Performance Review: GTX1070 FE + Vive

[Cmdr Menmnu]

TL: DR

GTX1070 FE runs ED 2.1 with the Vive in VR High comfortably at 90 FPS and has considerable headroom for improving the graphics in several areas. It won't allow Super-sampling >1.0 without dropping to 45 FPS. Mild over-clocking allows SSx1.5 at 90 FPS but there isn't much headroom left for anything else. SSx2.0 is still well out of reach even with mild over-clocking. Overall, performance seems similar to, or maybe slightly better than, a 980 ti. Suggests that a GTX 1080 FE should be able to run SSx1.5 at 90FPS out of the box and SSx2.0 with a mild over-clock.

Detailed post:

I spent a little while last night playing around with the graphics settings in ED 2.1 to see what the GTX1070 FE was capable of and this is a summary of the results. This was prompted by finding the text difficult to read well when I just dropped into it the night before and reading around this forum, reddit and (in particular) this wiki https://www.reddit.com/r/Vive/wiki/elitedangerous.

After a little work understanding it the Frame Timing panel accessible from Steam VRs 'performance' tab I found it really useful to see what was going on, so that's where the graphs below are from. The graphs show a 2000-frame window (x-axis) with the Frame Time (FT) on the y-axis (in ms). For 90 FPS the GPU and CPU have 11.1ms to compute what they need before the frame gets delayed. The top panel shows the CPU usage within each frame and the bottom panel shows the same for the GPU. The graphs have several colours. The ones the pay attention too are 1) dark purple - this is the amount of time used by the application (in our case ED) to calculate things on the CPU/GPU for each frame, 2) Green - this is the amount of time used by the compositor application which is the interface between the Vive and ED (this doesn't really change), and 3) pale yellow - this is the amount of the frame time that the GPU/CPU is idle for - essentially this shows how much overhead you have left to play with in each frame before the frames start getting dropped. If the purple+green segments are <11.1ms then the yellow segment will end at 11.1ms as each frame is drawn. If its >11.1ms then the FT doubles because the image is only updated every two frames.

For this initial testing I used the Elite Dangerous Lobby scene since this is stable, reproducible for others, and not subject to the vagaries of how many other ships/players there are around. For this scene with EDs 3D settings set to 'VR High' the GTX1070 FE Frame Time is solidly 11.1ms (confirmed by the displayed FR in the ED window of 90 FPS), with ~5.5ms taken up by the application and ~0.5ms taken by the compositor and ~5.1ms idle time. This was conformed by MSI afterburners showing ~50 GPU utilization.

What I did then was turn on, or increase, each of the available graphics settings under 'quality' and see what effect it had on the Frame Time usage. These results for the settings that made a significant difference to the performance in this scene were:


Ambient occlusion to High:

Total FT: 11.1ms -> 11.1ms
Application FT: 5.5ms -> 7.5ms
Idle FT: 5.1ms -> 3.1ms

https://drive.google.com/file/d/0B_5_Bo8SHplgcFV4SU1Ub0Vma3M/view?usp=sharing

Bloom to High:

Total FT: 11.1ms -> 11.1ms
Application FT: 5.5ms -> 6ms
Idle FT: 5.1ms -> 4.6ms

https://drive.google.com/file/d/0B_5_Bo8SHplgOG9uWTgxcDZnTms/view?usp=sharing

Shadows to Ultra:

Total FT: 11.1ms -> 11.1ms
Application FT: 5.5ms -> 6.25ms
Idle FT: 5.1ms -> 4.35ms

https://drive.google.com/file/d/0B_5_Bo8SHplgNkFsSndUUEdTMkE/view?usp=sharing

Anti-aliasing to FXAA:

Total FT: 11.1ms -> 11.1ms
Application FT: 5.5ms -> 6.25ms
Idle FT: 5.1ms -> 4.35ms

https://drive.google.com/file/d/0B_5_Bo8SHplgVWl1RlN6VklnMlE/view?usp=sharing

Anti-aliasing to SMAA:

Total FT: 11.1ms -> 11.1ms
Application FT: 5.5ms -> 6.5ms
Idle FT: 5.1ms -> 4.1ms

https://drive.google.com/file/d/0B_5_Bo8SHplgVnZLQjRNZS1zR2s/view?usp=sharing

Anti-aliasing to MLAAX4: [for some reason that is now unclear to me, last night I decided to go from SMAA to MLAAX4 for the graph below. The figures quoted are going from 'VR High' to MLAAX4 though]

Total FT: 11.1ms -> 11.1ms
Application FT: 5.5ms -> 6.7ms
Idle FT: 5.1ms -> 3.9ms

https://drive.google.com/file/d/0B_5_Bo8SHplgc3QxWmlDR0FfNm8/view?usp=sharing

Super-sampling to x1.5:

Total FT: 11.1ms -> 22.2ms (confirmed by 45 FPS FR in the ED window)
Application FT: 5.5ms -> 10.5ms
Idle FT: 5.1ms -> 11ms

https://drive.google.com/file/d/0B_5_Bo8SHplgNzlSdHJ0bUpEX28/view?usp=sharing

Super-sampling to x2.0:

Total FT: 11.1ms -> 22.2ms (confirmed by 45 FPS FR in the ED window)
Application FT: 5.5ms -> >14msms
Idle FT: 5.1ms -> <8ms

https://drive.google.com/file/d/0B_5_Bo8SHplgNzlSdHJ0bUpEX28/view?usp=sharing


Conclusion:
So, the long and show of this is that the out-of-the-box stock GTX1070 FE comfortably runs ED in VR High with plenty of headroom for improving several of the VR High graphics options. From what I read elsewhere here this seems to be (as you might expect) similar performance to a 980ti but it would be interesting to see some FT graphs of the same scene from 980ti users to compare. It's also worth noting that several options that may make a difference during actual game-play made no difference here, for example 'FX quality' (I'm planning on testing this later).


What the sock GTX1070 won't do is run ED in VR High with SS>1.0 without dropping below the 90 FPS target. SSx1.5 is right on the margin though so I thought I'd try a conservative over-clocking of the card and see if that improved things sufficiently to run either SSx1.5 or SSx2.0 at 90 FPS.

Over-clocking:

The OC settings (+225 GPU clock and +233 memory clock) are slightly conservative takes on the OC reported from here:

http://www.hardocp.com/article/2016...ders_edition_overclocking_review#.V2P38_iw6u8

This overclocking reduces the stock ED VR high FT from 5.5ms to ~4.3ms which is approximately a 25% improvement (not bad!) which would put the OC GTX1070 close to (but still below) a stock GTX1080 if the published benchmarks and these numbers are to be believed. Below are the resutls for SSx1.5 and SSx2.0 for the OC GTX1070:

Super-sampling to x1.5:

Total FT: 11.1ms -> 11.1ms
Application FT: 4.3ms -> 8.5ms
Idle FT: 6.3ms -> 2.6ms

https://drive.google.com/file/d/0B_5_Bo8SHplgaWFtLTRFVmtHTGs/view?usp=sharing

Super-sampling to x2.0: [again, for some reason that is now unclear to me, last night I decided to go from x1.5 to x2.0 for the graph below. The figures quoted are going from 'VR High' to SSx2.0 though]

Total FT: 11.1ms -> 22.2ms (confirmed by 45 FPS FR in the ED window)
Application FT: 4.3ms -> >13.5ms
Idle FT: 5.1ms -> 8.7ms

https://drive.google.com/file/d/0B_5_Bo8SHplgWTF2RDltM0NsSlk/view?usp=sharing

Conclusions:

So it looks like a gentle over-clock will get the GTX1070 FE over the line for running with SSx1.5 (and maybe even have a little space for another option?), but its not enough to run SSx2.0. This is a shame because SSx2.0 looks a *lot* better IMHO than SSx1.5 (which is better than no SS, but not so dramatically). Again, this seems similar to OC'd GTX980 performance. If this performance is similar (90% of) the performance of a stock GTX1080 FE then the fact that the FT for SSx2.0 is significantly over 11.1ms suggests that the GTX1080FE will not be able to run SSx2.0 at 90 FPS out of the box, but if a mild over-clock results in a similarly scale improvement for the GTX1080 the this should see it run SSx2.0 at 90 FPS.

Other Notes:

• Neither SSx1.5 or SSx2.0 seemed to make any difference to the quality of the menu text (this might not be the case for in-game text though - I'll test this tonight) but it made a dramatic difference to the quality of the image in general and specifically the jaggies. SSx2.0 seemed like a larger step-change in quality over SSx1.5 than SSx1.5 is over SSx1.0.
• None of the AA options seemed to improve things much for either the text or the jaggies; SMAA seemed the best but nothing dramatic and nowhere near as good as enabling SS). It's probably not worth enabling.

If anyone has any other suggestions on things worth trying to improve the image quality, let me know.

Other relevant system specs: Intel i5 3570K @3.4 GHz, Asus P8Z77-V PRO Z77, 16GB Corsair Vengence DDR3 RAM, Toshiba 240GB SATAII SSD.

 

[HeavyGroovez]

Good effort, repped.

The single most important thing with Pascal is keeping the card below the GPU Boost throttling threshold.

You need to stress it hard and find out where GPU Boost starts throttling and make sure you keep it below that temperature.

 

[Cmdr Menmnu]

Good effort, repped.

The single most important thing with Pascal is keeping the card below the GPU Boost throttling threshold.

You need to stress it hard and find out where GPU Boost starts throttling and make sure you keep it below that temperature.

I'm not quite sure how I'd test this - How do I find out if the GPU boost is on and/or being throttled? In these tests that card was using about 79% power and the temp was about 59 degrees (fan typically <45%). For the over-clock I upped the power threshold to the max 112% which rises the temp limit to 92 degrees. Its not getting anywhere near that..

 

[HeavyGroovez]

I'm not quite sure how I'd test this - How do I find out if the GPU boost is on and/or being throttled? In these tests that card was using about 79% power and the temp was about 59 degrees (fan typically <45%). For the over-clock I upped the power threshold to the max 112% which rises the temp limit to 92 degrees. Its not getting anywhere near that..

Take it to a High RES. Lots of vector geometry for the CPU to throughput and tons of textures for the rendering.

You need the card at ~95% usage to determine what clock speed you are getting flat out and if your cooling is sufficient. It will throttle like crazy if it runs hot at load.

With a decent fan curve you wont get it over 75 so the temp limit becomes irrelevant.

Just make sure you have power to 120%

Voltage appears to be locked on FE, you cant exceed factory.

Use Firestrike to benchmark it initially and get some heat metrics : http://www.3dmark.com/

 

[[HJ]-RedRaven]

Good writeup, repped! [up]

 

[Cmdr Mnumen]

nice!

 

[shadragon]

I'm not quite sure how I'd test this - How do I find out if the GPU boost is on and/or being throttled? In these tests that card was using about 79% power and the temp was about 59 degrees (fan typically <45%). For the over-clock I upped the power threshold to the max 112% which rises the temp limit to 92 degrees. Its not getting anywhere near that..

This vid may help.

https://youtu.be/OXUo1S55ZUM

Also, your older CPU may be holding you back. Here's a page from CPU boss. http://cpuboss.com/cpu/Intel-Core-i5-3570K

The CPU was introduced in 2012. "It's a decent step forward compared to the previous core, but it's still not a great gaming solution."

I can get 90 FPS with high / ultra settings on 1.5 SS in space, but do take a hit in stations. GTX 980Ti 6GB Gaming. No OC.

 

[Cmdr Menmnu]

This vid may help.

https://youtu.be/OXUo1S55ZUM

Also, your older CPU may be holding you back. Here's a page from CPU boss. http://cpuboss.com/cpu/Intel-Core-i5-3570K

The CPU was introduced in 2012. "It's a decent step forward compared to the previous core, but it's still not a great gaming solution."

I can get 90 FPS with high / ultra settings on 1.5 SS in space, but do take a hit in stations. GTX 980Ti 6GB Gaming. No OC.

The CPU is not holding anything back - the top graph shows the fraction of the frametime for each frame that the CPU us active for and it's typically idel for ~6ms of each 11ms frame so it really not under heavy load. This is supported by the low CPU fan speeds and temps I see. The front menu is (I think) equivalent to being in a station. Could you put up some similar graphs for yout 980ti in the staring menu on VR high for comparison?

- - - - - Additional Content Posted / Auto Merge - - - - -

Take it to a High RES. Lots of vector geometry for the CPU to throughput and tons of textures for the rendering.

You need the card at ~95% usage to determine what clock speed you are getting flat out and if your cooling is sufficient. It will throttle like crazy if it runs hot at load.

With a decent fan curve you wont get it over 75 so the temp limit becomes irrelevant.

Just make sure you have power to 120%

Voltage appears to be locked on FE, you cant exceed factory.

Use Firestrike to benchmark it initially and get some heat metrics : http://www.3dmark.com/

For the GTX1070 the power OC is limited to 112% (it's 120% ont he 1080 IIRC). I wasn't able to change the core voltage in afterburner 4.2, but 4.3beta allows me to change the voltage and adjust the boost curve too (not done this yet, looks scary). I'm planning to try out SSx1.5 in game with stations and low-number ships tonight and see how it goes - hopefully not too many dropped frames! I'll try a Haz Res thisevening if I can, but how can I read out the clock speed boost? I guess the clock speed must be in Afterburner somewhere.

I'll get the graphs for in-game performance toninght and stick them up when I get a mo.

 

[Cmdr Menmnu]

OK, so I tried out the Target Practice training mission in an Icy Ring, the Skirmish training mission (3 eagles combat in space), the docking training mission and the 'Wolf Pack Tactics' mission (3v1 combat in a rocky ring system). The latter was very useful in finding a good optimum. I have some graphs I can put up, but the basic result was that with SSx1.5 even with the OC the GTX1070 cannot maintain a constant 90 FPS in the Wolf Pack Tactics mission or inside the station and is pretty close to the edge in the Target Practice mission. in WPT, framerates regularly dropped to 50-60 FPS resulting in notable jerkyness when playing. This doesn't bode well for Haz. Res. sites! In the space skirmish the FT was about 7ms with SSx1.5 so that was fine. Dropping SS to 1.0 allowed me to enable/increase other optiones pretty much at will, with the exception of Ambient Occlusion. AO is quite a significant increase, particularly int he ring system and station and setting it to High along with the other increases (inc SMAA) resulted in dropped frames. Setting it to Medium was fine.

Interestingly, the Frame Time became considerably worse and was CPU bound when firing multicannons and when going through the slot regardless of SS. So even with SSx1.0 the Frame Time was being exceeded by the CPU here, not the GPU (thanks for making me check that Shadragon!) Over-clocking the CPU to 4.2GHz made a dramatic difference in limiting the impact of this although it doesn't save SSx1.5 in the ring system combat or stations sadly.

Playing through Wolf Pack Tactics a few times with different parameters on I decided that smooth gameplay was more important than SSx1.5 despite the graphical fidelity increase and the capacity to increase the other options, particularly maxing the draw distance FX quality, seemed to help with sense of immersion and scale. In the station the shadows were the most significant factor after SS ad setting them to High or ultra resulted in frametimes close to and often over the 11.1ms mark, resulting in framerates dropping as low as 60 FPS even with SSx1.0.

The settings I settled on were:

Max Draw Distance
Texture Quality: High
Shadow Quality: Medium
Bloom: High
Blur: On
AA: SMAA
SS: 1.0
AO: Medium
Environment Quality: High
FX Quality: High
Depth of Field: High
Reflections Quality: High
Material Quality: High
HMD Image Quality: High

Still to test: performance on planet surface, Haz Res.

 

[koop26]

I think that once game devs catch up and implement VR optimized code, you'll see the 1070 just perform better and better. I think launch cards are basically caught in trap of having to perform a magic show without all of their hats and bunnies ready. Pascal and the 1070 have long legs and i'm willing to bet the bottlenecks we are initially seeing are software.

With THAT said, I am glad I got a 1080, just in case i'm wrong and need that extra overhead for the next couple years.

 

[HeavyGroovez]

OK, so I tried out the Target Practice training mission in an Icy Ring, the Skirmish training mission (3 eagles combat in space), the docking training mission and the 'Wolf Pack Tactics' mission (3v1 combat in a rocky ring system). The latter was very useful in finding a good optimum. I have some graphs I can put up, but the basic result was that with SSx1.5 even with the OC the GTX1070 cannot maintain a constant 90 FPS in the Wolf Pack Tactics mission or inside the station and is pretty close to the edge in the Target Practice mission. in WPT, framerates regularly dropped to 50-60 FPS resulting in notable jerkyness when playing. This doesn't bode well for Haz. Res. sites! In the space skirmish the FT was about 7ms with SSx1.5 so that was fine. Dropping SS to 1.0 allowed me to enable/increase other optiones pretty much at will, with the exception of Ambient Occlusion. AO is quite a significant increase, particularly int he ring system and station and setting it to High along with the other increases (inc SMAA) resulted in dropped frames. Setting it to Medium was fine.

Interestingly, the Frame Time became considerably worse and was CPU bound when firing multicannons and when going through the slot regardless of SS. So even with SSx1.0 the Frame Time was being exceeded by the CPU here, not the GPU (thanks for making me check that Shadragon!) Over-clocking the CPU to 4.2GHz made a dramatic difference in limiting the impact of this although it doesn't save SSx1.5 in the ring system combat or stations sadly.

Playing through Wolf Pack Tactics a few times with different parameters on I decided that smooth gameplay was more important than SSx1.5 despite the graphical fidelity increase and the capacity to increase the other options, particularly maxing the draw distance FX quality, seemed to help with sense of immersion and scale. In the station the shadows were the most significant factor after SS ad setting them to High or ultra resulted in frametimes close to and often over the 11.1ms mark, resulting in framerates dropping as low as 60 FPS even with SSx1.0.

The settings I settled on were:

Max Draw Distance
Texture Quality: High
Shadow Quality: Medium
Bloom: High
Blur: On
AA: SMAA
SS: 1.0
AO: Medium
Environment Quality: High
FX Quality: High
Depth of Field: High
Reflections Quality: High
Material Quality: High
HMD Image Quality: High

Still to test: performance on planet surface, Haz Res.

Give the following a try and see how it goes :

1.3x SS in Debug Tool

Min Draw Distance
Texture Quality: High
Shadow Quality: Low
Bloom: Off
Blur: On
AA: Off
SS: 1.0
AO: Off
Environment Quality: Low
FX Quality: Low
Depth of Field: High
Reflections Quality: Low
Material Quality: Low
HMD Image Quality: High

 

[PvtTUX]

Give the following a try and see how it goes :

1.3x SS in Debug Tool

Min Draw Distance
Texture Quality: High
Shadow Quality: Low
Bloom: Off
Blur: On
AA: Off
SS: 1.0
AO: Off
Environment Quality: Low
FX Quality: Low
Depth of Field: High
Reflections Quality: Low
Material Quality: Low
HMD Image Quality: High

Debug tool ?

I think you are in Rift mode there !

 

[Yosharian]

If you haven't tested a RES site then it's misleading, RES sites are where I get all my juddering, it's absolutely obnoxious

 

[HeavyGroovez]

Debug tool ?

I think you are in Rift mode there !

Doh !

Lol yeah my bad

 

[Dar|{cyde]

Were you starting from VR Low or VR High when testing how much SS the game could handle? Shadows make a huuuge difference. I'm running a lowly 290 (OC'd on water, but still absolute min VR spec) and I need to turn shadows off to keep everything smooth.

Try everything low, shadows OFF, and see how much SS you can push. I'm pretty sure that turning off a few bells and whistles in order to get more supersampling will make the game look far better.

 

[Cmdr Menmnu]

Were you starting from VR Low or VR High when testing how much SS the game could handle? Shadows make a huuuge difference. I'm running a lowly 290 (OC'd on water, but still absolute min VR spec) and I need to turn shadows off to keep everything smooth.

Try everything low, shadows OFF, and see how much SS you can push. I'm pretty sure that turning off a few bells and whistles in order to get more supersampling will make the game look far better.

I was starting from VR High - you make a good point though - Ambient Occlusion and Shadows seem to be the biggest non-SS hit to the frame time. I'll try setting the shadows and AO to off and see what the performance is like with SS 1.5 & 2.0.

I tried the transition from supercruise to orbital cruise and on to glide and 'flight' over the weekend. Its a bit complicated to describe and my testing has not been exhaustive but it looks like a similar story on the face of it: GTX1070 cannot maintain SSx1.5 with other setting on VR High close to planet surfaces without hits tot he frame rate. Anecdotally it does seem to vary with the type of planet surface and, in particular, the depth and cragginess of the features which hints at the shadows being a significant time sink?

Interestingly, the card OC settings are quite stable in elite (in fact I've pushed the memory clock up to +650 and its stable), but its completely unstable for other VR titles.

 

[ArachAu]

I'm half tempted to see how VR Low goes with Super Samp 2x on the 1080. ED looks just amazing in SS2x. 1.5 is nice, but the half-step smudge on harsh orbit lines is a little too painful.

 

[Cmdr Menmnu]

Here are a few more graphs showing the performance of the GTX 1070 Fe with the Vive Docking, in rings and in combat in rings. For these tests I used the mild over-clocking described in my first post (+225 GPU clock and +233 memory clock), VR High presets with Supersampling set to SSx1.5. These settings perform well (8.0-9.0ms Frame Time giving a consistent 90FPS) in deep space but I wanted to see whether the different environments and action is sufficiently complex to damage this performance.

Docking - External to the station.

https://drive.google.com/file/d/0B_5_Bo8SHplgZUpLQTcwaVNQa0E/view?usp=sharing

Total FT: 11.1ms
Application FT: 8.5-9.5ms
Idle FT: 2.6-1.6ms

The performance is OK here and the station makes ~0.5ms difference in the used FT.

Docking - in the letterbox

https://drive.google.com/file/d/0B_5_Bo8SHplgQ2RDMjJreVdZczg/view?usp=sharing

Total FT: 22.2ms
Application FT: 11->14ms
Idle FT: <11.2ms

So even over-clocked the performance of the GTX 1070 here drops dramatically to 45 FPS. In part this is CPU bound (top graph) although the graphics load is high too.

Docking - in the station

https://drive.google.com/file/d/0B_5_Bo8SHplgREdFOEJabkk0ZzA/view?usp=sharing

Total FT: 22.2ms
Application FT: 12.7-13.5ms
Idle FT: <9.5ms

Again, even overclocked the GTX 1070 is unable to sustain 90FPS inside the station with SSx1.5. The CPU load is also relatively high and pretty spikey so my CPU is probably not helping the performance.

Icy Rings: no combat, other ships, little fog

https://drive.google.com/file/d/0B_5_Bo8SHplgR0pobGVac2U1enc/view?usp=sharing

Total FT: 11.1ms
Application FT: 9.5-10.0ms
Idle FT: <1.1-1.6ms

Performance here is consistently good.

OK, so now I wanted to try a more challenging situation. For this I initially dropped to SSx1.0 so I could then contrast with SSx1.5:

Wolf Pack tactics SSx1.0: rocky rings, fog, 3 other ships, combat (firing projectile weapons)

https://drive.google.com/file/d/0B_5_Bo8SHplgX3Z1Zl9jVUVSX2M/view?usp=sharing

Total FT: 11.1ms, 22.2ms when firing multicannons
Application FT: 4.5-10.0 - highly variable as the ships (and my head) move around.
Idle FT: <1.1-5.6ms

Performance was a consistent 90FPS in this test with SSx1.0. Interestingly, firing the projectile weapons causes the CPU to compute physics (I think) and the frame rate drop to 45 FPS is actually due to the CPU performance here not the GPU! Who knew. Preliminary testing suggests that beam weapons do not inclur this level of CPU load.

Wolf Pack tactics SSx1.5: rocky rings, fog, 3 other ships, combat (firing projectile weapons)

https://drive.google.com/file/d/0B_5_Bo8SHplgU0EwTld3RXA3X2M/view?usp=sharing

Total FT: 11.1ms, 22.2ms when firing multicannons
Application FT: 8.5-12.5 - highly variable as the ships (and my head) move around.
Idle FT: variable

Surprisingly, the performance is pretty good with SSx1.5 - most of the combat saw a consistent 90FPS with a similar drop in frame rate when firing the projectile weapons due to the CPU performance. It's much closer to the edge and the CPU is working harder causing occasions frame drops (red late starts in the graph), but its not uttery poor. The dropped frames are noticeable during game though as occasional stuttering.

Conclusions and to do..

As before SSx1.5 is right on the edge of what this card can cope with across the varies scenarios in E: D. I did a brief comparison of the graphics quality and VR-Low with SSx1.5 performance seemed similar to, and looked a little better than, VR-high SSx1.0 (with a few upped settings). I haven't tested this much yet though so thats something to try. I also have to try the performance on and close to planets .

 

[Cmdr Menmnu]

I've recently been experimenting with the vrsettings file in the Steam install (see this thread https://forums.frontier.co.uk/showt...e-pixel-density-VIVE-improve-text-readability). It seems like this makes a significant difference to image qualityand, in particular, text clarity. But it does cost some performance. Below are a few Frame Time graphs to show the performance cost. The graphs were taken, as before, in the main startup lobby, for consistencey, and used the mild overclock settings mentioned previously.

VR-High, SSx0.65, vrsetting renderTarget 2.0 (total SS 1.3):

Total FT: 11.1ms
Application FT: 8.7-9.3 ms
Idle FT: 1.8-2.4ms

https://drive.google.com/file/d/0B_5_Bo8SHplgZ1VyV0czOW9BZFk/view?usp=sharing

This level of performance is similar too, if slightly worse than VR-High with SSx1.5 (see original post). The visual quality and text are consideraby better though, in my opinion. I concluded in my previous post that I couldn't run VR-High SSx1.5 in game despite OK Frame Times in the lobby. In more complex environments (stations, ring systems with combat) the performance regularly dropped to 45 FPS. The results are similar for these settings.


VR-High, SSx0.65, vrsetting renderTarget 1.5 (total SS 0.98):

Total FT: 11.1ms
Application FT: 5.5-6ms.
Idle FT: 5.1-5.6ms

https://drive.google.com/file/d/0B_5_Bo8SHplgYVduQTVtNGFxbVk/view?usp=sharing

This level of performance is very similar to the settings I was actually using prior to these experiments to play the game (VH-High, SSx1.0, SMAA, Ambient Occlusion High, FX quality High effects, Bloom high, Environment Ultra). Again the visual quality is considerably better that standard VR-high, comparable with SSx1.5 if slightly worse than my in-game setting (IMHO). The advantage though is that the headroom in the Frame Time allows a few increases in some of the other settings (Ambient Occlusion Low, FX quality High effects, Bloom Low, Environment Ultra) without hurting the performance.



These setting give me a consistent 90FPS in ring systems, in the Wolf Pack Tactics test (except when CPU bound firing weapons) and in stations, but I do get a noticeable drop in frame rate around bases on planet surfaces - I guess the combination of the two is challenging the performance.

Conclusions...

VR-high, vrsetting 1.5 and SSx0.65 with a few other categories increased is now what I am playing the game with. Combined with the green NightVision colour scheme (which makes a huge difference for text it seems!) I am now pretty happy with the visuals, and the performance and I'm enjoying my E: D life. Still need to quantify the performance on planet surfaces.