"Like catastrophic brain damage"
"The move comes in response to an attack a team of academic researchers demonstrated against Nvidia’s RTX #A6000, a widely used #GPU for high-performance computing that’s available from many cloud services. A vulnerability the researchers discovered opens the GPU to Rowhammer, a class of attack that exploits physical weakness in DRAM chip modules that store data."
New #ZLUDA 5 Preview Released For #CUDA On Non-NVIDIA #GPU
For now this ability to run unmodified CUDA apps on non-#NVIDIA GPUs is focused on #AMD GPUs of the #Radeon RX 5000 series and newer, which is AMD Radeon GPUs with #ROCm. Besides CUDA code samples, GeekBench has been one of the early targets for testing.
https://www.phoronix.com/news/ZLUDA-5-preview.43
capitalism: Nvidia 1st public company w. $4 trillion valuation
https://apnews.com/article/nvidia-4-trillion-chipmaker-7947e86a7ee9a994b9f16c3c0779b74f
#AI #LLM #NVIDIA #GPU
#capitalism #billionaires #corporations #inequality
#capitalism #inequality
Collaborate, advocate, vote, educate, resist
#DemocraticSocialism #ecosocialism #solidarity #justice #equality
#Nvidia's newest top-tier #AI #supercomputers deployed for the first time — #GraceBlackwellUltra Superchip systems deployed at #CoreWeave
#Dell's and CoreWeave's initial rollout involves Dell Integrated Racks equipped with 72 Nvidia #BlackwellUltra #GPU, 36 Arm-based 72-core Grace CPs, and 36 BlueField DPUs per rack. Each #GB300 #NVL72 rack delivers 1.1 ExaFLOPS of dense FP4 inference and 0.36 ExaFLOPS of FP8 training performance, 50% higher compared to a GB200 NVL.
https://www.tomshardware.com/tech-industry/artificial-intelligence/nvidias-newest-top-tier-ai-supercomputers-deployed-for-the-first-time-grace-blackwell-ultra-superchip-systems-deployed-at-coreweave
Found updated NFB (New Feature Branch) of #nvidia #GPU driver sets 575.64.03.
Filed PR for #FreeBSD #ports as Bug 287984
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=287984
and opened corresponding review D51144
https://reviews.freebsd.org/D51144
Patch there is for "-devel" variant of ports like x11/nvidia-driver-devel.
Runnign on stable/14, amd64 until last night without new issue for me.
But as my GPU on hand is old (Quadro P1000 notebook), cannot confirm that the new version solves any of known problems on GPUs with GSP in them by myself.
#ZLUDA Making Progress In 2025 On Bringing #CUDA To Non-NVIDIA #GPU
ZLUDA #opensource effort that started half-decade ago as drop-in CUDA implementation for #Intel GPUs and then for several years was funded by ##AMD as a CUDA implementation for #Radeon GPUs atop #ROCm and then open-sourced but then reverted has been continuing to push along a new path since last year. Current take on ZLUDA is a multi-vendor CUDA implementation for non-NVIDIA GPUs for #AI workloads & more.
https://www.phoronix.com/news/ZLUDA-Q2-2025-Update
Say about Linux what you will, but getting this much detail about what your graphics card is doing is pretty cool.
`amdgpu_top --gui`
People continue to think about #AI in terms of #2010s computing, which is part of the reason everyone gets it wrong whether they're #antiAI or #tech bros.
Look, we had 8GB of #ram as the standard for a decade. The standard was set in 2014, and in 2015 #AlphaGo beat a human at #Go.
Why? Because, #hardware lags #software - in #economic terms: supply follows demand, but demand can not create its own supply.
It takes 3 years for a new chip to go through the #technological readiness levels and be released.
It takes 5 years for a new #chip architecture. E.g. the #Zen architecture was conceived in 2012, and released in 2017.
It takes 10 years for a new type of technology, like a #GPU.
Now, AlphaGo needed a lot of RAM, so how did it stagnate for a decade after doubling every two years before that?
In 2007 the #Iphone was released. #Computers were all becoming smaller, #energy #efficiency was becoming paramount, and everything was moving to the #cloud.
In 2017, most people used their computer for a few applications and a web browser. But also in 2017, companies were starting to build #technology for AI, as it was becoming increasingly important.
Five years after that, we're in the #pandemic lockdowns, and people are buying more powerful computers, we have #LLM, and companies are beginning to jack up the const of cloud services.
#Apple releases chips with large amounts of unified #memory, #ChatGPT starts to break the internet, and in 2025, GPU growth continues to outpace CPU growth, and in 2025 you have a competitor to Apple's unified memory.
The era of cloud computing and surfing the #web is dead.
The hype of multi-trillion parameter #LLMs making #AGI is a fantasy. There isn't enough power to do that, there aren't enough chips, it's already too expensive.
What _is_ coming is AI tech performing well and running locally without the cloud. AI Tech is _not_ just chatbots and #aiart. It's going to change what you can do with your #computer.
#NVIDIA #TensorCore Evolution: From Volta To Blackwell Amdahls Law, Strong Scaling, Asynchronous Execution, Blackwell, Hopper, Ampere, Turing, Volta, TMA
They introduce core features of major #datacenter #GPU, first explaining important first principles of performance engineering. Then trace evolution of Nvidia’s Tensor Core architectures and programming model, highlighting motivations behind evolution. End goal is to provide a resource for understanding Nvidia’s GPU arch
https://semianalysis.com/2025/06/23/nvidia-tensor-core-evolution-from-volta-to-blackwell/
Review D50697 for -devel versions of #nvidia #GPU #driver sets (latest NFB 575.64) on #FreeBSD now landed on main branch of #ports tree as commit c7cde11f842b33bb36b85b91400bec795430c421.
From now on, whichever the newer version of NFB (New Feature Branch) and Production branch is tracked on -devel branch. Beta Branch is not planned to be tracked, and just being a trigger to preparing for upcoming NFB or Production Branch.
Note that as I'm not an insider of nvidia, I cannot start investigating until nvidia releases new driver sets.
#AMD's Instinct #MI355X accelerator will consume 1,400 watts. #CDNA4 challenges #BlackwellUltra.
AMD's #MI350X-series #GPU are based on #CDNA 4 architecture that introduces support for FP4 and FP6 precision formats alongside FP8 and FP16.
Both SKUs will come with 288GB HBM3E memory that will offer up to 8 TB/s of bandwidth, but MI350X will offer maximum FP4/FP6 performance of 18.45 PFLOPS, whereas the MI355X is said to push the maximum FP4/FP6 performance to 20.1 PFLOPS.
https://www.tomshardware.com/pc-components/gpus/amds-instinct-mi355x-accelerator-will-consume-1-400-watts
The Orb · Part III · Confusion
“If the path before you is clear, you’re probably on someone else’s.”
Carl Jung
· · ·
Man must I be on my own path then...
· · ·
Opened review for #FreeBSD Bug 287268 on Phablicator as Differential Revision D50697, which adds latest New Feature Branch (NFB) of #nvidia #GPU drivers as x11/nvidia-driver-devel, x11/linux-nvidia-libs-devel and graphics/nvidia-drm[|-510|-515|-61|-66]-kmod-devel.
Once accepted and landed, it would track the latest version of whichever NFB or Production Branch of drivers. Beta Branch of drivers wouldn't be tracked.
https://reviews.freebsd.org/D50697
The #ComputerFrontiers25, co-organised by our colleagues Josef Weidendorfer and Amir Raoofy, starts today in Cagliari/Sardinia. It's all about new #computer technologies and system architectures.
Josef and Amir will present a tool for the systematic comparison of different #GPU programming models, which was developed at the Technische Universität München and the LRZ. You can read more about it in the slides of this post.
Find out more about #CF25:
https://www.computingfrontiers.org/2025/program.html
Any #Linux #kernel ,
#graphics or #GPU people out there?
I'm trying to understand the relationship between the #amgdpu driver shipped with the kernel; and the "andgpu-dkms" driver that comes with #ROCm .
Specifically, with a recent enough kernel, do we really need to install the ROCm version of the driver? Does the ROCm version contain stuff the general driver does not? Or is the ROCm stack (esp. libhsa) tightly tied to a very specific version of the driver?
An #Nvidia spox said the company was evaluating its "limited" options. "Until we settle on a new product design & receive approval from the #US govt, we are effectively foreclosed from #China's $50 billion #DataCenter market."
China remains a huge #market for Nvidia, accounting for 13% of its sales in the past financial year. It's the 3rd time Nvidia has had to tailor a #GPU for the world's 2nd-largest #economy after restrictions from the US who is keen to stymie Chinese #tech development.
The #GPU or graphics processing unit will be part of #Nvidia's latest generation #Blackwell-architecture #AI processors & is expected to be priced between $6,500 & $8,000, well below the $10,000-$12,000 the #H20 sold for, acc/to 2 of the sources.
The lower price reflects its weaker specifications & simpler manufacturing requirements.
It will be based on Nvidia's RTX Pro 6000D, a server-class graphics processor & will use conventional GDDR7 memory instead of more advanced high bandwidth memory.
Examining #RDNA4's out-of-order memory accesses in detail, and investigating with testing
#RDNA 4's memory subsystem enhancements are exciting and improve performance across a variety of workloads. #AMD specifically calls out benefits in #raytracing. But RDNA 4's scheme for handling memory dependencies isn't fundamentally different from that of #GCN many years ago. RDNA 4’s makes the most significant change to AMD’s #GPU memory subsystem since RDNA launched in 2019.
https://chipsandcheese.com/p/rdna-4s-out-of-order-memory-accesses
Jennifer 
