AMD spills the beans on Radeon Pro chips inside the new Macs

Yesterday's refreshes of the iMac and MacBook Pro were highlighted by the switch from Intel sixth-generation Skylake Core processors to seventh-generation Kaby Lake CPUs, but models with discrete graphics cards also get a boost to AMD Radeon Pro 500-series graphics chips. Let's have a look at the details of these refreshed Polaris parts.

Most MacBook Pro machines come with Intel IGPs, but higher-end 15" units come with AMD Radeon Radeon Pro 555 or Radeon Pro 560 graphics chips onboard. Both chips are likely based off the Polaris 21 chip inside the desktop Radeon HD 560. For reference, that chip has 1024 stream processors at a 1275 MHz boost clock.

The Radeon Pro 555 packs 1.3 TFLOPS of peak compute power courtesy of 768 SPs packed into 12 compute units. Those figures would imply a boost clock of 850 MHz. The bigger Radeon Pro 560 has 1024 SPs in an array of 16 CUs, offering up a total of 1.9 TFLOPS, indicating that the boost clock is presumably set at 925 MHz. Both graphics options list 81 GB/s of memory bandwidth, which works out to 5 GT/s GDDR5 memory on a 128-bit-wide memory bus. The Radeon Pro 555 comes with 2 GB of its own memory, while the Radeon Pro 560 doubles that figure at 4 GB.

Updated iMacs are available with three new AMD graphics chip options: the Radeon Pro 570, Radeon Pro 575, and Radeon Pro 580. All three appear to be built around the same Polaris 20 chip found in the desktop Radeon RX 570 and Radeon RX 580 cards. Memory bandwidth is set at 217 GB/s for the whole trio, a measure that works out to 6.8 GT/s GDDR5 on a 256-bit bus. For reference, the standard Radeon RX 570 packs 2048 SPs boosting to 1244 MHz and the Radeon RX 580 has 2304 at a 1340 MHz boost clock.

The Radeon Pro 570 has been knocked down to 1792 SPs (in 28 CUs) offering up to 3.6 TFLOPs, which should mean a 1.0 GHz base clock according to a little napkin math. The Radeon Pro RX 575 delivers up to 4.5 TFLOPS from its 2048 SPs or 32 CUs, suggesting a 1100 MHz boost clock. Both those chips get 4 GB of RAM. Lastly, the Radeon Pro 580 has 2304 SPs (thus 36 CUs) cranking out up to 5.5 TFLOPS, indicating a 1.2 GHz boost clock. Much like its desktop counterpart, the Pro 580 has 8 GB of RAM on tap.

The 768 SP configuration in the Radeon Pro 555 is not entirely novel. The same figure was found in the Radeon RX 455, though the 455 was built around older Bonaire silicon. Similarly, the 1792 SP layout in the Radeon Pro 575 has also been seen before in the Polaris 10 Radeon HD 470D sold exclusively in East Asian markets. All Radeon Pro 500-series chips appear to be Apple-exclusive, at least for now.

Comments closed
    • ermo
    • 2 years ago

    Could someone knowledgeable on the matter please explain why Apple *insists* on doing Metal even now that Vulkan exists and will incorporate OpenCL?

    What are the benefits for Apple?

      • tipoo
      • 2 years ago

      I agree in wishing they did add it, but I assume their reason is that Vulkan is a much more explicit API than Metal, you’re responsible for everything, while Metal, like older APIs, does a lot of the behind the scenes stuff for you.

      Between efficient and convenient, Vulkan hit the first, while Apple wanted both, so Metal is convenient to write while wringing out most of the efficiency of lower level APIs.

      It’s also now interesting that they’re going to be making their own GPUs without Imagination Tech, so maybe by forcing their own API they can reduce considerations for other APIs, focus on just one.

      Now that they have engine partners on board, whatever gets rid of their 6 year old OpenGL faster, man.

    • ronch
    • 2 years ago

    What amazes me about Apple is how they can put literally ANYTHING inside their stuff and people would still be all over them. They could put a Mediatek SoC in an iPhone and sell it for $800 and the RKOI would still be all over it (hope they at least put an X20 in there tho). They could buy up all the unsold stock of FX chips, put them in the latest Macs and nobody would notice the loud fan. OK, maybe reviewers will catch these things and cry foul like there’s no tomorrow but then again I suppose the Apple crowd doesn’t read tech sites. Maybe they’ll see them in an FHM or lifestyle magazine and buy one right away and swear they’re faster and more amazing, 8 cores being better than their old quad i7 Macs.

    • ronch
    • 2 years ago

    Polaris 2.0 just became amazing because it’s now inside Apple computers.

      • End User
      • 2 years ago

      Not amazing. Magical.

        • ronch
        • 2 years ago

        It’s all thanks to the RDF™ chip!!

    • heinsj24
    • 2 years ago

    So anyone else can squeeze a full-sized Polaris 10 into a laptop with no more than a down-clock; Apple has to cripple them to fit them into monitor.

      • Spunjji
      • 2 years ago

      The iMac cooling system is as dumb (and cheap) as a brick. They pretty much admitted so by making the revisions they for the iMac Pro.

      Knock the overpriced Xeon nonsense out of that and you’d have a decent high-end system, but nooooo. Bad for the margins.

    • NoOne ButMe
    • 2 years ago

    And the kicking of Nvidia to the curve finalizes at Apple.

    Or I suppose you could disagree and say Polaris beats Pascal on performance per watt. I think it is close when Polaris is binned for power. But not better.

      • tay
      • 2 years ago

      This has to do with OpenCL. Polaris is good enough that Apple doesn’t want to switch to CUDA/Nvidia.

        • jts888
        • 2 years ago

        Can somebody remind me again what happens after Khronos finishes rolling OpenCL completely into Vulkan, which at this point is notably only [i<]not[/i<] supported by Apple?

          • chuckula
          • 2 years ago

          Ironically for all the OpenCL talk Apple really doesn’t support OpenCL 2.0.

            • Namarrgon
            • 2 years ago

            Yeah you know their heart isn’t in it when even Nvidia has better OpenCL 2.0 support (on Windows at least). Seems it’s Metal all the way from here on.

      • Voldenuit
      • 2 years ago

      I think you’d need a lot of RDF to claim that Polaris is better than Pascal at Perf/Watt.

    • ET3D
    • 2 years ago

    Look to be exactly the same as the Radeon Pro 455 and 460. Perhaps the 560 is a tad (20MHz) faster than the 460, but the 555 and 455 have the same TFLOPS value. So these are basically pure rebadges.

    • Jtaylor1986
    • 2 years ago

    Nothing like a super expensive laptop with a low end graphics card in it.

      • Shobai
      • 2 years ago

      It kinda looks very much like that, actually…

    • Tumbleweed
    • 2 years ago

    “are are”? Grammar checkers are your friend, my friend. 🙂

      • EzioAs
      • 2 years ago

      Arr, shave ya belly with a rusty razor! We’re celebratin pirate’s day thar.

      • ronch
      • 2 years ago

      Arrrrrrrr!!!! /pirate voice

    • Chrispy_
    • 2 years ago

    The thing I found amusing about the new announcements from Apple is that they’re making a big song and dance about VR support on the iMac and MBP.

    Underclocked equivalents of a $100 entry-level card are not going to power the HTC vive that Apple prominently displays alongside the VR section of their announcement.

    The iMacs should be a little better off at least, but the chips are still slower than 2016’s 470 and 480 due to the lower clocks, and the full-speed RX480 used to be the minimum recommended GPU for the HTC Vive, [i<]back in 2016[/i<].

      • nico1982
      • 2 years ago

      They also anounced a TB3 external enclosure with a Radeon 580 🙂

        • defaultluser
        • 2 years ago

        Yeah, this is the only real benefit of this whole exercise, the first OFFICIAL GPU upgrade that Apple has allowed in four years!

        The rest of this will be quickly forgotten after Apple loses interest and stops the money flow. Apple just treading water as-usual, while trying to convince a larger number of developers to jump into the Metal pond. You know, that same pond where the let OpenCL and OpenGL rot from disuse.

          • Chrispy_
          • 2 years ago

          The OpenCL and OpenGL rot is causing a lot of strife among my game-dev circle of friends. Most of them just mutter “f*** Apple” at every opportunity.

          The one thing that comes from all this is that Apple might be the first company to design an external Thunderbolt GPU dock that isn’t a joke. I’d buy one if I could use it with a modern ultrabook.

            • Spunjji
            • 2 years ago

            Want to bet it costs $699?

            • NoOne ButMe
            • 2 years ago

            [url<]https://developer.apple.com/development-kit/external-graphics/[/url<] $599 included a Rx 580 8GB and USB hub. So $350 or so (hub is around 30-40, 580 around 200-225). You could call it under 300 if you use current 580 prices, but that is stupid for price prediction. $300-400 retail, maybe like $350 with no accessories, $400 including USB hub.

        • slowriot
        • 2 years ago

        Ahan extra cost solution to reach the bare minimum VR requirements. Sounds awful to me.

      • Voldenuit
      • 2 years ago

      [quote<]The iMacs should be a little better off at least, but the chips are still slower than 2016's 470 and 480 due to the lower clocks, and the full-speed RX480 used to be the minimum recommended GPU for the HTC Vive, back in 2016.[/quote<] And don't forget HardOCP's VR scorecard, which showed that the 480 was inadequate for VR due to the number of dropped and reprojected frames. Then again, it's possible (likely, even) that Apple is targeting VR usage other than gaming. Which I think VR needs to grow beyond niche use.

      • ImSpartacus
      • 2 years ago

      Remember the 290 (and 970) is the official minimum.

      The 480 exceeds that minimum by a small margin, so that’s worth noting.

      Practically speaking, I agree that none of these GPUs are ideal for an $800 Vive. However, when we’re talking about GPUs that might barely meet the official spec, we want to remember the original source of the spec.

    • chuckula
    • 2 years ago

    TR may have another article but on the Vega side Apple also announced that the version of Vega coming to the iMac Pro has in HBM2 providing 400GB/sec of memory bandwidth in addition to the “22Tflop” number (half-precision) performance number.

    It appears to be a small step down from the highest-end Vega parts but still pretty impressive for something that’s being crammed into an iMac chassis next to a huge CPU.

      • tipoo
      • 2 years ago

      Yup, sounds like they just turned the clocks down a bit to hit a more favorable efficiency point to fit it in that chassis, but otherwise it’s the high end hardware.

      • ptsant
      • 2 years ago
    • DragonDaddyBear
    • 2 years ago

    If you boot this into Windows would you get the “Pro” features of an AMD GPU?

      • tipoo
      • 2 years ago

      It shows up as Radeon Pro in Windows. There’s no pro “features” so much as software being tested against the driver though.

      • nico1982
      • 2 years ago

      Mine is still seen as 460 Pro into Bootcamp. The only “Pro” feature I’m aware of is the peculiar combination of SP and clock – more and lower, respectively – over the plain model.

        • DragonDaddyBear
        • 2 years ago

        OK. I thought there were certain features enabled in drivers for some applications when the card is a “professional grade” card. I guess this is just a branding marker, not an actual “pro” card. Like the old 9700 Pro.

          • tipoo
          • 2 years ago

          It’s just tested against some professional software titles, but there’s nothing different on a silicon feature level and nothing software locked out

            • DragonDaddyBear
            • 2 years ago

            Thanks. I’m not into that kind of work so i honestly don’t know. Does the green team do that out am i just that confused?

            • tipoo
            • 2 years ago

            They both do in the regular people tiers of cards, only once you get high end does the silicon start being different, some support ECC memory, Nvidia has some (more?) fully enabled chips that haven’t become Geforces iirc. AMD unified them more and the consumer cards have a fair amount of compute grunt, but I assume they too have some top tier chips not in Radeons.

            If you look at say a Firepro M5100 it’s identical to a 8870M practically, with minor clock differences. The D700s in the Mac Pros just showed up as Radeons under Windows!

    • tipoo
    • 2 years ago

    So 0.1Tflop gain vs the previous RP 460.

    Hopefully the new models push refurb models of the old one lower, as there’s not a whole lot of an upgrade for hundreds more.

Pin It on Pinterest

Share This