Anyone run gfxbench yet with metal 3?

[dugbug]

Just wondering. May need to wait for a venture native build of their app

 

[Xiao_Xi]

Do the Metal 3 enhancements apply to current hardware or only to future hardware?

 

[Andropov]

Just wondering. May need to wait for a venture native build of their app

Doubt there's going to be any difference.

Do the Metal 3 enhancements apply to current hardware or only to future hardware?

A13 Bionic or newer for iOS, some of the latest Intel/AMD graphics cards and any Apple Silicon Mac for macOS.

 

[Frantisekj]

I found this article today saying:

MetalFX will improve the graphics and apply timed antialiasing. The idea is to improve gaming performance with faster framerates than with pure hardware-based rendering.Resident Evil Village will use this technology when it comes to running on the Mac. Capcom says the game will run at 1080p easily on the MacBook Air and at 4K on the Mac Studio. When it arrives on Mac and iPad, Reeds No Mans Sky uses MetalFX upscaling. To complete the task we find Grid Legends which have to land on Mac.

No Mans Sky, RE Village and Grid Legends arrive on Mac with MetalFX - Game News 24

Apple is serious in its attempts to open the whole world of gaming world of hardware, and so, bynnouncing that No Mans Sky, Resident Evil Village and Grid

game-news24.com

 

[leman]

Metal 3 doesn’t bring any performance improvements (except for ray tracing which is not used for Mac games at all). I don’t understand where this idea even came from.

Now, Metal 3 does include some new APIs that can improve performance if you use them a d your game benefits from it. Like the resource loading APIs or obviously Apples new temporal AA/upscaler.

 

[Frantisekj]

Do the Metal 3 enhancements apply to current hardware or only to future hardware?

I have found this article where they claim that it does:

https://venezuela.detailzero.com/technology/44542/Apple-greatly-improves-the-performance-of-its-GPUs-using-Metal-with-the-latest-beta-of-macOS-Ventura.html

 

[dugbug]

Thanks Frantisekj

From the article above:

...with the Mac Studio with M1 Max 32 cores I was playing Total War: Warhammer III at QHD resolution and custom low-medium graphics settings that allowed me to play at 50 frames. After the update, it now moves the game over 90 f/s, which allows me to switch to the quality setting medium preset and play at 60 f/s.

 

[galad]

Apple might be as well improving the drivers and the performance of Metal, even if not using the new API. But that will need some better benchmark with a variety of games to get some actual useful data.

 

[leman]

I have found this article where they claim that it does:

https://venezuela.detailzero.com/technology/44542/Apple-greatly-improves-the-performance-of-its-GPUs-using-Metal-with-the-latest-beta-of-macOS-Ventura.html

The games mentioned in that post don’t use Metal 3 so it’s unclear how they could benefit from the new APIs. It is entirely possible however that Apple has optimized their drivers or maybe even I tricked game-specific optimizations to work around performance pitfalls if common games (like all GPU vendors do).

 

[diamond.g]

The games mentioned in that post don’t use Metal 3 so it’s unclear how they could benefit from the new APIs. It is entirely possible however that Apple has optimized their drivers or maybe even I tricked game-specific optimizations to work around performance pitfalls if common games (like all GPU vendors do).

Has Apple ever done game specific optimizations in an OS update? I don't recall that being a thing for macOS.

 

[thenewperson]

Has Apple ever done game specific optimizations in an OS update? I don't recall that being a thing for macOS.

IIRC they've made app-specific ones before, not sure if any games were part of that though.

 

[jdb8167]

Has Apple ever done game specific optimizations in an OS update? I don't recall that being a thing for macOS.

Apple doesn't seem to have done any optimizations for Shadow of the Tomb Raider considering that the 10 GPU core M2 MacBook Air runs at the same FPS as the 8 GPU core version. There aren't too many Metal optimized games anyway. SotTR doesn't seem to be particularly optimized for Metal.

I plan to run some benchmarks for the 10 core GPU when I get a chance. What I'm seeing with Unigine Valley is that the M2 10 GPU core MacBook Air starts to throttle right at the end of the benchmark at default full screen resolution. If you run the benchmark again, it throttles very heavily. It goes from peak package power of ~22 W down to about 8.5 W. The GPU cores throttle even more heavily. From a peak power of about 10.5 W down to the low 3 W range. Valley is a circa 2013 OpenGL benchmark so that might influence the level of throttling but I don't think it does.

 

[diamond.g]

Apple doesn't seem to have done any optimizations for Shadow of the Tomb Raider considering that the 10 GPU core M2 MacBook Air runs at the same FPS as the 8 GPU core version. There aren't too many Metal optimized games anyway. SotTR doesn't seem to be particularly optimized for Metal.

I plan to run some benchmarks for the 10 core GPU when I get a chance. What I'm seeing with Unigine Valley is that the M2 10 GPU core MacBook Air starts to throttle right at the end of the benchmark at default full screen resolution. If you run the benchmark again, it throttles very heavily. It goes from peak package power of ~22 W down to about 8.5 W. The GPU cores throttle even more heavily. From a peak power of about 10.5 W down to the low 3 W range. Valley is a circa 2013 OpenGL benchmark so that might influence the level of throttling but I don't think it does.

Have you thought of using GravityMark? It is a Metal bench that seems to push the hardware fairly hard.

 

[jdb8167]

Have you thought of using GravityMark? It is a Metal bench that seems to push the hardware fairly hard.

I’ll check it out.

 

[jdb8167]

Have you thought of using GravityMark? It is a Metal bench that seems to push the hardware fairly hard.

Doesn't seem to launch. It's an iPad app anyway. Not exactly how I want to test macOS and the M2 GPUs.

Edit: pulled a beta version from their website and it works but doesn't really stress the M2. It never went above about 65 °C and didn't throttle.

 

[Frantisekj]

Apple doesn't seem to have done any optimizations for Shadow of the Tomb Raider considering that the 10 GPU core M2 MacBook Air runs at the same FPS as the 8 GPU core version. There aren't too many Metal optimized games anyway. SotTR doesn't seem to be particularly optimized for Metal.

10 core throttle more so Tomb Raider score is moreover identical to 8 core on Air. Not sure some tested MBP M2.

 

[jdb8167]

10 core throttle more so Tomb Raider score is moreover identical to 8 core on Air. Not sure some tested MBP M2.

I think I saw somewhere that that the M2 13" MacBook Pro doesn't throttle.

 

[diamond.g]

Doesn't seem to launch. It's an iPad app anyway. Not exactly how I want to test macOS and the M2 GPUs.

Edit: pulled a beta version from their website and it works but doesn't really stress the M2. It never went above about 65 °C and didn't throttle.

I'll have to look at my setup when I get home. I could have sworn it pulled more power, especially with 2+ million asteroids.

 

[jdb8167]

I'll have to look at my setup when I get home. I could have sworn it pulled more power, especially with 2+ million asteroids.

I didn't change the default number of asteroids. Maybe that is it.

Edit: Slowed down the FPS to about 5-6 but didn't make much difference otherwise. No throttling and the temperatures didn't rise.

 

[leman]

I didn't change the default number of asteroids. Maybe that is it.

Edit: Slowed down the FPS to about 5-6 but didn't make much difference otherwise. No throttling and the temperatures didn't rise.

You need to check the power. Use powermetrics or one of the utilities designed around it (asitop/mx power Gadget)

 

[jdb8167]

You need to check the power. Use powermetrics or one of the utilities designed around it (asitop/mx power Gadget)

I was monitoring it. The power was about the same. Peak of about 6.5 W for the GPU and total package power of about 11 W. GPU was at 100% @ 1398 MHz. Almost no CPU with E cores at 40% at about 1000 MHz. Temperatures at about 60-70 °C (take with a grain of salt since the temperature monitor is written by me and not necessarily fully debugged.) No throttling.

 

[diamond.g]

I was monitoring it. The power was about the same. Peak of about 6.5 W for the GPU and total package power of about 11 W. GPU was at 100% @ 1398 MHz. Almost no CPU with E cores at 40% at about 1000 MHz. Temperatures at about 60-70 °C (take with a grain of salt since the temperature monitor is written by me and not necessarily fully debugged.) No throttling.

Change from Default hardware rasterization to CS (computer shaders) jumped from ~15w to ~20w on my Mac.

To get even more of a load I dropped the asteroid count to 1 million but upped the load bias to +6.

I tried 2 million with +6 load bias but it hardlocked the Mac, I had to restart the Mac. RT mode works, but it doesn't appear to stress the hardware much. MS (mesh shaders) doesn't work at least on this version of the OS, once Ventura is out I assume it should work there.

 

[jdb8167]

Change from Default hardware rasterization to CS (computer shaders) jumped from ~15w to ~20w on my Mac. View attachment 2041930 To get even more of a load I dropped the asteroid count to 1 million but upped the load bias to +6.
View attachment 2041931
I tried 2 million with +6 load bias but it hardlocked the Mac, I had to restart the Mac. RT mode works, but it doesn't appear to stress the hardware much. MS (mesh shaders) doesn't work at least on this version of the OS, once Ventura is out I assume it should work there.

My M2 MBA didn't lock up but my it got slow. Several seconds between clicks and it started eating up memory. I didn't try beyond the compute shaders (CS) and the +6 load bias with 500,000 asteroids. I was seeing consistently 14.5 W from the GPU. It still didn't throttle though—at least according to powermetrics.

I used Render: CS, FullScreen, 500,000 asteroids, LOD Bias +3 (x8) and I got the GPU up to 12 W and finally the M2 MacBook Air did throttle at the end of the benchmark.

 

[diamond.g]

My M2 MBA didn't lock up but my it got slow. Several seconds between clicks and it started eating up memory. I didn't try beyond the compute shaders (CS) and the +6 load bias with 500,000 asteroids. I was seeing consistently 14.5 W from the GPU. It still didn't throttle though—at least according to powermetrics.

I used Render: CS, FullScreen, 500,000 asteroids, LOD Bias +3 (x8) and I got the GPU up to 12 W and finally the M2 MacBook Air did throttle at the end of the benchmark.

That is interesting, my Mac dropped the GPU frequency a little bit, then returned to "normal" Eventually the fans ramped to 100% to keep the GPU temp around 90ºC.

 

[Frantisekj]

Some new comments about Metal 3

Metal 3 and WoW

Thanks for the replies! Hmm. I understand what you guys are saying. Hopefully it provides a performance boost and Blizzard gets around to implementing it if it’s any better? I found a random article looking around, claiming some sort of uplift from just using MacOS Ventura beta (which includes...

us.forums.blizzard.com

Mac App Store submissions now open for macOS Ventura - Latest News - Apple Developer

macOS Ventura makes the things you do most on Mac even better, with powerful new ways to get more done, share and collaborate in your favorite apps, immerse yourself in next‑level games, and take full advantage of your other devices. Machine learning enhancements make it even easier to provide...

developer.apple.com