ChefKissInc / NootRX

The rDNA 2 dGPU support patch kext. No commercial use.
https://chefkissinc.github.io/applehax/nootrx/
Other
593 stars 110 forks source link

Lower Geekbench score Than Older version #71

Closed colorzpe closed 2 weeks ago

colorzpe commented 3 months ago

macOS Version

Monterey

What is your CPU's model?

8700k

What is your GPU's model?

6700xt Ref

Please describe the behaviour in detail.

Geekbench score (opencl,metal) lower than nootrx older version and lower than 6600xt

What should've happened instead?

Same score or a bit lower or better

If applicable, attach the .gpuRestart, .panic, etc file related to this issue.

No response

github-actions[bot] commented 3 months ago

Greetings. It seems like this is the first issue you open on this repository.

We are letting you know that these are for bug reports or feature requests. Most of the reports we receive in this GitHub Organisation are user errors.

For the sake of saving time, here are the most common cases:

We will never support the use of configurator software, solutions like OCLP/BFixup, or using macOS beta versions.

This issue will be looked over by the respective maintainer when they can. In the meantime, look if you can resolve this yourself via checking the above.

Be patient, we are hobbyists.

VisualEhrmanntraut commented 3 months ago

The workflow policy was changed because on some GPUs it was causing coil whine.

kadaxo commented 3 months ago

The workflow policy was changed because on some GPUs it was causing coil whine.

Coil whine is normal for RX 6000 series. Talking about my GPU on Windows on high load it also whines. So why cut the performance?

VisualEhrmanntraut commented 3 months ago

Well I didn't know that. I'm a Mac user, I have an M3 MacBook Pro. 😆

kadaxo commented 3 months ago

I made some tests with Cinebench, and Cinebench score is normal.

VisualEhrmanntraut commented 3 months ago

Well, I still think the reason it's lower is not because it's wrong but because it was being boosted due macOS' value of COMPUTE_WORKLOAD edit: might be VR_WORKLOAD instead edit 2: it's COMPUTE_WORKLOAD, but on big sur 16 would be VR_WORKLOAD. weird

VisualEhrmanntraut commented 2 months ago

NootRX-1.0.0-DEBUG.zip NootRX-1.0.0-RELEASE.zip

is this build ok

kadaxo commented 2 months ago

NootRX-1.0.0-DEBUG.zip NootRX-1.0.0-RELEASE.zip

is this build ok

Nope. Same metal score. I get 60k+ more metal score with [bfff2a8] commit btw.

VisualEhrmanntraut commented 2 months ago

The previous high scores were a mistake, it was using Compute workload, but macOS uses VR workload. The compute workload will suck out too much power and cause coil whine on some of the GPUs.

VisualEhrmanntraut commented 2 months ago

The current changes use the default workload. These builds use the correct workload.

VisualEhrmanntraut commented 2 months ago

(You should also make sure to use the "Shaneee" PAT patch and not the "algrey" one for better performance if you are on an AMD CPU)

VisualEhrmanntraut commented 2 months ago

@00x0A I remember you had coil whine, can you try the builds in https://github.com/ChefKissInc/NootRX/issues/71#issuecomment-2313540776 and tell me if they have coil whine?

VisualEhrmanntraut commented 2 weeks ago

Merging with #69 as I theorise this is the primary cause now.

colorzpe commented 2 weeks ago

Sorry for the late response if new version is correct workload and score seem like performance worse than native support 6600xt even Vega 56 Screen Shot 2567-11-09 at 01 34 27

and this is the version I was referring to previously (normal score) NootRX.kext.zip

colorzpe commented 2 weeks ago

Update I'm testing with Cinebench 2024 ,Heaven Benchmark Heaven Bench Score Same Screen Shot 2567-11-09 at 02 08 45 But Cinebench not

VisualEhrmanntraut commented 2 weeks ago

Merging with #69 as I theorise this is the primary cause now.

.