Is RAM usage equivalent between silicon and Intel macbooks?

[pshufd]

I have a 16" M1 Max MacBook Pro with 32GB RAM and a PC with 32GB RAM. I really don't see how Apple believes that 16GB of RAM = 32GB RAM on PC.

In fact, it seems that my PC can push more "RAM" as it also has 16GB of VRAM from the RTX 5080 additionally available.

32 GB on my Windows laptop > 32 GB on my MacBook Pro because one of my Windows programs needs ten times as much ram on macOS.

I also use the Logi Bolt dongle, which, for some reason, uses 810 MB of RAM on macOS but doesn't use any RAM on Windows.

 

[mr_roboto]

I have a 16" M1 Max MacBook Pro with 32GB RAM and a PC with 32GB RAM. I really don't see how Apple believes that 16GB of RAM = 32GB RAM on PC.

In fact, it seems that my PC can push more "RAM" as it also has 16GB of VRAM from the RTX 5080 additionally available.

VRAM doesn't really substitute for main memory. It's there for performance, not capacity expansion - it's a bit like a giant cache for data that started out sitting in main memory.

In principle, once you've gotten a graphics data buffer copied over to the GPU you can throw away the original copy in main memory, freeing that RAM for other things. I'm not sure how extensively video drivers get to take full advantage of that, though.

 

[Zest28]

VRAM doesn't really substitute for main memory. It's there for performance, not capacity expansion - it's a bit like a giant cache for data that started out sitting in main memory.

In principle, once you've gotten a graphics data buffer copied over to the GPU you can throw away the original copy in main memory, freeing that RAM for other things. I'm not sure how extensively video drivers get to take full advantage of that, though.

PC’s can bypass the copying from SSD —> RAM —> VRAM and go straight from SSD —> VRAM too

Games like Diablo 4, Ghost of Tsushima Director's Cut, …… do this.

It’s basically up to the software developers implementation.

Xbox can do it also.

For “professional applications” where the CPU and GPU have to work on the same data, then a copy of the same data needs to be present in both VRAM and RAM, as the CPU cannot access the VRAM on the GPU.

And it’s not a bad thing, because the GPU has a super fast high-bandwidth VRAM to use (faster and higher bandwidth than the RAM of Apple Silicon), which cannot be slowed down by any operation the CPU does.

 

[mr_roboto]

PC’s can bypass the copying from SSD —> RAM —> VRAM and go straight from SSD —> VRAM too

Games like Diablo 4, Ghost of Tsushima Director's Cut, …… do this.

It’s basically up to the software developers implementation.

Xbox can do it also.

For “professional applications” where the CPU and GPU have to work on the same data, then a copy of the same data needs to be present in both VRAM and RAM, as the CPU cannot access the VRAM on the GPU.

And it’s not a bad thing, because the GPU has a super fast high-bandwidth VRAM to use (faster and higher bandwidth than the RAM of Apple Silicon), which cannot be slowed down by any operation the CPU does.

None of that addresses what I was talking about, which is that in discrete GPU systems, GPU VRAM isn't useful in the same way as system RAM. So, it doesn't make a lot of sense to imply that your 32GB RAM + 16GB VRAM computer is exactly equivalent to a 48GB Apple Silicon system. Sometimes, one is better, sometimes the other. If I had to pick one for the best general purpose results, it'd be unified memory.

 

[Zest28]

None of that addresses what I was talking about, which is that in discrete GPU systems, GPU VRAM isn't useful in the same way as system RAM. So, it doesn't make a lot of sense to imply that your 32GB RAM + 16GB VRAM computer is exactly equivalent to a 48GB Apple Silicon system. Sometimes, one is better, sometimes the other. If I had to pick one for the best general purpose results, it'd be unified memory.

You can hit the 32GB + 16GB if developers opt not to keep a copy in RAM. It can be done. It’s a choice of implementation, unless the application requires that both the CPU and GPU work on the same data. There are games that are not using the RAM when loading the data into VRAM as I have mentioned.

And unified memory is not better at all as you will hit the memory bandwidth much faster. It’s why Apple Silicon will never beat NVIDIA in high-performance computing.

Apple Silicon is only useful where latency matters, like scrubbing during video editing. And who cares about video editing, it will soon be done by AI too, powered by NVIDIA.

 

[mr_roboto]

You can hit the 32GB + 16GB if developers opt not to keep a copy in RAM. It can be done. It’s a choice of implementation, unless the application requires that both the CPU and GPU work on the same data. There are games that are not using the RAM when loading the data into VRAM as I have mentioned.

And unified memory is not better at all as you will hit the memory bandwidth much faster. It’s why Apple Silicon will never beat NVIDIA in high-performance computing.

Apple Silicon is only useful where latency matters, like scrubbing during video editing.

"There are games..." pretty much summarizes the entire 'benefit' to separate RAM and VRAM, for a particular subset of the gaming audience. If you're obsessed with frames, and are willing to spend a thousand watts and many thousands of dollars on a very high end DGPU, sure, it's great. If that's not your mania, unified RAM tends to be far more useful.

Even the people building modern gaming consoles agree - Playstations and Xboxes use unified RAM, just like Apple Silicon. Turns out that what makes for a fun gaming experience isn't obsessing over having the absolute most powerful GPU possible to build, regardless of all other considerations.

As for bandwidth, Apple provides plenty. M4 Max is 546 GB/s, which is far more than similar laptop oriented unified memory SoCs from other manufacturers (such as Qualcomm Snapdragon X2 Elite).

 

[Zest28]

"There are games..." pretty much summarizes the entire 'benefit' to separate RAM and VRAM, for a particular subset of the gaming audience. If you're obsessed with frames, and are willing to spend a thousand watts and many thousands of dollars on a very high end DGPU, sure, it's great. If that's not your mania, unified RAM tends to be far more useful.

Even the people building modern gaming consoles agree - Playstations and Xboxes use unified RAM, just like Apple Silicon. Turns out that what makes for a fun gaming experience isn't obsessing over having the absolute most powerful GPU possible to build, regardless of all other considerations.

As for bandwidth, Apple provides plenty. M4 Max is 546 GB/s, which is far more than similar laptop oriented unified memory SoCs from other manufacturers (such as Qualcomm Snapdragon X2 Elite).

Bandwidth is a limitation on Mac. My Mac can choke if you do many things at the same time while running a game in the background. My PC, never does it. And this isn’t even high-performance computing.

The entire market agrees that dedicated GPU’s is the way to go. NVIDIA is the largest company in the world now and companies are begging to buy more of their “inferior dedicated GPU solution”.

And gaming consoles are performing way worse than dedicated GPU’s from NVIDIA, so how does that proof that unified memory is better than dedicated GPU’s from NVIDIA?

 

[Azedquery]

In fact, it seems that my PC can push more "RAM" as it also has 16GB of VRAM from the RTX 5080 additionally available.

Exactly what I found on Mac Book Pro, despite absolute insistence from Apple that my 64GB Intel with 8 dedicated VRAM would be far slower than M3 Max with 48GB. Tested it and it was rubbish, exchanged for 64GB, better, but still not as fast as old Intel on certain graphics tasks. Apple RAM isn't magic, and VRAM counts, regardless of Apple claims.

 

[Azedquery]

This sounds very odd to me. There is no reason why the same file should occupy more RAM on Apple Silicon Mac compared to the Intel Mac, unless the software is different. Do you observe this behavior across different software suites?

Thanks for comment and apologies for very late response on re-visiting this thread! (I gave up on it all once I knew I was stuck with the config I had (still have). In answer to your question:

I've no reason to believe the same file occupies more RAM on the M3, only that the M3 takes far longer to do the same operation on the same file, using the same app, than the Intel did (when it was working). I”m assuming the file needs all 64GB to complete the task, and having no VRAM on the M3 Max is enough to reduce efficiency, but I’m not a technician which is why I was looking for input originally.

No, I don't notice it on any other apps, only Photoshop and Topaz - but then no other apps I use can do this task, and no other apps put as much stress on the hardware, so I have nothing to compare it with.

What I do notice is that Photoshop and Lightroom come to a standstill if Carbon Copy Cloner kicks in to backup changes. I don’t recall this being an issue on the Intel either.

 

[Azedquery]

See my video in this thread ; this is how low memory impact on performance )

Where is your vid?

 

[leman]

Thanks for comment and apologies for very late response on re-visiting this thread! (I gave up on it all once I knew I was stuck with the config I had (still have). In answer to your question:

I must say, I kind of like that this conversation is taking place over multiple years — brings a certain feeling of stability into my life.

I've no reason to believe the same file occupies more RAM on the M3, only that the M3 takes far longer to do the same operation on the same file, using the same app, than the Intel did (when it was working). I”m assuming the file needs all 64GB to complete the task, and having no VRAM on the M3 Max is enough to reduce efficiency, but I’m not a technician which is why I was looking for input originally.

No, I don't notice it on any other apps, only Photoshop and Topaz - but then no other apps I use can do this task, and no other apps put as much stress on the hardware, so I have nothing to compare it with.

What I do notice is that Photoshop and Lightroom come to a standstill if Carbon Copy Cloner kicks in to backup changes. I don’t recall this being an issue on the Intel either.

This sounds very odd to me indeed. The M3 is a much more capable machine every possible aspect, so what you are experiencing is very puzzling. It also does not appear to mirror the common experience. My (uneducated) guesses would be that you are either running into some obscure corner case with either the configuration or the software. Do you maybe use some third-party plugins or an older piece of software on your Mac that might be creating problems?