NVIDIA GeForce 536.67 WHQL Drivers Released

[btarunr]

NVIDIA today released the GeForce 536.67 Game Ready drivers. The drivers introduce support for the GeForce RTX 4060 Ti 16 GB graphics card that's being launched today. The drivers also add optimization for "Portal: Prelude RTX" Among the couple of issues fixed with this release are a bug with GeForce Experience Freestyle filters that were causing games to crash; and increased DPC latency observed in Latencymon for RTX 30-series "Ampere" GPUs. Grab the drivers from the link below.

DOWNLOAD: NVIDIA GeForce 536.67 WHQL



Gaming Technology

• Introduces support for the GeForce RTX 4060 Ti 16 GB.

Game Ready

• Portal: Prelude RTX

Fixed

• Applying GeForce Experience Freestyle filters cause games to crash [4008945]
• Increase in DPC latency observed in Latencymon for Ampere-based GPUs [3952556]

Known Issues

• For notebook computers, issues can be system-specific and may not be seen on your particular notebook.
• [Halo Infinite] Significant performance drop is observed on Maxwell-based GPUs. [4052711]
• [Battlefield 2042] Game stability can decrease when applying GeForce Experience Freestyle filters. [4170804]
• This driver implements a fix for creative application stability issues seen during heavy memory usage. We've observed some situations where this fix has resulted in performance degradation when running Stable Diffusion and DaVinci Resolve. This will be addressed in an upcoming driver release.

View at TechPowerUp Main Site

 

[Slizzo]

Wooooooow. DPC fix is finally in!

 

[Verpal]

DPC fixed yay.

Only on Ampere though..... would have guessed NVIDIA focus on fixing Ada first, I guess the root cause is different?

 

[P4-630]

NVIDIA Graphics Firmware Update Tool for DisplayPort 1.3 and 1.4 Displays

Download English (U.S.) drivers for NVIDIA hardware - , , ,

www.nvidia.com

​

NVIDIA Graphics Firmware Update Tool for DisplayPort 1.3 and 1.4 Displays​

Release Highlights​

To enable the latest DisplayPort 1.3 / 1.4 features, your graphics card may require a firmware update.

Without the update, systems that are connected to a DisplayPort 1.3 / 1.4 monitor could experience blank screens on boot until the OS loads, or could experience a hang on boot.

The NVIDIA Firmware Updater will detect whether the firmware update is needed, and if needed, will give the user the option to update it.

If you are currently experiencing a blank screen or hang on boot with a DP 1.3 or 1.4 monitor, please try one of the following workarounds in order to run the tool:

• Boot using DVI or HDMI
• Boot using a different monitor
• Change boot mode from UEFI to Legacy; or Legacy to UEFI.
• Boot using an alternate graphics source (secondary or integrated graphics card)

Once you have the tool downloaded, please run the tool and follow the on-screen instructions.

Supported Products​

NVIDIA TITAN Series:
TITAN X (Maxwell), TITAN X (Pascal), TITAN XP

GeForce 10 Series:
GeForce GT 1030, GeForce GTX 1050, GTX 1050Ti, GTX 1060, GTX 1070, GTX 1070Ti, GTX 1080, GTX 1080Ti

GeForce 900 Series:
GeForce GTX 950, GTX 950Ti, GTX 960, GTX 970, GTX 980, GTX 980Ti

GeForce 700 Series:
GeForce GTX 745, GTX 750, GTX 750Ti

Quadro Series:
Quadro Maxwell and Pascal products may be impacted. For support and additional details, contact OEM/Channel partner. If further assistance is required, visit https://support.nvidia.com.

 

[wolf]

Interesting content on RTX IO / GPU Decompression, keen to test Portal prelude myself and R+C soon after launch

 

[evernessince]

Wooooooow. DPC fix is finally in!

Not really, they only purport to do so for Ampere cards. That's still leaves Ada and every other Nvidia GPU generation for a long long time with the issue.

 

[trparky]

It might just be a placebo effect, but it seems that with this updated version of the driver, animation in Windows seems to be a whole lot smoother.

 

[Solaris17]

DPC fixed yay.

Only on Ampere though..... would have guessed NVIDIA focus on fixing Ada first, I guess the root cause is different?

This is the correct and important differentiation. It’s ONLY fixed for 3xxx series cards. The problem still exists

 

[Deleted member 232534]

Fixed

• Increase in DPC latency observed in Latencymon for Ampere-based GPUs [3952556]

i see no difference in Latencymon

RTX A2000 6 GB

 

[THU31]

i see no difference in Latencymon

RTX A2000 6 GB

What kind of latency are you seeing?

 

[Deleted member 232534]

BTW This is after setting manually the affinity to a less used cpu. By default it would load cpu 0 to the extend the bar in latencymon would be even red.

 

[Mussels]

What the heck Nvidia

Fast sync caps the game FPS to the monitors maximum refresh rate [4114157]

Be nice if they explained the latency issue, because it doesnt affect everyone

 

[THU31]

Be nice if they explained the latency issue, because it doesnt affect everyone

I mean, I don't really know in what use cases latency of ~450 us is a problem. I had 600-700 on my 3080 and I didn't notice any issues.
The only problem I had was with adaptive/optimal power management. Every time the power mode changed, the latency was spiking to 2000-3000 us. Those power management options were causing huge stutters in games with low GPU load, and even during video playback when the memory clock kept changing constantly between 405 and 810 MHz (P8 and P5 modes).

On my 4070, the latency is usually below 300 us, most of the time below 200. But I have Prefer max performance forced constantly, even for internet browsing.

The Network Driver Interface Specification spikes to ~800 sometimes. I don't think it causes anything bad, though.

 

[chrcoluk]

Wonder if the DPC fix will fix the slowdown to windows desktop that the Nvidia drivers cause vs using iGPU.

 

[Mussels]

Oh this is gold
The higher DPC latency? It's an intel problem and windows 11 is putting in the work to minimise it (possibly at the expensive of other CPU designs, in outlier cases)

Windows 11 vs 10: Intel CPUs apparently sluggish, and it's worse without Microsoft's help - Neowin

11/12/13 gen suffer the issue since the 'thread director' was adjusted in the CPUs, apparently


inter-core latency is believed to be the issue

Cause the 13th gen... doesnt do so well. Data within the one core is faster, as you'd expect with the newer gen CPU - but all core to core latency is worse at every step, averaging about 50% worse

 

[Deleted member 232534]

On my 4070, the latency is usually below 300 us, most of the time below 200. But I have Prefer max performance forced constantly, even for internet browsing.

what is power consumption at desktop with this setting?

for A2000 (Ampere-based GPU) it changes from 15watt to 30watt. (2 monitors connected to GPU)

 

[THU31]

what is power consumption at desktop with this setting?

for A2000 (Ampere-based GPU) it changes from 15watt to 30watt. (2 monitors connected to GPU)

30 W idle, 35-40 when watching videos in a browser, MPC-HC with madVR can push it to 50 W with low-res content (3080 was doing 110 W).

It's the only way to get perfect stutter-free video playback. The power increase is irrelevant to me. I only care about power during gaming because of heat output.

 

[Mussels]

30 W idle, 35-40 when watching videos in a browser, MPC-HC with madVR can push it to 50 W with low-res content (3080 was doing 110 W).

It's the only way to get perfect stutter-free video playback. The power increase is irrelevant to me. I only care about power during gaming because of heat output.

thats rough

my 3090 is 12W idle and my entire PC with a 32" monitor doesn't use that for 4K playback

Looks good here on AMD/3090

seems i can be smug that i'm faster than 10th and 13th gen intel in DPC latency

 

[trparky]

Oh this is gold
The higher DPC latency? It's an intel problem and windows 11 is putting in the work to minimise it (possibly at the expensive of other CPU designs, in outlier cases)

Windows 11 vs 10: Intel CPUs apparently sluggish, and it's worse without Microsoft's help - Neowin

I read about this this morning and watched some videos on YouTube from TechYesCity. I didn't know what to make of it. Does Ryzen suffer from the same latency penalties?

 

[Tomorrow]

View attachment 305361
View attachment 305362

Lucky you. Mine is as bad under Win10 as it is under Win11:

Spoiler: 4 big images

RTX 2080 Ti. Latest WHQL drivers. No manual affinity.

 

[THU31]

Lucky you. Mine is as bad under Win10 as it is under Win11:

RTX 2080 Ti. Latest WHQL drivers. No manual affinity.

I would assume this was under a very light load or typical desktop usage, not under heavy gaming load?

Set "Prefer max performance" power management in the NV Control Panel, restart and check the latency then. I'm almost sure you'll get nowhere near 1000 us. Or check a game with Prefer max performance, so that the power state is always P0.

 

[Tomorrow]

I would assume this was under a very light load or typical desktop usage, not under heavy gaming load?

Set "Prefer max performance" power management in the NV Control Panel, restart and check the latency then. I'm almost sure you'll get nowhere near 1000 us. Or check a game with Prefer max performance, so that the power state is always P0.

Correct. Mostly light tasks on desktop. I already have prefer max perf set. Also all tweaks that come with nvcleanstall are enabled such as MSI interrupts etc.
I will try a game with latencymon running in the background...

 

[Mussels]

I read about this this morning and watched some videos on YouTube from TechYesCity. I didn't know what to make of it. Does Ryzen suffer from the same latency penalties?

Not that i ever saw - theres always spikes from various activities, but i never experienced what others were complaining about - now i know why
The key is that intel 12/13th gen have 50%+ latency between cores, even if its just within the single set of 8 P-cores, and it's showing up in places like Nvidias drivers, that apparently relied on the low latency

Educated guess that it's now relying on a software scheduler and not a hardware one (that parts not a guess), so theres a time penalty where it has to process where to move threads over to - so you get good performance in multi threaded tasks, but theres a % chance of a slowdown/stutter if things go to sub-optimal places and need to get moved around.

Lucky you. Mine is as bad under Win10 as it is under Win11:

Spoiler: 4 big images

View attachment 305642
View attachment 305643
View attachment 305644
View attachment 305645

RTX 2080 Ti. Latest WHQL drivers. No manual affinity.

you don't have system specs, but if you wanna post a zentimings screenshot i've got a theory about what might be going on there
mostly because of this comment

Mostly light tasks on desktop

You arent on the windows power saving plan are you? Latest BIOS?

 

[Tomorrow]

you don't have system specs, but if you wanna post a zentimings screenshot i've got a theory about what might be going on there
mostly because of this comment

You arent on the windows power saving plan are you? Latest BIOS?

Im on the hidden Ultimate Performance plan. BIOS is pretty much the latest. There is F37e released just a few days ago but it's based on the same AGESA version as F37b.

 

[Mussels]

that ultimate plan is not a good idea, it's hidden for a reason.
You're breaking the cores being able to sleep, which harms their boost performance.


You should avoid things like that in the future, as they break a lot more than they fix - they often come paired with lots of other tweaks that only help reduce the damage tweaks like that power plan cause in the first place.



As someone running even higher IF speeds than you, it does sound like the traditional infinity fabric dropout - which component drops depends on the board and CPU, but it can be CPU, PCI-E lanes, USB or SATA.

Your VDIMM is really high, which adds the risk of simple RAM instability being the cause too. B-die doesnt like heat at all, and can get unstable as low as 40c with such high voltages - it only takes a few correctable errors to make a stutter, or the high power draw to cause an IF dropout
I'm pretty sure you need to raise other voltages at the same time as VDIMM, but 1.60v is so high i'm not sure it's even safe to do so.

first google result for B-die on AM4 confirmed that idea

There are also people with similar RAM that's hynix, which behaves entirely differently
Bought Viper Steel 4400MHz c19 and turns out to be HYNIX die. Have anyone experienced the same thing? : overclocking (reddit.com)

B-die should be able to do up to 1.55v on decent cooling. That’s at least the point at which I start see negative scaling

Your problems not GPU driver related, it's that you're running some extremely uncommon and not recommended tweaks in hardware and software.