EXPERIMENTAL: Resizable BAR in DCS for NVIDIA RTX 30 Series GPUs

[Buzzer1977]

If theres no data that can be accessed thru BAR how do you think is BAR going to be used ?

[ShaunX]

I'm not an expert and have no idea, which is why I'm asking for an explanation.

 

I've only read..

"

How does Resizable BAR work? In a nutshell, the feature, set via the system BIOS, determines how much of the graphics memory, or VRAM, on your video card is made available to be mapped for access by the CPU. Generally, this is limited to just 256MB of the card's onboard VRAM—which is to say, not much of it. A motherboard with Resizable BAR activated, however, can boost the limit to the full capacity of the VRAM buffer.

"

so you're saying that the card's VRAM buffer is represented by the shared gpu memory graph in task manager, so if it only remains at 0.1 then it's still only accessing the 256MB and therefore resizable BAR is not being used ?

 

So would you say that this means that setting the 3 flags is not enough to enable it's use in DCS, or that the CPU simply doesn't utilise it for DCS (on my rig at least) ?

 

Really appreciate any light you can shed.

 

[ShaunOfTheFuzz]

Shared GPU memory is different. Shared GPU memory refers to the case where your VRAM is at maximum capacity and the application has to dip into your system RAM. You never want to see this happening, and with a 3090, you won't.

 

Rebar is basically about using PCIe bandwidth to maximize access to VRAM. Currently the standard is to access it in small chunks of 256MB as stated above. Rebar is designed to maximize VRAM throughput.

Edited April 15, 2021 by ShaunOfTheFuzz
capitalization

[Hoirtel]

I'd be interested to hear if 2.7 has changed the way in which rebar is or isn't utilized in DCS. There seems to be lots of performance changes!

[Woona]

  

21 hours ago, Buzzer1977 said:

Have you verified that Resizable Bar is not only enabled but used ?
To check if your system has BAR enabled (Supported CPU, BIOS, BIOS settings, VBIOS, driver, ...)
open the NVIdia control panel (right click on the desktop) and check the system informations in the help menue.

To see if the system is actually using BAR open the GPU view in the task-manager and watch the shared memory segment while running the application.
0.1 GB is standard on a idle system and no sign of actally using BAR in the application.

 

Hi, as Shaun mentions, shared GPU memory is not indicative towards utilization of resizable BAR but rather indicates a usage of system memory by the GPU in the case that your GPU's VRAM is fully utilized. As you can see in your screenshot, that is also why it says 64GB - unless you have a GPU with 64GB of VRAM, of course. This tech predates even PCI-E 4.0 by several years.

https://www.extremetech.com/computing/251763-microsoft-shares-new-details-gpu-monitoring-capabilities-windows-10-fall-creators-update#:~:text=Shared memory represents system memory,memory used by that process.

Edited April 15, 2021 by Woona

[Woona]

So slightly off-topic but I thought it'd be worth a mention, although not worth a thread. I don't know if it's the enabling of ReBAR or the update to 2.7 that's done this, but just now I was running the good ol' NVCP tests to make sure nothing in the Nvidia Control Panel had changed. As usual, most everything didn't work and is a waste of time, but one thing that used to give some deltas is now giving yuge deltas. Check it out.

 

Wish I'd checked for this on 2.5.6 but I'm not bothering rolling back just for that. One has under 1/3 the missed frames, the other higher averages and half the frametime... I think the only way to really pick here is actually trying it out in the headset. Do any of you have experience between these two settings in NVCP?

Edited April 15, 2021 by Woona

[ShaunX]

I've never set it to anything but off.

 

I'll give it a whirl against the benchmarks I've been taking sine I did the ReBAR tests a in this thread on Tuesday

 

..ran it twice and saw no difference. Then again noticed my setting was actually "Use the 3D application setting" rather than off

Edited April 15, 2021 by ShaunX

[Hoirtel]

For the last few months I have always used fast in VR. Really felt that it made things much smoother and cleanly locked my fps to reprojected or full refresh rates. Good to see some data to support that!

[imacken]

Despite the result above, I have always believed and been told that VSYNC has no effect on VR. I have it permanently ON for monitor purposes. 

[Woona]

I think we should end the conversation on Vsync Fast vs Off with this post so we don't end up getting a hijack-situation, unless someone has a ReBAR addition to it. Sorry to kick it into a weird direction all of a sudden.

 

I just ran my Batumi benchmark with my headset, switching between the two. My experience kinda cements the results above. By going full aft, pointing to the sky, then quickly jinking down towards Batumi, i'd induce a stutter. Sometimes a stutter would just happen. I was not able to induce it on any of my Vsync Off runs. However, my Vsync fast setting bumped up my refresh to full rate more often. I fly with my Index locked to 48fps anyway (fellow Index users take notes, it's awesome) so I'll take the decrease in microstutters any day.

 

And here's the kicker: My global default was Vsync Fast, and I haven't touched my global defaults. Might poke around other people's DCS VR systems later and see if theirs is the same and if they experience the same decrease in stuttering on Vsync Off - if they were on Fast at all, of course.

[Twwhitey]

Just to add another unquantified data point seeing that ASUS released a BIOS update for the Z390 Hero XI MBs I went ahead and took the plunge this AM working through the process of activating ReBAR for DCS (ASUS Hero XI Wi-Fi MB, i9900K OC’d 5GHz, ASUS Tuf 3090 OC, Reverb G2).   No real immediately noticeable improvement (or obvious degradation for that matter)... which I suppose is expected given it sounds like any perf gains won’t appear to be demonstrably overwhelming.  Going to leave it activated and see if I notice anything worth reporting....

[jurinko]

Well since I like to mess with things which work, I went through the ReBAR procedure and it is a sure thing that it helps significantly. I got my RTX3070/5600x/32GB combo half+ back to 2.5.6. smoothness. My benchmark is rather unscientific, free flight in F-5E starting in a forest valley, then breaking right to the open country and buzz over some city. I know exactly when the game becames 100% fluid (after leaving the mountains behind) and with ReBAR it becames fluid almost as soon as 2.5.6. Microstutters over the city are half of 2.7. Basically, with ReBAR I get the same fluidity as if downgrading Pixel Density from 1.0 to 0.8-0.9. If you don't mind messing with BIOS, worth trying.

[dburne]

16 minutes ago, jurinko said:

Well since I like to mess with things which work, I went through the ReBAR procedure and it is a sure thing that it helps significantly. I got my RTX3070/5600x/32GB combo half+ back to 2.5.6. smoothness. My benchmark is rather unscientific, free flight in F-5E starting in a forest valley, then breaking right to the open country and buzz over some city. I know exactly when the game becames 100% fluid (after leaving the mountains behind) and with ReBAR it becames fluid almost as soon as 2.5.6. Microstutters over the city are half of 2.7. Basically, with ReBAR I get the same fluidity as if downgrading Pixel Density from 1.0 to 0.8-0.9. If you don't mind messing with BIOS, worth trying.

 

I thought Nvidia would have to provide support for DCS and ReBAR in order for this to function?

[imacken]

23 minutes ago, dburne said:

 

I thought Nvidia would have to provide support for DCS and ReBAR in order for this to function?

See post #1 here. 

[dburne]

3 minutes ago, imacken said:

See post #1 here. 

 

Yeah I had seen that previously but seems to be based on some guesswork.

Is it confirmed it actually works to enable it properly for DCS?

 

Edited April 20, 2021 by dburne

[ShaunX]

I started this thread with some optimism but I think the jury's still out on a) whether the Nvidia Profile Inspector changes actually enable its use in DCS and b) if they do, whether it has any effect.

 

I've performed a number of benchmark comparisons on and off. Sometimes I measure a noticable improvement and sometimes none at all with the same benchmark track file. I'd been doing benchmarks against 2.7 and different Kegetys' variations without it, so I turned it on again after those yesterday...no improvement.

 

All I can conclusively say at this point is that it doesn't appear to have a detremental effect.

[jurinko]

I tested it again on 2.5.6. installation which I set up for research purposes. ReBAR definitely works. My empiric method is to fly above one city, look sideways and roll up and down. 

 

2.5.6 ReBAR ON no stutter at all > 2.5.6 clean very slight stutter > 2.7 ReBAR ON slight stutter > 2.7. serious stutter. 

 

 

[dburne]

2 hours ago, jurinko said:

I tested it again on 2.5.6. installation which I set up for research purposes. ReBAR definitely works. My empiric method is to fly above one city, look sideways and roll up and down. 

 

2.5.6 ReBAR ON no stutter at all > 2.5.6 clean very slight stutter > 2.7 ReBAR ON slight stutter > 2.7. serious stutter. 

 

 

 

Interesting, I typically don't get stutters but may give it a try with 2.7 just to see if I can see any improvements.

[JaBoG32_Prinzartus]

On 4/5/2021 at 8:17 PM, speed-of-heat said:

Where did you get the BIOS to support your Maximus X Hero, last i could see ASUS have not yet released either their X  or XI MB, which you would require in order to get any different perf 

https://rog.asus.com/motherboards/rog-maximus/rog-maximus-x-hero-model/helpdesk_bios

 

There is a BETA BIOS available. I am currently using this, enabled reBAR and I am very grateful for this feature in combination with my 3080 FE and VR. I (subjectively) noticed a an improvement with reBAR enabled. My advice, go for it!

 

Attention: I had to completely redefine my BIOS OC setting. So, you might want to do phone screenshots as for easy reconfiguration, if neccessary.

Edited April 23, 2021 by JaBoG32_Prinzartus

[speed-of-heat]

I have the XI https://dlcdnets.asus.com/pub/ASUS/mb/LGA1151/ROG_MAXIMUS_XI_HERO/ROG-MAXIMUS-XI-HERO-ASUS-1901.ZIP the BIOS is here... turning REBAR on cost me about 1.3ms of GPU frame time and about 0.10ms of CPU time on that BIOS,  REBAR is turned off for me until the BIOS matures and I will give it another go when it gets out of beta. 

[ShaunX]

On 4/21/2021 at 6:47 PM, jurinko said:

I tested it again on 2.5.6. installation which I set up for research purposes. ReBAR definitely works. My empiric method is to fly above one city, look sideways and roll up and down. 

 

2.5.6 ReBAR ON no stutter at all > 2.5.6 clean very slight stutter > 2.7 ReBAR ON slight stutter > 2.7. serious stutter. 

 

 

 

I tried to test this, but as I don't see any stutter performing those manouvers in 2.7 without it I can't go any further. Still not convinced one way of the other.

[Hoirtel]

Hey all. I have just tried to enable ReBar in Nvidia profile inspector (following Luka's YT vid). For my driver, 466.11 I don't see the list of unknown flags at the bottom. There is only one entry, this is the same on the global profile. Nv inspector does seem to be working as everything else is there and I can see my DCS specific settings. I'm pretty sure I'm doing everything else right. Has this process now changed with a the driver update? I could roll the driver back but I wanted to check to see if anyone else has seen this or if I'm missing a step. Thanks

[Twwhitey]

1 hour ago, Hoirtel said:

Hey all. I have just tried to enable ReBar in Nvidia profile inspector (following Luka's YT vid). For my driver, 466.11 I don't see the list of unknown flags at the bottom. There is only one entry, this is the same on the global profile. Nv inspector does seem to be working as everything else is there and I can see my DCS specific settings. I'm pretty sure I'm doing everything else right. Has this process now changed with a the driver update? I could roll the driver back but I wanted to check to see if anyone else has seen this or if I'm missing a step. Thanks

easy overlooked button you need to click to see the “unknown” profiles, did you click that button?

    Click "Show unknown settings from NVIDIA Predefined Profiles" - second button from right on the toolstrip.
    

    
    Note the 3 entries which list 17 profiles.
    
    On the first 2, click on them and amend 0x00000000 to 0x00000001
    On the 3rd, click on its dropdown and select the only item.
    You should now see this..

[Hoirtel]

Thanks @Twwhitey That was what I had missed!

 

I have now tested my DCS with and without Rebar and I can honestly say it is a good improvement for me. I unfortunately haven't got any helpful frame times to provide quantitative data but I am 100% sure I have seen an improvement in overall latency. I run a Quest 2 (yes.. groan) on a AIB OC'd 3090 with an i9 9900ks. The CPU is overclocked at 5.2ghz on eight cores, 8 threads. Switching off HT allowed me to get a very stable and cool 5.2 overclock with no AVX offset. My system is primarily for DCS so 8 cores is fine.

 

My test is an F16 taking off from Incirlik with ground statics and doing a low right hand pass over Adana before going on to do some low flying across the countryside it then pulls up and does the same at height. I have a cloud pre-set on now too. My main issues with DCS is the ground judder in these situations and while not what I normally do I have found this helpful for tuning and tend to find that the better this is, the better the rest is.

 

Oculus is pretty rubbish at any helpful monitoring, the oculus overlay won't show in DCS and the framerates are always capped at half or full even when ASW is disabled. I have found that with the Q2 I can get this ground judder even when locked at 40fps (my reprojection level), which isn't always what I have found with other headsets. So I am left with a fairly subjective way of testing. As my framerate seems ok I was thinking my bottleneck would be not my CPU or GPU but something else - I have mostly put it down to the link encoding and decoding which is an unfortunate additional bottle neck for this VR (as if I needed another one!). Rebar has definitely improved this for me. I wasn't necessarily tuning to have this run without any judder, but rebar made it very very close. I was very surprised. On the third run I took control of the track and immediately did fast alerion rolls, some minor judder but wow. Pretty good. Of course I then had to see how much more I could get, so I upped viewing distance and MSAA to x4 (x2 is my norm) in various runs. This did bring back the judder but still vastly reduced on previous rebar off runs. I returned to my normal settings -  all fine again.

 

Additional the whole game feels smoother now, I have gone on to try other scenarios including one of my other tests which is the supercarrier cold start mission. I don't usually take off but this has always been a very taxing scenario, particularly when the AI jets are taking off and I look toward the island. This again was very smooth. So all in I'm pretty pleased, I'm sure rebar has eased a bottleneck and I will definitely be leaving it on now. 

 

Thanks to everyone who has been working on this!!

[jurinko]

I downloaded new nvidia drivers and though they are rubbish since DCS ran like crap. Then I remembered that ReBAR, checked it in nvidia inspector and turned it on. Yeah.

Everyone with 30X0 card, give it a go.