Osvaldo Pinali Doederlein

Just heard Bill Kristol, former chief of staff for Dan Quayle, say he thinks ICE should be defunded and that the entire DHS should probably be disbanded, which puts him to the left of maybe the majority of elected Democrats. Truly wild times we’re living in.

DOOM: Path Tracing and Bechmarks DOOM: The Dark Ages finally brought together all my computer hobbies: rendering tech, benchmarking, and killing monsters. The game at…

New Blog! DOOM: Path Tracing and Bechmarks. Built-in benchs, RT vs PT, upscaling. The perfect fix for the long US holiday!

cc/quotes: @xdadevelopers.bsky.social @techspot.com.web.brid.gy @digitalfoundry.bsky.social @dsogaming.com.web.brid.gy @pcghde.bsky.social

medium.com/@opinali/doo...

Not fair to Giuliani, any mafia boss would vouch for his experience.

Clinton, Clyburn, Summers, Palantir, Bloomberg, NYT, Wall St, AIPAC...all tried their best.

The entire establishment went full racist and Islamophobic in the end.

Still Mamdani won.

Biggest F U to the Democratic Establishment and 1% since AOC win.

Wow.

9070 XT and RTX 5080: Overclocking This is Part V of a series of reviews: Part I for the RTX 5080, Part I for the 9070 XT, combined Part III with Microbenchmarks, Part V for…

New blog! 9070 XT and RTX 5080: Overclocking

And that ends this dual-GPU series I guess. This one has a more limited scope since OC is very SKU-dependent, but the main goal was probing the limits of each GPU.

medium.com/p/9070-xt-an...

GravityMark Revisited One year and a half ago, and one GPU-generation ago, in Dec 2013 I published RDNA 3, GravityMark and Linux. TLDR: this new benchmark had…

New Blog! GravityMark Revisited. A refresher on this interesting benchmark with new GPUs (9070 XT and RTX 5080) and improvements in the benchmark itself.

cc/quotes: @chipsandcheese.bsky.social, and Sir Isaac Newton

medium.com/p/gravitymar...

*Squish*

Ant? Where? I only see a small, dark-red blotch on the floor

ALERT: I had not seen this article until this morning. It is a pretty big deal. Mike Waltz isn’t actually using Signal. He’s using a copy or third party version of Signal created by a company called TeleMessage the point of which is to allow archiving of messages. The app claims to preserve …

Totally called it. Not an impressive prediction of course, the easiest way to predict the future is to bet that Fascists will do Fascist things.

bsky.app/profile/opin...

Millions of Catholics waiting, again.

And even the unfaithful among us... photo by my wife from our travel in 2015, Piazza San Pietro.

Here's one term that aged like milk. It's curious the all-time peak is early March, 2X previous peak of Jul'18 but collapsed already.

We folks with "TDS" have been calling Trump a criminal, Fascist, autocrat, incompetent, Putin's puppet, etc. since 2016. And we were 100% right.

9070 XT and RTX 5080: Undervolting This is Part IV of a series of reviews: Part I for the RTX 5080, Part I for the 9070 XT, combined Part III with Microbenchmarks.

New Blog! 9070 XT and RTX 5080: Undervolting.

Warning: GPU gore. Some long rants (skipping allowed).

cc/quotes: @techspot.bsky.social @gamersnexus.bsky.social @techpowerup.com

medium.com/p/9070-xt-an...

Larry David writing a NYT guest essay as a man having dinner with Hitler, just to roast Bill Maher as a smug, self-important clown, is the stuff of comedy legend.

www.nytimes.com/2025/04/21/o...

And that leaves an open question, is there anything equivalent for NVIDIA GPUs? Cool Windows UI, one-click install, one-click finding and adding models, no searching and following long tutorials to install/config Python crap. But offering optimal performance with CUDA/TensorRT.🔚

BTW I don't focus much on Amuse by itself but from my non-AI-expert taste this v3 is fantastic. It makes using generative AI models incredibly easy, it's feature-packed, fast and reliable. Nothing else I know is even close in UX; AMD's marketing scored with this partnership.

AI Hardware vs. AI Software RDNA 4 is a big milestone for AMD’s efforts in AI on client: for gaming with FSR 4.0 and more coming later, but also for general AI…

New blog! AI Hardware vs. AI Software: Here, I take a look at the @Amuse_AI app to check how much RDNA 4 moves the needle in general AI performance for Radeon. Spoiler alert, things look very promising for AMD here. 🧵

medium.com/p/ai-hardwar...

Hey NVIDIA where's my bugfixes
Hey AMD where' my next batch of FSR4 enablement

Bonus 1: Hey Microsoft where's the promised DX12 update with DXR 1.2, Neural Rendering, DirectSR

Bonus 2: Hey MachineGames where's promised Indy patch with RTX Hair, was that just Blackwell vaporware

PREDICTION: Before the end of 2026 / midterms, there will be public school districts in red states forced to teach children that the 2020 election was "rigged" against Trump as the Dear Leader continues to insist, claim publicly, and demand agreement as fealty test from everyone working for him.

The logic for all the people in that 80% bar is simple: it would be great if my city had an extra 10K manufacturing jobs, lifting the economy with more consumers = more opportunity for jobs that *I* would accept, anything from a dentist's secretary to a hedge fund manager depending on my luck.

And that's even assuming MSRPs are real or comparable, when real street price is a higher premium over MSRP for NVIDIA not only now near launch with bad supply but very likely to maintain that gap for the whole generation otherwise AIBs would have no margin.

Navi 48 also has two full encoders now, not sure of details but they claim 2X AV1 throughput. Those uarch comps are very hard for other reasons, for one thing NV spends a lot of extra silicon on each Tensor Core and each RT Core. There's a good argument that MSRP is the only fair measure of "tier".

I know the naming is a mess on NV. Stil I wouldn't say 5080 & 9070XT are same tier. There's a huge gap in bandwidth, +6% die not nothing, the tx density factor is unclear at this point (e.g. whether dark silicon is counted on each side), AMD's shader count is way lower if you don't count dual-issue.

You forget that the 9070 series is not in the 80 class? In terms of hardware resources and price the XT its a good match for the 5070's where only the Ti (i.e. cut-down 5080) is a solid win and in RT only. But tbf the 5080 is also kinda fake 80-class 😁

I ignore engagement-spam "Likes", but this one gave me pause. Someone selling Tarot stuff.... not really the focus of posts🤣 BUT it's Tarot with AI and Quantum Software! Maybe they genuinely like my gaming/tech content.

(Nah, just spam triggered by keywords like "magic" in my last post.)

BTW that's the nemesis of all upscalers: movement without motion vectors. I suppose the flashlight is largely a shading effect, it's not a "thing" with its own geometry and proper motion. Non-AI upscalers like FSR2/3 used to suffer a lot with this but even top AI can struggle, esp in low contrast. 🔚

Here's ghosting, in the middle of flashlight's projection.

These issues are now arguably rare enough and small enough to not matter anymore.. still, I noticed those things while just playing normally, without "looking for it" I swear. 4K on 32" = you see every minor fault lol.

This is a good reminder that AI ≠ magic, the Transformer models (both DLSS4 and FSR4) are awesome but not perfect. I was recently playing TLoUp2 at 4K with DLSS4/Quality i.e. "easy mode" for any upscaler, and I could see minor issues like ghosting and moire, both on snow:🧵

But it's still early for a precise verdict. RDNA 4 is on its launch driver if you only count WHQL; god knows NVIDIA also needs to work on drivers. Both will also have a fine-wine curve ahead with many features that depend on game integrations, all of them with NVIDIA one step ahead in adoption.

Same game/test, confirms 9070 XT = RTX 5080 but also shows in prev gen Radeon was a full tier behind with 80-class SKUs losing to NVIDIA's 70-class etc. At the top the XTX only beats lower-tier 4080S because this game likes bandwidth, but even that doesn't help enough for the 7900 XT vs 5070 ti.

True in RT, also aspects of AI-even Raster to a lesser extent by fixing some weak points-but not everything, there are areas where RDNA 4 overperforms both gen-on-gen & vs Blackwell. Here's a good example not from me; Raster-only mode but that's one test where you can't say result is just catch-up.