Jump to content

R9 Nano

Brumbaer

By Brumbaer,
in ATi

 Share
129 posts in this topic

Recommended Posts

Brumbaer
Brumbaer
Posted
Posted

I tested a R9 Nano in my  "Sierra" system.
It is an ITX system and I can only fit a small card in the case.
Up to now I used an Asus GTX970 Mini OC.
 
The R9 Nano runs louder and hotter than the 970.
 
The values for Nano and GTX970 are from the same computer, the values for the 980TI from a comparably equipped computer.
The values are always in the same order:  R9 Nano, Asus GTX970 Mini OC, GTX980TI
CineBench R15 OpenGL 142, 138, 151
CineBench R15 OpenGL 1032, 1036, 1095

Heaven, Benchmark, Basic, 4269, 4103, 4921
Heaven, Benchmark, Extreme 1391, 1520, 2047

Bruce 19.75, 20, 15

Cinema 4D model first render (data is cached) 9.3, 9.4, 8.9

Single test (as in tested only once) for hibernation worked.
Those benchmarks are not relevant for all applications. When using Aperture the Nano seemed more responsive than the 970 - that's an impression, not the result of an scientific test :)

All in all I'm disappointed. 970 and Nano seem to be on par. I somehow expected the Nano to be faster.

 

EDITED:

See answer from

 

 

Link to comment
Share on other sites
Ciro82
Ciro82
Posted
Posted

I tested a R9 Nano in my  "Sierra" system.

It is an ITX system and I can only fit a small card in the case.

Up to now I used an Asus GTX970 Mini OC.

 

The R9 Nano runs louder and hotter than the 970.

 

The values for Nano and GTX970 are from the same computer, the values for the 980TI from a comparably equipped computer.

The values are always in the same order:  R9 Nano, Asus GTX970 Mini OC, GTX980TI

CineBench R15 OpenGL 142, 138, 151

CineBench R15 OpenGL 1032, 1036, 1095

 

Heaven, Benchmark, Basic, 4269, 4103, 4921

Heaven, Benchmark, Extreme 1391, 1520, 2047

 

Bruce 19.75, 20, 15

 

Cinema 4D model first render (data is cached) 9.3, 9.4, 8.9

 

Single test (as in tested only once) for hibernation worked.

Those benchmarks are not relevant for all applications. When using Aperture the Nano seemed more responsive than the 970 - that's an impression, not the result of an scientific test :)

 

All in all I'm disappointed. 970 and Nano seem to be on par. I somehow expected the Nano to be faster.

 

Do the test in OpenCL and you will see the Nano kicking the 2 nvidia cards butts.

  • Like 1
Link to comment
Share on other sites
Brumbaer
Brumbaer
Posted
  • Author
Posted

I believe it when I see it.


In my experience the great OpenCL advantage of ATI cards is a thing of the past, but still fondly retold. Like Skylakes are hard to turn into Hackintoshes etc. If I remember correctly the fastest OpenCL cards in Compubench were NVidias.


FCP is supposed to use OpenCL, but the difference in Bruce is negligible to the 970 and noticeable behind the TI.


Anyway if the software you run, doesn't use a certain technology (Cuda, OpenCL etc) you will not realise any advantage by it. 


I'm away for some days, I will look into it, once I'm back. Especially I will look whether my impression regarding Aperture is valid and I will run some OpenCL benchmarks.


See you than.

Link to comment
Share on other sites
Dwarfy
Dwarfy
Posted
Posted

Correct me if I am wrong. AppleGraphicsDevicePolicy is mostly for AMD GFX and Intel iGFX. MacOS/OS X will always work better with AMD and Intel, as Apple is only optimising for those GPU's. Yes Nvidia works, but if you see now, new Mac Profiles, they are not with Nvidia. Besides, we'd have to rely on web drivers. Sorry for getting off-topic. 

  • Like 2
Link to comment
Share on other sites
MattsCreative
MattsCreative
Posted
Posted

 

I believe it when I see it.

In my experience the great OpenCL advantage of ATI cards is a thing of the past, but still fondly retold. Like Skylakes are hard to turn into Hackintoshes etc. If I remember correctly the fastest OpenCL cards in Compubench were NVidias.

FCP is supposed to use OpenCL, but the difference in Bruce is negligible to the 970 and noticeable behind the TI.

Anyway if the software you run, doesn't use a certain technology (Cuda, OpenCL etc) you will not realise any advantage by it. 

I'm away for some days, I will look into it, once I'm back. Especially I will look whether my impression regarding Aperture is valid and I will run some OpenCL benchmarks.

See you than.

 

AMD when it comes to opencl is king they always will be and nvidia has poor metal support also which you will need in the future as 10.12's ui and system is rendered on it. nvidia has many many issues and needs a few fixes my 290x beats the 970 in everything but for a 1440p benchmark now the issue i have is have you set up the nano to work 100% in os x? and have you flashed a custom bios with a proper fan curve to remove any throttling it can have?

  • Like 2
Link to comment
Share on other sites
  • 2 weeks later...
  • 3 months later...
okrasit
okrasit
Posted
Posted

I'll make patches...


Quick and dirty:

 

SLgCAAAAAQAAAEiJQ1THQ3wIAAAASI0FCUk/AP+QSAUAAGbHg5wAAAABAUiDxAhb

SLgEAAAAAQAAAEiJQ1THQ3wgAAAASI0FCUk/AP+QSAUAAJCQkJCQkJCQkEiDxAhb

This is where the magic is, parameters for the parent AMDHardware class.

 

RTnBcp+Lg4AAAACLS3w5yA9CyImLgAAAAESIs5kAAABEiHMgW0FeXcOQVUiJ5UFXQVZBVUFUU1BJif6/YAEAAA==

RTnBcp+Lg4AAAACLS3w5yJCQkImLgAAAAESIs5kAAABEiHMgW0FeXcOQVUiJ5UFXQVZBVUFUU1BJif6/YAEAAA==

There's a "cap" field in the static structure, that limits the cu count to max 32, this one just ignores it.
 
6B1nAgBIid/oTuQAAEiJ3+hJhf7/vkgBAABMiffoQkf//4XAdCVIid/o8R0BAL5T
6B1nAgBIid/oTuQAAEiJ3+hzAgEAvkgBAABMiffoQkf//4XAdCVIid/o8R0BAL5T
Call _Cail_Fiji_InitFuctionPoiter instead of Baffin one. It does work without this, although I rather init my card the Fiji way. (mostly some power management stuff).
 
 
AMDRadeonX4100.kext
  • Like 5
Link to comment
Share on other sites
okrasit
okrasit
Posted
Posted

I would like to ask  if your dual Gpu scale well or a single Gpu have a greater value than 20550...

thank you

 

Yeah, it's exactly half of that 41k!

These things (Nanos) do tend to throttle due to temperature. I've got to find water blocks for these...

Brumbaer, can you re-run those benchmarks? Just to see if it helps with the ogl performance.

Also, in AMD9000Controller.kext's Info.plist

 

<key>PP_DisablePowerContainment</key>

 

<integer>1</integer>

 

for obvious reasons.

Link to comment
Share on other sites
Brumbaer
Brumbaer
Posted
  • Author
Posted

The patches do not work with 12.1, but they work on the latest Developer Beta(12.2.6). Didn't try anything in between.

 

The LuxBall Score is 19834.

 

This is some data I compiled in the hackintosh-forum.de.

 

It's sadly lacking the LuxMark scores from the GTX980. It would be great, if somebody could provide them (LuxMark 3.1, Ball, Microphone and Hotel - GPU only).

 

Theses are scores sorted by LuxBall - GPU only and Heaven Score Preset:Extreme

 

post-1634720-0-20348400-1481581003_thumb.png

 

Table as Image. Editor doesn't understand HTML or BBC Tables

Link to comment
Share on other sites
okrasit
okrasit
Posted
Posted

The patches do not work with 12.1,  but they work on the latest Developer Beta(12.2.6). Didn't try anything in between.

 

The LuxBall Score is 19834.

 

This is  some data I compiled in the hackintosh-forum.de.

 

It's sadly lacking the LuxMark scores from the GTX980. It would be great, if somebody could provide them (LuxMark 3.1, Ball, Microphone and Hotel - GPU only).

 

Theses are scores sorted by LuxBall - GPU only and Heaven Score Preset:Extreme

 

 

Table as Image. Editor doesn't understand HTM or BBC Tables

 

 

There's definitely something weird about OpenGL & AMD on OS X. It performs so poorly... GFXBench Metal would be interesting, though.

The patches are a bit excessive, they could be cut shorter to make them work on other Sierra versions too.

Link to comment
Share on other sites
Brumbaer
Brumbaer
Posted
  • Author
Posted

The average-ratio of OpenCL Score to OpenGL Score is about 12 for AMD and about 6 for NVidia (more like 7.5 for newer cards).

The LuxMark Scores are in the same realm on Windows and MacOS. Didn't find any score list for Heaven under Windows.

 

As soon as you patch code not tables, the chances are high that it will not work after a recompile, as register use and offsets will most likely change, so I wouldn't bother about making them shorter.

Link to comment
Share on other sites
okrasit
okrasit
Posted
Posted

You delete EFI part of bios? How did you do that? 

 

For example:

https://www.techpowerup.com/vgabios/152284/asus-r9290x-4096-131206

 

It's pci-id is 67b0:1002

 

At offset 0x230:

00000220   43 10 6C 04  2C 02 1C A0  C2 A0 A0 00  50 43 49 52  C.l.,.......PCIR

00000230   02 10 B0 67  00 00 18 00  00 00 00 03  80 00 29 0F  ...g..........).

00000240   00 00 00 00  41 4D 44 20  41 54 4F 4D  42 49 4F 53  ....AMD ATOMBIOS

replace B0 67 with something that doesn't match e.g. B5 67
 
And the EFI blob at offset 0x10020

00010000   55 AA 71 00  F1 0E 00 00  0B 00 64 86  01 00 00 00  U.q.......d.....

00010010   00 00 00 00  00 00 58 00  1C 00 00 00  50 43 49 52  ......X.....PCIR

00010020   02 10 B0 67  00 00 18 00  00 00 00 03  71 00 00 00  ...g........q...

Same thing B0 67 -> B5 67

 

Flash it to the card and you'll lose boot screen but it'll work with Sierra.

  • Like 2
Link to comment
Share on other sites
 Share
Go to topic listing
×
×
  • Create New...