Clarification on possible conflicting wisdom from DynDOLOD site and STEP Guide 2.3?

[S-Matrix]

Hi! I typically use the current STEP Guide iteration as a baseline for everything pertaining to LODs, and I had a question regarding the optional DynDOLOD toggle "Upgrade NearGrid Large References to FarGrid". The DynDOLOD site (on this page: Advanced Mode | DynDOLOD) explains that, if the large reference system is being used, the rule of thumb is to leave this unchecked. However, STEP Guide 2.3 instructions involve setting the large reference grids to 9 in the BethINI section but also ticks this box in both of the DynDOLOD config instructional images.

I know it's possible that one source might be less up to date than the other and as a result may not reflect the most current technical details of DynDOLOD; additionally, the principle is theoretically just a rule of thumb as opposed to a strict necessity. If it's the latter case though, what makes it a good idea to do this for the STEP Guide load order as opposed to following the rule of thumb?

Edited February 26 by S-Matrix

[DoubleYou]

You can uncheck it if you want more performance. If doing so, you may consider using a higher large reference grid value. I don't recall if we did much testing on the setting's performance impact. 

[z929669]

The "rule of thumb" is actually to tick the box "in case the LR system is not used" ...not untick the box. Also, with the BethINI Ultra preset, uLargeRefLODGridSize = 11 (it's 9 in the guide, because the guide recommends the BethINI High preset as a baseline before testing performance impact).

We didn't find any performance impact with "Upgrade NearGrid Large References to FarGrid" ticked or unticked in our setup back in 2018--all else being equal. We did not do any extensive testing though, and opted for the "better visuals" option at the time. That said, it would be useful to do more thorough testing to determine if it matters visually and if there is any significant performance impact if it does. Keep in mind that we're talking about dynamic LOD, so you need to have large refs that are impacted by dynamic LOD to test visuals. Performance impact should be testable regardless though.

Let us know what you find, but I agree that we should 'theoretically' not have this setting toggled in our GUI reference image. It's one of many things that are on the back burner.

[S-Matrix]

  On 2/27/2025 at 3:01 AM, z929669 said:

The "rule of thumb" is actually to tick the box "in case the LR system is not used" ...not untick the box.

Expand  

Sorry if I'm being obtuse, but just in case, I want to point out that the following is also written in the section where the rule of thumb is detailed:
"Vice versa, do not check this option in case the large referene system is used - uLargeRefLODGridSize > uGridsToLoad >= 5"

I've actually spent a fair bit of today doing general performance tuning that extends beyond tweaking DynDOLOD, so I'd need to do more controlled testing to know anything conclusively. That being said, after finishing up the DynDOLOD gen process earlier with the box left unticked, my exterior performance in the Rift had pretty noticeably improved during the play session I started up immediately afterwards (in regards to both average FPS, and smoothness/consistency over fast camera rotations). I personally didn't notice any change in visuals whatsoever (just eyeballing, so of course this also isn't conclusive) - I have my Large Ref grids setting at 9, and as you probably figured, I'm using the Large Reference workarounds implemented via the NG DLL.

Again, it could definitely just be that it coincided with unrelated and more significant changes I'd made outside of DynDOLOD (at the moment I don't remember what else I might have changed specifically before that session), so this is really just me saying that my findings, as of now, aren't exactly reliable and that I'd need a fair bit of time to set up a good testing environment (regenerating LOD, running a test save, etc).

Edited February 27 by S-Matrix

[z929669]

  On 2/27/2025 at 5:12 AM, S-Matrix said:

Sorry if I'm being obtuse, but just in case, I want to point out that the following is also written in the section where the rule of thumb is detailed:
"Vice versa, do not check this option in case the large referene system is used - uLargeRefLODGridSize > uGridsToLoad >= 5"

I've actually spent a fair bit of today doing general performance tuning that extends beyond tweaking DynDOLOD, so I'd need to do more controlled testing to know anything conclusively. That being said, after finishing up the DynDOLOD gen process earlier with the box left unticked, my exterior performance in the Rift had pretty noticeably improved during the play session I started up immediately afterwards (in regards to both average FPS, and smoothness/consistency over fast camera rotations). I personally didn't notice any change in visuals whatsoever (just eyeballing, so of course this also isn't conclusive) - I have my Large Ref grids setting at 9, and as you probably figured, I'm using the Large Reference workarounds implemented via the NG DLL.

Again, it could definitely just be that it coincided with unrelated and more significant changes I'd made outside of DynDOLOD (at the moment I don't remember what else I might have changed specifically before that session), so this is really just me saying that my findings, as of now, aren't exactly reliable and that I'd need a fair bit of time to set up a good testing environment (regenerating LOD, running a test save, etc).

Expand  

You are not being 'obtuse' in the least. You are being pedantic, which I think is a good thing in these situations

Again, I agree that we should NOT be ticking that box, so we will likely add change this in a future update or guide release. I would like to quantify the performance impact at least to some degree. Unfortunately, I have a fairly powerful PC, so I may not see much impact. I'd love to know the impact on a low to mid-end PC, but reason dictates that it would be more apparent.

[z929669]

Just following up on this. I did some testing and confirmed that "Upgrade NearGrid Large References to FarGrid" doesn't have any consistent performance impact on my system. Some areas seem to gain a few FPS, while other areas seem to lose a few.

I will continue testing this and a few of the other settings before arriving at a refined recommendation. Either way, we will recommend that "Upgrade NearGrid Large References to FarGrid" not be ticked.