Yes, it’s system wide. You can set your prefered way in Nvidia control panel->global settings-> cuda systemem fallback policity.

Driver default is prefer systemmem fallback, which means it’s going to offload to RAM instead of crashing when VRAM is full.

No System Mem fallback is basically the old memory management, it crashes once your VRAM is full.

Great test!

Unfortunately the Llama 2 Chat template is completely broken in SillyTavern. It not only uses a new line as separator instead of the correct one, but also ends the prompt after the system prompt with the input sequence [INS] instead of [/INST] if you are using the vector storage or an example dialogue. You can see for yourself by comparing the output to what the format should look like.

So these Airoboros 3.1.2 tests are unfortunately borked. Still though, interesting result for the other models.

That is indeed pretty interesting. Yes, usually the 2080Ti is a little bit below a 3070 but i this game, it’s 14% slower. I guess some refinements were made with Ampere and the highter throughput probably matters too. Still, not bad at all for a 5 year old card and it handily beats the 6700XT as well. I think buyers got their money’s worth with the 2080Ti. It’s longevity is kinda insane, especially for Raytracing and due to its 11 GB frame buffer.

They truly over exaggerate it, for whatever reason. DLSS at 1080p is perfectly fine and I think in many cases, it’s superior to TAA as it has more detail and better motion stability, even the lower presets.

The only reason its blurry in Alan Wake 2 is because the sharpening pass is disabled. But you can easily add sharpening yourself.

That is a very good guess. I think that’s exactly what is happening here. Definately interesting.

RDNA1 is definately better at async compute.

Why are you complaining about Remedy when AMD was the one selling a GPU with an outdated featureset 4 years ago? Remember Turing is 5 years old and supports mesh shading.

This entirely AMDs fault.