It feels like we’re living through a sibling rivalry in the GPU world. You know, like those brother-sister fights where each side keeps pulling out flashier tricks to outshine the other. AMD and NVIDIA — the dominant forces in the GPU space — constantly spar using their magic spells: DLSS, FSR, frame generation, super resolution, AI upscaling… all while Intel, the smaller kid in the family, is quietly waving its own wand with XeSS.
At first glance, it’s exciting. New technologies, big promises, and visual wonders. But underneath all that glitter, I can’t help but miss the raw — sorry, raster — performance era. The days when a graphics card’s worth was based on how well it rendered, not how well it faked.
Let me explain.
Back When GPU Power Meant Raster Muscle
There was a time when cards like the GeForce GTX 480, GTX 980 Ti, or the legendary Titan X (Maxwell) made waves not because they had AI cores or Tensor this and Frame Gen that — but because they could brute-force pixels with unmatched raster performance. You paid for horsepower, and you used that horsepower directly.
Even AMD’s Radeon HD 7970 or R9 Fury X had a certain charm. They came packed with raw compute and memory bandwidth — no illusions, no smoke and mirrors. Just straight-up rendering muscle.
Sure, GPUs ran hot and consumed loads of power, but they stood their ground. Plug it in, fire up Crysis, and see how far your silicon would take you. No DLSS to save you, no FSR to blur things over.
Enter the Magic Tricks: DLSS, FSR, XeSS
Fast forward to today, and it’s a different world. NVIDIA’s DLSS (now in version 3.5) is practically a requirement for playing new AAA games at 1440p or higher with RTX effects on. DLSS started as a neat AI upscaling trick, but now it creates frames that were never actually rendered. It’s not even pretending to be about resolution anymore — it’s about synthetic frame generation.
AMD followed suit with FSR (FidelityFX Super Resolution). While it’s less AI-focused than DLSS, it’s still doing something similar — taking a lower-res image and making it look high-res. FSR 3 even adds its own frame generation.
And then there’s Intel, the little one in this family feud, pushing XeSS — its answer to DLSS and FSR. Surprisingly competent, XeSS is compatible with other GPUs too. But again, it’s another magic trick layered on top of what should ideally be optimized rendering.
These techniques are helpful — no doubt. But they also mask poor optimization. Which brings me to the core of what’s been bothering me.
Are We Relying on Tricks Because Devs Stopped Optimizing?
I’ve noticed a trend. Games and even apps are being released unfinished or unrefined, with the assumption that upscaling or frame generation will fix everything.
Look at Starfield. Even on high-end cards like the RTX 4090, it struggles at native resolutions. The Last of Us Part I on PC launched in an embarrassing state. Jedi Survivor? Same story. They all lean heavily on DLSS or FSR just to feel playable.
It’s like studios aren’t targeting native performance anymore — they’re designing around AI helpers from the start.
Even worse, creators feel this too.
It’s Not Just Games — Creative Apps Suffer Too
In video editing or 3D work, raw raster power still matters. I’ve seen performance inconsistencies in apps like:
- Adobe Premiere Pro: Some GPUs struggle with simple 4K timelines unless proxies are used.
- DaVinci Resolve: Fusion tasks still choke on certain cards despite high TFLOPs on paper.
- Blender: Cycles rendering is better now with OptiX and HIP support, but some features rely heavily on driver-side improvements, not GPU grunt.
These apps don’t benefit from DLSS-style tricks. They still rely on how well the GPU pushes compute, memory, and rendering capability. No magic can save you here if the hardware is weak or poorly supported.
New Cards Dominate Easy Games but Still Fail Some Old Ones
What really breaks my brain is when modern GPUs dominate graphically intensive new games (with scaling) — but still choke on older titles without it.
Try running GTA IV maxed out with ENB and high traffic mods. You’ll watch your frame rate crawl, even with a 4080 or 7900 XTX. Same with:
- Arma 3
- Microsoft Flight Simulator (non-scaling mode)
- Crysis Remastered
- Kingdom Come: Deliverance
They don’t have DLSS or FSR baked in. And without optimization or scaling assistance, the GPU’s true raster strength is revealed — often to disappointing results.
Remember SLI? Or the Titans?
People used to flex multi-GPU setups. SLI and CrossFire were messy, yes, but they were also a statement. People wanted more raster power, and were willing to deal with profiles, heat, and power draw to get it.
And then there were the Titan cards — not for everyone, but legendary nonetheless. The Titan Z, Titan Xp, Titan V. These weren’t just graphics cards. They were statements of intent. Overkill, sure — but built to push boundaries.
Now? Even NVIDIA doesn’t talk about Titans anymore. They’ve been replaced by consumer versions of workstation cards (like the RTX 6000 Ada) and the idea of raster overkill has been erased — replaced by AI assist everywhere.
Final Thoughts: I Miss the Real Muscle
I’m not against DLSS, FSR, or XeSS. They’re engineering marvels. But I do miss when GPUs were judged by their brute force — not their ability to trick the eye.
Back then, performance was transparent. If your card couldn’t run a game at ultra, it wasn’t strong enough. Now, we have to ask: “Is it real FPS or generated FPS?”
We used to talk about fill rate, shader throughput, clock speed, and memory bandwidth. Now it’s Tensor cores, AI matrix ops, and upscaling profiles.
It’s all good progress — but I hope we don’t lose sight of what really matters: how strong a GPU is when it stands alone, without magic.
Because sometimes, the magic fades — and when it does, I want my card to still perform.