Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP All Models Tired of low Geekbench scores? Use RadeonBoost!
- Thread starter CMMChris
- Start date
- Sort by reaction score
That's normal. It's a codeless kext. You have to check if RadeonBoost is loaded indirectly.
e.g.
Bios number is from 1.3 or earlier, RadeonBoost is fine and means it’s working.
Hi - I am complete newbie here.
I got RX580 8gb with 35k metal result.
I got the kext installed but it shows me only VBIOs as radeonboost. Efi version does not exist in Graphics details.
Nothing has changed in the geekbench 5 tests.
Catalina 10.15.6 opencore.
Any idea what should I do?
I got RX580 8gb with 35k metal result.
I got the kext installed but it shows me only VBIOs as radeonboost. Efi version does not exist in Graphics details.
Nothing has changed in the geekbench 5 tests.
Catalina 10.15.6 opencore.
Any idea what should I do?
Last edited:
Hi there , very curious what the Radeon boost will do to the performance of my 5.1 12 core Mac Pro running mojave 10.14.6 and a Sapphire pulse radeon RX580/ 8gb while using real time generative software like Dericative Touch Designer. Would this allow me to use more particles in Touch Designer? Or is this ktext only beneficial for rendering speeds?
Nothing, the RX580 driver is quite optimised already. No much you can boost.
As long as you can see that RadeonBoost in VBIOS, it means the kext is working.
I've just found out that if I go to sleep - wake - run geekbench 5, it rocketed:
Metal 36 000 -> 56 000
Open Gl 31 000 -> 41 000
So it helped - but why it does not work after reboot? It only works after sleep/wake.
Please advise and thank you.
Last edited:
We have the same feature - rising Geekbench values from mid to high after first sleep - with mobile MXM slot cards of type AMD WX4130, WX4150, WX4170, and WX7100 in iMac systems. All cards are Polaris based.
I assume the boot process when switching from the white apple logo and progress bar to the login screen does not really initialize all drivers correctly. I can live with this, before using this extensions we had it the other way around, even worse, dropping from mid to low values. Same happend with OpenGL benches like Valley. Now these OpenGL results are constant and do not change.
I got RX-480 and with iMacPro identity through OC I get 40332 OpenCL Score and 48023 Metal Score without going through sleep. I don't use RadeonBoost, but device properties injection.
Since we use OpenCore with the 2011 iMacs too it might be an option to use the property injection. I will give it a try.
Thanks!
I tried with my WX7100 mobile, didn't change anything, though (with or without doubling PP,PP_xyz or CFG,CFG_xyz according to IORegistryExplorer)...
Only thing that worked was adding CFG,CFG_FORCEMAXDPM and CFG,CFG_FORCE_MAX_DPS as last parameter and setting them to TRUE (YES in Xcode). It seems to be important that the Falcon SMC FW is loaded before, otherwise those params have no effect for me. Maybe this also holds true for other params.
This results in max clocks for my card which is otherwise stuck at 300MHz SCLK under MacOS (no probs under Ubuntu though->likely a MacOS / OC / driver / config problem).
Regards,
roscho
Using some tools like HW Monitor and of course several benchmarks as Valley. We have this "feature" with MXM cards plugged into iMacs (late 2009 - mid 2011) of type WX4130/WX4150/WX4170/WX7100. Injecting a few values using OpenCore or a simplified version of the RadeonBoost makes it disappear.
[automerge]1596014711[/automerge]
Would you like to share your settings, please
A config.plist snap in text form would be great!I doubt if that 300MHz is reported erroneously. e.g. the software only report the idle clock speed.
However, the GPU itself actually can work to max boost clock speed.
First we installed these cards, then we figured out the performance is bad by running Valley (OpenGL) and GeekBench5 and other load generation tools.
Next step was cross checking the core frequency using some tools, followed by searching for a solution. Injecting some values from this RadeonBoost extension made it work and finally we got better performance and higher clock rates shown by the tools I just mentioned. @Nick [D]vB made the BIOS and gave me some hints about the direction.
Is there a specific reason you doubt the tools or the Valley benchmark results?
Using some tools like HW Monitor and of course several benchmarks as Valley. We have this "feature" with MXM cards plugged into iMacs (late 2009 - mid 2011) of type WX4130/WX4150/WX4170/WX7100. Injecting a few values using OpenCore or a simplified version of the RadeonBoost makes it disappear.
[automerge]1596014711[/automerge]
Would you like to share your settings, please
Please ;-)
Yeah, because you haven't mention that you determine that from the scores. And some software is know that cannot report the GPU clock speed correctly.
e.g. This is from my Radeon VII. Luxmark report 350MHz. People often complain the GPU stuck at that speed (in various forum). But if they also look at the score, they should able to tell their GPU was actually working correctly.
Anyway, obviously you know what you are looking for. And you did have GPU performance issue.
I do not use any benchmark software but Apples ioreg itself. This way:
Code:
roscho$ while sleep 1; do ioreg -l |grep \"PerformanceStatistics\" | cut -d '{' -f 2 | tr '|' ',' | tr -d '}' | tr ',' '\n'|grep 'Temp\|Fan\|Clock\|GPU Activity'; doneThis polls the actual clocks every second via ioreg.
Without the above mentioned parmeters the Core CLK is stuck at 300MHz but the Mem Clock risies under load until 1250MHz.
With ...FORCE... after Polarisboost the Core Clock rises to 1243MHz and OpenCL performance is significantly better (GB4 45000 vs 35000). By the way, if it comes to significance or standard error it is wiser to compare sets of repeated measures (e.g. t-test & S.E.).
Also the GPU load in Activity Monitor (CMD+4) is higher and the GPU heatsink tempt rises higher under load.
EDIT: And Luxmark reports correctly, although the result is disappointing. But maybe i have to play around with different parameters.
regards,
roscho
[automerge]1596022164[/automerge]
Would you like to share your settings, please
For sure, this is the relevant PolarisBoost Config.plist part, but the 2 params should also work in the same place in RadeonBoost:
Code:
<key>IOProviderMergeProperties</key>
<dict>
<key>ATY,EFIVersionB</key>
<string>PolarisBoost</string>
<key>CFG,CFG_NVV</key>
<integer>2</integer>
<key>CFG,CFG_PTPL2_TBL</key>
<data>ggAAAHwAAAB2AAAAcAAAAGoAAABkAAAAXgAAAFgAAABSAAAATAAAAEYAAABAAAAAOgAAADQAAAAuAAAAKAAAAA==</data>
<key>CFG,CFG_USE_CP2</key>
<true/>
<key>PP,PP_EnableLoadFalconSmcFirmware</key>
<integer>1</integer>
<key>PP,PP_Falcon_QuickTransition_Enable</key>
<integer>1</integer>
<key>PP,PP_WorkLoadPolicyMask</key>
<integer>16</integer>
<key>CFG,CFG_FORCE_MAX_DPS</key>
<true/>
<key>CFG,CFG_FORCEMAXDPM</key>
<true/>
</dict>Attachments
Last edited:
Yeah, Luxmark can report the max boost clock speed in Catalina regardless what the actual clock speed is. But in Mojave, it reports the idle clock speed.
TBH, I really don't expect the WX7100 has such poor support on the iMac. In the old Mac Pro, it's really just plug and play. And I actually downclocked my RX580 to 1243MHz to simulate WX7100 performance / power draw.
Sapphire PULSE RX580 8GB VBIOS Study
*** For info, this mod is NOT compatible with the OpenCore boot screen mod *** This thread is about how to mod the Sapphire PULSE RX580 8GB card's ROM image. If you worry about bricking the card, or void the warranty etc. STOP HERE! For those who has MVC flashed RX580. Do NOT perform this mod...
forums.macrumors.com
And this is the Luxmark result.
Even the cooling isn't that good on the iMac, but the performance can't be 65% lower. 12000-13000 make much more sense.
Anyway, wish you guys can find the key to unlock all the performance, good luck!
I changed the relevant section in my installation of radeonboost with this and from 48K in the geekbench metal test it went up to 54K+. That is impressive (I think). This is on a 5.1 macpro with OS 10.15.6 and an rx 580 pulse.
Has anyone noticed any improvements with the apps they use?
Some examples of working with Nick and my wx 4150 mxm card (note in xserve3,1).
example of the loop to check ioreg (without fan speeds since mxm has no fans)
Code:
$ while sleep 2; do ioreg -l |grep \"PerformanceStatistics\" | cut -d '{' -f 2 | tr '|' ',' | tr -d '}' | tr ',' '\n'|grep 'Temp\|Clock'; done|paste - - -
"Core Clock(MHz)"=1016 "Memory Clock(MHz)"=300 "Temperature(C)"=63
"Core Clock(MHz)"=964 "Memory Clock(MHz)"=1500 "Temperature(C)"=63
"Core Clock(MHz)"=998 "Memory Clock(MHz)"=1500 "Temperature(C)"=63
"Core Clock(MHz)"=1025 "Memory Clock(MHz)"=1500 "Temperature(C)"=63
"Core Clock(MHz)"=1008 "Memory Clock(MHz)"=1500 "Temperature(C)"=63Tried a few of his roms, some of which resulted in GPU clock being stuck at 214MHz (original rom idle speed was 300MHz) but Memory was able to hit 1750MHz (vs previous cap of 1500MHz) this of course killed metal bench back to 4k from nearly 14k.
Found windows compute numbers for the WX 4150 ... seems they have OpenCL around 15k-17k, so getting 13k/14k seems like a decent number considering 2x PCIe bus and a slow E5520 2.2Ghz CPU
Based on all of that i'd believe that it may get truly stuck at 300MHz depending on what was tweaked / not loaded properly.
Last edited: