Comments Locked

36 Comments

Back to Article

  • nicolaim - Tuesday, October 30, 2018 - link

    Missing spaces in table: "1.5Gbps HBM2 1.5Gbps HBM2 5.1Gbps GDDR5"
  • Ryan Smith - Tuesday, October 30, 2018 - link

    I'm not sure I follow. That's correct.
  • phoenix_rizzen - Tuesday, October 30, 2018 - link

    I think they're trying to say that it looks nicer/cleaner when there's a space between the number and the units.

    1.5Gbps vs 1.5 Gbps
    5.1Gbps vs 5.1 Gbps

    1300MHz vs 1300 MHz
    1185MHz vs 1185 MHz

    etc
  • nicolaim - Wednesday, October 31, 2018 - link

    Yeah. I should have been clearer.
  • s.yu - Thursday, November 1, 2018 - link

    That should be the technically correct way to write it, but it's not necessary as this is just a piece of news, not an academic publication.
  • silverblue - Tuesday, October 30, 2018 - link

    Small correction to first table - Polaris 10 was GCN 4, wasn't it?
  • Valantar - Tuesday, October 30, 2018 - link

    So that's where that went. Was wondering when (if ever) these chips would show up. Very interested in seeing how they perform, and the power draw (desktop Vega is quite efficient at 1100-1300MHz, but I suppose that depends on the DVFS curve too).
  • TEAMSWITCHER - Tuesday, October 30, 2018 - link

    So glad now I waited to buy that 2018 MacBook Pro. I'm going to "flip-the-script" this time and spend big money on my laptop - 32GB of RAM and Vega Pro graphics. My Intel HEDT PC might get a new GPU if Red Dead Redemption 2 is released for the PC soon. Otherwise, I'm going to neglect it.
  • ToTTenTranz - Tuesday, October 30, 2018 - link

    IIRC the previous Polaris 11/21 Radeon Pro GPUs had a 35W TBP, and it's unlikely that apple gave AMD more headroom.

    So a TBP of 35W on the same 14nm process between Polaris 21 and Vega 12 wasn't going to give enough headroom for AMD to come up with a significantly more powerful GPU.
    There's a single low-clocked 1024 HBM2 replacing 4*32bit GDDR5, Vega has extra transistors dedicated to allowing somewhat higher clocks, 4 more CUs and that's about what they could do. Maybe there are 32 ROPs too and that could make a large difference.
    Regardless, we're looking at ~80% more compute resources on Vega Mobile 20 vs. Radeon Pro 560. Plus, HBCC might make a considerable difference when VRAM isn't enough.

    And IMO this GPU not using GF's updated 12nm means it's late as hell. Especially considering Polaris 30 coming out next month on the updated process.

    I'd bet apple wanted this available for July when they launched the 6-core Macbook Pros, and by missing that deadline AMD was made to wait for apple's next reveal.
  • tipoo - Tuesday, October 30, 2018 - link

    Yeah the "X" GPU bumps very much seemed like they wanted Vega at that point but it wasn't ready, so they just took whatever refinements on Polaris they could get.
  • JasonMZW20 - Wednesday, October 31, 2018 - link

    DSBR on Vega also increases rasterization performance by about 10% over GCN4 and can save memory bandwidth too (anywhere from 5-33% on top of 25-40% savings from DCC).

    2:1 FP16 also helps. Polaris was only 1:1, so did FP16 at FP32 rates.

    Agree that these are late as hell. Guess it really depended on foundry slot times, ramp up abilities, and any prospective clients.

    Vega Mobile will probably all use molded packages too, and that adds assembly time, but protects against interposer failure during application of heatsink mounting pressure.
  • Targon - Wednesday, October 31, 2018 - link

    It is also possible that AMD had these ready, but Apple delayed the release due to another reason.
  • tipoo - Tuesday, October 30, 2018 - link

    Curious to get inside and look at the thermals. They used the same heatsink since 2016 right up to the current i9, which...Didn't go so well, so if Vega takes any more power and they didn't change the cooling it might get pretty constrained.

    Though, maybe it was waiting for fab refinements to not be a regression on perf/watt.
  • xype - Wednesday, October 31, 2018 - link

    AMD might also not be as late as it would appear. :D
  • trane - Wednesday, October 31, 2018 - link

    > however we’ve never seen what Vega is like in a more conservative configuration.

    Zhongshan Subor seems remarkably efficient. As is Raven Ridge.
  • beginner99 - Wednesday, October 31, 2018 - link

    I suspect the reason we are only seeing this now is because AMD found no other customer and then had to deal with apple to get enough money to actually make a launch profitable. The hbm is simply too expensive for the resulting performance and only Apple products are priced high enough to warrant this.
  • haukionkannel - Wednesday, October 31, 2018 - link

    And we can be quite sure that Apple is looking for 7nm upgrades of these to the next models! So next year They already have relative good base for these.
  • HStewart - Wednesday, October 31, 2018 - link

    couple of questions here

    1. What about 13in MacBook pro's - any GPU options
    2. What about "Vega M" - I could see it as ideal candidate for 13in MacBook Pro

    3. It sounds like that current it unsure which GPU will be used - but my guess because of production builds - Apple will only have one option for each size and possible no discrete GPU on the 13in.
  • tipoo - Wednesday, October 31, 2018 - link

    None for the 13". Seems their GPU story for the 13" is Thunderbolt.
  • Xajel - Wednesday, October 31, 2018 - link

    Usually, the 13in MBP doesn't have dGPU. So the only way for it to have Radeon is to have Raven Ridge APU's.

    And no, KabyLake-G is a 65W~100W TDP package, 13" MBP uses the U series 15W TDP CPU's, so even with Raven Ridge you only use the U series, the maximum is Ryzen 7 2700U, but the 8th gen Intel U series performs better in everything but gaming.
  • HStewart - Wednesday, October 31, 2018 - link

    That is not true - I have an older 13in MBP and it has a NVidia GPU in it.

    Raven Ridge is not an option for Apple - since ALL Mac's use Intel CPU..

    It appears that Apple has information on web site - and it use Integrate GPU 640 and 655 for the 13in and Radeon 555X or 560X and configure to Vega 16 and Vega 20 for 15 in

    https://www.apple.com/macbook-pro/specs/
  • HStewart - Wednesday, October 31, 2018 - link

    It looks like higher end 2.7Ghz i7 is 28W with i7-8559U cpu/
  • tipoo - Wednesday, October 31, 2018 - link

    That was a bit of a weird time, as Nvidia was making the chipset then, and the GPU was hanging off of that. They can no longer do that, and Apple is all about being as small as it can be.
  • HStewart - Wednesday, October 31, 2018 - link

    NVidia GeForce 9400M with 256MB shared memory

    https://support.apple.com/kb/SP579?locale=en_US
  • KateH - Thursday, November 1, 2018 - link

    the Geforce 9400M is a GPU but it's an integrated one in the sense that its part of the chipset and not a physically separate chip with its own memory. As tipoo said, the 9400M (or, MCP79 in Windows-land) is a single chip, uh, chipset, that contains the graphics processor, memory controller (northbridge), and SATA/USB controllers (southbridge) as well as, iirc, ethernet and audio. Internally the GPU portion is a connected via PCI-Express so to the OS it looks a lot like a regular GPU but due to the integration and shared memory, calling it a /dedicated/ graphics card would be a stretch. So yes it is a Geforce Graphics Processer but GPU =/= dedicated graphics. To be considered a dGPU it has to be 1) a separate chip not part of the chipset or CPU, 2) connect to the CPU via PCIe, 3) have its own memory. 9400m checks only #2. Conversely, Vega-M that's part of the Kaby Lake-G /is/ a dedicated GPU because it checks all 3 of those conditions. It just happens to have the GPU and memory on the same package as the CPU to save space on the motherboard but they're otherwise totally separate entities. Apple played up the Nvidia graphics in the 2008/2009/2010 13" MBPs because it was a marked improvement from the previous Intel GMA graphics but its no more dedicated graphics than, say, the similarly configured GMA950 was.
  • tipoo - Thursday, November 1, 2018 - link

    Yes.
  • iwod - Wednesday, October 31, 2018 - link

    I think I have given up for this generation of AMD 's Mobile GPU. I just hope they get their 7nm Mobile GPU right for 2019.
  • HStewart - Wednesday, October 31, 2018 - link

    Apple has official specs on there site for new models

    https://www.apple.com/macbook-pro/specs/
  • tipoo - Wednesday, October 31, 2018 - link

    So you have to move up to the upgrade model just to have the option to move up to the Vegas.

    Woof, this is going to hurt the wallet.
  • HStewart - Wednesday, October 31, 2018 - link

    We have to wait to November for price - pricing not available on site - but non-Vega models ship now.
  • eva02langley - Wednesday, October 31, 2018 - link

    I am having a 2400g with 11 CU. That thing handle everything at 720p at high-max setting. 1080p is a mixed bag, but that thing is two time more powerful than a 2400G.

    This GPU will handle almost everything at 1080p fine. Something between medium-high settings at 60Hz

    IMO, way better than any 1050TI or below out there, however with HBM2, it could be more pricier than expected.

    However, it is the only GPU fitting in the form format Apple was looking for.
  • noterage - Wednesday, October 31, 2018 - link

    This GPU has similar specs to a gtx 1060 max-q however it is probably going to perform almost exactly like a laptop 1050 ti
  • albert89 - Thursday, November 1, 2018 - link

    I've been following AMD's APU's for a decade now
    and I have to say they are the most versatile and
    powerful on earth. And things can only get better.
    Anantech please do a review of Apple's APU.
  • nerd1 - Thursday, November 1, 2018 - link

    Only apple is still using crappy AMD GPU on their machine... to maximize profit.
  • Smell This - Thursday, November 1, 2018 - link

    In case you missed it ... Apple and AMD have worked together for 10 years or so in heterogeneous computing and OpenCL.
  • Ankou - Monday, November 5, 2018 - link

    I hope in November the Macbook Pros are also updated with a non-touchbar model that can be fully spec'd equivalent to the top end touchable models as well as having the "fixed" keyboard

Log in

Don't have an account? Sign up now