NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Everything about latency. Tips, testing methods, mouse lag, display lag, game engine lag, network lag, whole input lag chain, VSYNC OFF vs VSYNC ON, and more! Input Lag Articles on Blur Busters.
User avatar
speancer
Posts: 193
Joined: 03 May 2020, 04:26
Location: Poland

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by speancer » 04 Oct 2020, 10:49

1000WATT wrote:
04 Oct 2020, 10:31
speancer wrote:
04 Oct 2020, 09:21
Just to let you know, I am especially curious how Reflex behaves in non-GPU-bound scenarios, so I'm looking forward to seeing your test results. I guess them results would give me the ultimate answer whether Reflex is completely useless for me or not, as I am never GPU-bound in my competitive play, my average framerate is 320-340 fps in competitive match with estimated average GPU usage between 35-40% (RTSS measurements).
Yes, it is interesting.
average GPU usage between 35-40%. Not an indication that the game is not tied to the GPU. With a GPU load of 30% and a heavy load on the memory of the video card, you can get delays above 300ms.
Video memory load is also low in my case, I checked it now. VRAM usage is just 1.1 GB max (out of 6GB of my GTX 980Ti), GPU load about 25-40%, average use would be around 30% (nothing too exact). Anyway, I'm surely not GPU-bound here at all.
Tested displays: Zowie XL2546KASUS VG259QMASUS VG279QMOmen X 25fZowie XL2546 MSI MAG251RXAcer Predator XB273 XAcer Predator XB271HU
Displays to test: ASUS VG258QM • Zowie XL2540K (currently testing)

User avatar
jorimt
Posts: 1484
Joined: 04 Nov 2016, 10:44
Location: USA

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by jorimt » 04 Oct 2020, 11:26

speancer wrote:
04 Oct 2020, 10:25
Are you able to estimate when would your Reflex and LLM-for-CS:GO test results come up?
I honestly have no idea at the moment. Nvidia is waiting on a second batch to be manufactured. All I know is I'm on that list. Rough ETA for delivery I was given is sometime this month.

Once received, I have to do some off-the-record testing to ensure I'm getting good data and performing best practices, but once I get the process down, everyone here can expect a semi-regular output of results in a variety of scenario and games.

Also, as a disclaimer, photodiode testing is limited where no sync is concerned (tends to equalize the differences between sync and no sync), so I will primarily be focusing on testing sync or no sync scenarios against secondary settings (like LLM, HAGS, Reflex, FPS limiters, etc), and not against each other.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: Acer Predator XB271HU / LG 48CX OS: Windows 10 MB: ASUS ROG Maximus X Hero CPU: i7-8700k GPU: EVGA RTX 3080 FTW3 UG RAM: 32GB G.SKILL TridentZ @3200MHz

blackstorm82
Posts: 10
Joined: 01 Jun 2020, 20:23

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by blackstorm82 » 06 Oct 2020, 13:15

jorimt wrote:
02 Oct 2020, 12:30
blackstorm82 wrote:
02 Oct 2020, 12:15
If it's not bound to the gpu, is it correct that the reflex doesn't work either?
Reflex eliminates the render queue. In GPU-bound situations, the render queue becomes saturated. In non-GPU-bound situations, the render queue is typically already near empty or empty.
blackstorm82 wrote:
02 Oct 2020, 12:15
[144HZ]
GSYNC ON + VSYNC OFF
141 FPS CAP GPU 50%-70%
REFLEX ON+BOOST
-------------------------------
Reflex doesn't help in this situation, right?
Right.
blackstorm82 wrote:
02 Oct 2020, 12:15
[144HZ]
GSYNC ON + VSYNC OFF
280 FPS [NOT CAP] GPU 95-99%
REFLEX ON+BOOST
95-99% Is it only turned on with a similar behavior to the low-latency mode?
When enabled, Reflex is "turned on" whether your system is GPU-bound or not. But since it removes the render queue, how much it reduces input lag depends on how full your render queue would be if Reflex was not enabled.

Reflex is a replacement to LLM. LLM only reduces the render queue. Reflex removes it.
LLM - ON/ULTRA
MPRF - Render queue "1" limit

REFLEX ON/BOOST
MPRT - Render queue "0" always

Can I think this way?

User avatar
jorimt
Posts: 1484
Joined: 04 Nov 2016, 10:44
Location: USA

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by jorimt » 06 Oct 2020, 14:12

blackstorm82 wrote:
06 Oct 2020, 13:15
LLM - ON/ULTRA
MPRF - Render queue "1" limit

REFLEX ON/BOOST
MPRT - Render queue "0" always

Can I think this way?
Yes.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: Acer Predator XB271HU / LG 48CX OS: Windows 10 MB: ASUS ROG Maximus X Hero CPU: i7-8700k GPU: EVGA RTX 3080 FTW3 UG RAM: 32GB G.SKILL TridentZ @3200MHz

Meowchan
Posts: 36
Joined: 17 Jun 2020, 02:06

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by Meowchan » 16 Oct 2020, 05:56

Are there other benchmarks for reflex other than this fortnight one and this? Image

I think NVidia might be trying to pass this off as better than it really is and I'd like to see some more hard numbers.

Meowchan
Posts: 36
Joined: 17 Jun 2020, 02:06

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by Meowchan » 16 Oct 2020, 06:48

Here's the thing:

Having 2 separate factories (CPU / GPU) for processing frames that are misaligned clearly leads to bad outcomes. Reflex addresses this by aligning the CPU factory just before the GPU one in cases where the GPU is the bottle neck. This takes care of some of the problems. But, and this is the issue I think NVidia is all too happy to gloss over in their marketing talk, the overall latency is still greater than CPU time + GPU time because you are delaying the CPU from starting to work on the task.

Here's an example with numbers:

In a GPU-heavy game the CPU frame processing time is 0, and the GPU frame processing time is 1 second per frame. At t = 0 the GPU start working on frame A. At t = 0.1s the user pressed the mouse button. With Reflex, the CPU will wait until just before t = 1 to start calculating frame B where the input can be taken into account, then at t = 1 the GPU will start processing frame B, and at t = 2 frame B can be displayed on screen. Total latency is 1.9s, which is significantly longer than CPU processing time + GPU processing time.

Without Reflex? The button press would still be displayed at t = 2, and Reflex had no effect.

In one sense the render queue has disappeared. But in another sense it merely moved to the front of the CPU processing.

slaver01
Posts: 18
Joined: 21 Sep 2020, 01:48

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by slaver01 » 16 Oct 2020, 07:48

From ignorant. What would be the best setting on Fortinite Competitive with DX12 (240hz)?
(Gpu 50% - Cpu 60%)

G-sync: ON
V-sync: OFF
LLM: OFF
Pre-rendering Frame: 0
Reflex: ON +Boost
Limit FPS in game: 237

There is something wrong?

bungles
Posts: 9
Joined: 26 Oct 2020, 11:32

Re: NVIDIA Reflex Low Latency - How It Works & Why You Want To Use It

Post by bungles » 26 Oct 2020, 17:29

When you are posting using VSYNC are you using regular VSYNC or VSYNC-Fast? for 144 and 240hz?

Post Reply