Thread: The New Mac Pro is Here!!!

Reply to Thread
Page 19 of 32 FirstFirst ... 915161718192021222329 ... LastLast
Results 181 to 190 of 314
  1. #181  
    Senior Member Blair S. Paulsen's Avatar
    Join Date
    Dec 2006
    Location
    San Diego, CA
    Posts
    5,136
    FWIW, I think RedUsers are the bullseye of the target market for the 7,1 MacPro. I fully expect a $12-20K MP 7,1 config to be fabulous for the workflows most of us use. I would further suggest that folks needing more compute power than that will need machine/server room racks anyway.

    If the mMP 7,1 is as fundamentally sound as it projects to be - easy access case design, premium 1400w PSU, support for 68 PCIe lanes, etc - and the ability to upgrade regularly - they should pay for themselves long before they are obsolete. Looking forward to seeing just what the Afterburner card can actually do, on paper its a beast.

    Cheers - #19
    Reply With Quote  
     

  2. #182  
    Senior Member
    Join Date
    Apr 2007
    Location
    NJ
    Posts
    609
    Quote Originally Posted by Marc Wielage View Post
    I have it on good authority it will be able to speed up other camera formats, too, before the end of the year. Possibly at product launch.
    As Spock would say...



    I didn't think they would base the card on FPGAs if they didn't plan on
    expanding functionality beyond the ProRes family.


    Brian Timmons
    BRITIM/MEDIA
    Reply With Quote  
     

  3. #183  
    Senior Member Antony Newman's Avatar
    Join Date
    Mar 2012
    Location
    London, UK.
    Posts
    1,579
    I think JL alluded to Proxy-Free R3D processing on the MMP / and a Metal optimised R3D SDK.
    +) I presume the first part of this refers to Apple breaking their own rules (by extending the OS to include 3rd party FPGA code)
    +) That RED will leverage the new OS feature to let it load balance Metal workload over multiple GPU / possibly even use the new ‘GPU channel’ (PCIe 3 based Infinity Fabric)

    If Apple believe that the GPU-GPU fabric is going to have a huge impact on the performance of the MMP, how would a MMP with 2xNAVI perform compared to 2xMDX modules (each with Dual Vega)?
    +) Each MDX connects to a single PCIe 3 x 16. Two x MDX (4 Vega) would have the same bandwidth as 2 x NAVI connected directly to the motherboard
    +) NAVI run ?50% faster than VEGA.
    +) NAVI Infinity Fabric (based on PCIe 4) is twice as fast as VEGA

    If 2xNAVI is 75% the performance of 4 Vega in MDX modules - this will make purchasing decisions interesting.

    And if the Speed-up is attributed to the Infinity Fabric - this will be a big win for Apple & AMD, who will benefit from 3rd party development for Metal and the IF respectively.

    AJ
    Reply With Quote  
     

  4. #184  
    Senior Member
    Join Date
    Jun 2017
    Posts
    1,756
    Quote Originally Posted by Marc Wielage View Post
    I have it on good authority it will be able to speed up other camera formats, too, before the end of the year. Possibly at product launch.
    At the higher end of the spectrum, the only codec that needs speeding up is .R3D at a low compression ratio. X-OCN is not that heavy, ARRIRAW is uncompressed logged, so also pretty light, Canon RAW is light and Panasonic VRAW is also light.

    In the midrange BMD's BRAW doesn't need it, Canon RAW light doesn't need it, Sony's FS7 XAVC-I doesn't need it, etc..

    Long GOP codecs(h264, h265, VP9) are pretty well handled by NVidia Turing and AMD Navi.

    So that leaves us kind of with .R3D at the camera side and AV1 at the delivery side(at their current resolutions) that might benefit from a speedboost.

    It will come down to the cost of the afterburner vs. the cost of CPU's and GPU's that currently handle these codecs.


    A fully specced Mac Pro incl. Afterburner card can only handle 3x ProResRAW(HQ?) 8k streams in realtime. ProResRAW HQ is atleast 4 times lighter on the CPU than R3D at a 3:1/5:1 compression ratio. With a Threadripper 2950x + 3 RTX titan X's you can achieve that today with 3 streams of 8k.R3D 5:1 24 fps. All the software that will see an optimalisation for Metal is already optimized for CUDA.
    Last edited by Misha Engel; 06-19-2019 at 06:41 AM.
    Reply With Quote  
     

  5. #185  
    Junior Member Daniel Jocher's Avatar
    Join Date
    Oct 2018
    Location
    Munich, Germany
    Posts
    27
    I didn't fully understand by now if this new R3D format will only benefit of the Mac Pro afterburner card or all Metal Based Macs with AMD GPUs? Did they clearly state that?
    Reply With Quote  
     

  6. #186  
    Senior Member Antony Newman's Avatar
    Join Date
    Mar 2012
    Location
    London, UK.
    Posts
    1,579
    Misha,

    Algorithms with few variables, small memory footprint and mathematics that resemble the native assembler of a CPU can run incredibly efficiently on a CPU.

    That same algorithm may or may not run efficiently on an GPU (depending if the algorithm can be expressed natively).

    If an algorithm can be expressed with higher level operations (eg FFT / massive matrix dot products / Wavelet processing) with fewer loops - it may end up being More efficiently expressed on an FPGA than a simpler algorithm that uses basic maths and large loop counts.

    https://www.researchgate.net/publica...elet_transform

    RED likely already knows what FPGA's can do in realtime on 28nm silicon [RRX]. Scaling that 14nm .. or if Apple have pulled out all the stops.. a 7nm monster .. would inform them of what the Afterburner could do in the world of Decrypt - Decompress - and Debayering.

    AJ
    Reply With Quote  
     

  7. #187  
    Senior Member
    Join Date
    Jun 2017
    Posts
    1,756
    Quote Originally Posted by Antony Newman View Post
    Misha,

    Algorithms with few variables, small memory footprint and mathematics that resemble the native assembler of a CPU can run incredibly efficiently on a CPU.

    That same algorithm may or may not run efficiently on an GPU (depending if the algorithm can be expressed natively).

    If an algorithm can be expressed with higher level operations (eg FFT / massive matrix dot products / Wavelet processing) with fewer loops - it may end up being More efficiently expressed on an FPGA than a simpler algorithm that uses basic maths and large loop counts.

    https://www.researchgate.net/publica...elet_transform

    RED likely already knows what FPGA's can do in realtime on 28nm silicon [RRX]. Scaling that 14nm .. or if Apple have pulled out all the stops.. a 7nm monster .. would inform them of what the Afterburner could do in the world of Decrypt - Decompress - and Debayering.

    AJ
    I pretty much know the ins and outs of CPU's, GPU's, FPGA's etc...

    The thing is that Apple showcased a fully specced MacPro, that can handle 3x 8k streams of ProResRAW.
    In the ProResRAW whitepaper Apple showed that ProResRAW is ~ 4 times lighter than R3D 3:1/5:1.

    Let's assume that 8k 24fps ProResRAW HQ needs the same compute power as 8k .R3D 5:1 24 fps.

    Estimating that the MacPro used in the demo cost around ~$30k+
    where a similar powered Threadripper + 3 RTX Titan X cost ~$15k.
    (256 GB DDR4 3200 cl14~ $2,800, cpu $850, AIO-cooler $250, MB with 2x 10G $500, 3 RTX Titans $7500,1 sata SSD as systemdrive 2TB $500, 2x 2TB NVMe SSD's $1,200, PSU 1500 Watt $ 350, Case $300, small material $150 ~ totals $15k)

    Both systems can handle the same workload at the same speed (3x 8k streams at the highest quality).
    All software mentioned during Apples WWDC is already(or in process of being) optimize for the NVidia GPU's, we just have to wait for NLE's implementing the RED+NVidia SDK(which is already available) for GPU decoding and I still haven't heard of any camera producing 8k ProResRAW.
    Reply With Quote  
     

  8. #188  
    Senior Member
    Join Date
    Sep 2009
    Location
    Northern VA
    Posts
    1,628
    Quote Originally Posted by Misha Engel View Post
    I still haven't heard of any camera producing 8k ProResRAW.
    The new Atomos Neon recorders can be upgraded to do 8K ProRes RAW with the appropriate cameras. That was announced with the new Mac Pro design and Afterburner cards.
    Reply With Quote  
     

  9. #189  
    Senior Member
    Join Date
    Jun 2017
    Posts
    1,756
    Quote Originally Posted by L. Langer View Post
    The new Atomos Neon recorders can be upgraded to do 8K ProRes RAW with the appropriate cameras. That was announced with the new Mac Pro design and Afterburner cards.
    I still haven't heard of any appropriate camera with enough SDI or HDMI outputs that can output an 8k RAW signal.
    Reply With Quote  
     

  10. #190  
    Senior Member Blair S. Paulsen's Avatar
    Join Date
    Dec 2006
    Location
    San Diego, CA
    Posts
    5,136
    With the Tokyo Olympics roughly a year away, I would expect the universe of shipping 8K cameras to expand rather quickly. Will any of them have 8KRAW spigots? Perhaps, but how many? I do expect several to have live 8K RGB (likely 4,320 x 7,680 16:9 raster) outputs for broadcast, probably only 8bit subsampled color to start.

    Over the last decade many have decried the focus on more pixels and wished for better pixels (starting with bit depth). Since sports coverage is leading the way to 8K at 60fps (or even 120) the first wave of devices will likely decimate color information rather aggressively to facilitate "manageable" signal transport data rates. AFAIK, RED DSMC2 is the only current camera platform that offers 8K and legit color information.

    Perhaps the more interesting discussion is what DSMC3 will offer. Will we break through the 300MB/s cap on DSMC2 mini-mag write speeds? My guess is 600MB/s with potentially a short cache based "burst" rate. We'll see. IAC, writing R3Ds is one thing - what about internally processed RGB signals? Some flavor of 8K ProRes seems likely, perhaps with frame rate limitations and 422 color, but still very useful on many assignments. I think low latency 10bit 4K RGB monitoring will come standard, but 8K???

    Back in DSMC1 days, RED made a 4K (actually UHD spec) broadcast module that included some processing muscle and a SMPTE fiber connector for hard lined EFP production. Total speculation alert - it seems like an 8K module designed to integrate with live broadcast infrastructure might be on the road map for DSMC3 to expand the market. If RED decides to create such a thing, it might have the secondary benefit of having enough added processing capacity to support 8K monitoring. While the engineers at RED might be able to get enough oomph inside the new brain to push out live 8K RGB, it seems like a stretch.

    FWIW, for the first time since the release of the RedOne, the road map for RED cameras is not obvious. Odds are that DSMC3 will be an evolution of DSMC2 - but there are other ways to go...

    Cheers - #19
    Reply With Quote  
     

Posting Permissions
  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts