Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Yes. Since FSR is open source, MetalFX is using a combination of FSR 2 and Apple's tech.

Since CP is coming with Path-Tracing on MacOS, i would assume Metal is upgrading to FS3 3 w/FG internally.
Yeah FG is going to be a problem. It is hot garbage (artifacts and increased input latency) when your framerate is below 60. I feel like a lot of folks are going to turn it on when their base framerate isn't high enough and get a worse experience.

AMD's version is more sensitive to base framerates than nvidia's (though technically the new DLSS4 version isn't hardware based and probably is also affected by base framerate).
 
  • Like
Reactions: Huntn
Yeah FG is going to be a problem. It is hot garbage (artifacts and increased input latency) when your framerate is below 60. I feel like a lot of folks are going to turn it on when their base framerate isn't high enough and get a worse experience.

AMD's version is more sensitive to base framerates than nvidia's (though technically the new DLSS4 version isn't hardware based and probably is also affected by base framerate).
Blame Nvidia

They will sell you a problem: Low fps with Ray Tracing
Solution: FG and DLSS with DLSS

New problem: Input lag and latency

New solution: More powerful GPU

🤡
 
Blame Nvidia

They will sell you a problem: Low fps with Ray Tracing
Solution: FG and DLSS with DLSS

New problem: Input lag and latency

New solution: More powerful GPU

🤡
Kinda. I think there is a reason they are pushing DLSS Performance with the transformer model (especially at 4k). It allows you to get your base framerate up enough that (M)FG can do a better job smoothing motion and not producing too many errors. Game should still feel like you are playing at 60+ fps. And it should still look good (CNN Performance is a noticeable visual downgrade).

Now I'll be honest, I have no clue how this if going to work out with the 5060/5060ti those GPUs have targeted 1080P/1440P (respectively) and DLSS Performance really lowers the render resolution.
 
  • Like
Reactions: Irishman
Kinda. I think there is a reason they are pushing DLSS Performance with the transformer model (especially at 4k). It allows you to get your base framerate up enough that (M)FG can do a better job smoothing motion and not producing too many errors. Game should still feel like you are playing at 60+ fps. And it should still look good (CNN Performance is a noticeable visual downgrade).

Now I'll be honest, I have no clue how this if going to work out with the 5060/5060ti those GPUs have targeted 1080P/1440P (respectively) and DLSS Performance really lowers the render resolution.
Yupp Ray tracing is an issue for lower end GPUs. For Macs, Apple silicon will need strong FSR4 support. Or else Path, let along ray tracing won't be viable on resultions above 1440p unless you dial quality settings down to low-medium. Even a 5080 can struggle to achieve meaningful FPS with ray tracing if DLSS is not switch on.
 
Based on the AC:S showing (with RT being required and them using it for uh something even when "off") I don't have high hopes for Apple Silicon with RT Overdrive.
 
  • Like
Reactions: Irishman
Based on the AC:S showing (with RT being required and them using it for uh something even when "off") I don't have high hopes for Apple Silicon with RT Overdrive.
No, we shouldn’t expect too much from current-gen hardware.

One thing - other members speculate the first M5 Mac will appear before the next version of MacOS in September.

I wouldn’t be surprised if Mac-flavored CP2077 doesn’t even ship till then.
 
  • Like
Reactions: Irishman
Someone in the Philippines says it showed up on the app store. I'm not clear if the game comes out on the 25th or 26th.
 
  • Like
Reactions: Irishman
Blame Nvidia

They will sell you a problem: Low fps with Ray Tracing
Solution: FG and DLSS with DLSS

New problem: Input lag and latency

New solution: More powerful GPU

🤡

You're close there. Nvidia's new solution is actually "Generate 4 fake frames at a time instead of one and tell everyone that's the native framerate, then double the MSRP of the card while only providing around a 10% performance increase compared to the prior generation card. Then deliberately restrict supply so that scalpers scoop up inventory and resell at 2-5x MSRP if you're lucky."

In reality, Nvidia doesn't give a damn about the consumer market anymore, because they are going all-in on datacenter and AI/ML workloads now.
 
  • Like
Reactions: Irishman
You're close there. Nvidia's new solution is actually "Generate 4 fake frames at a time instead of one and tell everyone that's the native framerate, then double the MSRP of the card while only providing around a 10% performance increase compared to the prior generation card. Then deliberately restrict supply so that scalpers scoop up inventory and resell at 2-5x MSRP if you're lucky."

In reality, Nvidia doesn't give a damn about the consumer market anymore, because they are going all-in on datacenter and AI/ML workloads now.
Hey, come on! It's only 3 fake frames! 4 is for 60 series...

I think they kinda designed themselves into a corner. They've been increasing performance by increasing die size and power and process nodes haven't kept up. I don't think there's much room to make better chips for a reasonable price and power usage anymore.
 
Hey, come on! It's only 3 fake frames! 4 is for 60 series...

I think they kinda designed themselves into a corner. They've been increasing performance by increasing die size and power and process nodes haven't kept up. I don't think there's much room to make better chips for a reasonable price and power usage anymore.

I didn't even get into the whole melting connectors issue with the 50 series - Nvidia learned nothing from the 40 series issues, and now you have 50 series cards with uneven power distribution pushing too many amps down 1-2 wires instead of evenly distributing amperage as intended. This is largely because the 12v connector itself was still designed by a moron, despite being a second (third?) generation version of the connector.
 
  • Like
Reactions: Irishman
You're close there. Nvidia's new solution is actually "Generate 4 fake frames at a time instead of one and tell everyone that's the native framerate, then double the MSRP of the card while only providing around a 10% performance increase compared to the prior generation card. Then deliberately restrict supply so that scalpers scoop up inventory and resell at 2-5x MSRP if you're lucky."

In reality, Nvidia doesn't give a damn about the consumer market anymore, because they are going all-in on datacenter and AI/ML workloads now.
There's no going back.The majority of performance gains going forward will be AI generated. This is cheaper for the manufacturers than providing bigger and more sophisticated silicon dies.

A great analogy is turbo charging. It became cheaper for car manufactuers to turbo charge 4 cylinder engines than naturally aspired 6-8 cylinder engines.
 
  • Like
Reactions: Irishman and Huntn
MetalFX predates FSR 2.
It's a misconception that MetalFX is based on FSR.
So why does Apple reference FSR in their legal notices then? And FSR 1 predates MetalFX and is also open source. Later down the line they could have moved to FSR2.

 
So why does Apple reference FSR in their legal notices then? And FSR 1 predates MetalFX and is also open source. Later down the line they could have moved to FSR2.

Because it's included in the crossover 2023 licence which is the foundation of GPTK.

Notebookcheck simply took a random persons reddit post and made a story out of it without doing any actual research. Other people then picked up on the notebookcheck story and ran with it.

You'll note the date of the FSR reference is 2021 which again predates FSR2. MetalFX launched with a temporal upscaler.
 
Because it's included in the crossover 2023 licence which is the foundation of GPTK.

Notebookcheck simply took a random persons reddit post and made a story out of it without doing any actual research. Other people then picked up on the notebookcheck story and ran with it.

You'll note the date of the FSR reference is 2021 which again predates FSR2. MetalFX launched with a temporal upscaler.
Why does MetalFX have a spatial upscale at all then?

Do we know if Apple is using a CNN or a Visual Transformer?
 
Why does MetalFX have a spatial upscale at all then?

Do we know if Apple is using a CNN or a Visual Transformer?
Because spatial upscaling is extremely cheap.

MetalFX spatial is pretty terrible though. The temporal is good.
No idea on the temporal algorithm, other than it does make use of the neural engine on Apple Silicon.
 
Because spatial upscaling is extremely cheap.

MetalFX spatial is pretty terrible though. The temporal is good.
No idea on the temporal algorithm, other than it does make use of the neural engine on Apple Silicon.
I guess the only way to tell is if there is a visual (image quality) difference between Performance and Quality modes. Visual Transformer (at least for DLSS) seems to retain IQ while improving performance (in "lower" modes).
 
I guess the only way to tell is if there is a visual (image quality) difference between Performance and Quality modes. Visual Transformer (at least for DLSS) seems to retain IQ while improving performance (in "lower" modes).
I don't know how it works now, but in the early days of MetalFX, Quality used termporal and Performance used spacial. Digital Foundry had a deep dive into this on RE:Village a few years ago.


I notce that more recent games seem to let you set the output and render resolutions separately (e.g. Control), which is probably more similar to the DLSS/FSR settings where it is the scaling factor that changes from quality down to ultra performance, rather than changing the spacial/temporal models.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.