Why generic nvidia card(like gtx1080) can't virtualize? - gpu

I read only grid 1.0 and 2.0 products can virtualize.
I read this product share cuda and dedicate framebuffer.
Is it not possible generic nvidia card virtualize like grid product only using software method(iommu, vfio-pci, ovmf, mmu ...) ?
Or any project exist?

It looks like marketing stuff. You need to pay a LOT of money to use "professional" adapters like GRID. AFAIK, you even need to pay for virtual video drivers that will be installed on your virtual machine. So you can't do virtualization with generic videocard just because nvidia needs a little bit more of money.
Unfortunately no projects exists that allows us to use consumer videocard in virtualization. The only thing you can do with generic card is to passthroug it completely to one of virtual machines.

Related

How can I determine the generation/codename of AMD gpu in Linux?

I want to detect the AMD gpu deneration in python code. My case is that to run specific application (davinci resolve), it is required to use amdgpu pro drivers for gpu cards before Vega. And amdgpu pro drivers are not required when AMD gpu is Vega or newer generation. See the list of amd gpus in wikipedia. I am writing a script (davinci-resolve-checker) that tells user if he/she need to use that driver.
The question is, how do I differentiate between gpu generations/chip codenames of a gpu? I am using pylspci to get information of the presented gpus. Is there a list of generations that I can check with?
There is a pci id list published here: https://pci-ids.ucw.cz/v2.2/pci.ids
In that list for Advanced Micro Devices, Inc. [AMD/ATI] (1002 vendor id) you can see their devices. For example, for AMD Radeon PRO W6600 GPU there is 73e3 Navi 23 WKS-XL [Radeon PRO W6600] line.
There you can see if the device name contains the codename substring. In this case it is Navi.
For the specified question, the codenames that currently describes vega and above are: Vega, Navi. So if the device name does not contain that substring, I consider it as "older than vega".
For programming that, you do not actually need the list, as you can just take the device name from VerboseParser device.device.name. But just in case, this list is located in /usr/share/hwdata/pci.ids in the system.
Probably, it is not a very reliable way. But I did not yet found a better way. Any suggestions are welcome.

Embedded Board Support Package

As I understand, a BSP (Board Support Package) contains bootloader, kernel and device driver which help OS to work on HW. But I'm confused because OS also contains a kernel. So what is the difference between the kernel in OS and the kernel in BSP?
What a BSP comprises of depends on context; generically it is code or libraries to support a specific board design. That may be provided as generic code from the board supplier for use in a bare-metal system or for integrating with an OS, or it may be specific to a particular OS, or it may even include an OS. In any case it provides board specific support for higher-level software.
A kernel is board agnostic (though often processor architecture specific), and makes no direct access to hardware not intrinsic to the processor architecture on which it runs. Typically an OS or application will require a Hardware Abstraction Layer (HAL); the HAL may well be built using the BSP, or the BSP may in fact be the HAL. A vendor may even package a HAL and OS and refer to that as a BSP.
The term means what it means to whoever is using it - context is everything. For example in VxWorks, WindRiver use the term BSP to refer to the layer that supports the execution of a VxWorks based application on a specific hardware design. A board vendor on the other hand may provide a complete Linux distribution ported to the board and refer to that as a BSP.
However and to what extent a particular vendor or developer chooses to support a board is a board support package regardless of how much or how little it may contain.
BSP definition is broad. It is a supporting software package for a specific board. BSP for a tiny microcontroller probably just contains HW drivers for its peripherals. On the other hand, for an embedded CPU it may contain HW drivers, bootloader and OS kernel and what not.
So the kernel in a BSP (board support package) is just a specific version of an OS kernel that has been ported to your board.
Im probably just saying the same things already said.
You have a chip and/or board product you want to sell to other (software) developers. A reference design (board) with the chip(s) in question are used. The BSP is a vague term to mean the software that is provided to you as a software developer to ideally make your life easier in using that product (chip and/or board) and adding your software to it or developing for it. So if it is a linux or rtos or other operating system capable platform and the vendor (providing the bsp) believes that users want an operating system and a specific operating system, then instead of you having to port the os to that target, they do it for you. If something like linux that is open source, then you either are told which linux sources to download then the patches made by the bsp are added and/or the bsp contains the complete sources for the whole thing already patched. Drivers, applications as deemed necessary by the vendor etc. Multiple operating systems may be supported if the vendor feels that is needed in order to attract customers to buy that board/chip product.
The whole package of software that you get from them to make that chip/board into your own product, is the BSP.
vxWorks kernel which you can run on a Board contains vxWorks core kernel and "other components" which may change from one environment.
Core kernel contains essential programs such as Scheduler, Memory manager, Basic File systems, security features etc.
These "other components" which are part of BSP may be optional or may vary from system to system, and helps the core kernel features.
In simple words, the image dislays the defination of BSP. Please correct me if I'm wrong
I would say for a well structured code base, the application layer should be abstracted from lower layers by the HAL layer. This would allow the app layer to be portable if we want to migrate the system to a new board. If you see you have board/CPU specific logic in your app layer, you know you have broken the portability.
The HAL layer functions' bodies should contain board specific code, here is where the BSP layer code comes into play. When we want to port the system to a new board, code changes should happens in the HAL functions' bodies, while the HAL functions' declaration should not change, which leads to the app layer remains the same.

Getting a pointcloud using multiple Kinects

I am working on a project where we are going to use multiple Kinects and merge the pointclouds. I would like to know how to use two Kinects at the same time. Are there any specific drivers or embedded application?
I used Microsoft SDK but it only supports a single Kinect at a time. But for our project we cannot use multiple PCs. Now I have to find a way to overcome the problem. If someone who has some experience on accessing multiple Kinect drivers, please share your views.
I assume you are talking about Kinect v2?
Check out libfreenect2. It's an open source driver for Kinect v2 and it supports multiple Kinects on the same computer. But it doesn't provide any of the "advanced" features of the Microsoft SDK like skeleton tracking. But getting the pointcoulds is no problem.
You also need to make sure your hardware supports multiple Kinects. You'll need (most likely) a separate USB3.0 controller for each Kinect. Of course, those controllers need to be Kinect v2 compatible, meaning they need to be Intel or NEC/Renesas chips. That can easily be achieved by using PCIe USB3.0 expansion cards. But those can't be plugged into PCIe x1 slots.
A single lane doesn't have enough bandwidth. x8 or x16 slots usually work.
See Requirements for multiple Kinects#libfreenect2.
And you also need a strong enough CPU and GPU. Depth processing in libfreenect2 is done on the GPU using OpenGL or OpenCL (CPU is possible as well, but very slow). RGB processing is done on the CPU. It needs quite a bit of processing power to give you the raw data.

How to find an embedded platform?

I am new to the locating hardware side of embedded programming and so after being completely overwhelmed with all the choices out there (pc104, custom boards, a zillion option for each board, volume discounts, devel kits, ahhh!!) I am asking here for some direction.
Basically, I must find a new motherboard and (most likely) re-implement the program logic. Rewriting this in C/C++/Java/C#/Pascal/BASIC is not a problem for me. so my real problem is finding the hardware. This motherboard will have several other devices attached to it. Here is a summary of what I need to do:
Required:
2 RS232 serial ports (one used all the time for primary UI, the second one not continuous)
1 modem (9600+ baud ok) [Modem will be in simultaneous use with only one of the serial port devices, so interrupt sharing with one serial port is OK, but not both]
Minimum permanent/long term storage: Whatever O/S requires + 1 MB (executable) + 512 KB (Data files)
RAM: Minimal, whatever the O/S requires plus maybe 1MB for executable.
Nice to have:
USB port(s)
Ethernet network port
Wireless network
Implementation languages (any O/S I will adapt to):
First choice Java/C# (Mono ok)
Second choice is C/Pascal
Third is BASIC
Ok, given all this, I am having a lot of trouble finding hardware that will support this that is low in cost. Every manufacturer site I visit has a lot of options, and it's difficult to see if their offering will even satisfy my must-have requirements (for example they sometimes list 3 "serial ports", but it appears that only one of the three is RS232, for example, and don't mention what the other two are). The #1 constraint is cost, #2 is size.
Can anyone help me with this? This little task has left me thinking I should have gone for EE and not CS :-).
EDIT: A bit of background: This is a system currently in production, but the original programmer passed away, and the current hardware manufacturer cannot find hardware to run the (currently) DOS system, so I need to reimplement this in a modern platform. I can only change the programming and the motherboard hardware.
I suggest buying a cheap Atom Mini-ITX board, some of which come with multi - 4+ RS232 ports.
But with Serial->USB converters, this isn't really an issue. Just get an Atom. And if you have code, port your software to Linux.
Here is a link to a Jetway Mini-Itx board, and a link to a 4 port RS232 expansion module for it. ~$170 total, some extra for memory, a disk, and a case and PSU. $250-$300 total.
Now here is an Intel Atom Board at $69 to which you could add flash storage instead of drives, and USB-serial converters for any data collection you need to do.
PC104 has a lot of value in maximizing the space used in 19" or 23" rackmount configurations - if you're not in that space, PC104 is a waste of your time and money, IMHO.
The BeagleBoard should have everything you need for $200 or so - it can run Linux so use whatever programming language you like.
A 'modern' system will run DOS so long as it is x86, I suggest that you look at an industrial PC board from a supplier such as Advantech, your existing system may well run unchanged if it adheres to PC/DOS/BIOS standards.
That said if your original system runs on DOS, the chances are that you do not need the horsepower of a modern x86 system, and can save money by using a microcontroller board using something fairly ubiquitous such as an ARM. Also if DOS was the OS, then you most likely do not need an OS at all, and could develop the system "bare-metal". The resources necessary just to support Linux are probably far greater than your existing application and OS together, and for little or no benefit unless you intend on extending the capability of the system considerably.
There are a number of resources available (free and commercial) for implementing a file system and USB on a bare-metal system or a system using a simple real-time kernel such as FreeRTOS or eCOS which have far smaller footprints than Linux.
The Windows embedded site ( http://www.microsoft.com/windowsembedded/en-us/default.mspx )
has a lot of resources and links to hardware partners, distributors and development kits. There's even a "Spark" incubation project ( http://www.microsoft.com/windowsembedded/en-us/community/spark/default.mspx )
What's also really nice about using windows ce is that it now supports Silverlight as a development environment.
I've used the jetway boards / daughter cards that Chris mentioned with success for various projects from embedded control, my home router, my HTPC front end.
You didn't mention what the actual application was but if you need something more industrial due to temperature or moisture constraints i've found http://www.logicsupply.com/ to be a good resource for mini-itx systems that can take a beating.
A tip for these board is that given your minimal storage requirements, don't use a hard drive. Use an IDE adapter for a compact flash card as the system storage or an SD card. No moving parts is usually a big plus in these applications. They also usually offer models with DC power input so you can use a laptop like or wall wart external supply which minimizes its final size.
This http://www.fit-pc.com/web/ is another option in the very small atom PC market, you'd likely need to use some USB converters to get to your desired connectivity.
The beagle board Paul mentioned is also a good choice, there are daughter cards for that as well that will add whatever ports you need and it has an on board SD card reader for whatever storage you need. This is also a substantially lower power option vs the atom systems.
There are a ton of single board computers that would fit your needs. When searching you'll normally find that they don't keep many interface connectors on the processor board itself but rather you need to look at the stackable daughter cards they offer which would provide whatever connections you need (RS-232, etc.). This is often why you see just "serial port" in the description as the final physical layer for the serial port will be defined on the daughter card.
There are a ton of arm based development boards you could also use, to many to list, these are similar to the beagle board. Googling for "System on module" is a good way to find many options. These again are usually a module with the processor/ram/flash on 1 card and then offer various carrier boards which the module plugs into which will provide the various forms of connectivity you need.
In terms of development, the atom boards will likely be the easiest if your more familiar with x86 development. ARM is strongly supported under linux though so there is little difficulty in getting these up and running.
Personally i would avoid windows for a headless design like your discussing, i rarely see a windows based embedded device that isn't just bad.
Take at look at one of the boards in the Arduino line, in particular the Arduino Mega. Very flexible boards at a low cost, and the Mega has enough I/O ports to do what you need it to do. There is no on-chip modem, but you can connect to something like a Phillips PCD3312C over the I2C connector or you can find an Arduino add-on board (called a "shield") to give you modem functionality (or Bluetooth, ethernet, etc etc). Also, these are very easy to connect to an external memory device (like a flash drive or an SD card) so you should have plenty of storage space.
For something more PC-like, look for an existing device that is powered by a VIA EPIA board. There are lot of devices out there that use these (set-top boxes, edge routers, network security devices etc) that you can buy and re-program. For example, I found a device that was supposed to be a network security device. It came with the EPIA board, RAM, a hard drive, and a power supply. All I had to do was format the hard drive, install Linux (Debian had all necessary drivers already included), and I had a complete mini-computer ready to go. It only cost me around $45 too (bought brand new, unopened on ebay).
Update: The particular device I found was an EdgeSecure i10 from Ingrian Networks.

How can I program a wireless adapter?

Is it possible to program a wireless adapter attached to a computer?
I need to modify how they work, not just using them to perform a task such as scanning or connecting.
I have already tried the Native Wifi API, but that library is too high level. I cannot modify how exactly the wireless adapter works.
Any solution in any programming language in any operating system is very welcomed. (Sounds so desperate lol)
You need an open-source operating system then. Hardware varies in how programmable it is, but for example, Atheros wireless cards do not have an on-board processor, and therefore they do the absolute minimum of the 802.11 protocol in hardware, leaving everything else to the device driver. More info in these places: http://linuxwireless.org/ http://git.kernel.org/?p=linux/kernel/git/linville/wireless-testing.git;a=summary;
If you really need to go further that what commodity hardware can do, look in to GNU Radio and the USRP/USRP2: http://gnuradio.org/redmine/wiki/gnuradio
And yes, you do have to be careful about the legal implications of this stuff, but then if you don't turn off the regulatory framework, there is software to help with that.
Generally speaking, the manufacturer will attempt to prevent you from doing this. Since what you're working with is really a radio transceiver, its operation is regulated. In the US, for example, such things fall under the purview of the FCC. Depending on the country, changing how it operates (and then operating it) is likely to be illegal.
If you have an atheros chipset on your WLAN card then load up linux and install ath5k/ath9k or madwifi and you can do some interesting things with the driver.