Osvaldo Pinali Doederlein

Sci-Fi, Gaming, PC hardware, Tech | Unoptimized code is the root of all evil | |

[Osvaldo Pinali Doederlein @opinali]

Hey, time to test your tech🔮. We now know an SDK update is coming July to deliver the first FSR4 backport. Let's call it FFX SDK 2.3.0.

In addition to FSR4 for RDNA3, do you expect/hope this release to have...: 🧵

[AMD GPUOpen @GPUOpen]

Big news for developers! FSR Upscaling 4.1 is coming to AMD RDNA 3 architecture GPUs in July! 🎉

🖥️ Expanding beyond RDNA 4 to Radeon RX 7000 Series.
🔧 FSR SDK update coming in July.
📰 Stay tuned to GPUOpen for more info.

Learn more about the SDK ⬇️

https://t.co/PCMeNx5Kuh

[Osvaldo Pinali Doederlein @opinali]

@paulbatistta Próxima vez tenta ler o artigo antes de falar merda 😉

[Osvaldo Pinali Doederlein @opinali]

@HtooMyatLin3 Whoa forgot to check the Vulkan relnotes. Descriptor heap & device addr exts are huge, with Kilometer-long list of contributors lol everyone and their dog wanted this change in programming model (reminds of of @SebAaltonen's rants). Long vector more directly relevant for FSR4.

[Osvaldo Pinali Doederlein @opinali]

Some interesting changes in Adrenalin 26.6.1, usual story with stuff not in the release notes.
- FurMark2 Knot OpenGL bug gone: 2160p Bench Preset works again (broken in 26.5.1 & 26.5.2)
- GeekBench: OpenCL -3.5%, Vulkan +14.5%
So another round of tweakin' those secondary APIs.

[Osvaldo Pinali Doederlein @opinali]

@MCH2024 I never heard of that? I used to run PR for years, from the 6800XT to early days of 9070 XT and I never thought it misrepresented its RT perf. It scores 83% of my RTX 5080, which is the right ballpark for that moderate level of RT (that bench is from 2019...).

[Osvaldo Pinali Doederlein @opinali]

This big 3DMark update documents a regression for Radeon. I don't even use Port Royal anymore, it's obsolete, but I checked now and indeed my score went from 88fps last time I ran it to 80fps = -9%. Same for drivers 26.5.2 and 26.6.1. Checking later if anything else is impacted.

[Osvaldo Pinali Doederlein @opinali]

@MeyerRants @LateNightCandi_ For a long series like this IMO we get too attached to the protagonist. Again I never played the Greek trilogy but form what I saw in videos, I think the gameplay will change much less from the latest Kratos to Laufey than it did from that old stuff to the Nordic games...

[Osvaldo Pinali Doederlein @opinali]

@MeyerRants @LateNightCandi_ Whoah now you're losing me. A new Tomb Raider with another tomb-raidin' hero would be legit. So what you're saying is that in a series strongly associated with the protagonist, that can never change... OK but then the title is irrelevant, just admit you want to play Kratos again.

[Osvaldo Pinali Doederlein @opinali]

@MeyerRants @LateNightCandi_ And Kratos is not "the" God of War in Nordic pantheon / series. He doesn't even want given his change of mind. Even in the Greek stories the real GoW is Ares so you could say Kratos was only "one" GoW not the owner of the title. (Never played the greek series so no idea of lore)

[Osvaldo Pinali Doederlein @opinali]

@MeyerRants @LateNightCandi_ Oh also The Witcher IV coming, people already complaining of changed protagonist; in this case the title was his title and Ciri is a Witcher so the name is still 100% kosher. Laufey is a God(dess) and already stablished by lore that she's a war badass as much as Kratos... /

[Osvaldo Pinali Doederlein @opinali]

@LupintheI Yes and no?... They have tests co-produced by NVidia, Intel & AMD (upscaling tests), and AFAIK all major tests like SpeedWay, Steel Nomad etc have significiant input from all major HW vendors. Upcoming Path Tracing bench is actually sponsored by Thermal Grizzly. TimeSpy was Galax

[Osvaldo Pinali Doederlein @opinali]

Adrenalin 26.6.1 is out, nothing special apparently, support for a couple new games and some fixes.

There's also a big update of 3DMark and it's all about Windows-on-ARM which reminds me that the whole industry needs to divert some effort to support that turd.

[Osvaldo Pinali Doederlein @opinali]

@MCH2024 Apple made that transition but it's a comically tiny ecosystem in both SW & HW compared to Wintel. And single-vendor monopoly, once Apple migrates there's no new computers or popular first-party apps supporting old CPU. Windows will never have that luxury, x86 will rule forever.

[Osvaldo Pinali Doederlein @opinali]

@MCH2024 Strictly = BS. ISA ≠ Impls. Agree it's worth-have on laptops... on Macs, maybe Linux too. Win32 = 33yr legacy. Windows on ARM for clients will always be a tradeoff. Apps that need emulation, HW that doesn't have a good native driver, random obscure stuff that doesn't work.

[Osvaldo Pinali Doederlein @opinali]

Hey @GPUOpen @AMDGaming, this was a cool slide and nice to see this moving.

But frustratingly, your catalog of FSR support is frozen since December with 260 games on FSR4. That should be updated periodically, maybe as part of Adrenalin release process.
https://t.co/YCgFNT8ma5

[Osvaldo Pinali Doederlein @opinali]

@SebAaltonen Note also the IRL kernel code returns -ENOMEM if it succeeds to allocate memory and even the IRQ but fails to remap the IRQ. That's shoddy error reporting, "-ENOMEM" is not adequate for all those errors. Easy to fix but shows this code style leads to undisciplined error handling.

[Osvaldo Pinali Doederlein @opinali]

@SebAaltonen But now look at this SEH alternative. Makes some assumptons like the dev.* fields coming zero/null-inited (but your're screwed with leaks if they weren't) and the other helper functions all throwing exceptions. In my judgment this is *fantastically* simpler and more readable. /