Recent content by steen2

S
Discussion RDNA4 + CDNA3 Architectures Thread

IIRC, UHBR13.5
- steen2
- Post #8,881
- Feb 28, 2025
- Forum: Graphics Cards
S
Discussion RDNA4 + CDNA3 Architectures Thread

DLSS 3/4 CNN/Trans are mixed precision down to fp16 for backwards compatibility. Ada/Blackwell support fp8 with the latter also supporting fp4 with additional quantization issues. The real question is what model does FSR4 use? CNN, transformer, fp8 weights only...? Do tell.
- steen2
- Post #8,868
- Feb 28, 2025
- Forum: Graphics Cards
S
Discussion RDNA4 + CDNA3 Architectures Thread

DLSS had the added benefit of looking good in static scenes. Motion is blur/occlusion fest. FSR is naturally worse. FSR 3 has had the option for MFG, but not enabled in release drivers. There's some regression J<->K. Work in progress with FP8 depending on E4M3/E5M2 datatypes for AD/GB.
- steen2
- Post #7,445
- Feb 1, 2025
- Forum: Graphics Cards
S
Discussion Nvidia Blackwell in Q1-2025

There's a push-back ready for the lower tier SKU VRAM belly aching. DLSS4 (TNN/MFG) reduces VRAM usage so 8-16GB is "fine". Just wait for the <=12GB SKUs. Tag team effort from technical marketing.
- steen2
- Post #2,934
- Feb 1, 2025
- Forum: Graphics Cards
S
Discussion Nvidia Blackwell in Q1-2025

PR perceives the fake frames backlash as damaging the brand. The damage control push back is that "all frames are fake, so what does it matter if they're generated & not rendered". The usual suspects are peddling the same trope. There are a number of other interesting talking points.
- steen2
- Post #2,893
- Feb 1, 2025
- Forum: Graphics Cards
S
Discussion Nvidia Blackwell in Q1-2025

This is an observed issue with the 4090/80 FE series, & esp seen when PCIE risers are used. Reviewers have been reporting instability to their Nvidia reps. It may be the off PCB signalling due to the modular I/O & PCIE PCBs.
- steen2
- Post #2,813
- Jan 30, 2025
- Forum: Graphics Cards
S
Discussion RDNA4 + CDNA3 Architectures Thread

And it's being unlaunched/relaunched as the 9080XT...
- steen2
- Post #7,110
- Jan 24, 2025
- Forum: Graphics Cards
S
Discussion Nvidia Blackwell in Q1-2025

It was rhetorical. You're conflating reduced precision OPs with Nvidia's measure of "AI TOPS". For ADA it's FP8 vs FP4 for Blackwell. Just like cf DLSS3/FG vs DLSS4/MFG for FPS charts. Marketing well done.
- steen2
- Post #1,919
- Jan 14, 2025
- Forum: Graphics Cards
S
Discussion Nvidia Blackwell in Q1-2025

Isn't that just FP4 vs FP8?
- steen2
- Post #1,891
- Jan 14, 2025
- Forum: Graphics Cards
S
Discussion RDNA4 + CDNA3 Architectures Thread

IIRC, I was contacted 1yr prior via Guru3D forums. Another Guru3D member actually posted the email. I ignored it. The story didn't blow up until it hit B3D, AT, [H], etc. Heh, Gary Tarolli & Simon Fenney discussing the merit (or otherwise) of benchmarketing on usenet trumps anything on B3D...
- steen2
- Post #3,714
- Aug 21, 2024
- Forum: Graphics Cards
S
Discussion RDNA4 + CDNA3 Architectures Thread

Nope. You forget Nvidia's AEG program, the rot had already set in. AT wasn't immune, either... Nvidia had a craw in their throat about B3D pre-dating the DP/BB days. B3D discussion was mainly ImgTec/3dfx/Rendition focused while Nvidia's products/practices usually received scorn even back then...
- steen2
- Post #3,711
- Aug 21, 2024
- Forum: Graphics Cards

RESOURCES

Top Bottom