Hm, not happy with SPIR-V to ISA on my RDNA3 that is reordering loads from shared memory in some ways that it is actually slightly less efficient...
Is there a way to instruct the ISA compiler via GLSL to not reorder loads from shared mem? (Not workgroup/subgroup barriers but simple no-reordering) I haven't yet tried if volatile could do the trick 
Anyone has some thoughts?
Hey #vulkan Community,
Isaak fliegt seine Drohne grade über den neuen Ausbruch bei Sundhnúkur auf #Island
#vulkanausbruch #eruption
https://www.youtube.com/watch?v=_GAjSpf8toU
Historical strategy game Old World switches to Vulkan for Linux / Steam Deck
Updated node connections in the video mixer / synthesizer.
Before, on changing the output size of a node, the connection broke and had to be recreated. Changing the resolution of many nodes was an annoyance. Especially with mixer connections, which have some own values.
It is now streamlined much more and the resolution of any node can be changed at any time without the need to reconnect.
Me looking at performance counters in Vulkan: oh!, look at that, VK_KHR_performance_query, so nice! 
Me checking https://vulkan.gpuinfo.org/listextensions.php, oops, nevermind... 
It is so unfortunate. I have to use proprietary API to get them (e.g. GPUApiPerf for AMD)
Ported https://salykova.github.io/sgemm-gpu to Vulkan (nice article!) 
it's 2x slower than Cuda. That one was tricky to port, (e.g. need to alias shared buffer to allow LDS/STS.128), half of it with AI, 2nd half going over lines 1-by-1
Ported the Kernel 5 from https://seb-v.github.io/optimization/update/2025/01/20/Fast-GPU-Matrix-multiplication.html to Vulkan 
It's only 15% slower than Cuda (and it works on AMD)
In both cases, it's quite difficult to reason about SPIR-V 
ISA, apart the AMD GPU Analyzer that is helping!
Digging deeper 
#Hawaii ist wieder im Fokus! 
Der Vulkan Kilauea spuckt erneut Lava und sorgt für Aufsehen. Experten beobachten die Situation genau, während Anwohner und Touristen die beeindruckenden Naturschauspiele aus sicherer Entfernung bestaunen. 
Weitere Informationen gibt es hier: https://de.euronews.com/2025/03/27/hawaii-vulkan-kilauea-spuckt-wieder-lava
#Vulkan #Kilauea #Naturereignis
Bleiben Sie sicher und informiert! 
Does anyone know what Vulkan dynamic rendering + multiview into a cubemap looks like? I can't find any examples anywhere and what I have (just setting VkRenderingInfo::viewMask) doesn't seem to be working
Boosts for visibility appreciated!
#gamedev #graphics #3dgraphics #vulkan #gpu #rendering
Nvidia's Open GPU Kernel Modules: A Leap for Haiku OS and Vulkan Support
Nvidia's recent release of their open GPU kernel modules under the MIT license has opened new avenues for Haiku OS, enabling the initial port of the NVRM driver. This development not only enhances the...
https://news.lavx.hu/article/nvidia-s-open-gpu-kernel-modules-a-leap-for-haiku-os-and-vulkan-support
NEWS FOR #HAIKU: #NVIDIA GPU support coming soon!
Developer @X512 has successfully ported Nvidia kernel drivers to Haiku. The driver will support Turing+ GPUs and already includes Vulkan integration via Mesa's NVK.
Initial tests are working and show potential for future uses, including AI acceleration with llama.cpp.
A major step forward for the Haiku ecosystem and hardware compatibility!
Der Ätna Südostkrater ist wieder aktiv! 
Vulkane.net berichtet, dass er aus mehreren Schloten feuert. Strombolianische Eruptionen & ein Lavastrom wurden beobachtet. INGV gab eine Warnung heraus. Keine Gefahr für den Flugverkehr. Mehr dazu hier: https://www.vulkane.net/blogmobil/aetna-suedostkrater-feuerte-aus-mehreren-schloten/ #Ätna #Vulkan #Eruption #Italien 
#newz
#Google wants to make #Android a top tier gaming OS, investing in #Vulkan to reach that goal.
https://www.androidauthority.com/porting-pc-games-to-android-3534575/
Starting to think that API likes Vulkan are on purpose limited 
Testing a RDNA3 RX 7600 XT, it's supposed to have 128KB of LDS per WGP. They say in the ISA, "A single work-group may allocate up to 64kB of LDS space". Also later, if the dispatch is done in WGP Mode (but I can't tell how to achieve this in Vulkan), it can access the full 128KB.
Now, Vulkan reports only 32KB accessible! Why so little? I don't think ROCm is limited for instance
Anybody from AMD around?
Is there any difference between computing AI workloads in Vulkan, OpenCL and CUDA?
I know that some people say that NVIDIA doesn't support (quite well) OpenCL or Vulkan, performance is achieved by using CUDA. But what is the story for other vendors (Intel, AMD, QualComm, Apple) ?
I tried Nsight Graphics today for GPU Compute profiling (Vulkan not CUDA) and it's amazing what we get these days for GPU optimizations, great improvements! (I used it a lot many years ago)
My only main complain so far that makes it less useful is that you don't have access to ISA disassembly, unless you are a company and have signed an NDA (!)
So, credits to the red, at least with the AMD Radeon GPU Analyzer, I can get ISA disassembly with the register pressure 
dxvk-nvap 0.9 is out with support for #NVIDIA Reflex for #Vulkan titles, support for reporting driver setting (DRS) to titles, support for permanently setting NGX debug registry settings, improved (non-) SLI detection, and various bug fixes https://github.com/jp7677/dxvk-nvapi/releases/tag/v0.9.0
So, I just received an AMD RX 7600 XT to test VK_KHR_COOPERATIVE_MATRIX on it and for now, it seems a bit disappointing: only 25-30 TFLOPS with fp16 - while the specs says that it can go up to ~45 TFLOPS. My current Nvidia RTX 4700 can do 100 FTLOPS
Tried the Radeon GPU Profiler, but it seems not working in my setup with multiple non-AMD GPU, what a shame (and frankly, this profiler is just so weird to use). Will try to see If I can disable my main display GPU, super annoying
Worked on #OpenXR #Vulkan support inside #Blender3d. After a couple of days hacking it is clear how the final solution would look like.
The rendering and displaying part is working. But need external memory to give the offscreen render result to OpenXR swapchain.
Time to add support for VK_KHR_external_memory_*.
