in the two. Even when the Rocket-X arrives, these two cards will keep working and stay right where they are now. At this point, I'm into two Rocket cards for about what my wife spends at Starbucks every day and still going. And I shouldn't even look at it that way because they both paid for themselves within the first couple months.

I think the biggest flaw with the current Rocket is the fan assembly. It's just not robust enough and RED has unwittingly created a device that a huge number of users decided to take mobile or to rent out. I wince in pain every time I see mention of Rocket cards being shipped about or transported between jobs to be installed in other peoples' workstations. OMFG... Which also leads me to point #2. Why has RED not released a ruggedized portable Rocket with Thunderbolt interface??? And I'm not talking about placing the fragile PCIe card in a PCIe to TB enclosure. No, I'm talking about a streamlined, purpose-built Rocket device with Thunderbolt and power connectors on one side and a full array of HDMI, SDI and DisplayPort connectors on the other side.

Given the design of the new Mac Pro, I'm hoping RED will take a swing at this idea.
 
Jeff Kilgroe said:
The Rocket is a necessity, IMO. These systems will perform without it for 1/2 res and lower for editing/ grading, but when it comes time to do full-quality renders for finishing or any intermediate full quality work, such as for compositing, the Rocket is a must-have. Just no way around it..
Click to expand...

Actually, I feel just the opposite. The render quality with the Rocket is not equal to that of a software only render, even though the software approach takes a LOT longer. But for the final product, you want the best, and that is not what you get using the Rocket. The place for the Rocket in my world is for faster turnaround of viewing and editing material in a form that's "no questions asked" regarding any noticeable aliasing, stairstepping on edges, or other artifacts typical of 1/2, 1/4, and 1/8 resolution debayering. But for a final delivery, well, that's exactly where you don't want to be using the Rocket.

As for the Rocket "appliance" approach, I've been advocating that for a long time, and now so is Apple, if a bit indirectly, with their new Mac Pro design. I have a feeling, though, that Red needs the higher bandwidth of TB2 to make this work reliably. So perhaps later in the year...
 
What we don't know yet is whether the second gen Rocket X will have enough processing muscle to produce the more refined render typical of a software based decode. If so, we get the best of both worlds, if not, Mike's note stands in my book.

Cheers - #19
 
Jeff Kilgroe said:
I should also add to my notes above... 1/2 res is totally do-able on that Z820 box, and even the 12-core Mac Pro, if enough RAM is available and the sequence on which you're working can be fully cached.

The Rocket is a necessity, IMO. These systems will perform without it for 1/2 res and lower for editing/ grading, but when it comes time to do full-quality renders for finishing or any intermediate full quality work, such as for compositing, the Rocket is a must-have. Just no way around it.

And, while I'm still not a fan of the expensive and proprietary accelerator card, I totally agree with everything Blair just said. I look at the current Rocket card, of which I own two. I have never had any troubles with either of mine <knock wood>. And the first one I received was out of RED's first shipment. Those cards have paid for themselves time and time again. At over 3 years in lifespan and a price, due to MX sensor upgrade bundle and buying my second card used, I have less than $7000 invested in the two. Even when the Rocket-X arrives, these two cards will keep working and stay right where they are now. At this point, I'm into two Rocket cards for about what my wife spends at Starbucks every day and still going. And I shouldn't even look at it that way because they both paid for themselves within the first couple months.

I think the biggest flaw with the current Rocket is the fan assembly. It's just not robust enough and RED has unwittingly created a device that a huge number of users decided to take mobile or to rent out. I wince in pain every time I see mention of Rocket cards being shipped about or transported between jobs to be installed in other peoples' workstations. OMFG... Which also leads me to point #2. Why has RED not released a ruggedized portable Rocket with Thunderbolt interface??? And I'm not talking about placing the fragile PCIe card in a PCIe to TB enclosure. No, I'm talking about a streamlined, purpose-built Rocket device with Thunderbolt and power connectors on one side and a full array of HDMI, SDI and DisplayPort connectors on the other side.

Given the design of the new Mac Pro, I'm hoping RED will take a swing at this idea.
Click to expand...

I very respectfully disagree. First of all, red Rocket debayer is less clean and sharp - it seems to line-skip or something. Side by side on a 1080p master, you can see the difference. its' not gigantic, but it's there. Skin tones and texture especially seem to suffer

As such, I would never use Red Rocket for a final output.

Second, on a 12 core Mac pro, full debayer of FF 5K is about 6/1. So a one hour show will take 6 hours, a 30 second commercial will take 3 minutes This is far from the end of the world, unless you work in news, and even if you do...

The working resolution of 1/4 looks more than fine, while working, in both Premiere and DaVinci, so you only have to worry about rendering at the end.

Avid will be another story, as will Final Cut 07.
 
M Most said:
Actually, I feel just the opposite. The render quality with the Rocket is not equal to that of a software only render, even though the software approach takes a LOT longer. But for the final product, you want the best, and that is not what you get using the Rocket. The place for the Rocket in my world is for faster turnaround of viewing and editing material in a form that's "no questions asked" regarding any noticeable aliasing, stairstepping on edges, or other artifacts typical of 1/2, 1/4, and 1/8 resolution debayering. But for a final delivery, well, that's exactly where you don't want to be using the Rocket.
Click to expand...

While that's most certainly a valid point and we get a softer image out of the Rocket with no options for processing filters, DRX, NR, etc.. I've found the difference to be negligible when oversampling from 5K down to 4K. My work isn't destined for the big screen, though. 4K and multi-display is a huge part of what I do these days, but it's all corporate, much of it destined to video walls or environmental projections in corporate lobbies. Marketing and promotional or corporate documentary type production. So I guess the Rocket for final render output works for me most of the time.

As for the playback and working abilities during editing for the Rocket, I've found it to be a stumbling block at 1/4 res and lower. It's about a toss-up at half-res. On the current crop of 12 and 16 core systems, the system itself can crank out 1/4 res frames fast enough that the Rocket is just in the way. Most software handles this fine and the Rocket shows its worth here and there for when I need to move to a full-res image here and there for a frame or two as a spot-check or something.

As for the Rocket "appliance" approach, I've been advocating that for a long time, and now so is Apple, if a bit indirectly, with their new Mac Pro design. I have a feeling, though, that Red needs the higher bandwidth of TB2 to make this work reliably. So perhaps later in the year...
Click to expand...

I know it's been something they have discussed. I think with the new Mac Pro design as well as the increasing numbers of PCs with Thunderbolt ports, this will happen. And probably over a year away at this point... I think the old/ current Rocket would have done just fine here. It runs decently much of the time on a standard TB port, even though its bandwidth is theoretically halved. When it comes to running a mobile Rocket on a Macbook Pro or similar, the rest of the system isn't powerful enough to keep up with a Rocket running at full I/O capacity anyway. With Rocket-X, it will definitely need Thunderbolt 2, and that will be a bottleneck, but I still think it would be a device that makes a lot of sense. Thunderbolt 2 allows for port aggregation as well. RED could go crazy and put 2 or 3 TB interface ports on the thing to maximize its throughput. Thunderbolt 2 is just a re-allocation of Thunderbolt 1 port resources and an expansion of its abilities -- aggregation and bonding, mostly. Thunderbolt 3.0 draft notes were [intentionally] leaked about 10 days ago to the developer community. It's Thunderbolt2 with a significant increase in actual speed, but not due for over 2 years... Falcon Ridge 2 host to have TB3 in Q3 2016...
 
Jeff Kilgroe said:
...My work isn't destined for the big screen, though. 4K and multi-display is a huge part of what I do these days, but it's all corporate, much of it destined to video walls or environmental projections in corporate lobbies. Marketing and promotional or corporate documentary type production. So I guess the Rocket for final render output works for me most of the time.

As for the playback and working abilities during editing for the Rocket, I've found it to be a stumbling block at 1/4 res and lower. It's about a toss-up at half-res. On the current crop of 12 and 16 core systems, the system itself can crank out 1/4 res frames fast enough that the Rocket is just in the way. Most software handles this fine and the Rocket shows its worth here and there for when I need to move to a full-res image here and there for a frame or two as a spot-check or something.
Click to expand...

From what you said, it seems that for you, the dailies part is all about speed. For a number of us, that's only part of the requirement. We have to deal with clients that expect the dailies to look very much like the final product, and also to provide a mechanism for determining whether there are any issues with the original material. The moment they see something like stairstepping on edges - even if they see it on a 200% blowup - they start questioning whether those issues are in the original material. Besides, in many cases, especially with features, the dailies are used directly to create preview screening elements, which magnify those kind of things significantly. Even in television, studio execs and producers are sometimes looking at cuts on large LCD and plasma monitors, both in the cutting room and in their offices. So in our scenario, 1/4 res and lower is usually an unacceptable level, even for dailies. Different requirements, different needs.
 
YES PLEASE!

YES PLEASE!

Jeff Kilgroe said:
(snip) With Rocket-X, it will definitely need Thunderbolt 2, and that will be a bottleneck, but I still think it would be a device that makes a lot of sense. Thunderbolt 2 allows for port aggregation as well. RED could go crazy and put 2 or 3 TB interface ports on the thing to maximize its throughput. (snip)
Click to expand...

I strongly encourage RED to offer a RedRocket in a box product that includes at least two TB2 ports for use with the upcoming MacPros. Oh, and could I have it by next Tuesday :smilewinkgrin:

Cheers - #19
 
M Most said:
From what you said, it seems that for you, the dailies part is all about speed. For a number of us, that's only part of the requirement. We have to deal with clients that expect the dailies to look very much like the final product, and also to provide a mechanism for determining whether there are any issues with the original material. The moment they see something like stairstepping on edges - even if they see it on a 200% blowup - they start questioning whether those issues are in the original material. Besides, in many cases, especially with features, the dailies are used directly to create preview screening elements, which magnify those kind of things significantly. Even in television, studio execs and producers are sometimes looking at cuts on large LCD and plasma monitors, both in the cutting room and in their offices. So in our scenario, 1/4 res and lower is usually an unacceptable level, even for dailies. Different requirements, different needs.
Click to expand...

Speed is a concern, but it only needs to keep up with as fast as myself and one, occasionally 2, other people need to work. We don't really mess with dailies in a traditional sense. For smaller projects, I'm piecing my edits as I shoot.

I would say 90% of our projects edit from the native camera files. Takes longer to set up an off-line workflow and round-trip pipeline on a project than it does just to go to work and get it done. I don't typically have paranoid clients, producers, agencies, etc.. breathing down my neck while I work or scrutinizing imagery before it's finished.

For times when I do need the absolute best in render quality for final, and that does happen on occasion or a client will specifically request we do an software render from R3D for the reasons mentioned above, then we have a small render farm for that. It's not something you see mentioned on Reduser much and we really don't use it hardly at all, ever, for R3D processing, but it's there. I still feel that a Rocket render of 5K footage, delivered as 4K, is indistinguishable in most situations.

I'm championing 4K and mastering most everything in 4K/ UHD. None of my clients want 1080p anymore, but they're not pushing for 4K/ UHD. As I said above, a lot of it is corporate video type stuff. The iPad and Galazy Tab devices are big deals and usually a focal point for most our delivery, often involving custom app development. The Sony and other 4K UHDTV's are popping up in high-end corporate circles and I have a lot of clients asking me about them. But these are gadget-people with money, they're the types who will personally buy the 65" Sony 4K for their family room or their office and ask me about content. Yet they're still hiring me to deliver a custom app with native pixel sized content for the iPad.

That's just some of where I'm coming from, but it changes regularly. The only thing that seems to be a constant is that I'm not making movies destined for the cinema. Most of my stuff is short form, but occasionally we have the lengthy corporate production. I guess I just go where the mostly local market takes me. And my background is still in game development, CGI, some VFX as are many of my bigger industry contacts, so I occasionally pick up jobs along those lines.
 
Thank you all for the hints and for the discussion.
I think for now we will go for upgrading the CPUs and being happy with 1/4 res playback. A red rocket doesn't seem to make sense at the moment with the new Mac Pro and hints about GPU accelerated debayering around the corner.

Cheers
 
You must log in or register to reply here.
Back
Top