Apple's Mac Pro gains new Radeon Pro MPX graphics card options

Posted:
in Current Mac Hardware edited August 2021
Customers can add new Radeon Pro W6800X and Radeon Pro W6900X modules to the Mac Pro, adding up to $11,600 to the base price.

The Mac Pro can be configured with new Radeon Pro graphics cards
The Mac Pro can be configured with new Radeon Pro graphics cards


Recent macOS updates have added support for the latest Radeon Pro graphics card modules to the Mac Pro and eGPU installs on Intel Macs. Customers can add these graphics cards when configuring a new Mac Pro, and a few hours after debut, made them available for upgraders as well.

New PCI-E and MPX module graphics cards available for the Mac Pro at time of purchase.

  • Radeon Pro W6800X with 32GB of GDDR6 memory costs $2,400

  • Two Radeon Pro W6800X with 32GB of GDDR6 memory each costs $5,200

  • Radeon Pro W6900X with 32GB of GDDR6 memory costs $5,600

  • Two Radeon Pro W6900X with 32GB of GDDR6 memory each costs $11,600

  • Radeon Pro W6800X Duo with 64GB of GDDR6 memory costs $4,600

  • Two Radeon Pro W6800X Duo with 64GB of GDDR6 memory each costs $9,600
Upgraders costs are $2800 for a Radeon Pro W6800X MPX module, $5000 for the W6800X Duo, and $6000 for the W6900X.

The Mac Pro uses standard PCI-E connections, so customers are able to purchase graphics cards separately at a lower cost. Though, due to market conditions, customers may find it difficult to locate Radeon Pro GPUs at retailers at regular price. However, the street price for two 6900XT video cards is still notably less than Apple's MPX module with the pair of GPUs.

AMD broke down some of the advantages of the new Radeon Pro graphics cards, citing the 7nm manufacturing and RDNA 2 architecture.

"We developed the AMD Radeon PRO W6000X series GPUs to unleash professionals' creativity and help them bring more complex and compute-intensive projects to life, from animating 3D film assets to compositing 8K scenes to game development," said Scott Herkelman, corporate vice president and general manager, Graphics Business Unit at AMD. "The new AMD Radeon PRO W6000X series is packed with remarkable energy efficiency, enhanced compute units and a new visual pipeline, enabling Mac Pro users to do more in less time across a broad range of pro applications."

The company boasts that the GDDR6 memory has a speed of up to 512GB/s for data-intensive professional apps. The Radeon Pro W6900X is capable of operating at 22 teraflops alone, and the Radeon Pro W6800X can achieve up to 30.2 teraflops when used in the duo configuration.

The update was first spotted by CNN Underscored editor Jake Krol. He notes that the new graphics cards will provide as much as 23% faster performance in DaVinci Resolve and 84% performance improvements in Octane X.

Read on AppleInsider

Comments

  • Reply 1 of 9
    crowleycrowley Posts: 10,019member
    Seems odd that the better card is only available with the lesser amount of memory.
  • Reply 2 of 9
    zimmiezimmie Posts: 598member
    crowley said:
    Seems odd that the better card is only available with the lesser amount of memory.
    All of the new options are 32 GB per GPU. The ones which show 64 GB have two GPUs on one card.

    The single-card prices are right in line with retail for the Radeon Pro line. It's more than a little weird that one the pairs are all $400 more expensive than buying two cards individually, though. Maybe that's the cost of the "Infinity Fabric Link connector"?
    dysamoriawatto_cobra
  • Reply 3 of 9
    mcdavemcdave Posts: 1,904member
    Modest brute force gains over the Pro Vega II but this Barefeats article shows even the non-X version holding its own against the consumer(?) cards;
    https://barefeats.com/pro-w6800-versus-other-gpus.html

    It’ll be interesting to see what Metal magic Apple has to pull to compete with its own Silicon.
    watto_cobra
  • Reply 4 of 9
    crowleycrowley Posts: 10,019member
    zimmie said:
    crowley said:
    Seems odd that the better card is only available with the lesser amount of memory.
    All of the new options are 32 GB per GPU. The ones which show 64 GB have two GPUs on one card.

    The single-card prices are right in line with retail for the Radeon Pro line. It's more than a little weird that one the pairs are all $400 more expensive than buying two cards individually, though. Maybe that's the cost of the "Infinity Fabric Link connector"?
    Got it, I missed the word Duo in the latter two W6800X options.  That's a whole lot of GPU.  Apple Silicon has a fair bit of work to do.
  • Reply 5 of 9
    zimmie said:
    crowley said:
    Seems odd that the better card is only available with the lesser amount of memory.
    All of the new options are 32 GB per GPU. The ones which show 64 GB have two GPUs on one card.

    The single-card prices are right in line with retail for the Radeon Pro line. It's more than a little weird that one the pairs are all $400 more expensive than buying two cards individually, though. Maybe that's the cost of the "Infinity Fabric Link connector"?
    No stats on power - perhaps a PSU in the mix...?

    nichegamer.com/2021/07/27/high-end-gaming-pcs-banned-in-six-us-states-after-california-energy-bill-limits-sales-on-high-performance-pcs/
  • Reply 6 of 9
    crowleycrowley Posts: 10,019member
    zimmie said:
    crowley said:
    Seems odd that the better card is only available with the lesser amount of memory.
    All of the new options are 32 GB per GPU. The ones which show 64 GB have two GPUs on one card.

    The single-card prices are right in line with retail for the Radeon Pro line. It's more than a little weird that one the pairs are all $400 more expensive than buying two cards individually, though. Maybe that's the cost of the "Infinity Fabric Link connector"?
    No stats on power - perhaps a PSU in the mix...?

    nichegamer.com/2021/07/27/high-end-gaming-pcs-banned-in-six-us-states-after-california-energy-bill-limits-sales-on-high-performance-pcs/
    Anyone buying the Mac Pro as a gaming machine needs their head checked.  Apple doesn't market it as one.
  • Reply 7 of 9
    MplsPMplsP Posts: 3,609member
    crowley said:
    zimmie said:
    crowley said:
    Seems odd that the better card is only available with the lesser amount of memory.
    All of the new options are 32 GB per GPU. The ones which show 64 GB have two GPUs on one card.

    The single-card prices are right in line with retail for the Radeon Pro line. It's more than a little weird that one the pairs are all $400 more expensive than buying two cards individually, though. Maybe that's the cost of the "Infinity Fabric Link connector"?
    Got it, I missed the word Duo in the latter two W6800X options.  That's a whole lot of GPU.  Apple Silicon has a fair bit of work to do.
    That’s why I wonder if Apple won’t develop a PCI-e interface. Apple Si’s graphics abilities are hardly sub-par and are fine for the majority of people, but for the people who truly need high end graphics, it seems like it would make more sense to let AMD do the development rather than focus a large amount of resources to develop an Apple-only solution that only a handful of customers would buy. 
    muthuk_vanalingamnadrieldysamoria
  • Reply 8 of 9
    MarvinMarvin Posts: 14,735moderator
    MplsP said:
    crowley said:
    zimmie said:
    crowley said:
    Seems odd that the better card is only available with the lesser amount of memory.
    All of the new options are 32 GB per GPU. The ones which show 64 GB have two GPUs on one card.

    The single-card prices are right in line with retail for the Radeon Pro line. It's more than a little weird that one the pairs are all $400 more expensive than buying two cards individually, though. Maybe that's the cost of the "Infinity Fabric Link connector"?
    Got it, I missed the word Duo in the latter two W6800X options.  That's a whole lot of GPU.  Apple Silicon has a fair bit of work to do.
    That’s why I wonder if Apple won’t develop a PCI-e interface. Apple Si’s graphics abilities are hardly sub-par and are fine for the majority of people, but for the people who truly need high end graphics, it seems like it would make more sense to let AMD do the development rather than focus a large amount of resources to develop an Apple-only solution that only a handful of customers would buy. 
    Unified memory benefits high-end tasks, other manufacturers are using complex copying, caching and high bandwidth link systems to get the same benefits:

    https://developer.nvidia.com/blog/beyond-gpu-memory-limits-unified-memory-pascal/

    Bandwidth between CPU and GPU is why Nvidia built a custom ARM CPU for the data center:

    https://www.nvidia.com/en-us/data-center/grace-cpu/



    If they have external, 3rd party GPUs, they have to keep copying data between system memory and the GPU memory. It saves memory capacity and bandwidth sharing the memory.

    While they could just switch the Intel CPU out in the Mac Pro with an M1X style chip and connect it to some AMD GPUs, they'd be paying AMD thousands of dollars to resell a slower GPU than they can make themselves and have to deal with driver issues.

    They don't need a lot of chip designs, it's not like they need 20 different CPU models like AMD/Intel per generation.

    M1 covers mini, MBA, 13" MBP, 24" iMac = 2.6TFLOP (10-20W)
    M1X covers 27" iMac, 16" MBP = 5/10 TFLOP (40-60W)
    M1X Duo/Quad covers high-end 27" and Mac Pro 20/40 TFLOP (150-300W) (on 3nm, this can go up to 60TFLOP or more)

    I think they'd have to do a custom chip for a Mac Pro anyway just for the CPU performance. 2x the M1 CPU isn't near fast enough for the Mac Pro (needs over 5x) and even 4x would use too much power for the MBP. As soon as they have to do a custom part, it makes sense to design the whole chip package to suit the target power.

    Apple could even put multiples of the M1X chip on a board, that's how old machines like the G5 Quad were setup and have fast communication links between the chips but it should be easy enough for them to build a bigger single chip package. I don't see any reason to support 3rd party GPUs any more. On 3nm Apple can build a 60TFLOP chip (this would be 2x a Nvidia 3090), on 2nm 90TFLOP. For real-time use, there's no need for anything faster than this, for offline processing there is but people buy multiple machines for offline processing and the 3rd party options won't be significantly faster and they won't have the benefit of unified memory.
    edited August 2021 watto_cobrafastasleep
  • Reply 9 of 9
    This news is enough to propel AMD for a $6 gain. LOL
Sign In or Register to comment.