LLM On vs ultra

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: 241
Joined: 03 May 2020, 04:26
Location: EU

Re: LLM On vs ultra

Post by speancer » 30 Sep 2020, 12:28

jorimt wrote:
30 Sep 2020, 10:15
Unlike LLM, Reflex is guaranteed to work in the game it supports, and it eliminates the render queue. LLM only reduces it, and only if the game allows override.

As for it being "useless" in non-GPU-bound scenarios, not necessarily. With reflex disable, the render queue still exists, and can still be filled at any given point in non-GPU-bound scenarios, just not as much as it would be when the system is GPU-bound.

Reflex also ensures that if your system does become GPU-bound at any point (even for just a few frames in a more demanding scene), input lag/buffering won't increase due to the render queue.
I see. So, let's say I'm running a game on lower settings that is not GPU-bound and it relies on the CPU mostly...

Firstly: if the GPU is basically sleeping, let's say it uses about 30% (like my GTX 980Ti does in CS:GO) at all times, I guess the render queue is always empty then? I play on lowered settings and 1024x768 resolution. And if it's always empty, then Reflex would be completely useless?

Secondly: what advantage in latency reduction would Reflex enabled give me in non GPU-bound situation in which render queue fills up occasionally, like you mentioned? I'm aware this might be too general question and it could depend on given circumstances.

Do you think Counter-Stike would eventually get Reflex support, or perhaps it would be completely unnecessary in such CPU-dependent game? On the other hand, it could help players with low-end systems to reduce their latency I reckon? Anyway, main focus here is serious competitive play on a high-end PC. I only play CS:GO competitively, it's my main game and only e-sport title I focus on, so I don't care that much about new techs that are not related to CS:GO when it comes to latency reduction.
Main display (TV/PC monitor): LG 42C21LA (4K 120 Hz OLED / WBE panel)
Tested displays: ASUS VG259QM/VG279QM [favourite LCD FPS display] (280 Hz IPS) • Zowie XL2546K/XL2540K/XL2546 (240 Hz TN DyAc) • Dell S3222DGM [favourite LCD display for the best blacks, contrast and panel uniformity] (165 Hz VA) • Dell Alienware AW2521HFLA (240 Hz IPS) • HP Omen X 25f (240 Hz TN) • MSI MAG251RX (240 Hz IPS) • Gigabyte M27Q (170 Hz IPS) • Acer Predator XB273X (240 Hz IPS G-SYNC) • Acer Predator XB271HU (165 Hz IPS G-SYNC) • Acer Nitro XV272UKV (170 Hz IPS) • Acer Nitro XV252QF (390 Hz IPS) • LG 27GN800 (144 Hz IPS) • LG 27GL850 (144 Hz nanoIPS) • LG 27GP850 (180 Hz nanoIPS) • Samsung Odyssey G7 (240 Hz VA)

OS: Windows 11 Pro GPU: Palit GeForce RTX 4090 GameRock OC CPU: AMD Ryzen 7 7800X3D + be quiet! Dark Rock Pro 4 + Arctic MX-6 RAM: 32GB (2x16GB dual channel) DDR5 Kingston Fury Beast Black 6000 MHz CL30 (fully optimized primary and secondary timings by Buildzoid for SK Hynix die on AM5 platform) PSU: Corsair RM1200x SHIFT 1200W (ATX 3.0, PCIe 5.0 12VHPWR 600W) SSD1: Kingston KC3000 1TB NVMe PCIe 4.0 x4 SSD2: Corsair Force MP510 960GB PCIe 3.0 x4 MB: ASUS ROG STRIX X670E-A GAMING WIFI (GPU PCIe 5.0 x16, NVMe PCIe 5.0 x4) CASE: be quiet! Silent Base 802 Window White CASE FANS: be quiet! Silent Wings 4 140mm PWM (3x front, 1x rear, 1x top rear, positive pressure) MOUSE: Logitech G PRO X Superlight (white) Lightspeed wireless MOUSEPAD: ARTISAN FX HIEN (wine red, soft, XL) KEYBOARD: Logitech G915 TKL (white, GL Tactile) Lightspeed wireless HEADPHONES: Sennheiser Momentum 4 Wireless (white) 24-bit 96 KHz + Sennheiser BTD600 Bluetooth 5.2 aptX Adaptive CHAIR: Herman Miller Aeron (graphite, fully loaded, size C)

Boop
Posts: 134
Joined: 08 Nov 2018, 22:10

Re: LLM On vs ultra

Post by Boop » 30 Sep 2020, 13:35

speancer wrote:
30 Sep 2020, 12:28
jorimt wrote:
30 Sep 2020, 10:15
Unlike LLM, Reflex is guaranteed to work in the game it supports, and it eliminates the render queue. LLM only reduces it, and only if the game allows override.

As for it being "useless" in non-GPU-bound scenarios, not necessarily. With reflex disable, the render queue still exists, and can still be filled at any given point in non-GPU-bound scenarios, just not as much as it would be when the system is GPU-bound.

Reflex also ensures that if your system does become GPU-bound at any point (even for just a few frames in a more demanding scene), input lag/buffering won't increase due to the render queue.
I see. So, let's say I'm running a game on lower settings that is not GPU-bound and it relies on the CPU mostly...

Firstly: if the GPU is basically sleeping, let's say it uses about 30% (like my GTX 980Ti does in CS:GO) at all times, I guess the render queue is always empty then? I play on lowered settings and 1024x768 resolution. And if it's always empty, then Reflex would be completely useless?

Secondly: what advantage in latency reduction would Reflex enabled give me in non GPU-bound situation in which render queue fills up occasionally, like you mentioned? I'm aware this might be too general question and it could depend on given circumstances.

Do you think Counter-Stike would eventually get Reflex support, or perhaps it would be completely unnecessary in such CPU-dependent game? On the other hand, it could help players with low-end systems to reduce their latency I reckon? Anyway, main focus here is serious competitive play on a high-end PC. I only play CS:GO competitively, it's my main game and only e-sport title I focus on, so I don't care that much about new techs that are not related to CS:GO when it comes to latency reduction.
LLM and Reflex sounds like it would be pretty worthless for CS:GO with your current setup.

User avatar
jorimt
Posts: 2481
Joined: 04 Nov 2016, 10:44
Location: USA

Re: LLM On vs ultra

Post by jorimt » 30 Sep 2020, 14:01

speancer wrote:
30 Sep 2020, 12:28
I see. So, let's say I'm running a game on lower settings that is not GPU-bound and it relies on the CPU mostly...
CS:GO is an exception.

Even though it has a multicore option, it primarily relies on single-threaded CPU performance, and has very low GPU requirements, which is why even a potato can run it at relatively high framerates.

However, since it is so dependent on single-threaded CPU performance, while high-end systems can also achieve very high framerates in it, they're bottlenecked from achieving even higher framerates because of the fact that CS:GO doesn't take enough advantage of the GPU or multiple CPU cores/threads, so basically, the better the system, the more diminishing returns you get in that game (this is the same reason Crisis is still difficult to run at very high framerates consistently; e.g. single-threaded priority on performance).

What I'm saying is, CS:GO "lag" is mostly limited by its netcode, so LLM/Reflex is the least of your worries, especially since a GPU-bound situation is unlikely to occur 99.9% of the time.

That said, yes, the render queue implementation can vary heavily by game. I've known some to queue up pre-rendered frames regardless of whether the game was CPU or GPU-limited, so LLM/Reflex could still theoretically be beneficial where a game continually queues frames into the render queue when it doesn't need to.

So there's really no harm in leaving LLM to "On," in CS:GO. Not 100% sure if LLM works in CS:GO though, as I haven't tested it.

Finally, concerning Reflex, I wouldn't be surprised if Nvidia ultimately adds it to that game, but I wouldn't expect much improvement if they did.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: ASUS PG27AQN, LG 48CX VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)

User avatar
speancer
Posts: 241
Joined: 03 May 2020, 04:26
Location: EU

Re: LLM On vs ultra

Post by speancer » 04 Oct 2020, 10:34

jorimt wrote:
30 Sep 2020, 14:01
So there's really no harm in leaving LLM to "On," in CS:GO. Not 100% sure if LLM works in CS:GO though, as I haven't tested it.

Finally, concerning Reflex, I wouldn't be surprised if Nvidia ultimately adds it to that game, but I wouldn't expect much improvement if they did.
Thanks again for your answers! I guess eventually you will know for sure, as we both know now you decided to test LLM for CS:GO yourself :lol: And concerning Reflex, probably it wouldn't change much as you say, but some improvement would surely be better than none.
Main display (TV/PC monitor): LG 42C21LA (4K 120 Hz OLED / WBE panel)
Tested displays: ASUS VG259QM/VG279QM [favourite LCD FPS display] (280 Hz IPS) • Zowie XL2546K/XL2540K/XL2546 (240 Hz TN DyAc) • Dell S3222DGM [favourite LCD display for the best blacks, contrast and panel uniformity] (165 Hz VA) • Dell Alienware AW2521HFLA (240 Hz IPS) • HP Omen X 25f (240 Hz TN) • MSI MAG251RX (240 Hz IPS) • Gigabyte M27Q (170 Hz IPS) • Acer Predator XB273X (240 Hz IPS G-SYNC) • Acer Predator XB271HU (165 Hz IPS G-SYNC) • Acer Nitro XV272UKV (170 Hz IPS) • Acer Nitro XV252QF (390 Hz IPS) • LG 27GN800 (144 Hz IPS) • LG 27GL850 (144 Hz nanoIPS) • LG 27GP850 (180 Hz nanoIPS) • Samsung Odyssey G7 (240 Hz VA)

OS: Windows 11 Pro GPU: Palit GeForce RTX 4090 GameRock OC CPU: AMD Ryzen 7 7800X3D + be quiet! Dark Rock Pro 4 + Arctic MX-6 RAM: 32GB (2x16GB dual channel) DDR5 Kingston Fury Beast Black 6000 MHz CL30 (fully optimized primary and secondary timings by Buildzoid for SK Hynix die on AM5 platform) PSU: Corsair RM1200x SHIFT 1200W (ATX 3.0, PCIe 5.0 12VHPWR 600W) SSD1: Kingston KC3000 1TB NVMe PCIe 4.0 x4 SSD2: Corsair Force MP510 960GB PCIe 3.0 x4 MB: ASUS ROG STRIX X670E-A GAMING WIFI (GPU PCIe 5.0 x16, NVMe PCIe 5.0 x4) CASE: be quiet! Silent Base 802 Window White CASE FANS: be quiet! Silent Wings 4 140mm PWM (3x front, 1x rear, 1x top rear, positive pressure) MOUSE: Logitech G PRO X Superlight (white) Lightspeed wireless MOUSEPAD: ARTISAN FX HIEN (wine red, soft, XL) KEYBOARD: Logitech G915 TKL (white, GL Tactile) Lightspeed wireless HEADPHONES: Sennheiser Momentum 4 Wireless (white) 24-bit 96 KHz + Sennheiser BTD600 Bluetooth 5.2 aptX Adaptive CHAIR: Herman Miller Aeron (graphite, fully loaded, size C)

User avatar
axaro1
Posts: 627
Joined: 23 Apr 2020, 12:00
Location: Milan, Italy

Re: LLM On vs ultra

Post by axaro1 » 23 Jan 2021, 10:05

jorimt wrote:
30 Sep 2020, 08:34
I recently tried to switch back to Radeon Anti-Lag ON with the newest drivers on a 5700xt/ryzen machine and I don't know if Amd actively improved Radeon Anti-Lag but I feel like even in high fps/low gpu usage (<75%)/ engine or cpu limited scenarios the engine is noticeably more responsive and consistent, I tried DX11 and DX9 games.

I'm starting to wonder if Battlenonsense's results are actually correct, I've never doubted the reliability of his testings, especially since he agrees with many statements about VRR + fps caps combos coming from Blurbusters itself but it seems like he is the only tester who stated that he measured negative impacts when testing ULL and Anti-Lag.

I'll wait for someone like Fr33thy to make the final input lag tests to prove if there is some actual difference between ULL/Anti-Lag ON/OFF since he is using a more precise way of testing.

I highly recommend RDNA and RDNA2 owners playing at high refresh rates to give Anti-Lag a try, it truly feels like my input are being processed more correctly, less floaty input feeling, the fps reduction penalty(-5% in the games I tested so far) seems absolutely worth the tradeoff.
XL2566K* | XV252QF* | LG C1* | HP OMEN X 25 | XL2546K | VG259QM | XG2402 | LS24F350[RIP]
*= currently owned



MONITOR: XL2566K custom VT: https://i.imgur.com/ylYkuLf.png
CPU: 5800x3d 102mhz BCLK
GPU: 3080FE undervolted
RAM: https://i.imgur.com/iwmraZB.png
MOUSE: Endgame Gear OP1 8k
KEYBOARD: Wooting 60he

mossfalt
Posts: 37
Joined: 23 Nov 2020, 08:43

Re: LLM On vs ultra

Post by mossfalt » 25 Jan 2021, 05:18

axaro1 wrote:
23 Jan 2021, 10:05
jorimt wrote:
30 Sep 2020, 08:34
I recently tried to switch back to Radeon Anti-Lag ON with the newest drivers on a 5700xt/ryzen machine and I don't know if Amd actively improved Radeon Anti-Lag but I feel like even in high fps/low gpu usage (<75%)/ engine or cpu limited scenarios the engine is noticeably more responsive and consistent, I tried DX11 and DX9 games.

I'm starting to wonder if Battlenonsense's results are actually correct, I've never doubted the reliability of his testings, especially since he agrees with many statements about VRR + fps caps combos coming from Blurbusters itself but it seems like he is the only tester who stated that he measured negative impacts when testing ULL and Anti-Lag.

I'll wait for someone like Fr33thy to make the final input lag tests to prove if there is some actual difference between ULL/Anti-Lag ON/OFF since he is using a more precise way of testing.

I highly recommend RDNA and RDNA2 owners playing at high refresh rates to give Anti-Lag a try, it truly feels like my input are being processed more correctly, less floaty input feeling, the fps reduction penalty(-5% in the games I tested so far) seems absolutely worth the tradeoff.
There is a new video from a5hun on utube called Input Lag Revisited: V-Sync Off and NVIDIA Reflex

User avatar
jorimt
Posts: 2481
Joined: 04 Nov 2016, 10:44
Location: USA

Re: LLM On vs ultra

Post by jorimt » 25 Jan 2021, 09:37

axaro1 wrote:
23 Jan 2021, 10:05
I recently tried to switch back to Radeon Anti-Lag ON with the newest drivers on a 5700xt/ryzen machine and I don't know if Amd actively improved Radeon Anti-Lag but I feel like even in high fps/low gpu usage (<75%)/ engine or cpu limited scenarios the engine is noticeably more responsive and consistent, I tried DX11 and DX9 games.

I'm starting to wonder if Battlenonsense's results are actually correct, I've never doubted the reliability of his testings, especially since he agrees with many statements about VRR + fps caps combos coming from Blurbusters itself but it seems like he is the only tester who stated that he measured negative impacts when testing ULL and Anti-Lag.

I'll wait for someone like Fr33thy to make the final input lag tests to prove if there is some actual difference between ULL/Anti-Lag ON/OFF since he is using a more precise way of testing.

I highly recommend RDNA and RDNA2 owners playing at high refresh rates to give Anti-Lag a try, it truly feels like my input are being processed more correctly, less floaty input feeling, the fps reduction penalty(-5% in the games I tested so far) seems absolutely worth the tradeoff.
mossfalt wrote:
25 Jan 2021, 05:18
There is a new video from a5hun on utube called Input Lag Revisited: V-Sync Off and NVIDIA Reflex
Indeed, beat me to it. I was ultimately going to post that here in a reply as supplement, as there's just not a lot of formal testing available on render queue-related input lag at the moment:

phpBB [video]


Bottom-line though, is with driver-level settings (such as LLM, where supported) in gpu-bound scenarios, you'll typically get about 1 frame reduction, and with game-level settings such as Reflex, you'll get closer to 2 frames reduction (but you can achieve this with a standalone manual FPS limit that prevents max GPU usage as well).

As for non-gpu-bound situations, it can depend, but you're usually not going to see a meaningful increase or decrease in input lag reduction when using any of these settings that manipulate the render queue, as it typically doesn't become saturated enough to make a difference.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: ASUS PG27AQN, LG 48CX VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)

Alpha
Posts: 132
Joined: 09 Jul 2020, 17:58

Re: LLM On vs ultra

Post by Alpha » 27 Jan 2021, 12:49

jorimt wrote:
25 Jan 2021, 09:37
axaro1 wrote:
23 Jan 2021, 10:05
I recently tried to switch back to Radeon Anti-Lag ON with the newest drivers on a 5700xt/ryzen machine and I don't know if Amd actively improved Radeon Anti-Lag but I feel like even in high fps/low gpu usage (<75%)/ engine or cpu limited scenarios the engine is noticeably more responsive and consistent, I tried DX11 and DX9 games.

I'm starting to wonder if Battlenonsense's results are actually correct, I've never doubted the reliability of his testings, especially since he agrees with many statements about VRR + fps caps combos coming from Blurbusters itself but it seems like he is the only tester who stated that he measured negative impacts when testing ULL and Anti-Lag.

I'll wait for someone like Fr33thy to make the final input lag tests to prove if there is some actual difference between ULL/Anti-Lag ON/OFF since he is using a more precise way of testing.

I highly recommend RDNA and RDNA2 owners playing at high refresh rates to give Anti-Lag a try, it truly feels like my input are being processed more correctly, less floaty input feeling, the fps reduction penalty(-5% in the games I tested so far) seems absolutely worth the tradeoff.
mossfalt wrote:
25 Jan 2021, 05:18
There is a new video from a5hun on utube called Input Lag Revisited: V-Sync Off and NVIDIA Reflex
Indeed, beat me to it. I was ultimately going to post that here in a reply as supplement, as there's just not a lot of formal testing available on render queue-related input lag at the moment:

phpBB [video]


Bottom-line though, is with driver-level settings (such as LLM, where supported) in gpu-bound scenarios, you'll typically get about 1 frame reduction, and with game-level settings such as Reflex, you'll get closer to 2 frames reduction (but you can achieve this with a standalone manual FPS limit that prevents max GPU usage as well).

As for non-gpu-bound situations, it can depend, but you're usually not going to see a meaningful increase or decrease in input lag reduction when using any of these settings that manipulate the render queue, as it typically doesn't become saturated enough to make a difference.
What a well done video, thank you for dropping the link to the find.

There are a few things that I question and I'm grateful for your expertise. I am going to point these questions directly at Warzone scenarios, the PC I am referencing has a 5950X, 3090 FTW3 Ultra w/ KingPin Bios. I never see more than low 60c. Mem @ 3733 CL14 1T. I play @ 1080p on a 360hz display. I leave GSync on no VSync. I know it's supposed to be on but it didn't feel right to me lately. I used to use Vsync Fast with Gsync and it seemed great. My personal opinion has changed and I just leave VSync off now.

1. My understanding was LLM's set and forget was On, but truly game dependent. If this is changed to Ultra or left to On does that impact Reflex since one is driver level and the other sounds like its built into the game? I would think leave LLM to Ultra and just use whatever boost option.
2. Aren't most competitive games CPU bound @ 1080p? Should I be using Reflex Enabled + Boost? Nvidia seems to suggest on newer arch Boost can deliver better results.

I pose these questions because I have noticed that for whatever reason, Rebirth Island I can easily see a 200+ FPS drop specifically top of prison. On main map, Down Town I'll see major latency with the only thing that is standing out is my CPU Time jumping from 2-3ms to being around 9ms. For those that may stumble across this making sure in Bios above 4g and rebar as a result, is disabled helped a bit and I have not looked to see what my GPU load is but what do you make of this? I have come to the conclusion its a trash game, paid to win, activision is snakes with their legit cheating patents but I still have fun memeing and it is what it is in the everyone gets a trophy era. No tot to digress lol, it does make me curious on why this performance hit happens. I have never seen anything like it on any other game. This happens with my 5900 and same with my 5950X. I don't recall seeing this on my 2080 Ti ever but also the game has steadily declined in terms of performance maybe due to train, subway and other changes. I played with the Adv_Options.ini for memscale and worker count but it makes little to no difference. Even newly imaged machine shows same behavior. Part of me was wondering if its just buggy Nvidia drivers that have been giving a lot of people issues lately especially those creating content or a setting in game maybe bios that I guess only impacts Warzone. COD is also hot garbage so not discarding that.

User avatar
jorimt
Posts: 2481
Joined: 04 Nov 2016, 10:44
Location: USA

Re: LLM On vs ultra

Post by jorimt » 27 Jan 2021, 14:44

Alpha wrote:
27 Jan 2021, 12:49
1. My understanding was LLM's set and forget was On, but truly game dependent. If this is changed to Ultra or left to On does that impact Reflex since one is driver level and the other sounds like its built into the game? I would think leave LLM to Ultra and just use whatever boost option.
Reflex overrides LLM settings, regardless of what they're set to. It's intended as a replacement, so it doesn't matter what you have LLM set to if Reflex is engaged in the game.
Alpha wrote:
27 Jan 2021, 12:49
2. Aren't most competitive games CPU bound @ 1080p? Should I be using Reflex Enabled + Boost? Nvidia seems to suggest on newer arch Boost can deliver better results.
Typically yes, especially at lowest settings. As for "Boost," it's the same thing as enabling "Prefer Maximum Performance," and tries to ensure the GPU doesn't downclock during gameplay.
(jorimt: /jor-uhm-tee/)
Author: Blur Busters "G-SYNC 101" Series

Displays: ASUS PG27AQN, LG 48CX VR: Beyond, Quest 3, Reverb G2, Index OS: Windows 11 Pro Case: Fractal Design Torrent PSU: Seasonic PRIME TX-1000 MB: ASUS Z790 Hero CPU: Intel i9-13900k w/Noctua NH-U12A GPU: GIGABYTE RTX 4090 GAMING OC RAM: 32GB G.SKILL Trident Z5 DDR5 6400MHz CL32 SSDs: 2TB WD_BLACK SN850 (OS), 4TB WD_BLACK SN850X (Games) Keyboards: Wooting 60HE, Logitech G915 TKL Mice: Razer Viper Mini SE, Razer Viper 8kHz Sound: Creative Sound Blaster Katana V2 (speakers/amp/DAC), AFUL Performer 8 (IEMs)

Post Reply