AMD Radeon finally back into laptops?

[silent_guy]

None of those mental gymnastics will make any of your aforementioned claims any less than pure conjecture.
Maybe that's enough for you to go e.g. write for fudzilla, semiaccurate or wccftech, or write posts on neogaf.

Here on B3D it's still pure conjecture and it will be valued as such.

The fact that it's conjecture doesn't make it less likely.

There haven't been a lot of reasons to choose AMD for laptops in the last 2 years, since power consumption happens to be a key factor, yet Apple continues to do so.

Conjecture as it may be, price is the most obvious factor.

For AMD, even a low price GPU may still be sufficient to increase their abysmal corporate gross margin percentage. For Nvidia, it'd result in a gross margin hit.

 

[Silent_Buddha]

The fact that it's conjecture doesn't make it less likely.

There haven't been a lot of reasons to choose AMD for laptops in the last 2 years, since power consumption happens to be a key factor, yet Apple continues to do so.

Conjecture as it may be, price is the most obvious factor.

For AMD, even a low price GPU may still be sufficient to increase their abysmal corporate gross margin percentage. For Nvidia, it'd result in a gross margin hit.

I doubt it's the price as much as it is that AMD continues to support OpenCL, and Apple is by all accounts the driving force behind OpenCL. Price will certainly be part of it, but it's more likely that OpenCL support is valued far more highly by Apple.

Regards,
SB

 

[MDolenc]

So... When do we get something newer then OpenCL 1.2 on Apple?

 

[Ryan Smith]

I doubt it's the price as much as it is that AMD continues to support OpenCL, and Apple is by all accounts the driving force behind OpenCL. Price will certainly be part of it, but it's more likely that OpenCL support is valued far more highly by Apple.

Regards,
SB

Apple is no longer the driving force behind OpenCL, at least as far as I can tell. (Admittedly, I don't see on the OCL committee) All signs point to OpenCL having been thrown out of 1 Infinite Loop for Metal (especially now that it has compute shaders).

They're still a top level (Promoter) member of Khronos, but if they're doing much these days besides funding Khronos, then they're doing their best to avoid notice.

 

[Deleted member 13524]

OpenCL's driving force should be Google, AMD and all the ARM GPU and SoC makers that are left out of the x86 ecossystem, and have no possible alternative for a heterogeneous compute API.

 

[Silent_Buddha]

Apple is no longer the driving force behind OpenCL, at least as far as I can tell. (Admittedly, I don't see on the OCL committee) All signs point to OpenCL having been thrown out of 1 Infinite Loop for Metal (especially now that it has compute shaders).

They're still a top level (Promoter) member of Khronos, but if they're doing much these days besides funding Khronos, then they're doing their best to avoid notice.

Seems, I'm a bit behind the times there. Thanks for the correction.

Regards,
SB

 

[Ryan Smith]

Seems, I'm a bit behind the times there. Thanks for the correction.

Regards,
SB

No problem. OpenCL is a bit like a red headed step child right now; Apple and NVIDIA have better things to do, and even AMD has Plan B (Boltzmann) just in case.

 

[Deleted member 13524]

New embedded Radeons using Polaris chips in MXM add-in format:

http://www.anandtech.com/show/10710/amd-announces-embedded-radeon-e9260-e9550


E9550 is a full Polaris 10 at 1100-1250MHz and 7Gbps GDDR5 with a 95W TDP using MXM-B.
E9260 is a Polaris 11 with 14 CUs enabled (out of 16) at up to 1.4GHz, 7Gbps GDDR5 and a 50W TDP, using the smaller MXM-A standard.

So the MXM cards are being produced. The consumer-equivalent E9550 should finally be replacing mobile Pitcairn (7970M + 8970M + R9 M290X) at the same TDP, as a GTX 1060 competitor.
Mobile consumer cards equivalent to the E9260 are usually soldered, though.





I do find very surprising that not a single Polaris 11 model was made for the back-to-school season so far, though.
That's definitely not good news.

 

[Picao84]

Not surprising at all considering the still large difference in perf./ watt against nvidia parts.. 95W Tdp for a chip that's significantly behind nvidia 100W one. Plus we have yet to see GP107, but judging from Polaris 11 abismal performance, barely faster than a cut down Gm206, it should run circles around it.

 

[gamervivek]

1.4Ghz on newer Polaris chip sounds promising for Vega. Strange that the full chip is still MIA, Apple don't allow for it?

 

[Deleted member 13524]

but judging from Polaris 11 abismal performance,

Besides the fact that gratuitously badmouthing every single AMD GPU release being the new black, what exactly is abysmal with Polaris 11's performance, compared to existing solutions within the same price range?

 

[ImSpartacus]

New embedded Radeons using Polaris chips in MXM add-in format:

http://www.anandtech.com/show/10710/amd-announces-embedded-radeon-e9260-e9550


E9550 is a full Polaris 10 at 1100-1250MHz and 7Gbps GDDR5 with a 95W TDP using MXM-B.
E9260 is a Polaris 11 with 14 CUs enabled (out of 16) at up to 1.4GHz, 7Gbps GDDR5 and a 50W TDP, using the smaller MXM-A standard.

So the MXM cards are being produced. The consumer-equivalent E9550 should finally be replacing mobile Pitcairn (7970M + 8970M + R9 M290X) at the same TDP, as a GTX 1060 competitor.
Mobile consumer cards equivalent to the E9260 are usually soldered, though.





I do find very surprising that not a single Polaris 11 model was made for the back-to-school season so far, though.
That's definitely not good news.

Am I missing something or is that 50W 14 CU Polaris 11 implementation boosting to significantly higher clocks than the 75ishW 14 CU Polaris 11 in the 460? I feel like I'm overlooking something.

That's encouraging though. Good to see that Polaris is capable of that, especially with the recent Vega rumors.

Besides the fact that gratuitously badmouthing every single AMD GPU release being the new black, what exactly is abysmal with Polaris 11's performance, compared to existing solutions within the same price range?

I think when someone uses a word like "abysmal" to describe Polaris 11, they are looking at the entire lifetime of the chip (which well be long if recent AMD chips are any indication), not just the relatively brief period before GP107.

You're right that Polaris 11 is competitive against today's competition, but only looking at today's competitive landscape is foolhardy when GP107 is a month or two away.

1.4Ghz on newer Polaris chip sounds promising for Vega. Strange that the full chip is still MIA, Apple don't allow for it?

I always liked the "Apple" theory to explain the lack of fully enabled Polaris 11 chips. I figure it could still be accurate in this situation.

 

[Deleted member 13524]

I think when someone uses a word like "abysmal" to describe Polaris 11, they are looking at the entire lifetime of the chip (which well be long if recent AMD chips are any indication), not just the relatively brief period before GP107.

Meaning: they are pulling stuff out of their ass, because GP107's release date, power consumption, price, die area and performance are a complete unknown at the moment.
Okay.

 

[Ryan Smith]

Am I missing something or is that 50W 14 CU Polaris 11 implementation boosting to significantly higher clocks than the 75ishW 14 CU Polaris 11 in the 460? I feel like I'm overlooking something.

The math is straightforward - AMD is promising 2.5 TFLOPs on 14 CUs - but I agree it's one of the weirder parts of the announcement. The official specs for the Embedded Radeons have been weird off and on over the years.

 

[ImSpartacus]

The math is straightforward - AMD is promising 2.5 TFLOPs on 14 CUs - but I agree it's one of the weirder parts of the announcement. The official specs for the Embedded Radeons have been weird off and on over the years.

Ok, I suppose if historic embedded parts have historically had weird specs, then this is at least consistent with that. I appreciate the context.

 

[Ryan Smith]

Ok, I suppose if historic embedded parts have historically had weird specs, then this is at least consistent with that. I appreciate the context.

For reference, when AMD originally announced Radeon E8950, they speced it for 3 TLFOPS @ 1000MHz with 2048 SPs.

 

[Picao84]

Meaning: they are pulling stuff out of their ass, because GP107's release date, power consumption, price, die area and performance are a complete unknown at the moment.
Okay.

Do you really expect GP107 to not be significantly faster when NVidia has such a big advantage on perf/watt? Polaris 11 barely reaches Maxwell perf/watt as it is! It is even worse than Polaris 10 at that! Remove your red tinted glasses and smell the coffee...

 

[Deleted member 13524]

I'm done feeding troll accusations of "abysmal performance" of a GPU that currently has pretty much no competition at its price point.
Feel free to cross the entire online-sphere to find a graphics card that goes for as low as $90 with the same kind of performance from the competition.
When you do, you'll have a point.

I shouldn't have fed the initial statement and now I regret it immensely.

 

[Picao84]

I'm done feeding troll accusations of "abysmal performance" of a GPU that currently has pretty much no competition at its price point.
Feel free to cross the entire online-sphere to find a graphics card that goes for as low as $90 with the same kind of performance from the competition.
When you do, you'll have a point.

I shouldn't have fed the initial statement and now I regret it immensely.

Why am I a troll for stating the obvious? AMD has to launch products that will match the SAME generation from the competition, not products that have been in the shelves for nearly two years (on an older 28nm process)! Why do you think AMD is in such a dire situation? Because they seem to follow your sort of reasoning, ever since the 6970 generation, which is was designed to simply match GTX480 and caught with their pants down when NVidia launched GTX580.

And, being even more on topic, we are talking about laptop GPUs. Those are not bought of the shelf by consumers, but by OEMs, meaning that they have access to new parts specifications and test chips quite a while in advance. Meaning that they have already seen GP107 for sure and compared it to Polaris 11. If you are not seeing laptops with Polaris 11, why do you think that is? Its not rocket science, it for sure is not trolling. Its called looking at reality outside of our own preferences. (example: I even said before that RX 480 is a better choice than GTX1060.. NVidia GX106 GPUs have been disappointing for a while now, ever since GX104 chips were converted to high end).

On the other hand, you would do well in not getting so offended every time someone criticises AMD, its not healthy. What is even more incredible is that you cite a very low price as a sign of success! If it was such a good product, AMD would be able to sell it for much closer to what NVidia charges. It seems that people here try again and again to show you how things work, but you cant see past your own assumptions and beliefs...

 

[ImSpartacus]

For reference, when AMD originally announced Radeon E8950, they speced it for 3 TLFOPS @ 1000MHz with 2048 SPs.

Yeah, it looks like the comparable Tonga-based M395X had a boost clock of only 909 MHz (not to mention the older M390X's 723 MHz boost), so the embedded variant had a healthy advertised bump from that. I suppose that helps explain the E9260's unusually high calculated boost clock. Maybe AMD marketing is rounding to the nearest half TFLOPS?

Well shoot, part of me was hoping for an example of GCN 4's higher clock headroom to support that recent 1.5ish GHz 64 CU Vega rumor. Oh well.