AMD Polaris IDs on Sierra / High Sierra

[MattsCreative]

Try to run a compute benchmark on geekbench 4 and you will see... 

My RX480 does 93.000 on Mac OS and 125.000 on windows, so there is a big difference

Cattura.PNG

ya that is a bit low my 290x gets 169k https://browser.primatelabs.com/v4/compute/188880

[Gigamaxx]

Matteo, of course it "works", but not the right way

And if you do a benchmark you will see low scores, because the GPU is not working properly. 

If you set your windows RX 480 up with integrated graphics and run the same benchmarks in open GL your scores will be very close.   I posted comparative benchmarks and there was 40-50 point difference between a "Working Card" as you would call it in windows and the sierra "working card" that I mention.        I ran a 1830 in Sierra and 1875 in Windows in open gl.     So to say it only has half a frame buffer 67ef, or baffin is baffling to me because my rx 460 should be matching these scores but it falls way below.    

 

So as far as working as well as it does in windows 10 with direct X12, no but comparing apples to apples it runs right up there with windows in Unigine Valley extreme hd in open Gl format which is what we're limited to in Sierra.    If you're waiting for the all inclusive "working' definition that may never be as it also effects other "working" AMD cards in Sierra.    Nvidia cards are also having trouble in Sierra are they not?   In fact, HD 530 has issues, Hd 4600 many graphics are not "Working" according to your high standard.

 

From the list I provided, can you get an MSI 480 to even work nominally with metal, an Asus?    Do you have a list of "Dysfunctionally working cards" in Sierra?    This list will help people that want to try an RX in Sierra, and if they buy a Gigabyte or Asus 480 they may get frustrated because nobody mentioned a lack of compatibility.    

 

I started another thread but you moved it so where do I post?     I remember you telling me that there was no Polaris 10 in sierra, and that the Baffin frame buffer was totally developed only to see the 4100ket show up a week later.    Vega10 is now here as well in the 9500ControllerKext.     I really can't wait to get a vega card "working" so you can tell me its not.    If I didn't know better I would think you don't want RX cards to work so you're prior predictions can be taken as correct.   I don't care about being riight or wrong, no progress was ever achieved without mistakes and more importantly admitting them.   Can you?

 

I bid you all ado!

[MattsCreative]

If you set your windows RX 480 up with integrated graphics and run the same benchmarks in open GL your scores will be very close.   I posted comparative benchmarks and there was 40-50 point difference between a "Working Card" as you would call it in windows and the sierra "working card" that I mention.        I ran a 1830 in Sierra and 1875 in Windows in open gl.     So to say it only has half a frame buffer 67ef, or baffin is baffling to me because my rx 460 should be matching these scores but it falls way below.    

 

So as far as working as well as it does in windows 10 with direct X12, no but comparing apples to apples it runs right up there with windows in Unigine Valley extreme hd in open Gl format which is what we're limited to in Sierra.    If you're waiting for the all inclusive "working' definition that may never be as it also effects other "working" AMD cards in Sierra.    Nvidia cards are also having trouble in Sierra are they not?   In fact, HD 530 has issues, Hd 4600 many graphics are not "Working" according to your high standard.

 

From the list I provided, can you get an MSI 480 to even work nominally with metal, an Asus?    Do you have a list of "Dysfunctionally working cards" in Sierra?    This list will help people that want to try an RX in Sierra, and if they buy a Gigabyte or Asus 480 they may get frustrated because nobody mentioned a lack of compatibility.    

 

I started another thread but you moved it so where do I post?     I remember you telling me that there was no Polaris 10 in sierra, and that the Baffin frame buffer was totally developed only to see the 4100ket show up a week later.    Vega10 is now here as well in the 9500ControllerKext.     I really can't wait to get a vega card "working" so you can tell me its not.    If I didn't know better I would think you don't want RX cards to work so you're prior predictions can be taken as correct.   I don't care about being riight or wrong, no progress was ever achieved without mistakes and more importantly admitting them.   Can you?

 

I bid you all ado!

you sir have everything 1000000% wrong let me help you out shall i : 

 

first the rx480 in os x on valley and heaven bench the same without or without the igpu as the igpu/rx480 do not stack in power or compute power

next metal support does not and will never be depended on the brand of gpu it's all hardware 

and last please don't make these half assed post just don't .....please 

[Gigamaxx]

Working for me with modified AMDRadeonX4100.kext and baffin framebuffer in clover config

 

What brand and model if you could please?

you sir have everything 1000000% wrong let me help you out shall i : 

 

first the rx480 in os x on valley and heaven bench the same without or without the igpu as the igpu/rx480 do not stack in power or compute power

next metal support does not and will never be depended on the brand of gpu it's all hardware 

and last please don't make these half assed post just don't .....please 

Do you know of anyone with a semi, or partially metal supported Gigabyte or asus rx 480?     There are some cards that have not been able to get metal support with the kext change, can you disprove this?     Nice of you to chime in though, do you have a RX card in testing?

[Nightf4ll]

Try to run a compute benchmark on geekbench 4 and you will see... 

My RX480 does 93.000 on Mac OS and 125.000 on windows, so there is a big difference

Cattura.PNG

damn, that seems hard.. its almost 30% difference.

 

for some reason geekbench gives me inconsistent results, variating from 60 k to 97..

 

[Ciro82]

damn, that seems hard.. its almost 30% difference.

 

for some reason geekbench gives me inconsistent results, variating from 60 k to 97..

 

As you can see, something is not working properly

[MattsCreative]

As you can see, something is not working properly

ya on windows i get 190k so even for me it's a bit off

[Matthew82]

Little off-topic...When I updated system to 10.12.2 iStat menu start showing gpu processor usage when rendering project in FCPX (earlier only memory usage was show)

 

 

 

[MattsCreative]

Little off-topic...When I updated system to 10.12.2 iStat menu start showing gpu processor usage when rendering project in FCPX (earlier only memory usage was show)

it's always done that for me i only use my gpu for video rendering 

[Gigamaxx]

damn, that seems hard.. its almost 30% difference.

 

for some reason geekbench gives me inconsistent results, variating from 60 k to 97..

 

Screen Shot 2016-11-02 at 23.17.59.png

 

Is it open CL?     

 

In open Cl Sierra only supports 16 compute units, whereas the RX 480 has 36.

But this second pic is a crossfire of Sapphire RX 470 nitro and XFX RX 480 reference card.

As you can see it doubled the score, but if all 36 compute units were running it would be much higher.

 

In open GL the support is much more even.

Windows DX11 Dedicated graphics card only

Windows open GL Igpu/RX

Sierra open GL igpu/RX

 

There is a 3% difference in performance in open Gl formats between Windows and Sierra.   3%   

 

Vega 10 the AMD Fury Pro? 

[Frisa]

Ellesmere is the Chip Family, it's not a Framebuffer

RX480/470 FB's are Dayman and Guariba (i guess).

RX470 has the same device id 67DF.  

 

@Slice can you add the 9500controller in ATI.h and ATI.c with related Framebuffers? Please

no,there are indeed something wrong in device list,check out the string for 280

  { 0x6798,    0x00000000, CHIP_FAMILY_TAHITI,    "AMD Radeon HD 7970X/8970/R9 280X",    kFutomaki    },

and here for 480:

  { 0x67DF,    0x00000000, CHIP_FAMILY_ELLESMERE,    "AMD Radeon RX480",         kNull    },

280 string was right which framebuffer for tahiti chips is futomaki, where all framebuffer setting for polaris chips are "Null"

 

@Gigamaxx i believe we define working as the system could be booted with discrete gpu alone without the igpu, no needing for edit kext files,basic functions like Quartz and opengl could be rendered with hard acceleration,booting with igpu definitely give you a partially accelerated system,but first two requirements are not satisfied currently,don't forget many people using Xeon E5 or X99 mobo

and even you got a accelerated system, the drivers tend to crash when under high load or complex operations,just trying enable opencl in FCPX

[Nightf4ll]

Is it open CL?     

 

In open Cl Sierra only supports 16 compute units, whereas the RX 480 has 36.

But this second pic is a crossfire of Sapphire RX 470 nitro and XFX RX 480 reference card.

As you can see it doubled the score, but if all 36 compute units were running it would be much higher.

 

There is a 3% difference in performance in open Gl formats between Windows and Sierra.   3%

 

so, there should be no difference between rx460 opencl tests?? Also, windows sounds more and more like the better option, too bad I installed it on the slower HD... os x development is really disappointing the last few years (since steve jobs died).

 

[Ciro82]

no,there are indeed something wrong in device list,check out the string for 280

  { 0x6798,    0x00000000, CHIP_FAMILY_TAHITI,    "AMD Radeon HD 7970X/8970/R9 280X",    kFutomaki    },

and here for 480:

  { 0x67DF,    0x00000000, CHIP_FAMILY_ELLESMERE,    "AMD Radeon RX480",         kNull    },

280 string was right which framebuffer for tahiti chips is futomaki, where all framebuffer setting for polaris chips are "Null"

 

Yesterday we added the missing infos in Ati.h and Ati.c, now Clover adds the correct Framebuffer (Dayman) and ports number (6), but same issue (black screen) happens w/o iGPU.

[moob]

What brand and model if you could please?

Do you know of anyone with a semi, or partially metal supported Gigabyte or asus rx 480?     There are some cards that have not been able to get metal support with the kext change, can you disprove this?     Nice of you to chime in though, do you have a RX card in testing?

 

Reference RX 480

 

[Pavo]

One thing all you guys are not getting on these benches is, if you have iGPU it will also use that with the GFX card for benching, which is why some people's score are higher than others without iGPU on same model of GFX card. There is literally no reason to be comparing Windows benches with macOS benches, they are done in 2 completely different ways. Also all RX 480s, doesn't matter what manufacture still requires kext edit and using the wrong accelerator as Ciro82 stated. So please bring this topic back to it intended purpose of testing different methods of getting RX 480 supported.

[ErmaC]

@Slice can you add the 9500controller in ATI.h and ATI.c with related Framebuffers? Please

 

Sorry if I'm not Slice

 

but you can do a quick test? I don't expect miracles...

(I just add what you suggest)

-> removed

ErmaC

[Frisa]

Sorry if I'm not Slice

 

but you can do a quick test? I don't expect miracles...

(I just add what you suggest)

-> removed

 

ErmaC

thanks,tested,still hanged on boot

btw,does slice knowing this bug?he did not reply the ticket in sourceforge

[firefly_can_fly]

thanks,tested,still hanged on boot

btw,does slice knowing this bug?he did not reply the ticket in sourceforge

kmon, sourceforge is open project, ant to review all opened tickets, you need to make couple clicks.

about slice knowing - more yes then no, but still, you can ask him by yourself - he is active on this forum.

[moob]

One thing all you guys are not getting on these benches is, if you have iGPU it will also use that with the GFX card for benching, which is why some people's score are higher than others without iGPU on same model of GFX card. There is literally no reason to be comparing Windows benches with macOS benches, they are done in 2 completely different ways. Also all RX 480s, doesn't matter what manufacture still requires kext edit and using the wrong accelerator as Ciro82 stated. So please bring this topic back to it intended purpose of testing different methods of getting RX 480 supported.

Absolutely wrong, there is not any technologies that support a mixed elaboration between two graphics card of different manufactures. Windows does not support it and not event macOS. The benchmark uses the GPU of the screen where it is running, is the SAME thing on Windows.

 

(I have three monitor, two connected to RX 480 and one to HD 4600)

 

Valley running on the screen connected to HD 4600:

 

Valley running on the screen connected to RX 480:

[Pavo]

Absolutely wrong, there is not any technologies that support a mixed elaboration between two graphics card of different manufactures. Windows does not support it and not event macOS.

 

You might want to do some research before saying someone is wrong because this little piece of technology that you say doesn't exist is call FDI. Flexible Display Interface is an interconnect created by Intel in order to allow the communication of the HD Graphics integrated GPU found on supported CPUs with the PCH southbridge. It provides a path between an Intel processor and an Intel southbridge on a computer motherboard which carries display data from the graphics controller (North Display) of the Intel processor package that are attached at some PCH (South Display) versions. Currently it supports 2 independent 4-bit fixed frequency links/channels/pipes at 2.7Gbit/s data rate. Which was introduced in 2010 on the i3 and i5 CPUs and chipsets.

[Frisa]

You might want to do some research before saying someone is wrong because this little piece of technology that you say doesn't exist is call FDI. Flexible Display Interface is an interconnect created by Intel in order to allow the communication of the HD Graphics integrated GPU found on supported CPUs with the PCH southbridge. It provides a path between an Intel processor and an Intel southbridge on a computer motherboard which carries display data from the graphics controller (North Display) of the Intel processor package that are attached at some PCH (South Display) versions. Currently it supports 2 independent 4-bit fixed frequency links/channels/pipes at 2.7Gbit/s data rate. Which was introduced in 2010 on the i3 and i5 CPUs and chipsets.

i though it have nothing to do with opengl performance,its just provide a channel for IGP video output

indeed its possible to collaborate the IGP and discrete GPU, but it required to be supported by both GPU, like AMD's hybrid graphics

[moob]

You might want to do some research before saying someone is wrong because this little piece of technology that you say doesn't exist is call FDI. Flexible Display Interface is an interconnect created by Intel in order to allow the communication of the HD Graphics integrated GPU found on supported CPUs with the PCH southbridge. It provides a path between an Intel processor and an Intel southbridge on a computer motherboard which carries display data from the graphics controller (North Display) of the Intel processor package that are attached at some PCH (South Display) versions. Currently it supports 2 independent 4-bit fixed frequency links/channels/pipes at 2.7Gbit/s data rate. Which was introduced in 2010 on the i3 and i5 CPUs and chipsets.

it has nothing to do with parallel computation between GPUs, it is only a way to connect two GPU into one PC but independently from each other. Actually there are some technologies that support this thing, like DirectX 12 or SLI or CrossFire and macOS does not support no one of these. Just inform yourself before posting to avoid misinformation on public forums.

 

From: https://en.wikipedia.org/wiki/DirectX#DirectX_12

Multiadapter support will feature in DirectX 12 allowing developers to utilize multiple GPUs on a system simultaneously, multi-GPU support was previously dependent on vendor implementations such as AMD CrossFireX or NVIDIA SLI.

Implicit Multiadapter support will work in a similar manner to previous versions of DirectX where frames are rendered alternately across linked GPUs of similar compute-power. Explicit Multiadapter will provide two distinct API patterns to developers. Linked GPUs will allow DirectX to view graphics cards in SLI or CrossFireX as a single GPU and use the combined resources. Whereas Unlinked GPUs will allow GPUs from different vendors to be utilized by DirectX, such as supplementing the dedicated GPU with the integrated GPU on the CPU, or combining AMD and NVIDIA cards. However, elaborate mixed multi-GPU setups requires significantly more attentive developer support.

 

From: https://en.wikipedia.org/wiki/AMD_CrossFireX

AMD CrossFire (also known as CrossFireX) is a brand name for the multi-GPU technology by Advanced Micro Devices, originally developed by ATI Technologies. The technology allows up to four GPUs to be used in a single computer to improve graphics performance.

Associated technology used in mobile computers with external graphics cards, such as in laptops or notebooks, is called AMD Hybrid Graphics.

 

From: https://en.wikipedia.org/wiki/Scalable_Link_Interface

Scalable Link Interface (SLI) is a brand name for a multi-GPU technology developed by NVIDIA for linking two or more video cards together to produce a single output. SLI is an algorithm of parallel processing for computer graphics, meant to increase the processing power available for graphics. The initialism SLI was first used by 3dfx for Scan-Line Interleave, which was introduced to the consumer market in 1998 and used in the Voodoo2 line of video cards. After buying out 3dfx, NVIDIA acquired the technology but did not use it. NVIDIA later reintroduced the SLI name in 2004 and intended for it to be used in modern computer systems based on the PCI Express (PCIe) bus; however, the technology behind the name SLI has changed dramatically.

[joe75]

Flexible Display Interface. Currently it supports 2 independent 4-bit fixed frequency links/channels/pipes at 2.7Gbit/s data rate. Which was introduced in 2010 on the i3 and i5 CPUs and chipsets.

 

FDI is a fancy name in an effort to distract from the facts of its own limitations.. Its called marketing! 

[Revod]

Hey guys, I'm newbie and noob with this, My RX 480 XFX Reference, 8gb was detected when I plug the HDMI cable, on start the the black screen but after power on I get the login image and my gpu is detected, but is so slow and cant do anything, doesn't have acceleration (I guess), what should I do to fix that?, I'm on the latest macOS version 10.12.1

 

Edit: hey @moob can you please can say to me what you do to make your card fully working please?

[Ciro82]

Hey, I'm newbie and noob with this, My RX 480 XFX Reference, 8gb was detected when I plug the HDMI cable, on start the the black screen but after power on I get the login image and my gpu is detected, but is so slow and cant do anything, doesn't have acceleration (I guess), what should I do to fix that?, I'm on the latest macOS version 10.12.1

Because you need to modify the X4100 kext by adding the 67DF id... Or by injecting via DSDT the rx480 with the fake ID 67FF

 

 

Inviato dal mio iPhone utilizzando Tapatalk