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 :)

The new iMac Pro able to handle 8K files with a breeze?

External GPU... That’s just more money spent to place a GPU onto an external interface that only provides 1/4 of its potential bandwidth and incurs a huge penalty in terms of latency. I’m personally not a fan... It’s an acceptable or even welcome solution to get a nice boost in GPU ability for a notebook when you return back to an office setup or similar. The iMac isn’t really a portable solution, it’s built for the desktop. If you need to place two or three external boxes next to your computer to get the desired GPU and storage capability, then why buy the iMac? Serious question, what makes the iMac so attractive to you that you’d want to keep spending money and cluttering up a desk with compromised additions? I’m asking because I’m trying to understand and help.

I’m not necessarily trying to steer you guys away from the iMac Pro. However, it’s not the ideal solution for a lot of needs in this community.
 
Jeff Kilgroe said:
External GPU... That’s just more money spent to place a GPU onto an external interface that only provides 1/4 of its potential bandwidth and incurs a huge penalty in terms of latency. I’m personally not a fan... It’s an acceptable or even welcome solution to get a nice boost in GPU ability for a notebook when you return back to an office setup or similar. The iMac isn’t really a portable solution, it’s built for the desktop. If you need to place two or three external boxes next to your computer to get the desired GPU and storage capability, then why buy the iMac? Serious question, what makes the iMac so attractive to you that you’d want to keep spending money and cluttering up a desk with compromised additions? I’m asking because I’m trying to understand and help.

I’m not necessarily trying to steer you guys away from the iMac Pro. However, it’s not the ideal solution for a lot of needs in this community.
Click to expand...

I'm not planning on getting an iMac Pro. Currently I'm running a Hackintosh, but it's a pain for upkeep and stability, I'm waiting on the new Mac Pro to arrive next year hopefully.

But there is a market for it. I can imagine people doing less-demanding editing and colouring jobs, i.e. non-8K, or ProRes work - which is the majority of editing work out there - would be able to do very well with the iMac Pro, and able to pay it off within a few months. The graphics card, while not the best, doesn't suck either. On my hack, I was able to edit and do basic colouring of 8K Helium footage at 1/4 quality with an i7 6700K and an RX480, so I'm sure this computer would still do great.

Editing 8K RAW at full quality, high-level colouring and VFX are entirely another kettle of fish, which I guess comes back to the original thread topic :crazy:
 
Even with TB3 it's still just 1/4 of its bandwidth?
My thought with eGPU for the iMac Pro is basically just to make it even faster.
When readings these tests there are somethings to gain when using eGPU + iMac + Davinci Resolve:
http://barefeats.com/imac17egpu.html

Thanks,
V
 
Vincent Larsson said:
Even with TB3 it's still just 1/4 of its bandwidth?
My thought with eGPU for the iMac Pro is basically just to make it even faster.
When readings these tests there are somethings to gain when using eGPU + iMac + Davinci Resolve:
http://barefeats.com/imac17egpu.html

Thanks,
V
Click to expand...

Also, just wondering, the bandwidth doesn't necessarily affect the compute speed? I thought PCIe bandwidth would be in regards to with memory fill, such as loading textures, which is more critical with gaming? I could be wrong, but wouldn't DaVinci/AE/Premiere be more dependent on the GFLOPS rather than PCIe bandwidth?
 
Homer Dulu said:
Also, just wondering, the bandwidth doesn't necessarily affect the compute speed? I thought PCIe bandwidth would be in regards to with memory fill, such as loading textures, which is more critical with gaming? I could be wrong, but wouldn't DaVinci/AE/Premiere be more dependent on the GFLOPS rather than PCIe bandwidth?
Click to expand...

Resolve leans heavy on mem. bandwith of the video card (even more than on the number of TFlops).
With R3D you first have to decompress etc. so with RED you would also need a fast CPU (more than with CinemaDNG or ARRIRAW).

Looking forward to a Resolve Benchmark with the TESLA GV100 with 900GB/s bandwith.
 
Misha Engel said:
Resolve leans heavy on mem. bandwith of the video card (even more than on the number of TFlops).
With R3D you first have to decompress etc. so with RED you would also need a fast CPU (more than with CinemaDNG or ARRIRAW).

Looking forward to a Resolve Benchmark with the TESLA GV100 with 900GB/s bandwith.
Click to expand...

Got it, thanks!
 
Quoting this from my response in the eGPU discussion thread in the hardware forum:

Jeff Kilgroe said:
Real world performance....

First of all nVidia GPUs are said to get support soon, but nothing official yet. Some support has shown in certain developer beta driver sets and same with OS 10.13 releases. Unfortunately, the only nVidia cards that are working are those with a modified BIOS to add specific OS X support.

Beyond that, assuming your eGPU enclosure can fully support your desired card, we have to accept the limitations of the Thunderbolt / PCIe bridge interface. Thunderbolt 3 ports are 40Gbps capable, but that is not always the case here. TB3 exists as either single or dual ports per header. The dual header is backed by 8 lanes of PCIe v3 and can supply either of the ports at a full 40Gbps of PCIe bandwidth, but not both at the same time. A single TB3 port is backed by 4 lanes of PCIe v3. 4 lanes = 3.94 GB/s which actually translates to 3.88GB/s after the 128b/130b transmission overhead. What that means is you have a real world 32Gb/s to work with and not the 40Gb/s advertised.

Don’t take that as a bad thing regarding the available bandwidth, that is the bandwidth of a PCIe v3 X4 slot.

Most all current GPUs are PCIe v3 X16. Only the latest, most powerful ones really utilize all that bandwidth and even then, only for large memory dumps, loading data sets to onboard memory, stuff that’s pretty darn quick and not computationally demanding. For computational tasks the X4 bandwidth is going to be fine in many cases. Typically the more complex the computation and longer it takes, the less of a bottleneck the PCIe interface. Some operations that the GPUs have become very good at handling like Bayer pattern processing, it’s possible for that X4 interface to become a bottleneck. Although, the bulk of computations people in this community are likely to run are not going to be an issue as long as current GPUs are kept to one per TB3 port.

The biggest hinderance to eGPUs is latency. For me the eGPUs feel sluggish for initial response. But once everything is transferring and flowing along, you’re fine, not a big deal that it took a handful of milliseconds longer to get started. If the software is not well suited to deal with this or you’re dealing with large data sets or evolving / real-time frame data that has to be loaded in and out of the GPU frame buffer from the PCIe bus, this can be a significant performance suck.
Click to expand...

As Misha points out, for R3D processing (in Resolve and elsewhere that CPU+GPU support exists) both a powerful CPU and GPU combination is ideal. If you have too much GPU your CPUs won’t be able to feed them. If you don’t have enough GPU power, then you won’t be able to use your CPU to its full potential. Even if you have plenty of CPU cores, if they operate at too low of a frequency, it may not be enough to drive your GPUs unless you pile on even more cores. It’s all very maddening at times and makes it very difficult to spec out an ideal computer that can handle a wide variety of tasks or workflows.

For the new iMac Pro, if you want to add two additional Vega 64 GPUs to it, plan on having each on their own TB3 port in their own eGPU enclosure, or one that takes two cards off two separate TB3 connections. Really becomes a head scratcher in terms of logical approach to the problem. Not the right system for the job.
 
Misha Engel said:
Resolve leans heavy on mem. bandwith of the video card (even more than on the number of TFlops).
With R3D you first have to decompress etc. so with RED you would also need a fast CPU (more than with CinemaDNG or ARRIRAW).
Click to expand...

If your monitor is connected to the external GPU box, then you get a significant savings in bandwidth usage for interactive work because the GPU doesn't need to send data back over the Thunderbolt connection in order to render it. For rendering though, the machine needs to send the image data to the GPU and get the processed frames back to save them out.
 
Okay this is probably a silly/stupid question: but is there a way to remove some of the "throttling back" of the Vega GPU in the iMac Pro? I guess I need explanation on what's going on with it. I did order an 18C, 64 vega machine. I'm not too worried...just wondering. :-)
 
Great thread, props to Jeff and the rest of the NERD army (that's a compliment, seriously) for diving into what the iMacPro is, and isn't.

IMO, if the 10c/64GB/Vega64/xTB SSD is not enough to meet your needs then take a different path. Pushing it up to 18 slower cores, adding eGPU boxes, etc in an attempt to make it something it's not doesn't make sense to me. I appreciate folks like Zeb who, like me, vastly prefer OSX, going full pimp to get the best user experience. For Mac folk, the other part of weighing options is the mMP due next year.

Example: I plan to get the mMP next year, but realize it could be late next year and expensive. So I see the iMacPro as a bridge to that day, and would rather spend $7K and save the other $5K-ish to put into the mMP down the road. If, OTOH, you plan to rock the iMacPro as your main machine for several years - then the specced up version might be right for you, even if you overpay a bit for certain upgrades.

Ultimately, a machine that serves your needs is the goal - not winning a bang for the buck bragathon on the internet.

Cheers - #19
 
Scot Yount said:
this....+1000
Click to expand...

You must be the lucky ones, I'm on a budget. So no Monstro, Optimo Ultra 12x or 5 full iMacPro's for us this year.
But still pretty happy with the cheap nerd stuff that sometimes even beats the happy few equipment.
 
Misha Engel said:
You must be the lucky ones, I'm on a budget. So no Monstro, Optimo Ultra 12x or 5 full iMacPro's for us this year.
But still pretty happy with the cheap nerd stuff that sometimes even beats the happy few equipment.
Click to expand...

Fair enough. Due respect to clever ways to keep costs down. In my use case, the advantages of a powerful, reasonably portable iMacPro offset the higher initial cost - easily. For other use cases, perhaps not. The value of this thread is to know more about the under the hood realities vs marketing-speak. To me, the takeaway is that the iMacPro makes sense for many RedUsers in it's mid-line (10c/64GB/Vega64) trim - particularly for owner/ops and small shops that have neither the budget for a big bomber workstation or the IT chops to manage alternatives the way Misha does.

If you have $10K to spend and need lotsa CUDA cores for key apps, the iMacPro is not the droid you're looking for. For RedUsers who are more artist than tech wizard...

Cheers - #19
 
Ronald Williams said:
Okay this is probably a silly/stupid question: but is there a way to remove some of the "throttling back" of the Vega GPU in the iMac Pro? I guess I need explanation on what's going on with it. I did order an 18C, 64 vega machine. I'm not too worried...just wondering. :-)
Click to expand...

Not likely. It's throttled for thermal reasons; in other words, it's basically a lot of wasted engineering solving the wrong problem. The focus should have been on a genuine macpro, not wasted on an inferior imitation of a Surface Studio (which is also in dire need of a refresh, and backed by a company that's still too dumb to adopt Thunderbolt).

Still, I don't understand why anyone cares. While editing, 8K scaled to fit the source and record windows even on a 4K monitor is indistinguishable from Redcode debayered at 1/16th resolution. Since it's easy enough to set the debayer resolution to 1/16th to edit and to whatever you want to render, it seems a bit ridiculous to worry about editing 8K.

Sure, if you zoom in a lot you'll get some pixilation, but since it won't be there in the final render, who cares?

The ability to debayer to basically any power of 2 fractional resolution of the original footage (within reason) is one of Red's biggest benefits. I'm color grading a feature film on an Ultrabook (a crappy Razer Blade Stealth 2016 edition) with an egpu right now, and it's working out. The render will be slow due to having just two cores, but for now this is due to budget constraints... longer term I'll get a workstation with a Threadripper or a Core i9 or Xeon and 2 Vega GPUs and have nice quick renders, but that will take a bit of time to save up for.
 
Scot Yount said:
Simple. Just make your project 2K, or 1080, or whatever. I do it all the time. We shoot 8KHD and I edit in a 1080 timeline. If I want more pixels when I export, i just change the settings in the project. And btw, I never transcode, or make proxies. I also leave background rendering off, so nothing is rendered until I finish my project. I just set the machine to "Better Performance" and go. The only time I have any trouble, is when I have like 8 callout graphics in one shot....My edit station is a late 2013 MacPro with 12 cores over Thunderbolt 1 to some old Promise Pegasus R6 drives set to RAID 5.
Click to expand...

I just got off the phone with Apple Pro Apps team. They told me if I import 8K footage as 2K and try to export to 8K that won't work. What they said is that I can import at 8K and create a proxy and work off that for faster editing and then export to whatever size I want at 8K or lower. Maybe I misunderstood what you all were describing.
 
I think the key phrase here is "import as 2K". Don't do that. You may wish (highly recommended) to scale those 8K images within an editorial timeline, something FCP-X and other NLEs typically manage for you in the background using terms like "optimized" or "proxy". You also have the option of authoring a low rez version manually for editorial, then simply linking back to the original R3Ds for output if you'd rather.

My advice is to import the R3Ds as they are and use the settings in the NLE to choose a "working" resolution independent of how large the raster might be in the R3D. Most NLEs default to creating output from the source files - not the lightweight editorial version - without any further intervention on your part. I think you're making it harder than it is.

Cheers - #19
 
adam brennan said:
I just got off the phone with Apple Pro Apps team. They told me if I import 8K footage as 2K and try to export to 8K that won't work. What they said is that I can import at 8K and create a proxy and work off that for faster editing and then export to whatever size I want at 8K or lower. Maybe I misunderstood what you all were describing.
Click to expand...

As Blair said below, it's easy. Either use the NLE's proxy workflow, which if you're in FCPX or Resolve is VERY easy, or set the de-Bayer resolution to 2K in the Red RAW settings. I don't know how to do that part yet in FCPX, but in Resolve it's a project settings option that you can override for a given clip if you need to.

By doing this I'm able to work with Redcode on an ultrabook. I was even able to work with 8K raw on my buggy Razer Blade Stealth with a pansy dual-cure CPU and the pitiful integrated Intel GPU. All I had to do was set the de-Bayer resolution to 1/16, which in a source or record window is indistinguishable from scaled 8K, especially while editing. Of course you'd want to set the de-Bayer resolution higher for delivery, but then you don't need it to be interactive any more.
 
Blair S. Paulsen said:
I think the key phrase here is "import as 2K". Don't do that. You may wish (highly recommended) to scale those 8K images within an editorial timeline, something FCP-X and other NLEs typically manage for you in the background using terms like "optimized" or "proxy". You also have the option of authoring a low rez version manually for editorial, then simply linking back to the original R3Ds for output if you'd rather.

My advice is to import the R3Ds as they are and use the settings in the NLE to choose a "working" resolution independent of how large the raster might be in the R3D. Most NLEs default to creating output from the source files - not the lightweight editorial version - without any further intervention on your part. I think you're making it harder than it is.

Cheers - #19
Click to expand...

This. Well put.
 
You must log in or register to reply here.
Back
Top