Use of RAM, maximising....

This is a Windows relevant question:

Is there a way of utilising more of the system Ram for Resolve, taking it off what appears to be reserved for Fusion?
I never use Fusion and notice that I start with -
System Memory of 32GB.
Max out Resolve to 24 GB.
Limit Fusion to "6GB".

I watch the Task Manager Memory usage when I'm working on a project and it never seems to go above 16GB/50%. I know that Resolve wont use more than 75% of Ram, but it never seems to get up to that point.
I have 1200-1300 clips in the media library, and have maybe 200+ clips on the Timeline. When in Edit mode I presume that Resolve is using/making Timeline clip icons/proxies and putting them in Ram.
Each time I zoom in and out of the Timelime I notice my CPU's max out making the icons, my Ram gets up to about 16GB use, but never seems to go beyond. Its like I have a wasted 15GB of Ram.

I'd love to be able to turn off Fusion and free up the extra Ram to stop the Timeline icons in Edit being recomputed(?).
I know "Bypassing Color Grades and Fusion Effects" means they are not computed, but it doesn't appear to free up the Ram. It's like I have 6GB Ram tied up going to waste.

I'm interested in the answer to this also. Replies to my recent posts asking about Video Cards and GPU RAM emphasized that GPU RAM was the most important thing, along with CPU.

I have a similar question to yours now because I just got a used desktop with 64GB RAM and am actually curious how much benefit there would be in adding more RAM when the computer will primarily only be running Resolve, and not opening lots of other programs, browsing internet, etc.

My sense at the moment is that, even though more RAM is always a nice thing in general, Resolve is mainly making use of GPU and CPU, so as long as you have adequate RAM (which seems to be 64GB for most uses), there will be very limited benefit in adding more. My understanding is it would be more likely to be helpful for multitasking when you have tons of other programs open, but in terms of rendering and playback and video processing within Resolve, it's mostly putting that burden on your video card. So maybe this is why you are only seeing 50% usage of even your 32GB of RAM. Others can confirm if my understanding is correct.

I don't wish to divert away from your original question, but I'm curious if there are any users who are working in certain situations (perhaps 8K or 16K resolutions) where they found 64GB wasn't enough and absolutely needed to upgrade to 128GB+ of RAM.

I don't believe the Fusion part is separated as you think. I believe that setting simply limits the amount of "Resolve memory" that Fusion can use, when it's used.

Muchly appreciated Stephan & Jim.
Seems to be a grey area that not many people know about, and I can't find an in-depth explanation in the manual.

It looks like the cpu appears to keep on having to recompute(via CPU?) Timeline icons, when Editing, when I go to different Timeline zoom sizes which intrigues me...... this may be a wrong conclusion on my part though.

During editing, the responsiveness of the GUI is absolutely critical to the creative process, hence all the wonderful(and complex) techniques DR uses to achieve this.

.........................................

EDIT:
I think I've found why my system was slowing down - my current project appeared to lose connection with the Optimised media, and it needs to generate it's icons (after Timeline zoom changes) from the UHD H.264 & H.265 source footage (My CPU's aren't that fast).
I now create 2 smart bins - one showing optimised media, the other un-optimised media and check at startup.
- and regularly check these bins during the day.
It took me a couple of goes re-connecting optimised media, plus deleting Render Cache to get smooth playback.

More RAM could also be used as part of GPU Shared Memory. How much that comes into play is entirely dependent on what you are doing, and BMD experts are the ones who know best how the resources are allocated. Resolve was digging deep into my RAM when I was severely taxing the system with a 16k timeline, and it wouldn't even finish rendering. I would be interested to know if more RAM could allow my system to overcome that hurdle and Resolve would utilize 128GB+ of RAM as GPU Shared Memory for really tough renders at extremely high timeline resolutions.

I don't believe there is shared memory in the way you mean, in Resolve. I think a moderately good 6/8 core i7 with 32gb ram and an 8/16gb gpu will outperform a 128gb ram i9 or dual xeon with only 4gb gpu, everytime in Resolve. Fusion is less optimised for gpu than color or the rest, at the moment. Maybe this is different in Linux and Mac, though probably not. Linux seems to be very popular in high end grading suites, since it can use multiple gpus.

It is good that Fusion is still standalone - I wish Fairlight was too, since it's audio sub-system is not currently consistent with the rest of resolve.

I really think this is the bottleneck for the millions who have downloaded the free Resolve. Ram and cpu is still a lot cheaper than gpus and with a lot of people who have laptops, far easier to upgrade. This may change as gpus are becoming increasingly important in image content creation.

I don't believe there is shared memory in the way you mean, in Resolve.

False. Every time I post about this issue, I state that GPU is by far the most important thing, so no one should be confused about that. But the area of curiosity here is related to GPU Shared Memory that is definitely being used, and sometimes significantly so, in Resolve. I'm observing this personally. I will describe how you can observe it too. My question is, when one already has a very powerful GPU, for example 16GB, and one also has adequate RAM, for example 64GB, is there benefit to more RAM? Or is it a waste of money? I don't have standalone Fusion, so all references to Fusion are the one integrated into Resolve.

Windows makes available up to half of its RAM to "GPU Shared Memory", so this means 32GB of the 64. When I maxed things out by rendering out of a 16k Resolve timeline into 4k output (with no Fusion elements, compositing all video elements using nested timelines in the Resolve Edit screen), all 16GB of GPU memory were being used, as well as over 18GB of the GPU Shared Memory! I'm not just talking about briefly, I'm saying in a sustained manner during the editing process and rendering as well.

If you have Windows, Open Task Manager and Click on Performance and GPU to see all this. It adds up your GPU Memory and GPU Shared Memory into a total, so in my case it was 48GB available to be used. This is not me adding it up in my head, it actually displays 48GB right there. It uses the dedicated GPU memory first of course, and then seems to draw on GPU Shared Memory as a reserve. My total during that operation was consistently reading that total GPU use (combining dedicated with shared) was above 30GB out of 48GB available.

No one is arguing in favor of a weak video card. Your hypothetical comparison positing an inadequate 4GB GPU isn't fair because the question I posed earlier in this thread isn't really about using RAM to compensate for inadequate GPU. It's more about, once you have enough GPU, CPU, RAM, will additional RAM be utilized in these processes? A more helpful comparison would be to make all things equal, for example, if someone has the same computer, same video card, but simply more RAM.

So, Xeon dual CPUs, 16GB Radeon GPU: 64GB RAM vs 256GB RAM. Will the additional RAM help in Davinci Resolve paid studio version during intense high resolution, multiple VFX displays and renderings? Or will it not help at all? I've personally been observing the use of GPU Shared Memory in different scenarios, so I'd like to hear from someone who has experience observing these things about which situations make use of it, and which don't.

In my observations, GPU Shared Memory was heavily used rendering out of Resolve, without Fusion elements, coming from a 16k timeline. However, when I recreated the exact same video elements using the Fusion screen, with a 4k timeline, but using a 16k canvas within Fusion, it barely used 3GB of GPU Shared Memory. It also didn't even max out the dedicated GPU, and was hovering just above 10GB used during the render. However, it activated all those display graphs (in the Performance monitor) in a more complex way.

In Resolve settings, you can specify how much RAM Resolve will use and Fusion (within Resolve) will use. When I did Fusion-heavy renders, RAM usage (not GPU Shared or Dedicated) was much higher, up to the maximum specified in my settings for Fusion. In addition, I noticed that the GPU itself was at a higher level of what Windows labels "Utilization".

By that I mean, in the earlier example where all 16GB dedicated GPU memory was used and it also 18GB of GPU Shared memory, it was still mysteriously reading that GPU "utilization" was at 10-20%. This was puzzling to say that utilization was super low, when clearly the available GB were totally maxed out. So I noticed that the graphs in the Performance tab weren't showing much activity under Video Encoding, Video Decoding, 3D, Copy, etc.

So after that, I rendered a Fusion-heavy project, and even though the total GB being used was less, all those graphs showed a lot of complex activity in those categories, and the "utilization" percentage went up to 60%. I'm interpreting this as being a more efficient and proper use of Resolve in terms of structuring one's project. The previous example perhaps was maxing out the GBs by overwhelming all resources with so many elements to process it was just trying to bear the load, but not necessarily efficiently.

So it's clear that composing the exact same video project only in the Edit screen of Resolve will be processed very differently than composing it in Fusion (within Resolve). They use memory very differently. This is why feedback from others who are able to monitor all these stats while they work would be valuable here. And this is why it's still an open question as to whether adding RAM above 64GB will help. I'm not asking this as an academic issue. I'm trying to decide a course of action to achieve a specific project, both with my hardware and the way I choose to structure my project within Resolve, since one way could lead to a render which takes 5 days to complete, but doing it another way could get it done in a few hours. One way could lead to a likely crash, and another way seems to be utilizing different aspects of the hardware in a more complete way that won't crash.

False. Every time I post about this issue, I state that GPU is by far the most important thing, so no one should be confused about that.

I'm no computer expert but 'false' is an extreme reaction. When you're rendering out, it is my understanding, exporting that is handled by cpu/ram, so it understandable, you will see the cpu and ram being used, especially so with 16k to 4k. I asked specifically one of the BMD people here, Dwight I think, 'Is it gpu or nothing for pixel image manipulation in Resolve' he confirmed emphatically it was. Maybe my example was a little excessive since there are many many benefits to good cpu and high ram other than just image pixel computing, such as codec encode decode. But nonetheless a large fast 8gb + gpu is essential to get optimum performance from Resolve with 4K+, inspite of the power of the machine in other ares.

.

Calm down it's only telly, internet or whatever. No need for the essay. Yes I'm saying my statement was not false. I accept the extreme example was misleading but. But no shared cpu/gpu between for pixel image processing is what BMD specifically stated, therefore to the best of my knowledge, from what I've been told 'false' is false.

No hard feelings Steve. If I could deliver my essay in my voice, it would be funny rather than mean-sounding. Anyway, perhaps that's true of pixel processing, but there may be other functions necessary to get to the processing. When I saw RAM being used as GPU Shared Memory the most during rendering and my workflow, it was when there were a lot of elements on a huge 16k timeline, so perhaps it was being used to process or hold certain things in order to funnel them into the GPU for pixel processing. We can revisit this topic later after I have more time to see how the 64GB of RAM does. And if I upgrade and get more, I'll be sure to post back and let everyone know if it helped. Take care.

From a Fusion standpoint only, the stand-alone version of Fusion can eat ram RAM like "The Ravenous Bugblatter Beast of Traal."

192GB of RAM can be used without much effort. Running 16K timeline in Resolve seems like it could only benefit from more RAM if you use ANY Fusion elements. Resolve Fusion with 32GB of RAM is like using stand-alone Fusion with 16GB in my experience.

-David

Maybe some useful info here:

From a Fusion standpoint only, the stand-alone version of Fusion can eat ram RAM like "The Ravenous Bugblatter Beast of Traal."

Hey great HHGG reference! For a bit there I felt like no one would lift a finger to answer my RAM question without orders signed in triplicate, sent in, sent back, queried, lost, found, subjected to public inquiry, lost again, and finally buried in soft peat for 3 months and recycled as firelighters. I found a special version of the radioplay when I was 8. No one seems to know that version even exists, but it's superior to the BBC version in 2 important ways. First, it is slightly cheaper...

Since I'm just beginning to learn Fusion, I'm trying to figure out the best approach to what I'm trying to do before I commit to being absolutely tied to needing a 16k timeline. I'm learning about 3D camera at the moment to see if this can eliminate the need for that big of a timeline.

I'm also trying to figure out if I can position my elements in Fusion at a lower resolution for speed of working purposes, and then when I'm ready to render, set a resize node to the 16k resolution to get the sharpness on zooms. I still have a lot to learn.

"GPU Shared Memory"

for my understanding, to evaluate and calculate a frame in the color tab (not fusion) only the GPU is used. and only the memory in the actually GPU card is used. (from what I know from BM).

that calculation if you have (let's say a 24fps timeline) need to be execute within 1/24th of a second. if that not happen you break real-time.

CPU is responsible to get the frame from disk, decode if necessary, pass it to the GPU with the metadata instruction of what to do, get the frame back, give it to the GUI and the video card out. This si the other memory usage.

it will be a waste of time to push the image out of the GPU to a slower BUS in attempt to use more RAM: you will never get realtime.

if you have two GPUs, each get the other frame and have 2/24 of a second to perform calculation, while the CPU task stay the same.

it will be a waste of time to push the image out of the GPU to a slower BUS in attempt to use more RAM

Thank you Walter. My specs say: CPU bus speed - QPI: Up to 8.0 GT/s.

Is this fast enough for my GPU to push the image out and benefit from more RAM, or is this considered a "slower bus"?

it will be a waste of time to push the image out of the GPU to a slower BUS in attempt to use more RAM

Thank you Walter. My specs say: CPU bus speed - QPI: Up to 8.0 GT/s.

Is this fast enough for my GPU to push the image out and benefit from more RAM, or is this considered a "slower bus"?

(what is a GT/sec?)

you cannot build a software based on one specific specs only, it's more of a philosophy: if you have one GPU, you go in and out twice, if you have two GPU is 4 times, if you used shared memory during the frame calculation, it's 8 times and so on.
in the Linux at work we have 5 GPU: one for GUI and 4 for computation, you can see that the bus can get very busy. (dual xenon... dual bus too...).

there was a version of resolve that due to a specific way the final "to the display" path was build, under some setting the image was going in and out of the GPU twice per allotted frame (1/24 of sec) and it was killing very powerful machines. (this has been solved in V15 IIRC).

I dont think the GPU ever pass the image to shared RAM within resolve.

now, if you use Fusion AND the images are NOT cached AND the tools you use are GPU accelerated, you might have a back and forth between GPPU and CPU and bus... that kills real-time for sure.

I'm not sure what GT/s is, but that's the bus speed according to the specs on my motherboard.

I'm definitely buying more RAM now that I'm working in Fusion more because I can see it consistently uses its maximum allotted. When using Fusion it does not dip into Shared GPU Memory very much at all. It uses the GPU fully, and RAM. So I anticipate when I get my new RAM and more than double what Fusion is allowed to use, my performance will improve. I can barely play as it is. I'm working essentially from still frames and have to render just to see a preview because I have so much going on. Hopefully RAM will help with playback. I'll report back when I get it.

It doesn't seem that Fusion caches so a preview is ready. Does this mean that Playback > Render cache is only for the Timeline? And my Playback > Fusion Memory Cache is set to "Auto". Any reason to change this? It would be nice if Fusion could pre-render so I could occasionally playback in close to realtime, but I'm not familiar with how to do this.

From the Beginner's Guide.

"You can assign more or less RAM for Fusion RAM playback in the Preferences panel. The amount of RAM assigned to Fusion RAM playback is taken from the total amount assigned to the DaVinci Resolve application." [emphasis added]

From the Beginner's Guide.

"You can assign more or less RAM for Fusion RAM playback in the Preferences panel. The amount of RAM assigned to Fusion RAM playback is taken from the total amount assigned to the DaVinci Resolve application." [emphasis added]

Given the above.. if I never use fusion, would a best practice be to set the Fusion assigned RAM to a lower number, thus freeing up more RAM to be utilized by the Color page?

No this is just RAM used when you are using it. If you aren't using it, it won't be used.

I think Stephen is right. The Fusion RAM setting doesn't exclude that amount form use on other pages, it simply limits the Fusion page to that amount.