How much memory do I really need ? 24, 36, 48 vs numbers of gpus

[colonelbutt]

I use lightroom extensively. And will use photoshop more. Using denoise and AI quite a lot. Otherwise I just use normally and play the occasional game. I would play more first person shooters if they actually made them for the mac!

I am deciding between 24, 36 and 48gb.
I think there is a relationship to the number of gpus which is not immediately obvious. Basically gpus use shared memory. That means the more you have, the more memory you need just to run normally.

Therefore, if an application is more sensitive to memory usage, one is better off having 48gb with 20 gpus rather then 32 gpus say, as there is more memory bandwidth per gpu and overall.

Of course there is a max amount of memory that the gpu would comfortably use with lightroom and photoshop. So over a certain amount of memory, there may be a limited (or diminishing) effect. Although I am just speculating here as I am not an expert, and you guys know a lot more then me about these things.

I would be interested in comments on the above, and also any ideas on real life performance of lightroom and photoshop.

For example, let’s say the options are as follows:
1. 24gb with 16 gpus
2. 24gb with 20 gpus
3. 48gb with 20 gpus
4. 36gb with 32 gpus
5. 48gb with 32 gpus

It is always tempting to just say the best, but if the performance of 5 vs 2, is just 20% more, even 50% more if an operation is 30 down to 20 seconds, is it worth it ? As I am looking for best value as well.

What I don’t want is freezes or going from seconds to minutes. But a number of more seconds is not a big issue.

Given my ramble - what do you guys recommend ? Any comments VERY welcome

 

[bzgnyc2]

I use lightroom extensively. And will use photoshop more. Using denoise and AI quite a lot. Otherwise I just use normally and play the occasional game. I would play more first person shooters if they actually made them for the mac!

I am deciding between 24, 36 and 48gb.

How big are yours images? What system do you have now and how much RAM does it have? Do you see memory pressure (see Activity Monitor) during your typical workflows?

I think there is a relationship to the number of gpus which is not immediately obvious. Basically gpus use shared memory. That means the more you have, the more memory you need just to run normally.

Therefore, if an application is more sensitive to memory usage, one is better off having 48gb with 20 gpus rather then 32 gpus say, as there is more memory bandwidth per gpu and overall.

I think you are conflating memory size with memory bandwidth. Memory bandwidth on the Apple Silicon architecture is mainly a function the tier and model of chip. All else being equal configuring a Mac with more RAM doesn't increase memory bandwidth. For example an M4 with 32GB has less memory bandwidth than an M4 Pro with 24GB. As far as memory bandwidth: M4 Max 16/40 > M4 Max 14/32 > M4 Pro > M4 regardless of RAM size selected (though Apple generally requires selecting the bigger chips in order to select larger RAM sizes).

Then the importance of memory bandwidth varies by the application. Many algorithms -- especially these days those for AI -- benefit from more memory bandwidth. On the other hand, many algorithms are insensitive to memory bandwidth. They won't go slower all else being equal but some smart programmer found a way to optimize them to minimize that dependency (or perhaps another way to look at them is that their bottlenecks lie elsewhere).

On the other hand, more GPUs don't generally require larger RAM configurations all else being equal. RAM needs are generally a function of the size of your data and your applications / related algorithms (which for commercial applications you won't know directly but have to infer from how they perform on various systems with different sized data). More GPUs can take advantage of more memory bandwidth (or to look at it another way, are less likely to be bottlenecked by more memory bandwidth) but as above Apple generally scales the memory bandwidth with the number of GPU via scaling the GPU with the tier of chip (which as above drives how much memory bandwidth is available).

Of course there is a max amount of memory that the gpu would comfortably use with lightroom and photoshop. So over a certain amount of memory, there may be a limited (or diminishing) effect. Although I am just speculating here as I am not an expert, and you guys know a lot more then me about these things.

I would be interested in comments on the above, and also any ideas on real life performance of lightroom and photoshop.

For example, let’s say the options are as follows:
1. 24gb with 16 gpus
2. 24gb with 20 gpus
3. 48gb with 20 gpus
4. 36gb with 32 gpus
5. 48gb with 32 gpus

It is always tempting to just say the best, but if the performance of 5 vs 2, is just 20% more, even 50% more if an operation is 30 down to 20 seconds, is it worth it ? As I am looking for best value as well.

What I don’t want is freezes or going from seconds to minutes. But a number of more seconds is not a big issue.

Given my ramble - what do you guys recommend ? Any comments VERY welcome

Your point about diminishing returns is important and a lot of people overlook. For a lot of work, the difference with larger chips, more GPU, etc has a diminishing returns. Maybe a bigger chip peaks at 2x faster but it's only 50% faster in practice and then that person's work don't go 50% faster just because the computer does.

It also varies by the nature of the work -- for some it might be 50ms/frame versus 33ms/frame which means 20 fps versus 30 fps and the former means they can't work real-time. For others 3 seconds versus 2 seconds or 30 seconds versus 20 seconds doesn't matter as you note.

In any case, I look at # of GPU and size of RAM as independent dimensions of system performance/capacity. For applications that can use GPU at all, having more GPU is generally better though as above there are diminishing returns.

Independently, you need enough RAM for your applications to work with the size data you plan to work with. Beyond that there is little to no return and it mainly goes to waste. If you can load all your applications with the files/datasets you normally work with in 16GB of RAM without ongoing swapping/lags, 32GB is unlikely to be noticeable better. On the other hand, if your application needs 24GB to do its business, 16GB will be painful if it works at all.

If all this sounds complicated, it is because modern computer systems are very complex systems. There are multiple dimensions of performance with tradeoffs and sometimes interactions among them. For better or worse however Apple simplifies this by offering just a few models with a few tiers and dimensions of performance that their engineers believe represent good combinations of capabilities for their target market segments.

What I recommend doing is determining how much RAM you foresee needing for the expected life of the system based on the applications you plan to use and the images/data you expect to work with. As above excess RAM isn't helpful but the penalty for slightly too little is worse than slightly too much so better to round up. Then get the system that can support at least that much RAM with the number of CPU/GPU/etc where the next tier up doesn't feel any faster.

 

[colonelbutt]

Thanks. Very interesting. I currently use a 45mp sensor, but can’t rule out 60mp next year.

I currently use a PC with 32gb and an i5-13xxxx and it does everything I need but it’s a bit sluggish. For example denoise and AI remote is arojnd 15-30s and export is a bit more. I also have a Microsoft Surface Laptop with 16gb and 12 arm cpus and 8 gpus. It’s actually a bit faster then the PC but it’s not my editing platform, I use it for other things, and basically wanted a portable editor which is noticeably faster then my PC.

 

[Fishrrman]

OP wrote (in the very first line):
"I use lightroom extensively. And will use photoshop more"

Then... you'll need plenty of RAM.
I suggest at least 32/36gb.
48gb might be better, if you're willing to pay.

I don't use either of those apps, but I'm going to take a GUESS and reckon that all those GPU's won't matter too much. The RAM matters more.

 

[bzgnyc2]

Thanks. Very interesting. I currently use a 45mp sensor, but can’t rule out 60mp next year.

I currently use a PC with 32gb and an i5-13xxxx and it does everything I need but it’s a bit sluggish. For example denoise and AI remote is arojnd 15-30s and export is a bit more.

Not quite following -- your original post suggested that going from 30s to 20s may not be worth it but here you're suggesting your relatively recent Intel processor is sluggish at around 15-30s.

Is the UI and basic operations fast enough there and just these batch type (denoise, etc) operations too slow? Or does the whole thing feel sluggish?

By the way, you can see how Lightroom is performing across a variety of hardware here:

My Recent GPU Experience (Gigabyte Nvidia RTX 4070 Ti 12GB)

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ "This is a relatively long discussion. It is worth reading the discussion in sequence, as posted." To assist people returning to this discussion, the following are direct links to frequently used posts...

www.lightroomqueen.com

Note it does appear that the GPU is important for Lightroom's denoise so if your PC doesn't have a high-performance Nvidia or AMD GPU (e.g. you are just using Intel's integrated GPU), it may be struggling here due to that not the core CPU.

I also have a Microsoft Surface Laptop with 16gb and 12 arm cpus and 8 gpus. It’s actually a bit faster then the PC but it’s not my editing platform, I use it for other things, and basically wanted a portable editor which is noticeably faster then my PC.

Based on what I see in the above forum, a MacBook Pro M4 Max would likely be substantially faster than your current PC. You could try running the benchmark in that thread on your current PC to get a sense of how much faster different Mac models would be.