It's true but notice Nvidia didn't make dlss 4.5 fp8 exclusive to Blackwell and allowed rtx hardware to run all models. Yes the newer Transformer models like m and l run more efficiently on newer hardware but I believe this is more of marketing the newer hardware for those on older hardware. When AMD brings fsr 4.1 or newer to older hardware, gamers will be on a cross road upgrade my hardware to run fsr 4.1 more efficiently or take the hit especially if they running the higher end 6800 to 6950XT gpus as well higher end 5000 series gpus. AMD gets street credit while making their newer Gen flagships more attractive to previous Gen owners. It's a win/ win situation imo.
DLSS 4 using preset K is still the best choice on the Nvidia side overall. RTX 2000/3000 has full support and very limited perf hit here. Preset M is mostly for 4K/UHD+ and optimized for performance preset, no 1080p/1440p gamers should be using preset M, unless they are clueless. RTX 4000/5000 is better at handling preset M.
However, 95% of PC gamers, uses 1440p or lower. Therefore they should be running preset K, which is optimized for DLSS Quality/Balanced mode. M is optimized for DLSS Performance mode. No-one should be using Performance mode at 1440p or lower, unless they seriously lack performance.
Transformer model is K too. Limited performance penalty on RTX 2000/3000 due to these have Tensor cores that do the calculations, not the GPU, like on the AMD side for Radeon 6000 (sometimes in 2027 you will see, FSR 4 on Radeon 6000 will likely make the GPUs buckle due to FP8 emulation work, using INT8 - Extremely taxing for the GPU).
Radeon 5000 and 6000 don't have any dedicated AI cores of any sort. They will be dirt slow with FSR 4. These GPUs only have regular GPU cores, with full rasterization focus.
7000 had matrix cores (WMMA instructions) and 9000 series had AI cores - These 2 generations has dedicated hardware to handle FSR 4 "well" - Nvidia had this since RTX 2000 series from 2018, almost 10 years ago. Tensor cores. That is why DLSS 4 works on any RTX card, since day one, while AMD is struggling and working hard to bring FSR 4 to older GPUs. However, AMD probably held FSR 4 exclusive to Radeon 9000 till now, to sell cards. With no FSR 4 exclusity, Radeon 9000 lineup would have looked miserable at the reveal.
Radeon 6000 owners don't need proper upscaling sometime in 2027, they needed it years ago. RDNA 2 GPUs are dirt slow at native today. Meanwhile RTX 2000/3000 owners are smooth sailing still, due to DLSS 4. DLSS has been worth using since 2020, when DLSS 2 released (DLSS 1 was pure garbage, just like FSR 1, 2 and 3). DLSS 2 came out almost 7 years ago now. FSR first became relevant with FSR 4, released in Q2 2025. Facts. 6 years too late really.
Good upscaling + RT capable (many games have forced RT elements and had for years), even with less overall VRAM, aged far better than: Bad upscaling (FSR 1 (completely useless) + FSR 2 and 3) and more VRAM.
Meaning RTX 3000 aged massively better than Radeon 6000 series, both are 6+ years old now. Upscaling secured good longevity on the Nvidia side. DLSS 2 was great on launch and only improved since then, DLSS 4 is pretty much upscaling-perfection with massive game support + mods)
16GB on Radeon 6800/6900 sounded nice, sadly the GPU arch itself was miserable and could not handle RT elements or deliver good upscaling. Aged like milk due to this. Game development changed approach. The arch was not ready for the future. AMD only focussed on raster perf with Radeon 5000/6000, therefore aged badly. Regardless of having "plenty of VRAM"
VRAM alone never saved a weak GPU. VRAM is not as important as people think, and memory compression improves gen to gen anyway. You just need "enough" and then you will be fine.
A good GPU these days, ticks all these boxes:
- Good rasterization perf.
- Enough VRAM with good memory compression and cache system.
- Good upscaling and frame gen, with actual support in games.
- Can handle RT without buckling, due to many newer games having forced RT elements.
VRAM is important, but not that important, you can always lower settings to get around this. A fast GPU with a solid feature set, and it will age like wine.
And don't forget about memory compression and bandwidth improvements. An 8GB GPU from 10 years ago, is not the same as an 8GB GPU today. Memory compression improved bigtime. Nvidias cache hit/cache miss system works very good. Hence why Nvidia overall needs less VRAM than AMD. In tons of games, AMD GPUs uses more VRAM
for the exact same settings - This is due to Nvidia having better memory compression and cache system.
Nvidia increased cache bigtime on RTX 4000/5000 series due to this. 10 times more cache or so, overall SKU vs SKU.
People generelly needs to wake up from their slumber - This is 2026 not 2016 - SO many people here still think RASTER PERF and VRAM AMOUNT is the only important metric. ITS NOT. Your GPU will age like MILK without good features like upscaling and frame gen + decent RT perf for the games that force RT elements (pretty much every Unreal Engine 5 game but many more engines has RT elements these days, had for years)
People needs to wake up and accept that game developers changed approach. AMD knows this, and the prime focus with RDNA 4 was improved upscaling and RT performance. AMDs own words. Search and you shall find, they officially said this. And delivered on this promise.
