Thoughts on system upgrade path

This is a question mostly for those who have been through upgrading their computer (windows machine specifically), and have some idea of what has the greatest impact on increased performance regarding Resolve. I am trying to limit the cost to $500, which I know is effectively nothing to work with.

So, my system. I'm running a win 10 machine. The upgrade paths are, RAM, CPU, or Graphic card.

RAM: I have 32GB RAM currently. My motherboard can upgrade to 64GB.

CPU: My CPU is an Intel i7 5820K 6-core 3.3GHz. This upgrade path would require not only upgrading the CPU, but also the motherboard...and possibly other components.

Graphics: I am running on an XFX Radeon R9 290 4GB graphics card.

Again, what I am interested in are thoughts on which upgrade path will have the overall maximal performance increase, and notable recommendations regarding specific hardware (if any happen to come to mind).

Thanks!

New CPU will require a new motherboard as you say due to the different socket type.
RAM on your system should already be DDR4 if I'm not mistaken. However if you want to go to 64 and your current 32GB is 4 sticks instead of 2 you're out of luck (assuming you have 4 RAM slots) unless you buy a full new set of 64GB. If it's 2x 16GB it can still be tricky to get the exact ones you currently have, you'd need to check if they are still available.

GPU is something you can pretty much always upgrade and I think getting one now with at least 8GB highest midrange to high end will have the biggest impact.
Would personally go for something like an RTX3070Ti or higher. You'll want to check your power supply though if it's at least 80Plus Bronze and enough Watts to power it.

[edit] totally glanced over the price limit sorry. Dunno what the prices are in your region but an RTX 3060 is probably also a pretty good upgrade which cost 499 here.

You would be correct that it is DDR4 RAM. The RAM upgrade is actually the cheapest route. Even though outdated now, I did pick a good motherboard when I originally built this machine in 2014. Thus, I have 8 available RAM slots (8GB sticks are the largest it will accept), thus the 64GB limit. This is one of the options on the list I had determined simply because I have seen it mentioned several times that more RAM could help. That said, I'm haven't yet been convinced the 32GB RAM is the most problematic bottle neck in my system.

From bits I have read, my guess was the CPU is the most problematic bottleneck. But, with having only used Resolve for a couple weeks now, I thought the worst problem might be the GPU.

It may help to know I am not working on anything larger than 1080p. This, of course, is what led me to the line of thinking that my worst issue may be the CPU. I have seen a few people comment on not needing anything more than 4GB VRAM having major impact if only working with 1080p footage. But...those people were also discussing Resolve 16... Likely, still relevant...but I figured it would be best to simply ask for opinions from people who are actively involved.

All that said, yeah, the RTX 3060 was looking to be a prime option for the GPU path. Glad to have that reaffirmed as a reasonable option.

I hadn't completely ruled out the CPU path since I have seen some motherboard & CPU combos with Ryzen 5 & 7 for less than $500 USD. Obviously, i would have to investigate the RAM compatibility. I would need a new cooler...maybe a good time to move to a very quiet air cooler??? But, I have Corsair AX860 80 plus platinum power supply, which may be usable with a much newer motherboard given the modular connectors. And, with my ATX mid tower case, there should be plenty of room.

Anyway...thanks for the feedback!

More GPU memory is always the most helpful thing in resolve.

Another opinion for GPU. RTX 3060 or better.

My choice here would be for the best nVidia RTX GPU I could manage.

Thanks for the feedback! I ended up going with an RTX 3060ti plus upgrading to 64GB of RAM originally, but that has proven seriously inadequate. So, next week, I will be installing an RTX 3090ti.

Case in point, I am currently rendering out a 3:10 long clip of a simple masked background with a modified sky (doing some mild gridwarp perturbation)...and with 5% completed the time remaining is showing at just short of 4 hours and 13 mins. smh... I wanted to have a continuous 30 minute long background...but the time required is just obscene to render that out. Plus, I have several fusion segments that will also need to be overlayed onto that.

How much GPU is using in that?

How much GPU is using in that?

GPU utilization stayed around 80-90%, with memory utilization also at about 80%. It was enough that I could still surf the net...with some very notable lag. I didn't spend much time online, but I wanted to see how adversely it was affecting the system, and it was very notable. So, not maxed, but close. The render finished in 4 hours 15 minutes, and 45 seconds. I rendered it to h.264 at 4k with no audio. So, I now have a decent baseline for the 3060ti to compare against the new hardware when it arrives.

I've decided I am going to install the 3090ti without changing anything else in the system, do the test. Then pull out both cards, and setup the 3090ti only to do the final test. With what I have heard thus far, I am expecting best results to come from the 3090ti only setup. If not, I will be more than happy to switch back to the previous dual set up.

Looking forward to the results of your test (per other thread)!!!

How much GPU is using in that?

GPU utilization stayed around 80-90%, with memory utilization also at about 80%. It was enough that I could still surf the net...with some very notable lag. I didn't spend much time online, but I wanted to see how adversely it was affecting the system, and it was very notable. So, not maxed, but close. The render finished in 4 hours 15 minutes, and 45 seconds. I rendered it to h.264 at 4k with no audio. So, I now have a decent baseline for the 3060ti to compare against the new hardware when it arrives.

I've decided I am going to install the 3090ti without changing anything else in the system, do the test. Then pull out both cards, and setup the 3090ti only to do the final test. With what I have heard thus far, I am expecting best results to come from the 3090ti only setup. If not, I will be more than happy to switch back to the previous dual set up.

Thanks. What encoder did you choose? NVENC(Nvidia) and quality?

Thanks. What encoder did you choose? NVENC(Nvidia) and quality?

Nvidia and best (automatic)

Ok. It seems a huge amount of time, wonder what is pushing it. Maybe you take every FX one by one and check and then try to find a more efficient option for the culprit.

Ok. It seems a huge amount of time, wonder what is pushing it. Maybe you take every FX one by one and check and then try to find a more efficient option for the culprit.

There isn't a whole lot I can do to reduce the complexity. Here is the node tree.

Capture.JPG (53.44 KiB) Viewed 966 times

Anthony,

How does it perform on standalone Fusion?

Can you share a sample project we could test to see how our systems perform?


Sent from my iPad using Tapatalk

Anthony,
How does it perform on standalone Fusion?

Heh! Funny you should ask that question...cause I only found out a few hours ago that my DR license was also good for Fusion Studio. I have spent the last few hours moving my current project into FS to compare it to how everything flows in DR.

First, the overall... FS feels like it flows smoother and the preview windows feel like they have a higher quality render. Not sure if there is any validity to the quality, but it does make sense for the workflow to feel smoother since it is a streamlined platform.

Transitioning the project from DR to FS was almost entirely a copy and paste procedure. I was using two reactor nodes (XGlow and Tintensity) which required manual adjustment. This tells me 100% for sure there are some processing differences under the hood. Still not sure exactly if there is a real quality difference though...as it could be that I just happen to like the result I ended up with better in FS (due to being forced to manually adjust XGlow and Tintensity).

Other than the things I pointed out, it was just like working in DR. I can say, from now on, any of the more complex fusion stuff WILL be done in FS, but otherwise the projects will be done entirely in DR.

Can you share a sample project we could test to see how our systems perform?

Not at the moment. I can tell you the mediaIn (MainBG) is a still image @7932x5291 and the entire project is 4K UHD @24fps, 00:03:04.00 duration. Instance_MainBG_1 is the primary background layer. MainBG is a masked sky (about half the image) which is gridwarped using a mild perturb modifier layered on top of the primary background. Instance_MainBG is the masked foreground layered on top of the grindwarped layer. It just a simple moving cloud...cinemagraph??? motion image??? i can't remember for sure what it is called when a person induces motion into a still image...without using video.

If there is interest, I may throw some tutorials (with project files) together, showing how I did everything in the final project. It seems super simple to me, but maybe it isn't as straightforward as I am perceiving it???

Did you try VFX Connect to make it easier getting the results into DR?

Did you try VFX Connect to make it easier getting the results into DR?

Never heard of it. I will have to look into it.

See chapter 54 in the manual (it used to be called Fusion Connect in older versions).

See chapter 54 in the manual (it used to be called Fusion Connect in older versions).

I just hunted down a video on it.

Resolve to Fusion via VFX Connect from the VFXstudy channel

Sounds nice to keep the workflow streamlined.

By some hack you can even use it with other compositing software.

I can tell you the mediaIn (MainBG) is a still image @7932x5291

Had you shared that info earlier, you would probably have been given advice to try changing the Media In to a Loader node (in Resolve's Fusion Page)... as I understand it, a 'Loader' node loads the still into memory only once, whereas a 'Media In' node reloads it on every frame. In the past, Folks have noted significant performance improvements when using a Loader node with stills in Resolve based Fusion compositions.

I can tell you the mediaIn (MainBG) is a still image @7932x5291

Had you shared that info earlier, you would probably have been given advice to try changing the Media In to a Loader node (in Resolve's Fusion Page)... as I understand it, a 'Loader' node loads the still into memory only once, whereas a 'Media In' node reloads it on every frame. In the past, Folks have noted significant performance improvements when using a Loader node with stills in Resolve based Fusion compositions.

Woah! That's a huge deal. Thanks for bringing that up. I will switch to using Loader nodes for all still images from now on.

So I gave using a loader node for still images a try....I created a Fusion comp "comp-1" I exported this comp and then created a new Fusion comp "comp-2" I imported the first comp into the new one and everything worked great. Does anyone know how to change the file path of the loader node? I noticed the the "browse" file area on the inspector panel in no longer there..... like it was when I created the loader node and pointed it at my initial image. Am I missing something? Any help would be greatly appreciated.