[TonyMacx86]Possible Polaris 12 GPU and Refreshed Polaris 10 spotted in macOS Sierra drivers.

[Glo.]

Nothing more to add to this at this moment. It is interesting that Apple has added non existing GPU kexts into the system.

2 GPUs may come: Polaris 12, and refreshed Polaris architecture. It would be baffling if the benchmarked GPU was Polaris 12.

And BTW. This framebuffer is linked to Trash Can Mac Pro. So somebody is at least testing hardware.

macOS Sierra 10.12.2 Beta 5.
https://www.tonymacx86.com/threads/...d-radeon-drivers.197273/page-118#post-1386344

And no, this is NOT fake.

 

[dacostafilipe]

Okay, now that's interesting!

It seems I'll have more choice when upgrading my Hackintosh next year

 

[Glo.]

I am currently trying to connect the dots.

And that Polaris 12 may be actually the RX 490 GPU.

 

[crisium]

Polaris 10 XT2?

 

[Glo.]

Polaris 10 XT2?

Tahiti XT2 was in 7970 GHz edition and in R9 280X.

Refresh might appear of Polaris 10. Otherwise there would be no kext here for it.

 

[Tweak155]

Can you explain to a n00b why this is interesting? Not criticizing, just not smart enough / in the know enough to get it.

 

[Glo.]

Neither Polaris 12, nor Polaris XT2 are existing in the real world. Yet, the kexts appear in unreleased, beta version of system from specific vendor which appears to have direct partnership with AMD, for GPU supplies.

Two possibilities. Polaris 12 is Apple specific GPU - non realistic.
Polaris 12 will appear in desktop computers - may be more realistic than we think right now.

Polaris 10 XT2 appears to be GPU with revised die, or the same die, on new, tuned process.
Better efficiency/higher performance in the same thermal envelope? Yes, please. Especially when we heard many months ago that Nvidia may be planning of revising their Pascal line.

 

[Valantar]

The question here is - given that these show up at some point - what the performance will be like. Polaris 10 XT2 sounds like a minor upgrade (process refinement?) to the RX480/Radeon Pro WX 7100. If their numbering scheme is to make any sense at all, though, the 12 ought to be even smaller than the P11 - i.e. <800SMs. While I'd gladly welcome an up-to-date low-end GPU like that for HTPC use, I don't quite see how AMD will make any money off it. Mobile, perhaps? Although the slow-and-wide approach that they're using in the RP 450/455/460 lineup in the current MBPs should in theory be more efficient than a narrow-and-fast architecture.

It also makes little sense for P12 to be a higher end part, given that Vega is arriving soon to take that place.


edit: typo.

 

[dacostafilipe]

I did read somewhere that Polaris 10 does not have all the power saving functions that Polaris 11 does, because it was created for the desktop.

Could this Polaris 12 be an updated P10 for mobile? Could also use the same technique to keep it's height small like P11 does?

 

[NTMBK]

Perhaps Polaris 12 is the new IGP for Zen APUs?

 

[beginner99]

If their numbering scheme is to make any sense at all, though, the 12 ought to be even smaller than the P11 - i.e. <800SMs.

As has been covered multiple times the naming is the order in which the uArch was designed. So Polaris 10 -> Polaris 11 -> Polaris 12 and has nothing to do with die size or performance. This is also why Vega 10 is small Vega and Vega 11 is big Vega, eg exactly opposite from the Polaris case.

So Polaris 12 can be anything from larger to smaller to mobile version of P10.

 

[Ancalagon44]

I would find it difficult to believe that Polaris 12 is a bigger GPU than Polaris 10, for the simple reason that we know there are two Vega GPUs incoming. The likelihood of their being 3 GPUs larger than Polaris 10 seems extremely low - Nvidia for instance only has 2 GPUs larger than the one used in the 1060 (I forgot the name). Counting consumer GPUs, not Quadros.

However, one possibility is that the Vega codename has been dropped, and what were going to be Vega GPUs will now be called Polaris 12 and 13. I think this is unlikely though, since Polaris and Vega are both internal codenames and not used for marketing.

My guess is that Polaris 12 will be a custom part made for Apple, and slots in below Polaris 11 in size. It will be mGPU only, optimized for low power consumption.

 

[Valantar]

As has been covered multiple times the naming is the order in which the uArch was designed. So Polaris 10 -> Polaris 11 -> Polaris 12 and has nothing to do with die size or performance. This is also why Vega 10 is small Vega and Vega 11 is big Vega, eg exactly opposite from the Polaris case.

So Polaris 12 can be anything from larger to smaller to mobile version of P10.

That might be, but it still leaves the problem of where to fit a larger Polaris chip in their lineup. If V10 is ~4000 SMs, it could reasonably be cut down to ~3000 or slightly more. This leaves a very narrow performance band for P12. Not to mention that Vega is supposed to deliver significantly better perf/W than Polaris (see the (admittedly vague and unlabeled) graphs in their public roadmaps), this would mean that a ~3000SM Polaris part would use noticeably more power than a ~3000SM Vega part. Makes little sense to me. Which is why I'm leaning towards a mobile part.

 

[Glo.]

Perhaps Polaris 12 is the new IGP for Zen APUs?

Zen APUs use Vega architecture.

The question here is - given that these show up at some point - what the performance will be like. Polaris 10 XT2 sounds like a minor upgrade (process refinement?) to the RX480/Radeon Pro WX 7100. If their numbering scheme is to make any sense at all, though, the 12 ought to be even smaller than the P11 - i.e. <800SMs. While I'd gladly welcome an up-to-date low-end GPU like that for HTPC use, I don't quite see how AMD will make any money off it. Mobile, perhaps? Although the slow-and-wide approach that they're using in the RP 450/455/460 lineup in the current MBPs should in theory be more efficient than a narrow-and-fast architecture.

It also makes little sense for P12 to be a higher end part, given that Vega is arriving soon to take that place.


edit: typo.

WX7100 is not Ellesmere XT2. It is just Ellesmere XT.

Polaris 12 in theory could be smaller part than Polaris 11. However for that part of the segment AMD can use 28 nm GPUs.

Polaris 12 can be GPU used in Project Scorpio. 6 TFLOPs - 3072 GCN cores at under 1000 MHz core clock.

 

[dacostafilipe]

Polaris 12 can be GPU used in Project Scorpio.

Inside a macOS kext? Isn't Scorpio using a custom APU?

For me, it looks like a mobile version of P10 for the next iMacs.

 

[Glo.]

Inside a macOS kext? Isn't Scorpio using a custom APU?

For me, it looks like a mobile version of P10 for the next iMacs.

What I mean is the same GPU design that Polaris 10, is in the Playstation 4 Pro. It would be very efficient in terms of money if AMD would reuse similar GPU design in their GPUs, because porting it one way or another would save a lot of money.

What we have seen about Project Scorpio is that it is 8 core CPU, 24 GB RAM(384 or 192 Bit memory bus), with 6 TFLOPs GPU.

 

[dacostafilipe]

What I mean is the same GPU design that Polaris 10, is in the Playstation 4 Pro.

No it's not. The PS4Pro APU is more advanced (has features that are not P10).

Whatever those new DeviceIDs are, they certainly have nothing to do with AMD's custom silicon business.

 

[Keysplayr]

So polaris 10 is bigger than polaris 11, yet polaris 12 is bigger than polaris 10??

What's the reasoning behind this belief?

is this confirmed?

is this a hopeful "come on AMD lets go!!!" thing?

Also If polaris 10 is larger than 11, why would anyone think Vega 10 is smaller than Vega 11?
can anyone give an example in history where chip family naming nomenclature departed from the standards both AMD and Nvidia has been using forever now?

And if AMD did deviatefrom the normal chip badging, then why? where is the advantage in doing so?

 

[Glo.]

No it's not. The PS4Pro APU is more advanced (has features that are not P10).

Whatever those new DeviceIDs are, they certainly have nothing to do with AMD's custom silicon business.

I know that PS4P APU is more advanced. But the core number is the same. And this is what I was referring to in my post.

So polaris 10 is bigger than polaris 11, yet polaris 12 is bigger than polaris 10??

What's the reasoning behind this belief?

is this confirmed?

is this a hopeful "come on AMD lets go!!!" thing?

Also If polaris 10 is larger than 11, why would anyone think Vega 10 is smaller than Vega 11?
can anyone give an example in history where chip family naming nomenclature departed from the standards both AMD and Nvidia has been using forever now?

And if AMD did deviatefrom the normal chip badging, then why? where is the advantage in doing so?

Because naming scheme has nothing to with the size of the chip, but it has all to do with the chronology of the chips being designed.
Vega 10 and Polaris 10 are the first GPUs that have been designed in particular uarchitecture. So either Polaris 12 can be smaller than Polaris 11, and either can be bigger than Polaris 10.

 

[Det0x]

So polaris 10 is bigger than polaris 11, yet polaris 12 is bigger than polaris 10??

What's the reasoning behind this belief?

is this confirmed?

is this a hopeful "come on AMD lets go!!!" thing?

Also If polaris 10 is larger than 11, why would anyone think Vega 10 is smaller than Vega 11?
can anyone give an example in history where chip family naming nomenclature departed from the standards both AMD and Nvidia has been using forever now?

And if AMD did deviatefrom the normal chip badging, then why? where is the advantage in doing so?

Despite 11 being a larger number than 10, Polaris 10 is actually a larger, higher end GPU than Polaris 11. Chief Architect & head of AMD’s Radeon Technologies Group confirmed at Capsaicin that the naming scheme is time-based & Polaris 10 had simply been designed before Polaris 11, hence the smaller numerical designation.

http://wccftech.com/amd-launching-polaris-10-gpu-june-1st-computex/

My (wishful) take on the situation:

Polaris 10 XT2 = RX 485 -> Basically a higher clocked 480 with better performance/watt
Remember this ? http://wccftech.com/amd-polaris-revisions-performance-per-watt/ (AMD Rolling Out New Polaris GPU Revisions With 50% Better Perf/Watt)

Polaris 12 = RX 490 ->6 TFLOPs - 3072 GCN like Glo said.
Remember this ? http://videocardz.com/64475/are-those-radeon-rx-490-vega10-benchmarks-results (to go up against GTX 1070++)

Vega 10 = 4096 CFX gen 9 stream processors -> small Vega (to go up against GTX 1080/1080TI)
Remember this ? http://www.guru3d.com/news-story/amd-vega-10-vega-20-and-vega-11-gpus-mentioned-by-cto.html (replace current Fury (X) series in Q1 2017)

Vega 11 = More then 4096 CFX gen 9 stream processors -> big Vega (to go up against Titan XP/uncut GP102+)
Remember this ? https://forums.anandtech.com/threads/vega-navi-rumors.2486940/page-9#post-38492317 (unknown release date)

Go big or go home

 

[Valantar]

WX7100 is not Ellesmere XT2. It is just Ellesmere XT.

Polaris 12 in theory could be smaller part than Polaris 11. However for that part of the segment AMD can use 28 nm GPUs.

Polaris 12 can be GPU used in Project Scorpio. 6 TFLOPs - 3072 GCN cores at under 1000 MHz core clock.

a) You're completely misreading my post. Read it again.

b) Sure, they could. But if power efficiency was a point, they wouldn't. Power efficiency is the main reason for modern process nodes and architectures to trickle down to the low-end.

c) Why on earth would the Project Scorpio GPU be referenced in OSX system files? This makes zero sense. Even if a similar GPU/APU setup were to be used in an upcoming Mac, this would make no sense. Scorpio is based on a semi-custom part, shared with no other products. As such, even if an iGPU with the same number of SMs and whatnot was integrated into a Mac APU, it would have a different codename. The only semi-rational explanation for this would be that some engineers at AMD wanted to see if they could Hackintosh a pre-production Scorpio for fun. I very, very, very much doubt this is the case. They certainly wouldn't be paid to do this, and they wouldn't have access to hardware to do this in their spare time. Heck, it would probably get them fired.

 

[SlickR12345]

Going from AMD's wording, bigger numbers means lower end products. So Polaris 12 is probably a new low end GPU for notebook and light laptops, something along the lines of the RX 460.

Polaris 10xt2 is probably an updated GPU. If the 14nm yields and process is more mature to consume less power, and AMD has managed to identify some weak spots on Polaris and improve them, we are going to see those GPU's as refreshes with higher clocks and a same watt usage.

 

[Det0x]

@ Valantar

If they already have "developed" a 6 TFLOPs - 3072 GCN part to be used in the Project Scorpio APU, why wouldn't they reuse what they already have made and release a GPU as "Polaris 12" with a few unneeded bits disabled (in relation to the GPU in the APU) ?

Would basically mean that they have 2 income sources from the same work. It would be stupid not to do so, when they already are manufacturing the chip/logic.

Glo only used the "Scorpio" name to describe the specs for "Polaris 12"... (3072GCN)

@ SlickR12345

2 posts above:

Despite 11 being a larger number than 10, Polaris 10 is actually a larger, higher end GPU than Polaris 11. Chief Architect & head of AMD’s Radeon Technologies Group confirmed at Capsaicin that the naming scheme is time-based & Polaris 10 had simply been designed before Polaris 11, hence the smaller numerical designation.

 

[tential]

Going from AMD's wording, bigger numbers means lower end products. So Polaris 12 is probably a new low end GPU for notebook and light laptops, something along the lines of the RX 460.

Polaris 10xt2 is probably an updated GPU. If the 14nm yields and process is more mature to consume less power, and AMD has managed to identify some weak spots on Polaris and improve them, we are going to see those GPU's as refreshes with higher clocks and a same watt usage.

It'd already been explained in the thread that amd simply named the chips based on the order they were developed.... You don't need to guess (probably wrongly so) that p12 is small if it even exists. It would just be the next chip amd developed in that lineup

 

[dacostafilipe]

If they already have "developed" a 6 TFLOPs - 3072 GCN part to be used in the Project Scorpio APU, why wouldn't they reuse what they already have made and release a GPU as "Polaris 12" with a few unneeded bits disabled (in relation to the GPU in the APU) ?

You guys are confusion technology with actual hardware ...

No, they can't reuse the GPU part of an APU for doing a desktop GPU.

Yes, they can reuse the technology to spin a new desktop GPU.

But by reusing the technology, they will need to re-optimize everything from scratch ... it's not like you can copy/paste from an APU. Certainly, some blocks could be reused 1:1 but not the entire GPU part.