Poll: Do you care about ray tracing / upscaling?

[WelshBloke]

I'm really struggling to benchmark BL4, mainly because I'm not enjoying it at all!

It runs badly at 3440x1440. If I set it to BadAss mode, no FG, no upscaling it runs at a craptacular 22fps. If I set it to the lowest setting, no FG, no upscaling it runs at a slightly less craptacular 48 FPS.

Basically you have to use upscaling (and that looks shitty) or FG and that feels shitty.
Tbh though it feels shitty anyway. Bullets move really slowly, gun aiming is sluggish and laggy, visually everything is pretty cluttered and you're going to get shot by people who you can't find, plus you have to shoot people a billion times to kill them.

 

[DAPUNISHER]

My son and I couch co-op the first 2 games so many hours that I think it burned me out on the franchise. I got BL3, BL The Pe-sequel, Tiny Tina's WL, all for free from Epic over the years. I started to play the pre-sequel but it was same old same old and I was bored in under an hour. Maybe the bug will bite me again someday.

Try the Hardware Unboxed optimized settings out.

 

[psolord]

I guess you've heard about the FSR4 leak, which led to the creation of a .DLL that can be used on non-RDNA4 gpus.

This guy is testing it on his 6900XT.

There's a lot of rambling as the title says, but the interesting part is at 09:00 where he shows the difference of FSR3.1 vs FSR4 and at 13:00 he gives some performance numbers.

Pirate Yakuza native 4k = 40fps
Pirate Yakuza at 4k with frs 3.1 = 100fps
Pirate Yakuza at 4k with fsr4= 68-75 fps

effin A

Also this guy posted a how to, but the important thing is that he used FSR4 on a freagin....GTX 1650!

I asked chatGPT to devise a table showing how older gpus may fair. It's estimation is based on INT8 performance per gpu generation. For example it explained that the GTX 970 while it can do INT8, each op ties up a full 32-bit ALU. Maybe that's why my 970runs like a dog if I try to use XeSS. But I WILL try the FSR4 mode for laughs. I guess its Radeon 6000 estimations are a bit off! I will test my 6600 as well ofc. xD

Year	NVIDIA	AMD
2006 – GeForce 8 (G80)	First CUDA GPUs, INT8 possible but slow.	R600 / HD 2000: INT8 via shaders only, very slow.
2010 – Fermi (GTX 400/500)	No INT8 acceleration.	TeraScale 2/3: no low-precision accel.
2012 – Kepler (GTX 600/700)	FP32 focus, no INT8 acceleration.	GCN 1.0 (HD 7000): FP16 emulated, no INT8.
2014 – Maxwell (GTX 900, e.g. 970)	INT8 works but extremely slow.	GCN 2/3 (R9 200/300, Fury): no fast FP16/INT8.
2016 – Pascal (GTX 10)	DP4a instruction → real INT8 acceleration in CUDA cores.	GCN 4 (Polaris, RX 400/500): no INT8/FP16 acceleration.
2017 – Volta (Tesla V100)	Tensor Cores (FP16 only). No INT8 yet.	Vega (RX Vega 56/64): rapid packed math (fast FP16). Still no INT8.
2018 – Turing (RTX 20)	Tensor Cores: INT8 & INT4 + FP16. First consumer GPUs with real AI acceleration.	RDNA 1 (RX 5000): FP16 supported, but no INT8/matrix ops.
2020 – Ampere (RTX 30)	Tensor Cores: FP16, BF16, INT8, INT4.	RDNA 2 (RX 6000): no matrix cores, INT8 via shaders only.
2022 – Hopper (H100)	Adds FP8 Tensor Cores.	RDNA 3 (RX 7000): WMMA instructions for INT8/FP16, partial acceleration.
2024 – Ada (RTX 40)	Refined Tensor Cores: FP8, FP16, BF16, INT8, INT4.	RDNA 4 (RX 9000): AI accelerators with real INT8/FP8 hardware.