Identical rigs, different performance

[Alonzo]

Hey everyone,

 

I'm a VR regular and have been playing in VR for 18 months. I regularly help out with the benchmarking that Chili does. I run the Combat Box server (along with a talented admin team) and I'm a tech guy for my day job. So in that context, you can imagine how frustrating this issue must be:

 

A fellow flier, @IdahoBookworm has been having problems with his Rift S. In War Thunder, he gets a rock-solid 80 FPS. In IL2, he's had a lot of issues, regularly ending up at 40 ASW. And when the ASW does kick in, it's often not just ghosting and double-planes, but some kind of weird oil painting effect. This started months ago, and Idaho was on an 8700 (non-K) and we recommended that was only just good enough for IL2. So he's been upgrading, to the point where he has a very capable rig:

• Intel 9600K @ 4.9ghz, zero AVX offset, 4.3ghz ring ratio
• 16GB 2666mhz RAM at 16-18-18
• RTX 2080 Super

 

Even with the latest patch that has wowed everyone with performance, he's having problems. I found this incredible, so I downclocked my machine to match his. Mine is this spec:

• Intel 8086K @ 4.9ghz, zero AVX offset, 4.3ghz ring ratio, HT off
• 16GB 2666mhz RAM at 16-18-18
• RTX 2080 (non-Super)

 

So my rig should be actually a shade slower than his.

 

What we're seeing, though, is that doing like-for-like activities, my rig has lower CPU frame times (as measured by Oculus performance HUD) than Idaho's rig. The difference is stark enough that while I'm getting ~9ms frame times on Combat Box training server in the dogfight zone, he's getting ~14ms+ and it's pushing him into ASW (ASW is off, actually, he's hitting 40 FPS ATW I think).

 

Spoiler

My rig, clocked 100mhz higher than his at time of filming, but basically identical. Combat Box training server, 10+ planes whizzing around.

 

Idaho's rig. You can see when he gets surrounded by planes (= more network traffic?) his CPU frame times spike.

 

 

So why would two almost identical rigs get different IL2 performance? We've checked CPU-Z, GPU-Z, x16 PCIe slot, memory latency... I'm at a loss. And Idaho's been upgrading his rig for the last few months so literally the only components that are still the same as when he started out are his SSD, RAM, and the Rift S itself. He might be able to borrow a different headset from someone else, but that'll take a while.

 

I'm at a loss. And I'm a tech guy so not being able to figure something out is annoying and it pokes at me. It seems that his CPU gets hit harder than mine by network traffic. Anyone got ideas? He's using wired ethernet from the motherboard.

[SCG_Fenris_Wolf]

Motherboard is my first guess. Faulty bridges, drivers, blown capacitors, a "dirty" lane? Something physical. I've had a bad MoBo before. This is all about latency in the feed maybe something's wrong there? 

 

I can only guess as well from here.

 

Did he freshly setup his OS? Must have. Second guess is RAM. Just intuition though too.

 

I'd start with the MoBo. Check, replace, check, go on. The old algorithm.

Edited May 23, 2020 by SCG_Fenris_Wolf

[SPUDS_IdahoBookworm]

Hey Fenris, thanks for the suggestion. I just got this motherboard. I was using a B360 board and getting poor performance, so I upgraded to a Z390 so I could overclock the 9600k. Performance did improve, but as you can see, it's still fairly bad.

 

With the new mobo I did a complete reinstall of Windows. I also moved IL-2 to a different (SATA) SSD. I've tried running it off the C drive (M.2 NVMe) before with no improvement there. 

 

Since I started trying to get into IL-2, I have upgraded everything except the hard drives (and actually, one of them is new, I just forget which) and the RAM. The only other thing that's the same is a single fan I moved out of my old case.

[Alonzo]

I was wondering if bad USB could somehow cause the Rift S to create more CPU load. Or that network packets could, somehow. In single player things seem reasonably similar between the two rigs, but in multiplayer with more network traffic (not necessarily more polygons on screen) the CPU gets hit hard on Idaho's rig.

[dburne]

Use task manager to check your running programs and see if anything is eating into your CPU resources.

Are you using GeForce Experience, or have it installed for your GPU?

You can also try a clean driver install by uninstalling graphic drivers with DDU ( Display Driver Uninstaller) and installing the latest Nvidia drivers clean in safe mode.

Just install the driver, PhysX, USB-C port driver. No need for GeForce Experience.

Also you might try running the Rift S USB through that C-Port on the 2080 GPU, assuming it has one.

[A_radek]

You checked running processes? No funky trojans you could have picked up already with the fresh install?

 

I remember some bloke on this forum had prime95 auto start with windows without knowing, complained about both performance and loud fan noise at ‘idle’.

[VR-DriftaholiC]

Make sure your on the latest chip-set drivers. Try disabling C-states in bios. Setting power plan "Ultimate Performance" for lowest latencys (can find how to enable via a quick google search) in windows and then turning off USB selective suspend to remove that as a worry.  Probably some driver or software in the background though.  

You can log with HWINFO and see if something is dipping occasionally

Edited May 23, 2020 by driftaholic

[SCG_Fenris_Wolf]

1 hour ago, Alonzo said:

I was wondering if bad USB could somehow cause the Rift S to create more CPU load. Or that network packets could, somehow. In single player things seem reasonably similar between the two rigs, but in multiplayer with more network traffic (not necessarily more polygons on screen) the CPU gets hit hard on Idaho's rig.

 

Hmm you mention in MP though mostly. Lots of packet loss maybe? 

 

What if you run the tests on the same flight record under otherwise equal conditions? You could rule out a local hardware issue that way as well. 

59 minutes ago, SvAF/F16_radek said:

You checked running processes? No funky trojans you could have picked up already with the fresh install?

 

I remember some bloke on this forum had prime95 auto start with windows without knowing, complained about both performance and loud fan noise at ‘idle’.

 

On funny stories, I've had helped a (non-techie) guy who hadn't noticed he had put his monitor into his onboard HDMI port... his 1080 sat in the Computer for around 3 months living the chilled life. 

 

Not even remotely suggesting such a thing would be it, just sharing a story.?

[chiliwili69]

There is one little detail going from 9600K to 8086K, they have the same number of cores (6) but the 9600K has 9Mb L3 cache and the 8086K has 12Mb (2Mb per core, same than 9900K or 10900K).

We still really don´t know how much matter L3 cache in IL-2 VR. But probably not enough to explain this difference.

 

You can try to compare performance using a recorded track in monitor SP, then monitor MP, then VR SP and finally VR MP. 

Along this way you will be able to discard Rift-S related issues, or network related issues, etc, etc.

 

I compiled a list of things in the past to review when people were not getting same performance than expected:

https://forum.il2sturmovik.com/topic/34107-items-to-review-if-you-have-low-performance-in-il-2-vr-test/

 

[firdimigdi]

One thing that comes to mind is to do some tests for DPC latency on the "slow" rig.

[E69_Qpassa_VR]

I would test with another video game to discard the computer matter or to reassure. Total war warhammer 2 has a benchmark where you could test it and if the results are equals it would mean is because of the servers or some situational reason. If they are not equals I would format his computer, check the overclock and do tests again.

 

 

Maybe is because the processor is not being properly ventilated?

Edited May 25, 2020 by E69_Qpassa_VR

[=RS=haikcube]

I'm just spitballing here, you mentioned that you checked the ram timings on CPU-Z, was XMP mode switched on and the ram modules installed in the correct sockets to enable dual channel mode?

[Alonzo]

3 minutes ago, [=PzG=]-haikcube said:

I'm just spitballing here, you mentioned that you checked the ram timings on CPU-Z, was XMP mode switched on and the ram modules installed in the correct sockets to enable dual channel mode?

 

Good question, RAM looks ok to me:

 

[=RS=haikcube]

that row refresh cycle time looks awfully long.

Is that according to specs? I'm not very familiar with that however so i cant be of much help, but everything else looks good save for not knowing the Bank cycle time(tRC)

Edited May 25, 2020 by [=PzG=]-haikcube
add more stuff

[WheelwrightPL]

He can use PassMark Performance Test (I think it is free) and run each of its benchmarks for: CPU, Video Card, Hard Drive and RAM. It should at least narrow it down.

Additionally he can also run FurMark for the GPU and "Intel Extreme Tuning Utility" and Prime95 for the CPU.

[Gomoto]

You surly have checked the GPU driver settings.

[Alonzo]

4 hours ago, [=PzG=]-haikcube said:

that row refresh cycle time looks awfully long.

Is that according to specs? I'm not very familiar with that however so i cant be of much help, but everything else looks good save for not knowing the Bank cycle time(tRC)

 

It's actually slightly longer on my rig at ~600 or so. So I think it's okay.

 

3 hours ago, WheelwrightPL said:

He can use PassMark Performance Test (I think it is free) and run each of its benchmarks for: CPU, Video Card, Hard Drive and RAM. It should at least narrow it down.

Additionally he can also run FurMark for the GPU and "Intel Extreme Tuning Utility" and Prime95 for the CPU.

 

Good idea - we actually did that and got super similar CPU and memory ratings from PassMark. So again, the two rigs look very similar.

 

1 hour ago, Gomoto said:

You surly have checked the GPU driver settings.

 

We're using the same driver version, we checked that. What else? "Prefer Maximum Performance" is one, I'll make sure we check that.

 

Thanks for all the suggestions folks, really appreciated.

[Indianer]

I don't know about VR, but do you both have windows game mode disabled?

I just turned it off, because I forgot after a fresh windows installation. Now I can finally have (nearly)steady 144 fps. Also turned off threaded optimization for il2 in NCP and now its all good. Was wondering why a friend of mine had always better performance than me, with nearly the same PC specs. And he has an ultrawide monitor vs. my 16:9 1080p..

[Gomoto]

threaded optimization off improves performance on your rig? strange.

Edited May 26, 2020 by Gomoto

[Indianer]

55 minutes ago, Gomoto said:

threaded optimization off improves performance on your rig? strange.

Found it strange too. I tried something from a thread about microstuttering with trackIR in combination with ffb (even I don't have ffb stick, but microstuttering in opentrack) and it worked. Before, my fps were oscillating between 80-144 and headmovement  went from fluid to stuttery all the time. After, I got a more constant fluidity when looking around. Turned off windows game mode then, and now my framerate never drops below 130.

 

 

[Gomoto]

will do a testrun myself. thanks for the idea

[=RS=haikcube]

6 hours ago, Alonzo said:

Good idea - we actually did that and got super similar CPU and memory ratings from PassMark. So again, the two rigs look very similar.

Well,if the ratings are similar,the only logical differences would be the game and vr settings. Are both of you running the same quality settings in the startup.ini, both running in open composite with similar supersampling settings? If not,perhaps both the occulus software and steamvr is also adding it's own supersample values which might be causing the heavy blurring and double vision mentioned earlier.

Edited May 26, 2020 by [=PzG=]-haikcube
Edited for more clarification

[Bernard_IV]

I live in Nampa which may be local to him if you want me to take a look or you can swap around some parts with mine to see if anything is broken.  What CPU cooler?

Edited May 26, 2020 by Bernard_IV

[SPUDS_IdahoBookworm]

I thought I'd bring some closure to this thread. I upgraded from 2666mhz RAM to 3600 mhz RAM and that seems to have brought me to performance parity with Alonzo and others who talk about their VR experience on this forum. I no longer get ghosting and smearing with every dogfight. I still have jutters and hitches, but that seems to be normal and it doesn't ruin the experience. 

 

It's strange that Alzono still got measurably better performance than me when he downclocked his ram to match mine at 2666. But the RAM upgrade is the only thing I've changed since we ran those tests, so it has to be it. Maybe I got a bad set of the 2666, or, more likely, Alonzo's slightly better processor made the difference. Anyway, that final upgrade of getting 3600 mhz RAM seems to have pushed me over the top at last, after more than a year of frustration and far too much money spent.

 

So if anyone is looking for what to upgrade to try to get better VR performance, don't ignore the RAM!