GPU upgrade

[doog442]

Hi folks

 

I'm currently running an i5 9600k @4.8 , 16mb DDR3, 1070 GPU on a Rift s.

 

I appreciate its very early days in relation to the 3080, however do you think I would see a suitable graphics upgrade from my 1070 to a 3080 (or even a 3070 when its released) ?

 

I wouldn't be expecting a night and day improvement but would look at it as an incremental upgrade to get the best out of the Rift S before considering perhaps a new headset next year. 

 

cheers 

[messsucher]

Not sure how much 3080 would help with that CPU. But in my opinion even 10 FPS more is good gain in VR mode. Maybe you could get 20 fps more? Just guessing here, but I would do the upgrade.

[dburne]

1 hour ago, doog442 said:

Hi folks

 

I'm currently running an i5 9600k @4.8 , 16mb DDR3, 1070 GPU on a Rift s.

 

I appreciate its very early days in relation to the 3080, however do you think I would see a suitable graphics upgrade from my 1070 to a 3080 (or even a 3070 when its released) ?

 

I wouldn't be expecting a night and day improvement but would look at it as an incremental upgrade to get the best out of the Rift S before considering perhaps a new headset next year. 

 

cheers 

 

 

Yes.

[chiliwili69]

With the 3070 or 3080 you will be able to run the game a higher settings without decreasing your actual performance. For example clouds at high, more Supersampling, more MSAA.

 

For other CPU related items you will not see a performance gain. (shadows, mirror, presets, dense plane scenarios, etc).

 

BTW, your RAM is DDR4 not DDR3. And RAM speed is also important for VR performance.

[doog442]

2 hours ago, messsucher said:

Not sure how much 3080 would help with that CPU. But in my opinion even 10 FPS more is good gain in VR mode. Maybe you could get 20 fps more? Just guessing here, but I would do the upgrade.

 

I guess the CPU was my worry as a possible bottleneck . That said I read some posts recently that pointed towards it having adequate single core performance but how good I'm not sure. I've been looking at a 10700k but that would also mean a new motherboard.  

 

1 hour ago, chiliwili69 said:

With the 3070 or 3080 you will be able to run the game a higher settings without decreasing your actual performance. For example clouds at high, more Supersampling, more MSAA.

 

For other CPU related items you will not see a performance gain. (shadows, mirror, presets, dense plane scenarios, etc).

 

BTW, your RAM is DDR4 not DDR3. And RAM speed is also important for VR performance.

 

Cheers mate, of course yes the ram is DDR4. Any thoughts on the I5 9600k ? 

[doog442]

3070 narrowly beating 2080ti in first performance results. I see how these pan out in real world testing before pulling the trigger. Seems a cost effective solution. 

 

https://www.tomshardware.com/uk/news/nvidias-geforce-rtx-3070-beats-2080-ti-in-first-performance-results

Edited October 5, 2020 by doog442

[HunDread]

16 hours ago, doog442 said:

 

I guess the CPU was my worry as a possible bottleneck . That said I read some posts recently that pointed towards it having adequate single core performance but how good I'm not sure. I've been looking at a 10700k but that would also mean a new motherboard.  

 

 

Cheers mate, of course yes the ram is DDR4. Any thoughts on the I5 9600k ? 

 

Check the below reply coming from one of the experts here. You should not be worrying about your CPU.

 

 

[SCG_Wulfe]

Yes absolutely you will see an increase going to the 3080. 

 

I currently have an Odyssey+

 

I also am running a 9600k. (Mine is at 5.0ghz)

I just upgraded from a 1080 to the 3080. 

 

Before the upgrade I had to run 124% Steam Super Sampling, FXAA, clouds on medium, shadows low and Mirrors low.... with these settings and my 1080 I would get roughly 90FPS unless I was on a server with heavy clouds or congested areas. In those circumstances, I would average around 70FPS. 

 

I now am running 150% Steam Super Sampling, 4X MSAA, Clouds on high, shadows on ultra and mirrors medium. With these settings and my 3080 I get roughly 90FPS even with heavy clouds. If I am in a congested area, my CPU becomes the bottleneck and FPS drops to 80+FPS

[E69_Qpassa_VR]

3 hours ago, SCG_Wulfe said:

Yes absolutely you will see an increase going to the 3080. 

 

I currently have an Odyssey+

 

I also am running a 9600k. (Mine is at 5.0ghz)

I just upgraded from a 1080 to the 3080. 

 

Before the upgrade I had to run 124% Steam Super Sampling, FXAA, clouds on medium, shadows low and Mirrors low.... with these settings and my 1080 I would get roughly 90FPS unless I was on a server with heavy clouds or congested areas. In those circumstances, I would average around 70FPS. 

 

I now am running 150% Steam Super Sampling, 4X MSAA, Clouds on high, shadows on ultra and mirrors medium. With these settings and my 3080 I get roughly 90FPS even with heavy clouds. If I am in a congested area, my CPU becomes the bottleneck and FPS drops to 80+FPS

Let's see your benchmarks with the G2 ?

[Charlo-VRde]

11 hours ago, SCG_Wulfe said:

Yes absolutely you will see an increase going to the 3080. 

 

I currently have an Odyssey+

 

I also am running a 9600k. (Mine is at 5.0ghz)

I just upgraded from a 1080 to the 3080. 

 

Before the upgrade I had to run 124% Steam Super Sampling, FXAA, clouds on medium, shadows low and Mirrors low.... with these settings and my 1080 I would get roughly 90FPS unless I was on a server with heavy clouds or congested areas. In those circumstances, I would average around 70FPS. 

 

I now am running 150% Steam Super Sampling, 4X MSAA, Clouds on high, shadows on ultra and mirrors medium. With these settings and my 3080 I get roughly 90FPS even with heavy clouds. If I am in a congested area, my CPU becomes the bottleneck and FPS drops to 80+FPS

 

If you fly multiplayer, how much of your 3080's VRAM are you using in VR?

[Alonzo]

On 10/4/2020 at 9:11 AM, doog442 said:

I'm currently running an i5 9600k @4.8 , 16mb DDR3, 1070 GPU on a Rift s.

 

I appreciate its very early days in relation to the 3080, however do you think I would see a suitable graphics upgrade from my 1070 to a 3080 (or even a 3070 when its released) ?

 

Are you sure it's DDR3 ? It's probably DDR4. What speed is it?

 

The 9600K is a very competent CPU. I'm not sure what the other guy is smoking. You should try to get it to 5.0ghz if you can, as well as get as much speed you can out of the memory. You will see a giant benefit going from the 1070 to a 3080, and a pretty big improvement by going to the 3070. The 3070 is a very good pairing for the Rift S. Given the poor availability of NVidia cards you could also wait until the AMD announcements on October 28th. Rumor is they will have competitive cards with more memory than the NVidia cards, and slightly cheaper. With a Rift S you don't need the absolute maximum performance, so AMD should be a consideration (if the rumors are true).

[NiiranenVR]

Yes , the 9600k is not a bad CPU - I have mine running around the 5.2 GHz 

I have water cooling installed ...and OC 

[doog442]

52 minutes ago, Alonzo said:

 

Are you sure it's DDR3 ? It's probably DDR4. What speed is it?

 

The 9600K is a very competent CPU. I'm not sure what the other guy is smoking. You should try to get it to 5.0ghz if you can, as well as get as much speed you can out of the memory. You will see a giant benefit going from the 1070 to a 3080, and a pretty big improvement by going to the 3070. The 3070 is a very good pairing for the Rift S. Given the poor availability of NVidia cards you could also wait until the AMD announcements on October 28th. Rumor is they will have competitive cards with more memory than the NVidia cards, and slightly cheaper. With a Rift S you don't need the absolute maximum performance, so AMD should be a consideration (if the rumors are true).

 

Yes my bad DDR 4 3200. I'll probably push the CPU a little further, I'm getting decent temps using air at 4.8ghz and a mahoosive Noctua fan. 

 

Thanks for the info on the AMD cards, I guess its a waiting game. I'd certainly like to upgrade to try and get the best out of the Rift S , nice to know the CPU is ok at this level. 

[SCG_Wulfe]

On 10/5/2020 at 11:38 PM, Charlo-VR said:

 

If you fly multiplayer, how much of your 3080's VRAM are you using in VR?

I will take a look next time I get a chance to fly. Unfortunately I've been so busy the last few weeks since I got the 3080, I haven't had time to use it much. 

[SCG_Fenris_Wolf]

@Charlo-VR  I have the same GPU as Wulfe, and an 8KX with 2x 4K screens. 

 

While playing IL-2, I had roughly 6GB VRAM load. More doesn't seem to be needed, even with these ultra-high resolutions.

 

 

P.S. Disregard any benchmarking done in Half Life Alyx - if you see anyone using Half Life Alyx Benchmarks, close the video of the Benchmarker: They are incompetent.

Reason: Half Life Alyx scales both sampling and details such as distant shadows, AO, textures, and more, as well as VRAM usage automatically up or down in the background to maintain framerate. It's one of the finest, best optimized games due to this - as it's hardly noticable. It is the reason you have seen 11.9GB used in HLA. It's actually just pre-loading beyond what it requires, highly dynamic and well-made engine. You cannot fix that without modifying the exec extensively. As thus, it falls flat regarding benchmarks though.

[Charlo-VRde]

4 minutes ago, SCG_Fenris_Wolf said:

@Charlo-VR  I have the same GPU as Wulfe, and an 8KX with 2x 4K screens. 

 

While playing IL-2, I had roughly 6GB VRAM load. More doesn't seem to be needed, even with these ultra-high resolutions.

 

You and Wulfe and Chili are three of the folk I read most closely for hardware advice.  Last week I upgraded to an i7-10700K OC @ 5.1 GHz on an MSI MPG Z490 Gaming Carbon WIFI motherboard, and 32 GB's of G.Skill Trident Z Royal DDR4-4000 RAM running at 4,000 MHz.

 

Now in my own typical usage of IL-2 in CombatBox running 4x MSAA on my Quest, 1.1 SS in OTT, ASW off, 4K textures, 3840 x 2160 full screen, GPU-Z tells me I am pretty much using all of my aging Titan X’s 11GB’s of VRAM.

 

What would you recommend I use to test that more precisely than GPU-Z?

[SCG_Fenris_Wolf]

That is a somewhat uncommon high usage of VRAM in IL-2. 

 

Thank you for mentioning this, I must add though: I am not familiar with the Quest and its link, and I only like to give advice if I am certain of what I am saying. In this case, I can hardly help you with the linked Quest - or do you use VD to cast it? 

 

But I wonder, 3840 x 2160 full screen on the Desktop while sending another to the Quest might increase the demand on VRAM considerably. MSAA certainly boosts it to some degree, but here you apply it to both the full screen Desktop mirror as well as to the two Virtual Reality scenes. That's exponential. That's like pulling at your horse into all directions, from multiple sides, while beating it to run faster. I suggest you disable MSAA alltogether and use VR's own SuperSampling instead, while lowering Desktop Window resolution, and then check again after a clean start, without any other App having run before (including Steam Home, Oculus Home, or similar). This way you apply the SuperSampling just for the 2 VR scenes. Also, @dburne is quite competent in all things Oculus, using OpenComposite (if possible) should be necessary. 

[Varibraun]

50 minutes ago, Charlo-VR said:

You and Wulfe and Chili are three of the folk I read most closely for hardware advice. 

 

I just want to echo this and send my continued thanks to @SCG_Fenris_Wolf, @chiliwili69, @SCG_Wulfe, @dburne (and many others) for all of the time they spend testing, and then unselfishly spending more time posting here.  It is a godsend to help those who may be like me and simply apply a "monkey see, monkey do" approach to get acceptable hardware/software performance results.  I freely admit to basically plagiarizing Don's whole system with the exception of my current Reverb G1 (Glad he is upgrading to G2 so I can eventually copy his config file too!)

 

Thank you all! 

Edited October 8, 2020 by Varibraun

[dburne]

2 minutes ago, Varibraun said:

 

I just want to echo this and send my continued thanks to @SCG_Fenris_Wolf, @chiliwili69, @SCG_Wulfe, @dburne (and many others) for all of the time they spend testing, and then unselfishly spending more time posting here.  It is a godsend to help those who may be like me and simply apply a "monkey see, monkey do" approach to get acceptable hardware/software performance results.  I freely admit to basically plagiarizing Don's whole system with the exception of my current Reverb G1 (Glad he is upgrading to G2 so I can eventually copy his config file too!)

 

Thank you all! 

 

 

Thanks for the kind words.

[Charlo-VRde]

3 hours ago, SCG_Fenris_Wolf said:

That is a somewhat uncommon high usage of VRAM in IL-2. 

 

Thank you for mentioning this, I must add though: I am not familiar with the Quest and its link, and I only like to give advice if I am certain of what I am saying. In this case, I can hardly help you with the linked Quest - or do you use VD to cast it? 

 

But I wonder, 3840 x 2160 full screen on the Desktop while sending another to the Quest might increase the demand on VRAM considerably. MSAA certainly boosts it to some degree, but here you apply it to both the full screen Desktop mirror as well as to the two Virtual Reality scenes. That's exponential. That's like pulling at your horse into all directions, from multiple sides, while beating it to run faster. I suggest you disable MSAA alltogether and use VR's own SuperSampling instead, while lowering Desktop Window resolution, and then check again after a clean start, without any other App having run before (including Steam Home, Oculus Home, or similar). This way you apply the SuperSampling just for the 2 VR scenes. Also, @dburne is quite competent in all things Oculus, using OpenComposite (if possible) should be necessary. 

 

Thanks  for that advice - I thought I was seeing unusually high VRAM usage, and based on Chili's recommendations I suspected running my monitor at 3840 x 2160 together with MSAA was really taxing beyond what I need.

 

I use OpenComposite and OTT based on advice I picked up from @dburne, and I'll experiment with all you suggested above - though the Oculus app appears to be a requirement for Oculus Link. I only run the game at such a high monitor resolution for the comfort of selecting game choices and startup sequences with the Quest off my head until I'm ready to start taxiing.

 

And of course, whenever my Reverb G2 on pre-order finally arrives on my doorstep will start some experimentation all over again

[SCG_Wulfe]

Thanks for the kind words guys, but I certainly have not spent nearly the time nor effort of the other VR gurus here. 

 

That said. I will be flying tonight and plan to check my VRAM usage. 

 

In terms of the MSAA or not question, I did actually test this out last week.

 

My base settings that give what I felt were the best graphics/performance with the 3080 on an Odyssey+ are the following...(with these I get a solid 90FPS with heavy cloud cover except in very busy areas or over big towns):

 

Steam SS: 150% 

 

Preset: High

Shadows: Ultra

Mirrors: Medium

Distant Landscape detail: x3

Horizon Draw Distance: 130km

Landscape Filter: Blurry

Grass Quality: Off

Clouds Quality: High

Dynamic Resolution: Full

MSAA: 4X

Gamma: 0.8

SSAO: Off

HDR: On

Sharpen: On

Use 4k textures: On

 

 

 

I then tried the same settings but turned off MSAA (MSAAx0) and cranked my Steam SS to 200% in an attempt to achieve a similar affect to the SS/MSAA combo I was previously running. The edge definition was similar but I did also wind up with more flickering roads etc. than with MSAA. *The big problem was that FPS dropped to approximately 65 average with high clouds* 

 

I then tried lowering to 180% Steam SS and 2XMSAA as a combo. With this I almost had the same performance as my baseline settings... though I felt like it would drop from 90FPS a touch more. That said, to my eyes, it didn't look quite as good as my base settings either. 

 

 

In conclusion,  I think the baseline settings work best with a non 4k VR headset and a 3080. However, I may find that once I receive my Reverb G2, The game will look best run at native resolution with either no AA or maybe even mild FXAA or something. 

 

I do think what Fenris said about lowering the display resolution of the game on your monitor is a wise idea. Displaying at 4K on your screen can't be helping performance. 

 

 

 

 

 

 

Edited October 9, 2020 by SCG_Wulfe

[HunDread]

8 minutes ago, SCG_Wulfe said:

My base settings that give what I felt were the best graphics/performance with the 3080 on an Odyssey+ are the following...(with these I get a solid 90FPS with heavy cloud cover except in very busy areas or over big towns):

 

Steam SS: 150% 

 

Shadows: Ultra

Mirrors: Medium

Distant Landscape detail: x3

Horizon Draw Distance: 130km

Landscape Filter: Blurry

Grass Quality: Off

Clouds Quality: High

Dynamic Resolution: Full

MSAA: 4X

Gamma: 0.8

SSAO: Off

HDR: On

Sharpen: On

Use 4k textures: On

 

Which preset are you using?

 

9 minutes ago, SCG_Wulfe said:

In conclusion,  I think the baseline settings work best with a non 4k VR headset and a 3080. However, I may find that once I receive my Reverb G2, The game will look best run at native resolution with either no AA or maybe even mild FXAA or something

 

Isn't the O+ with 150% SS runs roughly the same amount of pixels than the G2?

[SCG_Wulfe]

I am using the high preset. 

 

Yes it's approximately the same, though rendering at the native resolution should look far superior to SS. 

[SCG_Fenris_Wolf]

3 hours ago, SCG_Wulfe said:

I am using the high preset. 

 

Yes it's approximately the same, though rendering at the native resolution should look far superior to SS. 

Yes very true. Want to add, even Reverb needs to be a bit "above" native resolution, for distortion profiles need more, and we'll want headroom to soften edges using SS/MSAA

[HunDread]

4 hours ago, SCG_Wulfe said:

I am using the high preset. 

 

Yes it's approximately the same, though rendering at the native resolution should look far superior to SS. 

 

Oh yes, I meant you should probably have no issues with the G2 with your current settings. 

[SCG_Wulfe]

8 minutes ago, HunDread said:

 

Oh yes, I meant you should probably have no issues with the G2 with your current settings. 

 

Ah, gotcha. Ya. Hopefully.  though if I can get away with less AA once the resolution is truly higher, It would be nice to reclaim some FPS headroom. 

[SCG_Fenris_Wolf]

I have received both Pimax 8KX and RTX 3080, and am into testing performance. I won't review the 8KX without a metric against it yet (I will wait for the Reverb G2 to make an assessment), but I can already say that the 75Hz ultra-low persistence panels look like 90Hz and allow me to enjoy IL-2 on Ultra Preset (clouds: high) for the first time, without framerate drops or reprojection.

 

The magic number the engine can stomach is somewhere at 75 to 80Hz I would assume now, at least for my i7 7700K running at 4.8GHz without AVX offset. The frametime spikes of old have also vanished.

 

I use a diagonal FOV of 170° (normal), and 3080 serves 2x3156x2604 at 75Hz without complaining. The image is sharp, like on the widescreen monitor.

[chiliwili69]

On 10/8/2020 at 10:08 PM, Varibraun said:

It is a godsend to help those who may be like me and simply apply a "monkey see, monkey do" approach to get acceptable hardware/software performance results.

 

I am also a monkey following that approach with the benchmarks. Perhaps we go one step beyond with the Greif correlation and try to determine with facts what is the key hardware to maximize performance and the performance cost of every graphic setting.

 

The pitty is that it is difficult to maintain a common test over the IL-2 version. Most of the game have their own benchmarks and having an always playable flight record will help a lot to all community here when looking to upgrade or tweak the settings. But @Jason_Williams or dev team @Han never say a single word about that. It would be welcome some words from them in that sense, specially for the always growing IL-2 VR community. Specially when they were requesting help to detect stutters in SP and MP, and they wanted a more scientific method. It would take very little effort from their side to create an always playable track record of 2-3 minutes.

[NiiranenVR]

But .....what a World/ what to buy ...... ( Maybe/guess I will follow the little crowd ?) 

There diff prices to the 3080 .... There : extreme/gaming / OC  / so on ... 

What are you going for when buying ??

[NiiranenVR]

Rumor ?? 

I have just read ( Danish page ) that 3080 will come with 20GB vRam in December 

https://tweak.dk/grafikkort/rygte:-geforce-rtx-3080-20-gb-frigives-i-december

Edited October 10, 2020 by Niiranen-VR

[SCG_Fenris_Wolf]

Rumors, yes heard of them. The games released or in development use between 4-6, rarely 8GB of VRAM. I don't see how it is feasible to wait for one and pay even 200€ more for no performance benefit.

[messsucher]

5 hours ago, Niiranen-VR said:

But .....what a World/ what to buy ...... ( Maybe/guess I will follow the little crowd ?) 

There diff prices to the 3080 .... There : extreme/gaming / OC  / so on ... 

What are you going for when buying ??

 

I buy a card which perform well in noise levels. Counter intuively enough low noise level cards tend to still have better cooling and better performance, and you don't have a superjet hoover screaming and annoying you.

Edited October 10, 2020 by messsucher

[SCG_Wulfe]

Ok, I just had a chance to check on Combat Box with heavy cloud which I find to be the most taxing environment. VRAM usage hit about 7000 MB peak. 

 

*edit, never mind, after running IL2 significantly longer and jumping to other servers... I managed to reach a peak of 8000*

Edited October 10, 2020 by SCG_Wulfe

[SCG_Fenris_Wolf]

That is just allocation. The VRAM will also fill up over time with unused stuff as well, which does not matter. There is no shortage, so don't worry..

Edited October 10, 2020 by SCG_Fenris_Wolf

[airsheep_VR]

On 10/4/2020 at 7:38 PM, chiliwili69 said:

For other CPU related items you will not see a performance gain. (shadows, mirror, presets, dense plane scenarios, etc).

 

Are you sure the mentioned settings are cpu related? Does this  mean if I changed them to higher it affects my cpu usage mostly?

Just because I have Reverb, a 9700k~5ghz,  1080ti, and to get  60 fps I must run the game on balanced setting and almost everithying set to low or off.

 

preset balanced

shadow medium

mirrors off

distant landscape 2x

horizon draw 70km

grass off

cloud high

aa off

ssao off

hdr off

 

According to FPSVR my cpu usage is ~ 30% and my gpu is 100%. 

When I change one of those options to higher (shadows, mirror, presets, etc) my fps drops immediately but my cpu stands almost the same.

Fpr example now mirrors off ->mirrors simple = 10fps drop but cpu usage is still the same. 

It seems to me my gpu does not have the power to handle the gaphics while my cpu sleeps. 

 

 

[chiliwili69]

On 10/11/2020 at 10:07 AM, airsheep_VR said:

According to FPSVR my cpu usage is ~ 30% and my gpu is 100%

 

We should not take into account the % CPU usage (or even %core usage) reported by any monitoring application (including fpsVR, MSI afterburner, etc) since that calculation is misleading.

 

IL-2 VR has many threads running accross the cores of the CPU, they are jumping from core to core (this is a Window thing). The CPU% is just measuring the average utilization time of all %core utilization. And the %core usage is just measuring the utilization time  (working time/total time) of the core.

 

The "problem" with IL-2 VR is that from all those threads, there is one which is the heavy one and it is bottlenecked by the core during the fraction of the time it is running in that core.

 

So forget that %CPU number. Instead you can see the CPU frametimes repported by fpsVR while you are playing.

 

Here I was showing how every graphic option affects the CPU in a not GPU constrained system. It was done before referred rendering was implemented. So perhaps numbers are different now since we saw that referred rendering was unloading CPU and loading GPU. But still we are CPU bounded in many cases (dense secenarios).

 

You can just check that by putting a very low value in SteamVR SS, for example 50%. So GPU load will be low but you will not reach full fps because CPU is limiting the system.