AMD Polaris Rumors and Discussion

+ AMD needs dual-graphics solution for Bristol Ridge. It isn't only about 4k. Previous solutions (Mars/Oland, Cape Verde etc.) are really outdated in terms of power consumption and monitor outputs. In previous generations AMD always had sub-100mm² solutions.

In H2 2017 AMD is bringing Raven Ridge which is a Zen APU with up to 12 Vega CUs at a maximum TDP of 35W.
Why would AMD develop a discrete GPU to pair with Bristol Ridge, if a handful of months later (at best) they're launching a new APU that would get better performance than the hybrid combo?


Besides, the APU+GPU dual-graphics solutions AFAIK existed to increase the appeal of AMD's mid-range APUs in comparison to mobile i3 solutions, but they were never successful. Either the driver never worked very well or the OEMs screwed up by pairing the APU with single-channel RAM and/or preventing Hybrid Crossfire from happening altogether, those dual-graphics setups ended up never being popular or getting good reviews.
 
In H2 2017 AMD is bringing Raven Ridge which is a Zen APU with up to 12 Vega CUs at a maximum TDP of 35W.
Why would AMD develop a discrete GPU to pair with Bristol Ridge, if a handful of months later (at best) they're launching a new APU that would get better performance than the hybrid combo?
Also a rumoured HBM APU positioned above it that could very well scale pass Polaris 11.

But regardless of having it or not, AMD APUs might have a stronger position hopefully thanks to Zen and DDR4 at that point, and Polaris 11 is always there as the next step. I don't see why AMD would push dual graphics as it did before TBH. There isn't much room below Polaris 11 now, except for the bandwidth gap.

(I'd be really surprised if AMD managed to support LPDDR4 in Raven Ridge, and lands design wins with it. 50+ GB/s of bandwidth at the lowest rate (3.2GT/s). iPad Pro is still the first and only one)
 
(I'd be really surprised if AMD managed to support LPDDR4 in Raven Ridge, and lands design wins with it. 50+ GB/s of bandwidth at the lowest rate (3.2GT/s). iPad Pro is still the first and only one)
Assuming that this rumor is true and it is referring to Raven Ridge, then I expect it to support LPDDR4. Skylake and (almost certainly) Kaby Lake do not support LPDDR4, and Apple's use of LPDDR memory in the MacBook Pro is why the latest MBP doesn't support more than 16 GB of RAM. A Zen APU supporting LPDDR4 could be one reason why Apple would switch away from Intel.
 
That rumor is suggesting in the server market, which should be one of the Owls and not Raven Ridge. For Raven Ridge and compact APUs I'd agree LPDDR4 would be interesting.
 
(I'd be really surprised if AMD managed to support LPDDR4 in Raven Ridge, and lands design wins with it. 50+ GB/s of bandwidth at the lowest rate (3.2GT/s). iPad Pro is still the first and only one)
I don't know what the big deal is with lpddr4. The mobile (mobile as in phones) guys already have two generations of chips out using it. intel just kinda missed this I think... Even though with just about everyone now soldering down the memory anyway in the slim notebooks, it's quite an obvious choice.
Albeit Raven Ridge supporting it would not necessarily mean it would support 3.2GT/s. This is not the lowest possible rate by any stretch of the imagination (it's just the original jedec specification got up to that), in fact it gets down to 400MT/s if I see that right. Intel's atom chips (Apollo Lake), unlike their big brothers, support lpddr4 just fine, but only up to 2.4GT/s.
That said, supporting 3.2GT/s lpddr4 would surely be nice for that integrated graphics (albeit in a 15W package, the memory clock might not actually matter all that much, as long as it's using 128bit memory). And if it support such clocks for ddr4, why wouldn't it for lpddr4.
 
I don't know what the big deal is with lpddr4. The mobile (mobile as in phones) guys already have two generations of chips out using it. intel just kinda missed this I think... Even though with just about everyone now soldering down the memory anyway in the slim notebooks, it's quite an obvious choice.
Albeit Raven Ridge supporting it would not necessarily mean it would support 3.2GT/s. This is not the lowest possible rate by any stretch of the imagination (it's just the original jedec specification got up to that), in fact it gets down to 400MT/s if I see that right. Intel's atom chips (Apollo Lake), unlike their big brothers, support lpddr4 just fine, but only up to 2.4GT/s.
That said, supporting 3.2GT/s lpddr4 would surely be nice for that integrated graphics (albeit in a 15W package, the memory clock might not actually matter all that much, as long as it's using 128bit memory). And if it support such clocks for ddr4, why wouldn't it for lpddr4.

DDR4 is scaling up slowly, while LPDDR4 starts at 3.2 GT/s (only available as XMP/AMP profile in DDR4 platforms). Go look at the available LPDDR4 parts.

By the way, what matters is the rated max speed. Of course it would have a wide range of operating frequency for different power saving modes. So if you support this standard, you meant to support at least its baseline spec (i.e. min 3.2 GT/s).

Saying SoCs not necessarily supporting the max rate, as LPDDR4 could go down to 400 MHz, is like saying GPUs just needs DDR3 because GDDR5 can drop into DDR3 mode to save power. Huh.


Sent from my iPhone using Tapatalk
 
Raven Ridge landing LPDDR4 support would be nothing short of huge, IMO.
Kaby Lake doesn't support it because it's supposedly pin-compatible with Skylake, so AMD getting the upper hand on memory bandwidth for mobile APUs would mean a lot.
Of course, Intel can eventually compensate with Crystalwell, but those models haven't been enormously successful regarding performance or power consumption.
 
So if you support this standard, you meant to support at least its baseline spec (i.e. min 3.2 GT/s).
You are right that all parts you can buy support 3.2GT/s. But as said, Apollo Lake only supports 2.4GT/s lpddr4 regardless, so I suppose it makes _some_ sense supporting lpddr4 at lower speeds - presumably the memory controller simply isn't up to the task.
 
That rumor is suggesting in the server market, which should be one of the Owls and not Raven Ridge. For Raven Ridge and compact APUs I'd agree LPDDR4 would be interesting.
My post must have been unclear, I was referring to the tweet from Bits and Chips that says "According to our source, Apple will use Zen APU just in MacBook Pro and MacBook."
 
You are right that all parts you can buy support 3.2GT/s. But as said, Apollo Lake only supports 2.4GT/s lpddr4 regardless, so I suppose it makes _some_ sense supporting lpddr4 at lower speeds - presumably the memory controller simply isn't up to the task.
Apollo Lake targets low-end PCs, and higher speed LPDDR4 is apparently not a cheap part for those segments. It is more expensive than LPDDR3, which is supposed to be more expensive than DDR4. Moreover, what GPU is it shipping with?

Now let's get back to Raven Ridge — four Zen cores that can perhaps boost beyond 3.4 Ghz under 35W, and a GPU that is probably just a bit smaller in scale than Polaris 11. It is quite a bit of difference, really.
 
Assuming that this rumor is true and it is referring to Raven Ridge, then I expect it to support LPDDR4. Skylake and (almost certainly) Kaby Lake do not support LPDDR4, and Apple's use of LPDDR memory in the MacBook Pro is why the latest MBP doesn't support more than 16 GB of RAM. A Zen APU supporting LPDDR4 could be one reason why Apple would switch away from Intel.
Since this is brought up, the 2016 MBP thingy is not really about the support of LPDDR4, but the lack of 64Gbit x32 LPDDR3 (and apparently Apple isn't interested to call for a part that would have a super low economies of scale). At least when I last checked the public product catalogues, you can't even achieve 16GB in 128-bit DQ with what Micron and Samsung are offering in LPDDR4 (except SK Hynix, perhaps).
 
Im sorry for being a knob for asking this, but does anyone know if/whether a higher end GPU is going to be released any time soon? I was looking for a high end GPU from AMD preferably, but the RX 480 isn't quite fast enough.
 
Im sorry for being a knob for asking this, but does anyone know if/whether a higher end GPU is going to be released any time soon? I was looking for a high end GPU from AMD preferably, but the RX 480 isn't quite fast enough.

1H 2017 the vega line.

https://twitter.com/BitsAndChipsEng/status/805456477221715970
My mistake then, that's what I was responding to from the same guy/thread you linked.

Apple using an AMD APU for a server? did I missed something here?
 
Apple using an AMD APU for a server? did I missed something here?
Wasn't Apple, but another unspecified company from the context. The two were getting confused though. It was in regards to a major buyer switching from Intel to AMD based on rumor.
 
Im sorry for being a knob for asking this, but does anyone know if/whether a higher end GPU is going to be released any time soon? I was looking for a high end GPU from AMD preferably, but the RX 480 isn't quite fast enough.
Vega seems to be between the GTX1080 and Titan X Pascal in performance (probably in price too). And then there's an obscure "Polaris 12" in the drivers together with the RX490 name that has been appearing here and there but no detail has ever appeared about those.
 
With AMD not even mentioning DP performance of the MI125, is it safe to assume there won't be half DP?

Aren't these new cards for neural networks exclusively?
If so, is there any advantage in talking about DP?
 
Back
Top