RTX 3070 Driver recommendation

[crack]

Oh, right. I can test DOOM Eternal for Vulkan workload, D3D11 Borderlands GOTY Remastered, DX9 S.T.A.L.K.E.R. Shadow of Chernobyl and CoD1 for OGL. What time per testing session should I go for? 30 mins? more? less? right now I don't have time to test.

You don't need to do long sessions, shorter intervals are sufficient.
You'd set the duration to be exactly the length of the benchmark you're targetting.
The main problem is finding a repeatable testing environment / benchmark.
For example: CS2 (D3D11) benchmark maps, OW2 (D3D11 & D3D12) benchmark maps — HEZHJ & D3KVN can be good games to test driver performance in.

You could also find a D3D12, D3D9, OGL & VK title from here:
https://www.pcgamingwiki.com/wiki/List_ ... benchmarks
https://www.pcgamingwiki.com/wiki/List_of_OpenGL_games

Here's a few examples of games with a built-in benchmarks (taken from the lists above):
D3D12: Ashes of the Singularity, Shadow/Rise of The Tomb Raider, CoD MW3 (2023), RDR2
D3D9: Arma2, UT2004, GTA IV, CS:S, NFS Carbon, FEAR, Batman Arkham Asylum, Mafia II (original)
OGL: Xonotic, Minecraft, UT2004 (custom), Doom 3
VK: Wolfenstein Youngblood, Doom The Dark Ages

Remember that you are interested in the CPU overhead, meaning you don't want your GPU to be limitation here.
Use lowest graphical settings in-game & lowest possible resolution possible, as a baseline)
I would stick to D3D title tests only, as OGL/VK titles are on DXGI interop (replacing GDI) on Nvidia GPUs
https://old.reddit.com/r/nvidia/comment ... _in_55585/

Note: The DPC/ISR driver performance in 1 title may not apply to other titles of the same graphics API.

Got it. To note GTA IV for me, specifically the benchmark is extremely volatile, one time I start it and its stuck on a black screen, another time its infinitely loading, another time it loaded the benchmark, works halfway through and then freezes or crashes, sometimes it freezes or crashes right before results show, and finally it can just work.

It's not in this order, it's completely random.

The game itself and DLC work fine, sometimes I get really trippy graphical artifacts, maybe the game is just really picky about its hardware, on my old XP machine with a C2Q Q6600 and 8600GT it had no issues.

[soul4kills]

Adding a useful tip. Each driver version performs differently in regards to msi enabled/disabled. I found this out recently.

For example, 576.52 is the best firmware to use for my 2060. It's best with msi enabled with interrupt priority high. But with the newer 581.08, it performs better with msi disabled and high priority interrupt. 581.15, performed better with msi enabled, but performed worst than 581.08.

Right now i'm sticking with 581.08 as it's one of the latest ones that performs almost as good as 576.52.

[kyube]

Got it. To note GTA IV for me, specifically the benchmark is extremely volatile, one time I start it and its stuck on a black screen, another time its infinitely loading, another time it loaded the benchmark, works halfway through and then freezes or crashes, sometimes it freezes or crashes right before results show, and finally it can just work.

It's not in this order, it's completely random.

The game itself and DLC work fine, sometimes I get really trippy graphical artifacts, maybe the game is just really picky about its hardware, on my old XP machine with a C2Q Q6600 and 8600GT it had no issues.

I've only mentioned GTA IV due to it's popularity & the inclusion of a built-in benchmark. You don't need to use it in your tests
What you've described could be a symptoms of a larger issue.
Make sure to run these tests without any overclocking applied (disable EXPO/XMP) and on a fresh installation of Windows to rule out any potential HW concerns.

Adding a useful tip. Each driver version performs differently in regards to msi enabled/disabled. I found this out recently.

For example, 576.52 is the best firmware to use for my 2060. It's best with msi enabled with interrupt priority high. But with the newer 581.08, it performs better with msi disabled and high priority interrupt. 581.15, performed better with msi enabled, but performed worst than 581.08.

Right now i'm sticking with 581.08 as it's one of the latest ones that performs almost as good as 576.52.

You should never disable MSI mode on any single driver on your system.
https://www.techonline.com/wp-content/u ... 321070.pdf
From the link:

Results collected using this method across two classes of Intel platforms (Workstation and System on Chip (SOC)) show a 3x reduction in interrupt latency when using MSI as compared to IO-APIC and over a 5x reduction when compared to XT-PIC. MSI greatly reduces interrupt latency and the CPU overhead involved in servicing interrupts.

What does "performs best" mean? What tools were used to come to these conclusions? If it isn't Presentmon-based (e.g.: Frameview) with repeatable data, it's somewhat non-applicative.

[soul4kills]

I don't do benchmarks. If you're interested you can test out my claims for yourself, but I have nothing to provide.

How I know is by playing CS2, and doing a comparison to how it feels and how it looks. To me it's clear because my system is pretty optimized, and found the best driver and best settings that give me a good baseline to test other drivers against. And that's the 576.52 firmware with msi enabled with high priority interrupt. My 2060 is designed to work with msi disabled, only the more recent nvidia video cards are designed to work with msi enabled.

Sometimes things just don't work as they are expected to work. So your hard stop of saying you shouldn't disable msi is bad advice and may keep people from figuring out their latency issues. And that's why this forum exists because things are not working as they are expected to work. I know its subjective but everyone's setup and environment is different so it will always be subjective. What works for one person may not work for another. So it's better to provide advice pointing people in the right direction and encouraging them to test them out for themselves.

But back to how i know it performs better, well because of hit reg, mouse feel, and frame blurring. When it's good, my frame delivery is instant, my frames look crisp and sharp frame to frame if i shake my mouse real quick and don't feel delay to my mouse movement. Which results in me not having to correct my aim. When it's bad, i get blurring frame to frame and mouse feels delayed and constantly need to correct my aim.

But the only way to know is solving all your latency issues in the first place to be able to compare, if you don't have that you won't be able to tell.