AMD debuts Radeon M400 mobile GPUs with a host of rebrands

We've just noticed that AMD updated its website with a new line-up of laptop Radeon GPUs, the Radeon M400 series. The company hasn't made an official announcement about this line-up, but the specifications on their own are enlightening. Let's start with the R9-class parts in this lineup.

Radeon R9 M400 Series R9 M485X R9 M470X R9 M470
Stream processors 2048 896 768
Texture units 128 56 48
Render output units 32 16 16
Memory capacity up to 8GB GDDR5 up to 4GB GDDR5 up to 4GB GDDR5
Memory bus width 256-bit 128-bit 128-bit
Memory clock up to 5GT/sec up to 6GT/sec up to 6GT/sec
Memory bandwidth ≤160GB/sec ≤96GB/sec ≤96GB/sec

Those numbers look pretty familiar. The R9 M485X appears to be closely related to a previous high-end mobile Radeon, the R9 M390X. That product used the GCN 1.2-based Amethyst GPU—a clock-limited Tonga—in its most-complete incarnation, and we expect the same for the new product. Meanwhile, the two smaller GPUs resemble the R9 M385X and the R9 M380. Given the nomenclature and data at hand, it seems likely they are both based on the GCN 1.1 "Bonaire" silicon dating back to the Radeon HD 7790.

R7 M400 Series R7 M465X R7 M465 R7 M460 R7 M445 R7 M440
Stream processors 512 384 384 320 320
Texture units 32 24 24 20 20
ROPs 16 8? 8? 8? 8?
Memory capacity up to 4GB GDDR5 up to 4GB GDDR5 up to 4GB DDR3 up to 4GB DDR3 up to 4GB DDR3
Memory bus width 128-bit 64 or 128-bit 64-bit 64-bit 64-bit
Memory clock up to 4.5GT/sec up to 4.6GT/sec up to 2GT/sec up to 4GT/sec up to 2GT/sec
Bandwidth ≤72GB/sec ≤73.6GB/sec ≤16GB/sec ≤32GB/sec ≤16GB/sec

The R7 M400 series is less clear-cut. We're guessing the R7 M465X is most likely based on Cape Verde silicon, a GCN 1.0 design first seen in the Radeon HD 7770 over 4 years ago. It isn't a fully-enabled chip, as Cape Verde has 10 compute units versus this part's 8, but barring a completely new design, there's nothing else in the annals of GCN that fits this chip's specs. The rest of the new mobile R7 parts are more difficult to pin down with any certainty, but our best guess is that they are all yet more releases of the venerable Oland (aka Topaz) GPU, also a GCN 1.0 part.

Radeon R5 M400 Series R5 M435 R5 M430 R5 M420
Stream processors 320 320 320
Texture units 20 20 20
Render output units 8? 8? 8?
Memory capacity up to 4GB GDDR5 up to 4GB DDR3 up to 4GB DDR3
Memory bus width 64-bit 64-bit 64-bit
Memory clock up to 4GT/sec up to 2GT/sec up to 2GT/sec
Memory bandwidth ≤32GB/sec ≤16GB/sec ≤16GB/sec

The low-end R5 M400 series are odd beasts. To our eyes, these look like even more Topaz rebadges, but going by the available specifications, there is little to set them apart. Even more confusingly, the R5 M435 appears to be identical to the R7 M445, while the R5 M430 and M420 appear to be identical not only to each other, but to the R7 M440. We can only assume that these parts will eventually be set apart by GPU core clock and final memory configuration.

There's no word on clock rates for the new parts, but that's not the only thing missing from these charts. You, astute TR reader, have no doubt already noticed that the most recent processor in the list is the R9 M485X's Tonga GPU. That GCN 1.2 design debuted in late 2014 aboard the R9 285. Everyone reading this right now is no doubt aware that AMD has plans to unveil GPUs based on 14nm FinFET technology in the coming weeks, and their absence from these charts is conspicuous.

AMD has demonstrated the performance of their new hardware twice already, so we have little reason to doubt that Polaris is coming soon. Some holes have been conveniently left in the R9 400 series for the new hardware too, both above and below the M485X. Let's look forward to Computex and hope AMD fills those holes with some spicy new hardware.

Comments closed
    • BurntMyBacon
    • 3 years ago

    Given OEM pressure for new chips and the fact that end users can’t directly buy the chips, I’m decidedly [b<]less[/b<] aggravated about the idea of rebranding chips in the mobile space than the desktop space. The same OEMs who are demanding the rebrand are the ones that often look bad to the average consumer (if they even notice). [b<]HOWEVER[/b<], at the R9 end of the lineup, the rebranded models look to be stepping their way down the lineup as they should be, but the rebranding becomes parallel by the time we hit the R5s. There are too many products in the lineup filling effectively the same role to begin with. They should drop the bottom two SKUs entirely and shift the model numbers of everything below the R9s as makes sense to fill in the gaps. Also, it would have been nice if GCN 1.0 made an exit here.

    • Unknown-Error
    • 3 years ago

    Admit it! No matter what you paid intel/nvidia shills say, AMD totally dominates “rebranded” market.

    • ronch
    • 3 years ago

    “28nm still has a lot of legs left. Let’s use it for a few more years.”

    -Papermaster

    • Laykun
    • 3 years ago

    28nm so good you gotta go back for more. Might have to send AMD to the 28nm rehab clinic (yes it’s a very small building).

      • BurntMyBacon
      • 3 years ago

      [quote<]yes it's a very small building[/quote<] That's supposed to be the follow up line for the next poster. I dock you two points for esoteric forum posting etiquette [i<](While also giving your post a thumbs up)[/i<].

    • Kretschmer
    • 3 years ago

    Some of these rebrands are just plain depressing. Can these low-end R5 parts even match Intel’s iGPUs?

    • DeadOfKnight
    • 3 years ago

    So much for the top-to-bottom 14 nm refresh that was rumored.

      • chuckula
      • 3 years ago

      From the looks of it, it’s more of a lower-middle-to-upper-middle launch scheme.

      • ImSpartacus
      • 3 years ago

      Where was that rumored? We’ve known since like March (from official amd sources) that Polaris only has two gpus and they occupy only part of a full lineup.

    • swaaye
    • 3 years ago

    Gotta wonder how the low end chips sell these days with IGPs being about as useful/useless and included already.

      • tipoo
      • 3 years ago

      Wal mart specials that have “DEDICATED GPU” stickers on them. Seriously, that’s probably a huge part of it. The number of people I’ve heard say their system should be powerful because it’s a dedicated GPU, or has a “4GB video card” with no other description, is large.

    • Anovoca
    • 3 years ago

    There was a very confusing article the other day that purported leaked numbers for “most likely mobile” polaris chips.”

    Grain of salt here, but these could be those two missing chips in the line up mentioned above:

    [url<]http://wccftech.com/amd-r9-480x-470x-specs-allegedly-revealed/[/url<]

    • chuckula
    • 3 years ago

    Can somebody with more AMD expertise tell us about the actual differences between all these GCN point versions? Does it really make a practical difference in some way?

      • Deanjo
      • 3 years ago

      Of course it makes a difference. It’s .1 moar!!!!!

      • RAGEPRO
      • 3 years ago

      GCN 1.0 was the original release, with Tahiti (7970), Pitcairn (7870), Cape Verde (7770), and later, Oland (8670 / R7 250). (By the way, mobile Pitcairn is called Neptune, mobile Cape Verde is called Venus, and mobile Oland is called Oland or Topaz. There are a billion model names for this mobile stuff.)

      GCN 1.1 came with Bonaire (7790/260X) and Hawaii (290X) and brought TrueAudio, Mantle support, VSR, and Powertune improvements. It also slightly restructured the GCN design to be more modular.

      GCN 1.2 is used in Tonga (R9 285, fully-enabled-except-for-possibly-384-bit-memory in 380X/M295X) and Fiji (Fury/Nano). It is a larger change than 1.1 and has delta color compression to save memory bandwidth, a vastly improved tessellator, and a new video engine.

      I could be wrong about some of this, but it should be pretty accurate. As far as whether it makes a practical difference, well, yes; there’s various feature support (only GCN 1.1 and up have DX12 featurelevel 12_0), and GCN 1.2 has a significant performance advantage over older revisions, particularly with respect to memory performance.

        • chuckula
        • 3 years ago

        Thank you.

        • tipoo
        • 3 years ago

        1.2 also went from the old 2 compute queues (ACEs) with 2 commands each, to 8×8*.
        Interestingly the PS4 was the first to ship with the 8×8 for 64 compute commands. Similar to how Xenos was mid-gen with unified shaders in the x1800 era.

        *This thread is really interesting – GCN can maintain the same performance up to 128 command sources, as having 1 source

        [url<]https://forum.beyond3d.com/threads/dx12-performance-discussion-and-analysis-thread.57188/page-9#post-1869058[/url<]

          • RAGEPRO
          • 3 years ago

          Ah, thanks for that. I knew there was something to do with compute, but I couldn’t remember what it was.

        • namae nanka
        • 3 years ago

        The mobile chips were the first GCN1.1 or GCN2 as AMD likes to put it. Neptune being part of the solar system of chips that are OpenCL2.0 conformant. Venus likewise.

        Tonga has 4 shader engines to Tahiti’s 2 which leads to a doubling of the front end and hence better tessellation performance. Hawaii has same and is about on par with Tonga since the 3xx release drivers.

          • auxy
          • 3 years ago

          TPU GPU database lists Neptune and Venus as GCN1 parts. Pretty sure they are unmodified Pitcairn and Cape Verde silicon running different firmware… (´・ω・`)

            • namae nanka
            • 3 years ago

            That’s too bad. Look at the OpenCL2.0 conformant chips here.

            [url<]http://developer.amd.com/tools-and-sdks/opencl-zone/amd-accelerated-parallel-processing-app-sdk/system-requirements-driver-compatibility/[/url<]

        • cynan
        • 3 years ago

        Pretty sure GCN 1.0 (at least Tahiti) also supports Mantle, though perhaps the later GCNs have better support. What the early GCN cards don’t support is XDMA crossfire (which also means that frame passing in crossfire doesn’t work on these).

      • ImSpartacus
      • 3 years ago

      Anandtech came up with that naming scheme and it sorta became unofficially official during 28nm.

      It started with Bonaire. Anandtech explains in a section literally called “What We’re Calling GCN 1.1”:

      [url<]http://www.anandtech.com/show/6837/amd-radeon-7790-review-feat-sapphire-the-first-desktop-sea-islands/2[/url<] Hawaii is also 1.1. I think Tonga and Fiji are 1.2: [url<]http://www.anandtech.com/show/8460/amd-radeon-r9-285-review/2[/url<] Then amd's marketing department requested that the next gcn iteration (Polaris) be called "4th gen gcn": [url<]http://www.anandtech.com/show/9886/amd-reveals-polaris-gpu-architecture/2[/url<] So yeah, it was kind of a cool thing to watch through the years.

      • crabjokeman
      • 3 years ago

      I didn’t know google and/or wikipedia were down. Oh wait, they’re not: [url<]https://en.wikipedia.org/wiki/Graphics_Core_Next[/url<]

        • chuckula
        • 3 years ago

        Allow me to quote somebody you know: [url=https://techreport.com/news/26855/report-haswell-e-is-epoxied-to-its-heat-spreader?post=838848<]Quit while you're behind.[/url<]

          • crabjokeman
          • 3 years ago

          I feel the <3, but I downvoted you out of reflex.

          Edit: Of course, I’m a plebeian around these parts and I don’t have triplicate privileges.

    • DrCR
    • 3 years ago

    I’m waiting for the Navi architecture — it decreases player reaction time in DOTA2.

      • chuckula
      • 3 years ago

      Sure it does, but they didn’t tell you about the side effects did they?
      It ain’t easy having blue skin.

        • DrCR
        • 3 years ago

        With a Ukrainian accent as well, no less.

          • ForceEdge
          • 3 years ago

          so when do i get [A] gpus? D:

    • USAFTW
    • 3 years ago

    Well, I think the new upcoming Polaris cards are meant to fill the x90x and x80x positions if my intuition won’t deceive me.

Pin It on Pinterest

Share This