Disclaimer: As everything on this blog is, this is my own opinion, and not my employer’s. The conventional Unix security model is based around users, not applications. This implies that every application running in a given user context has access to all user data, even if that’s not related at all to what the app… Read More » Recall: random notes about the security model
Yesterday, Microsoft announced at their event a new generation of “Copilot+” PCs with new features, let’s focus on Recall, which looks to be the most controversial so far. Going back in time: Timeline Timeline was a Windows 10 feature that appeared late in its lifecycle, on the April 2018 release. It used a set of… Read More » Recall? Rewind? Timeline?
This is gathered from various sources online, including the FDT of the device, the comments on the announcement, Jeff Gerling’s thread, and more. Updated further on September 29 and October 1st. Pricing $60 for the 4GB of RAM model and $80 for the 8GB of RAM one. SoC The SoC is on TSMC’s 16nm process.… Read More » Gathering (some) Raspberry Pi 5 details
TL;DR: it does not affect NVIDIA driver users on the desktop. If you’re one of those, don’t worry, it doesn’t affect you. What the change actually does: It makes it impossible for a networking driver to integrate into the NVIDIA driver P2P interface (for RDMA) without having a complex install where the networking driver has… Read More » What this whole NVIDIA and the GPL drama is about this time
Heterogenous memory management on GPUs allows to use the full host process’s address space within GPU device code, without requiring pinning or the allocation to even be done within CUDA. Pages are dynamically migrated from the host to the device depending on use, without any programmer action being required. A warning to keep in mind… Read More » HMM in CUDA on the NVIDIA r530 drivers
On a number of platforms, the hypervisor is not controlled by the OS vendor but by the platform vendor. In those cases, how are virtualisation capabilities exposed? The platform hypervisor On Qualcomm platforms, this role is fulfilled by Gunyah. On the Arm Realms architecture, this role will be fulfilled by the Realm Management Monitor, which… Read More » Virtualisation when you don’t own the…
After PowerVR, let’s look at the state of the Vivante drivers for Windows, as part of the i.MX8M Windows BSP. Implemented APIs D3D11 using DXVK No D3D9, no D3D12 implementation provided. Note on Vulkan A Vulkan implementation is included but is an internal implementation. No ICD files are present, with Vulkan being provided through vulkan.dll… Read More » A quick look at Vivante Windows drivers on…
Since a while now, PowerVR-based PCIe GPUs appeared under the Moore Threads branding… let’s have a quick look to see what the drivers for those look like, for both Windows and Linux. Windows Utterly unsuitable compared to required modern GPU feature levels. This GPU driver is freely available for download at the Moore Threads website.… Read More » A look at modern PowerVR (MTT) drivers
Introduction As part of CUDA unified memory, the addresses used to access memory on both of the CPU and GPU side are identical. However, only memory allocated from CUDA (or registered as pinned memory) can be accessed to from the GPU. How could the GPU access to regular paged memory, including memory-mapped files, without needing… Read More » Beyond GPU unified memory: fully unified address spaces
In the prior post, we highlighted why PCIe BARs should be mapped as device memory on 64-bit Arm platforms. What is unified memory in a dedicated GPU context? Unified memory is memory that accessible from both the CPU and GPU from an application programmer’s perspective. This is often done using the same pointer address across… Read More » How does NVIDIA implement unified memory on PCIe GPUs and…