Earlier this week Apple announced their 2015 15” Retina MacBook Pro. Though Apple didn’t make any CPU changes, they did make some GPU changes on the high-end model, swapping out NVIDIA’s GeForce GT 750M for AMD’s Radeon R9 M370X. Since the M370X was not a published part number in AMD’s recently-announced 2015 Radeon M300 series refresh lineup, there have been some questions over just what M370X really is.

At the time of the rMBP’s launch, we suspected that it was an AMD Cape Verde GPU, based on the fact that this GPU is also in the M375, which is the next part above M370X. With the new laptop shipping immediately, M370X models have already begun arriving in buyers’ hands, finally giving us a chance to confirm the GPU inside.

Image courtesy Reddit user ootan

Thanks to Reddit user ootan, who posted a screenshot of the rMBP’s System Profiler, we can now confirm that the GPU in the rMBP is in fact AMD’s Cape Verde GPU. AMD has previously used the 6821 device ID on other mobile Cape Verde parts, so 6821 is already a known quantity.

AMD M300 Series GPU Specification Comparison
  R9 M375 R9 M370X (rMBP) R7 M360
Was Variant of R9 M270/M260 Variant of R9 M270X Variant of R7 M270/M260
Stream Processors 640 640 384
Texture Units 40 40 24
ROPs 16 16 4?
Boost Clock <=1015MHz 800MHz <=1015MHz
Memory Clock 2.2GHz DDR3 4.5GHz GDDR5 2GHz DDR3
Memory Bus Width 128-bit 128-bit 64-bit
VRAM <=4GB 2GB <=4GB
GPU Cape Verde Cape Verde Oland/Topaz
Manufacturing Process TSMC 28nm TSMC 28nm TSMC 28nm
Architecture GCN 1.0 GCN 1.0 GCN 1.0

As for Cape Verde itself, as our regular readers may already know, it’s one of AMD’s first-generation GCN 1.0 GPUs, launched back in 2012. Compared to the GK107 GPU found in the GT 750M, it is a larger, more powerful (and at least in the desktop, more power-hungry) GPU, designed for more powerful devices than GK107 was. As for why Apple opted to switch now and to this GPU, we’ll leave that to Apple, AMD, and NVIDIA, though it’s by no means surprising that after having run GK107 for a couple of years, they would want to upgrade to a more powerful GPU.

Meanwhile, though we don’t have M370X on-hand at the moment, at least in the desktop, where GPU performance is unrestricted by thermals, Cape Verde fares very well. The rMBP on the other hand does have thermal constraints to deal with, so performance won’t be the same, but I expect it to fare reasonably well as well. Though at the same time I’m also curious if the use of a higher performance part has impacted the rMBP’s battery life when the dGPU is active; AMD and NVIDIA both do heavy binning, so a simple extrapolation won’t work here.

Update: And no sooner do I post this then someone sends me a screenshot of GPU-Z from a 15" rMBP running Boot Camp.

GPU-Z, for those unfamiliar with it, uses register poking to identify GPUs, so if the device ID wasn't enough, this settles it. This also confirms the clockspeeds - 800MHz core, 4.5GHz VRAM - and that the M370X is using GDDR5, unlike it's M375 counterpart. Thank you DMDrew812.

Source: Reddit User ootan (via SH SOTN)



View All Comments

  • testbug00 - Saturday, May 23, 2015 - link

    Nvidia made their own solution? You mean using IBMs standard and calling it NVLink? Pretty sure that IBM made that spec... Reply
  • V900 - Saturday, May 23, 2015 - link

    CUDA has become a standard in some industries. And while AMD has a slight advantage in OpenCL, Nvidia is no slouch either.

    So it's basically: AMD: Great OpenCL, no CUDA. Nvidia: Good Open CL, Great CUDA.

    The only reason I can see for this, is that Apple threw some money AMDs way as an act of charity of sorts. They want two manufacturers, and want AMD to survive financially until they launch their Zen CPUs...
  • dragonsqrrl - Saturday, May 23, 2015 - link

    Probably one of the biggest misconceptions I encounter on a regular basis amongst AMD fanboys: OpenCL is dominating CUDA, and when they realize it's not there must be some industry wide conspiracy against AMD... Reply
  • halo37253 - Sunday, May 24, 2015 - link

    CUDA has been a flop on the Apple side though....
    Even Apple has been pushing OpenCL for years now.
  • crimsonson - Sunday, May 24, 2015 - link

    I don't think CUDA is a flop on the Mac platform as more of Apple pushed OpenCL for strategic reasons.
    OpenCL is their baby so it makes sense they would push it.
  • dragonsqrrl - Saturday, May 23, 2015 - link

    "Furthermore you're assuming they're the same when there's likely power management and thermal improvements in M370X, just like how all chip vendors make progressive improvements despite remaining on the same process size."

    But it's not just the same process, it's the exact same GPU. A lot of what you're suggesting would require at least a manufacturing revision, and a deeper architectural revision to see more substantial benefits.
  • testbug00 - Sunday, May 24, 2015 - link

    What version of the silicon is it? A0? B3? C2? (I have no clue how many spins Cape Verde has had, just making number numbers) Reply
  • dragonsqrrl - Monday, May 25, 2015 - link

    I don't think AMD or Nvidia are known to do this without updating the name of the GPU in some way, or renaming it entirely. GF11x were largely manufacturing revisions of GF10x. There were two revisions of GK110, each with an updated name. Heck, AMD has even renamed unchanged GPU's before :P. At this time even Ryan believes it's the exact same GPU, because there's really reason to believe otherwise. Reply
  • Kjella - Saturday, May 23, 2015 - link

    Mad? No. This is pure business. AMD is selling themselves cheap to get the PR from having the GPU of the top model Mac Book Pro, while the number of $2500 laptop sales of a single model is presumably rather slim. Same reason they're the ones in all the consoles and still hurting economically, if you lowball the offer you win at the expense of your margins. Their problem is that the halo effect isn't much, gamers read benchmarks and nVidia is making a killing - from January to April the share of GTX 970s in the Steam hardware survey has gone from 1.80% to 2.81%, in one quarter 1 in 100 Steam users have bought the same $300+ graphics card, most play on cheap old cards so that's massive. It's more than the total market share of AMD Radeon R9 200 Series of 0.94%, which is barely selling anymore with +0.11% in the last quarter. Add in the 960/980/Titan and nVidia has 90-95% of the high end. That HBM card really needs to be a killer if AMD wants to stay in the competition. Reply
  • MrSpadge - Saturday, May 23, 2015 - link

    It does seem strange, indeed. On the other hand:

    - AMD desperately needs money, so they probably made a very good price.
    - There has hardly been any progress design-wise from AMD, so apart from some features GCN 1.0, 1.1 and 1.2 perform pretty the same per clock and shader
    - GM107 based mobile GPUs are likely too fast (=expensive), whereas GM108 based mobile GPUs are cirppled by 64 bit DDR3 memory busses

Log in

Don't have an account? Sign up now