VR overhead

[BeastyBaiter]

Given the Oculus Rift sale, I am considering picking one up but I'm concerned about performance since it and a new GPU are definitely out of the budget. Currently I am running a 2560x1440 monitor and am able to maintain 90+ fps most of the time at ultra detail in BoX. DCS also runs pretty well, though with a lot more variance (some areas are 40 fps, others 140, graphics options no effect). I know Oculus Rift runs slightly lower total resolution (2100x1200 or so) and so should theoretically have slightly higher performance. But that assumes no overhead. Does Oculus add any substantial overhead and if so, what kind of FPS cut can I expect?

 

 My current system is:

CPU: R5 1600x

GPU: RX480 8GB

RAM: 16GB DDR4 2667MHz

OS: Win 10

Mobo: MSI B350 Tomahawk

[macnkill]

I think you should be fine with your setup. You don't really need the graphics on ultra when in the rift anyway since the overall clarity isn't nearly as good as your 1440p monitor, so if you were to experience stuttering due to framerate drop just knock it down a couple notches in quality for shadows etc. Also I recommend doing what other people have suggested and turning of ASW/ATW with the oculus debug tool when playing IL2.

 

The rift is an amazing experience, not just for IL2, but for other games too that incorporate touch.

[coconut]

 

 

I know Oculus Rift runs slightly lower total resolution (2100x1200 or so) and so should theoretically have slightly higher performance

Not really, the rendering resolution is higher than the hardware resolution. For me it's 1600x1904 for each eye. Total number of pixels will then be 3200x1904, somewhat close to 4K. Then there's also about twice the work to do for the processor, since there's two views to render.

[Urra]

Given the Oculus Rift sale, I am considering picking one up but I'm concerned about performance since it and a new GPU are definitely out of the budget. Currently I am running a 2560x1440 monitor and am able to maintain 90+ fps most of the time at ultra detail in BoX. DCS also runs pretty well, though with a lot more variance (some areas are 40 fps, others 140, graphics options no effect). I know Oculus Rift runs slightly lower total resolution (2100x1200 or so) and so should theoretically have slightly higher performance. But that assumes no overhead. Does Oculus add any substantial overhead and if so, what kind of FPS cut can I expect?

 

My current system is:

CPU: R5 1600x

GPU: RX480 8GB

RAM: 16GB DDR4 2667MHz

OS: Win 10

Mobo: MSI B350 Tomahawk

I would wait one more month. As Lenovo is going to release their new headset in August and has better resolution. There should be some comparisons online pretty soon afterwards. I have rift and for general flying its great, but to id a target gets frustrating after a while. They are lowering the price because they know that newer headsets are going to release right around the corner. That's my experience.

[javelina]

you could run the Oculus compatibility checker.  That should give you the system feedback you're looking for.

[BeastyBaiter]

I ran that and exceeded the recommended specs of course, but that isn't what I'm asking about. I know it will work, the question is how well compared to 1440p? Coconut, you're on track of what I'm asking about. Can you please elaborate a little? I assume you have a regular monitor you can compare it to and also, how does OR render to nearly twice the device's actual resolution? Is that just an up scaling feature you are using as an alternative to AA or is it always like that?

Edited July 11, 2017 by BeastyBaiter

[coconut]

By default VR apps render above the resolution of the hardware, I've read that it's to minimize artifacts when applying transformations to compensate for lens distortion. I've also seen explanations that mention that it helps convey the impression of movement even for very small sub pixel turns of the head. Then there's additional resolution you can add using supersampling, it helps with reading small text and provides antialiasing.

 

I have a 1440p monitor, typically rendering at 100fps or more. That's not enough for 90 fps at all times, but I still find it smooth enough, even with ATW disabled.

[A_radek]

If your willing to spend some time reading and testing, this is a good thread:

 

https://forum.il2sturmovik.com/topic/29322-measuring-rig-performance-common-baseline/page-1

 

There is a list of benchmark results done on various hardware. There are also "how to" benchmarking instructions and a link to the balapan track used for these tests.

 

As many VR users have benchmarked both their 2d and VR performance. You should be able to make a good VR performance estimation by testing your system with the same track.

 

Here's my assumtion I will take no responsibility for: VR is not comparable to 1440 or 4k resolution on a regular monitor - in terms of hardware requirements. In VR you are rendering the game twice, once for each eye, and from two different perspectives. From the benchmarks mentioned above a 1070 Gpu will do just fine for VR while very good cpu's are struggling to keep those 90fps. On a high resolution monitor I assume it would be an as fast a possible Gpu you want. While the load on the CPU would lessen.

[chiliwili69]

As a_radek said you can test the balapan record in FullHD resolution and with the same Balapan settings and then figure out what you should expect in VR.

I am curious to know how well the Ryzen performs in monitor vs intel

[BeastyBaiter]

I ran the test but will wait to post results in that thread until I have my Rift to compare to. I will share the 1440p results here though:

 

AVG: 113.7

Max: 151

Min: 70 (occurred when passing over heavily smoking bomber near the end)

 

Graphs:

 

 

 

 

Settings:

Balapan standard at 1440p.

 

Of interesting note is the CPU usage. BoS does appear to use 4 threads, but nearly all of it is piled onto one. Despite the high CPU usage on logical core 0 on the particular run pictured, it appears to be a GPU bottleneck as the dips in GPU usage do not coincide with FPS hits. Incidentally, I ran this test several times and in most cases the CPU usage was a bit lower as it was on something other than CPU 0 (second CCX was more common, cores 6-11). No significant performance difference occurred though, so it isn't important.

 

 

System specs:

CPU: R5 1600x at 4.0GHz @ 1.375v (set in Ryzen Master, down clocking still allowed for efficiency)

GPU: RX 480 8GB @ 1370MHz core and 2025MHz VRAM

RAM: 16GB TridentZ @ 2667MHz, 1.35v, timings 16-18-18-36-18

Mobo: MSI B350 Tomahawk

OS: Win 10 Home

[chiliwili69]

That´s a good avg number for being at 1440p!!

I think that your CPU will run quite well for VR. But you can run the test at 1080p if you want to compare to people who run the test with 7700K OCed.

In my case, the test at 1080p didn´t full loaded the 1070.

 

I also noticed that running exactly everything with the same parameters I can get +-1 fps diference, but I was not checking the load of the cores.

 

Take note that most of the people in VR use supersampling to get more detail of the render. Increasing the SS will load your GPU considerably.

 

In my case, with SS of SteamVR at 2.0 the average fps drops to 45!!  (with ASW off). But the Balapan track is complex and uses the HUD with icons (something that I normally don´t use).

So in my "normal" flight I am in the 60 to 90 range depending of the complexity. I prefer to loose some fps (above 60 is quite OK) but have better detail.

[BeastyBaiter]

I get a fair bit of variance on it. I just reran the test with BoS manually assigned to cores 6-11, got:

AVG: 106

Max: 141

Min: 71

 

That's inline with the other 4 tests run, so changing affinity doesn't seem to do anything despite the fact that I'm now downloading Elite Dangerous in the background (didn't do that for first 3 tests. Incidentally, here's the resource usage. Note that it seems to be jumping around cores 6-11 when the affinity is manually set, but it didn't harm performance any.

 

 

Edit: CPU  0, 2, 3 and 4 are doing non-BoS stuff in that screenshot. Yay for lots of cores. 

Edited July 14, 2017 by BeastyBaiter

[SCG_Fenris_Wolf]

As a_radeck said already, comparing 2d performance with VR does not work. It has different requirement in CPU and gpu, which scale differently. Your performance tests are thus irrelevant for estimating anything regarding VR - and by the way, should not be posted in the VR section of the forums, but in the general hardware section.

[chiliwili69]

As a_radeck said already, comparing 2d performance with VR does not work. It has different requirement in CPU and gpu, which scale differently. Your performance tests are thus irrelevant for estimating anything regarding VR - and by the way, should not be posted in the VR section of the forums, but in the general hardware section

 

Well, this is not totally true according to the tests.

We clearly have seen that the people who obtain an average of 167-175 fps in 2D fullHD, then are achieving an average of 84-87 fps in VR.

Although it is not a perfect linear correlation since there are other influences depending on graphic card and RAM.

In fact, if more people try the test and report their results in 2D(fullHD) and VR we will have a broader base to compare rigs and what we should expect in VR based in the performance obtained in 2D. But again, there will be other factor that could influence the final performance in VR.

Edited July 14, 2017 by chiliwili69

[dburne]

I found in the sims I run in VR, after going to the Rift my framerates approximately halved.

[SCG_Fenris_Wolf]

Well, this is not totally true according to the tests.

We clearly have seen that the people who obtain an average of 167-175 fps in 2D fullHD, then are achieving an average of 84-87 fps in VR.

Although it is not a perfect linear correlation since there are other influences depending on graphic card and RAM.

In fact, if more people try the test and report their results in 2D(fullHD) and VR we will have a broader base to compare rigs and what we should expect in VR based in the performance obtained in 2D. But again, there will be other factor that could influence the final performance in VR.

No you haven't. The amount of data gathered is not nearly sufficient to even open any comparison, not even by statistically relevant estimation.

 

 

That said, it's on a good way though. You however must scrap all data of people only testing either or, and only go with those who provide both. Exactly as you mentioned. I've seen the thread.

Edited July 15, 2017 by 1CAG_Fenris_Wolf

[BeastyBaiter]

Agreed, just one or the other doesn't really say much. I'll add 1080p, 1440p and VR stats to the main test thread once I have the OR to test. I was hoping this thread would help shed some light on what I could expect but it doesn't look like we have enough data yet.