Summary - To try to determine how much CPU is really required for the optimal Elite VR experience, I've tested my Core i7-2600K at various frequencies in various scenarios using the Oculus Debug tool to measure "CPU Render time". Ideally the CPU Render time should be < 11.1ms for the Oculus CV1 (or Vive consumer version).
I tested scenarios starting with roving in the SRV through to taking off and landing at a base on another planet. I do not have a way to record live data from the Oculus Debug tool so I cannot show absolutes but rather what I saw on the HUD at various times and took notes from (see below).
Preliminary take-aways:
- i7-2600K at stock clocks will experience occasional to moderate FPS drops* below 90 fps, and a faster CPU will improve the experience. This may be subjectively "good enough" for some players, however.
- An i7-2600K @ ~ 4.2-4.3 GHz (and faster DDR3) is fast enough to maintain 90 fps in almost all scenarios. This provides a good "objective" and "subjective" experience.
- (Extrapolating), the i7-3770K is about 5% faster at the same clock than Sandy Bridge; so a 3770K will need some overclocking. A 4770K should be good enough at stock speeds, but could probably benefit from a very mild OC. A 4790K or 6700K are definitely fast enough at stock speeds.
- *Orbital Cruise and Approaching Planetary bases are significantly more CPU intensive than super-cruise or flying/roving around a planet. For instance, a stock 2600K might never exceed 11.1ms "alone in the SRV", but will struggle to maintain target latency in Orbital Cruise.
..... Benchmarks:
System Config: i7-2600K, 16GB DDR3-1866; MSI Gaming 980Ti (out of box OC), Crucial MX100 512GB SSD, Windows 10 installed ~ 9 months ago. Oculus Rift CV1.
All Scenarios - Open Play. VR Low preset.
Scenario 0 - Horizons / VR Main Menu (Buggy and ship in space dock).
CPU @ 4.5 GHz - CPU Peak ~ 3.6ms (GPU ~ 5.1ms)
CPU @ 3.0 GHz - CPU Peak ~ 5.0ms (GPU ~ 5.7ms)
CPU @ 2.0 GHz - CPU Peak ~ 7.0ms; (GPU ~ 7.2ms).
2.0 GHz is fast enough for the main menu
.
Scenario 1 - Roving on a planet.
4.5 GHz - CPU Peak ~ 7.4ms (6.x typical); (GPU ~ 7.3ms)
3.0 GHz - CPU Peak ~ 9.9ms (9.0 typical); (GPU ~ 11ms)*
2.0 GHz - CPU Peak ~ 17ms (13.x typical); (GPU ~ 15ms)**
*Occasional framerate drops to ~ 84fps.
** Typically 50-60 fps; some drops 'lock' at 45 fps.
3.0 GHz is enough.
Scenario 1B - Roving on a planet. Recalled Ship.
4.5 GHz - CPU is now peaking at ~ 8ms.
3.0 GHz - CPU is now peaking at ~ 12ms, with frame drips sometimes down to mid-50 fps.
Notes - CPU time stabilized close after 30 seconds or so; something significant was happening CPU wise for 'a while' during and after ship landing. This was repeatable although the impact was less during subsequent recalls/dismissals.
3.0 GHz "good enough" with some occasional FPS drops depending on what is going on; Async. Timewarp addresses these scenarios promptly.
Scenario 2 - Supercruise from one planet to another (base landing).
4.5 GHz - CPU typically ~ 6-8 ms in space, 9-10 approaching planet typical.
GPU has light trouble maintaining 90 fps in orbital cruise (render time >11.1ms).
Notes - Did not test 3.0 GHz; but 3.0 GHz "OK" for supercruise, "Not good" for Orbital cruise. ATW may help 3.0 GHz here.
Scenario 3 - Flying and Docking at a ground based dock. Imperial Clipper.
4.5 GHz - CPU ~ 9 ms near base, with regular spikes to ~ 10.x ms.
GPU mostly stable at 90fps.
Once landed, the CPU load dropped a bit, and when inside hangar was down to ~ 6-7ms range.
Notes - 4.5 GHz 2600K was being pushed very hard here with ~ 1 ms (10%) of margin left. 3.0 GHz would clearly not be enough.
Q&A
What else is missing?
- I would love to be able to record data in Oculus Debug tool since lower frequencies tended to show more "App missed submit count" action. The Oculus debug tool also requires some CPU time.
Why no screenshots?
- I could not get shadowplay to record with the Oculus debug HUD up. Also,
Shadowplay adds CPU usage (although if I could get it to work I could test it's impact).
Why no tests at stock 2600K speeds (3.4/3.8 GHz)?
- There are a ton of variables that determine available CPU speed. Are you running a virus scanner? are web browsers open? etc. I'd recommend using the 3.0 GHz as a guide; a 2600K stock is 3.4 GHz when loading 4 cores.. this gives some margin for other applications to run. Oh, and if you're running a 2600K at stock speeds, you're really leaving a lot of performance on the table.
Can you simulate a Core i5?
- Not really. I can disable Hyperthreading and lower the clocks, but the i7 also has 33% more L3 cache (8MB vs 6MB) that may have substantial impact on the game code. Digitalfoundry has done videos showing that the i7 does a lot to stabilize minimum FPS vs. a core i5; likely due to a combination of that cache and hyperthreading.
To Do:
- Figure out how to get the Oculus Debug tool to log and record data.
- Figure out how to take a screenshot of Elite VR without something "heavy" like Shadowplay or FRAPs.
I tested scenarios starting with roving in the SRV through to taking off and landing at a base on another planet. I do not have a way to record live data from the Oculus Debug tool so I cannot show absolutes but rather what I saw on the HUD at various times and took notes from (see below).
Preliminary take-aways:
- i7-2600K at stock clocks will experience occasional to moderate FPS drops* below 90 fps, and a faster CPU will improve the experience. This may be subjectively "good enough" for some players, however.
- An i7-2600K @ ~ 4.2-4.3 GHz (and faster DDR3) is fast enough to maintain 90 fps in almost all scenarios. This provides a good "objective" and "subjective" experience.
- (Extrapolating), the i7-3770K is about 5% faster at the same clock than Sandy Bridge; so a 3770K will need some overclocking. A 4770K should be good enough at stock speeds, but could probably benefit from a very mild OC. A 4790K or 6700K are definitely fast enough at stock speeds.
- *Orbital Cruise and Approaching Planetary bases are significantly more CPU intensive than super-cruise or flying/roving around a planet. For instance, a stock 2600K might never exceed 11.1ms "alone in the SRV", but will struggle to maintain target latency in Orbital Cruise.
..... Benchmarks:
System Config: i7-2600K, 16GB DDR3-1866; MSI Gaming 980Ti (out of box OC), Crucial MX100 512GB SSD, Windows 10 installed ~ 9 months ago. Oculus Rift CV1.
All Scenarios - Open Play. VR Low preset.
Scenario 0 - Horizons / VR Main Menu (Buggy and ship in space dock).
CPU @ 4.5 GHz - CPU Peak ~ 3.6ms (GPU ~ 5.1ms)
CPU @ 3.0 GHz - CPU Peak ~ 5.0ms (GPU ~ 5.7ms)
CPU @ 2.0 GHz - CPU Peak ~ 7.0ms; (GPU ~ 7.2ms).
2.0 GHz is fast enough for the main menu
.Scenario 1 - Roving on a planet.
4.5 GHz - CPU Peak ~ 7.4ms (6.x typical); (GPU ~ 7.3ms)
3.0 GHz - CPU Peak ~ 9.9ms (9.0 typical); (GPU ~ 11ms)*
2.0 GHz - CPU Peak ~ 17ms (13.x typical); (GPU ~ 15ms)**
*Occasional framerate drops to ~ 84fps.
** Typically 50-60 fps; some drops 'lock' at 45 fps.
3.0 GHz is enough.
Scenario 1B - Roving on a planet. Recalled Ship.
4.5 GHz - CPU is now peaking at ~ 8ms.
3.0 GHz - CPU is now peaking at ~ 12ms, with frame drips sometimes down to mid-50 fps.
Notes - CPU time stabilized close after 30 seconds or so; something significant was happening CPU wise for 'a while' during and after ship landing. This was repeatable although the impact was less during subsequent recalls/dismissals.
3.0 GHz "good enough" with some occasional FPS drops depending on what is going on; Async. Timewarp addresses these scenarios promptly.
Scenario 2 - Supercruise from one planet to another (base landing).
4.5 GHz - CPU typically ~ 6-8 ms in space, 9-10 approaching planet typical.
GPU has light trouble maintaining 90 fps in orbital cruise (render time >11.1ms).
Notes - Did not test 3.0 GHz; but 3.0 GHz "OK" for supercruise, "Not good" for Orbital cruise. ATW may help 3.0 GHz here.
Scenario 3 - Flying and Docking at a ground based dock. Imperial Clipper.
4.5 GHz - CPU ~ 9 ms near base, with regular spikes to ~ 10.x ms.
GPU mostly stable at 90fps.
Once landed, the CPU load dropped a bit, and when inside hangar was down to ~ 6-7ms range.
Notes - 4.5 GHz 2600K was being pushed very hard here with ~ 1 ms (10%) of margin left. 3.0 GHz would clearly not be enough.
Q&A
What else is missing?
- I would love to be able to record data in Oculus Debug tool since lower frequencies tended to show more "App missed submit count" action. The Oculus debug tool also requires some CPU time.
Why no screenshots?
- I could not get shadowplay to record with the Oculus debug HUD up. Also,
Shadowplay adds CPU usage (although if I could get it to work I could test it's impact).
Why no tests at stock 2600K speeds (3.4/3.8 GHz)?
- There are a ton of variables that determine available CPU speed. Are you running a virus scanner? are web browsers open? etc. I'd recommend using the 3.0 GHz as a guide; a 2600K stock is 3.4 GHz when loading 4 cores.. this gives some margin for other applications to run. Oh, and if you're running a 2600K at stock speeds, you're really leaving a lot of performance on the table
.Can you simulate a Core i5?
- Not really. I can disable Hyperthreading and lower the clocks, but the i7 also has 33% more L3 cache (8MB vs 6MB) that may have substantial impact on the game code. Digitalfoundry has done videos showing that the i7 does a lot to stabilize minimum FPS vs. a core i5; likely due to a combination of that cache and hyperthreading.
To Do:
- Figure out how to get the Oculus Debug tool to log and record data.
- Figure out how to take a screenshot of Elite VR without something "heavy" like Shadowplay or FRAPs.