Hello, I am using a GPU (Tesla for the time being) for radar data processing. In this scenario some bespoke hardware, a PCIe plugin card captures and accumulates radar data. Once a certain amount of data is accumulated (4MB), it has to be transferred to the GPU for processing (every 8ms). The chall…

google works better than the forum search :) : https://www.google.com/search?hl=en&lr=…q=&gs_rfai= This is the topic: http://forums.nvidia.com/index.php?showtop…p;#entry1069704

Many thanks for the link; it looks interesting, for people using Infiniband though. It still does not address what I would like to achieve: Get a data acquisition card copy captured data directly onto the GPU card. I may be completely mistaken and this kind of peer to peer copy is not possible on P…

Did you read the post about changing the low-level allocator in the driver? There is a post about how the infiniband trick was already possible by changing some part of the driver. I think it should that way be possible to map a region of your capture card into (pinned) memory and use that pinned m…

Hello again, Well, I read the post that mentioned changing the low-level allocator in the driver. But I did not see any links to code, or instructions on how to actuall do it. Do you really mean to mmap a region of my capture card, e.g. by implementing the mmap file ops, i.e. the PCIe device mem…

Are you still in a planning stage, or do you have some initial results that are problematic? How far are you from reaching the 8ms deadline? Does it break consistently or only sometimes, in which case the RT patches might help (actually, I believe that would be a prerequisite.)

As a matter of fact, I am working on a the implementation of a smaller Proof of Concept system. Data have to be collected in smaller chunks from the FPGA card at ~ 2KHz. I started off by using a userIO based driver that catches an interrupt and wakes up a process having selected the corresponding d…

Thanks for the hint. I am not quite sure whether I understand the purpose of the extension; must be something around video capture from a video capture device, possibly a device on the system bus and not on the graphics card itself. I could not quite see how a data pipeline between two devices on a …

Think the other way around. If you have a pointer to (non-swappable) memory that actually points to the memory window of your pcie card, you can have your GPU DMA that memory in one go from the device to the GPU. I have no idea if you can map that pcie-card memory into the normal memory space, but…

Hello and many thanks for the mind opener! I always assumed that DMA transfer from the host to the GPU is controlled by the host, i.e. by the NVidia driver. But Yes, the GPU can DMA and be the bus master, too. If I understand you correctly, this is what’s happening in a cudaMemCpy() The PCIe-car…

Real-time GPU processing Peer 2 peer data copy, Linux kernel memory, kernels in kernel,

Accelerated Computing CUDA CUDA Programming and Performance

Nico June 14, 2010, 6:46pm 10

I’m not sure if it’s what you’re looking for, but I stumbled upon this OpenGL extension.
Never tried it myself, but if it works you could use cuda-gl interop.

Topic		Replies	Views
From NIC to GPU. CUDA Programming and Performance	42	14245	August 21, 2025
GPU Communication Protocol CUDA Programming and Performance	16	6704	May 17, 2010
GPU to GPU transfers most effective method? CUDA Programming and Performance	27	38544	March 3, 2011
PCIe Impact Give some examples of how PCIe impact your applications CUDA Programming and Performance	15	2469	October 17, 2010
NVidia GPUs in Embedded Computing Has the GPU computing and CUDA penetrated the embedded market? CUDA Programming and Performance	11	4045	August 3, 2010
Transfer video frames from a PCIe capture card to Jetson TX1 device memory (for RT video processing) Jetson TX1	20	6253	June 1, 2018
How to Optimize Data Transfers in CUDA C/C++ Technical Blog	12	1338	January 22, 2022
Device Memory Bandwidth CUDA Programming and Performance	17	8685	January 17, 2018
Can I use streaming to overlap kernels and data transfers in this scenario? CUDA Programming and Performance	13	546	July 5, 2024
Transfer camera data stream from FPGA to AGX via PCIe Holoscan SDK camera	2	956	January 8, 2024

Real-time GPU processing Peer 2 peer data copy, Linux kernel memory, kernels in kernel,

Related topics