i am a high school student who manages the school gpu server right now.
School server : Ubuntu 20.04, Nvidia A6000
Driver : 510
I am now reinstalling my school server and i have question. In the picture, all gpus are in p8 mode. I want to turn it to p0 mode but i can't find out how. Even though i thought it would be automatic and tried a very large sized model, the p8 was maintained.
nvidia-smi picture
can anyone help me please?
Related
I'm a developer for different mobile and backend systems and pretty new to network and hardware stuff. I want to build a system/network with 6 cameras placed 100m away in the field, which I want to control with a web interface. I know how to build such interfaces, but I have no clue how to connect the hardware. I thought about the following:
I need 6 cameras(*infos added below) standing side by side with ca. 1.5m space between. These should be connected to a switch, so a 100m wire (USB or LAN, I prefer LAN) goes to a RPI which can setup the web interface controlling the cameras like ".../whatever/camera-slot-ip-or-number".
As I said in the introduction, I have no clue how to start, because actually webcams using USB as a std, but does they provide wake on LAN features? Or is it better to do it with 6 USB-cams and several RPIs?
I hope someone with a better hardware understanding can help me.
Thanks a lot
Specification for the cameras:
HD is not needed, but it should recognize a 0,5cm round hole in a 50x50cm area properly. The distance between camera and object is 7-10m A color image should transmitted, but there only 2 main-colors.
EDIT:
draft 2.0:
Piping USB through a 100m cable is not easily going to work.
Some models of USB cameras can be used with the Raspberry pi, but the performance (speed of taking a picture, and image quality too) are better with the 'native' raspberry pi camera.
The Pi also has a built-in H.264 video encoder, so you can stream live video with relative ease if you want to. A quick and brute way of doing that is to pipe the output from the built-in raspivid application to your own application that then handles flow control and pipes the data further to a socket.
If wifi does nto work for you, then you could pick some other raspberry pi model with an ethernet interface and go that way.
Also, the cost of additional Raspberry Pis (especially the zero w) is so low that the easiest and most cost efficient thing might just be to one raspberry pi camera on 6 raspberry pi's. If connecting them with Wifi works in your application, you can use the Zero W model and then you just need to feed power to them via cable.
Thank you for the updated information. I am pretty much in agreement, I think, with Sami's answer but wanted to add a few more details that are a bit big and unwieldy for a comment.
If you look across the top of your diagram, you have 6 stations at 1.5m intervals, so the width of your diagram is 7.5m. That is easily within wifi range so I am thinking a wifi access point on any of the 6 stations and a 100m Cat 6 Ethernet cable down the length of your diagram to the front-end.
As your processing doesn't sound too involved, you can likely get away with just a Raspberry Pi Zero W and a v2 camera at each station and save a fair bit of money vis-à-vis Raspberry Pi 3B+.
One thing that does concern me is looking for 0.5cm from 7-10m. The lens on the Raspberry Pi camera is pretty wide angle and a 0.5cm hole is going to be awfully small at 10m in a wide angle shot unless at very high resolution. I haven't done the maths, but I think you will be looking for a telephoto lens if such a thing exists... the maths now follows.
The horizontal field of view (FOV) of v2 camera is 62 degrees, so half that is 31 degrees and the camera is 1000cm away. So:
tan(31 degrees) = half the FOV width / 1000
So, at 10m you will get 1200cm of stuff across your image and that will be imaged by 3,280 pixels on the sensor if you shoot at the very highest resolution. So, each pixel in your image will correspond to an area 0.3cm wide whereas you are looking for holes 0.5cm wide - so it will be pretty marginal as to whether you can make it out... maths is subject to revision after a glass of wine later.
I have a raspberry pi 3.
Whenever I want to shutdown, I always execute this command:
sudo shutdown -h now
And I will wait until only the red light is lighting. (And I always remove the power until only the red light is lighting)
But the problem is : every time I shutdown, I can't boot it again. I need to install the OS into the SD card again every time after I shutdown.
If anyone has the same problem ? Please help me. Thanks a lot.
I encountered this after my comment. As I mentioned, I perform the same steps to safely shutdown, unplug, and store.
Based on this page, you might be getting a corrupt SD card. Not sure if any of these apply.
https://www.raspberrypi.org/documentation/installation/sd-cards.md
If you are having trouble with corruption of your SD cards, make sure you follow these steps:
Make sure you are using a genuine SD card. There are many cheap SD cards available which are actually smaller than advertised or which will not last very long.
Make sure you are using a good quality power supply. You can check your power supply by measuring the voltage between TP1 and TP2 on the Raspberry Pi; if this drops below 4.75V when doing complex tasks then it is most likely unsuitable.
Make sure you are using a good quality USB cable for the power supply. When using a high quality power supply, the TP1->TP2 voltage can drop below 4.75V. This is generally due to the resistance of the wires in the USB power cable; to save money, USB cables have as little copper in them as possible, and as much as 1V (or 1W) can be lost over the length of the cable.
Make sure you are shutting your Raspberry Pi down properly before powering it off. Type sudo halt and wait for the Pi to signal it is ready to be powered off by flashing the activity LED.
Finally, corruption has been observed if you are overclocking the Pi. This problem has been fixed previously, although the workaround used may mean that it can still happen. If after checking the steps above you are still having problems with corruption, please let us know.
Hope that helps.
Cheers.
Ian
I am wondering if someone have an idea about the range of the wi-pi usb wireless adapter, while I am using it with Raspberry Pi and its transmission power is 20 dBm and working in 802.11b.
thanks!
Highly depends on the environment. In open area with line-of-sight it goes to 100m or even further. In indoor environment, WiFi signal may penetrate only a few brick walls.
My experience with the wi-pi is that it is a pile of junk. I have trouble getting 10m range through one brick wall that other adapters have no problem with. It regularly stops working and requires replugging. Completely unusable.
There are a few utilities out there that help you undervolt your cpu. For the PC, for example, there is RMClock. For the Mac, there is Coolbook.
On the Cookbook website, however, it states that it is incompatible with OX X 10.7 Lion, and the i3/i5/i7 processors. I am interested in replicating the functionality of Coolbook, which works with OS X 10.6 Snow Leopard, and the Core 2 Duo processors, but for the newer OS and CPUs. I really have no idea where to start. As far as I know, Coolbook holds a monopoly on the Mac platform, and I was wondering if there is even a hackerish way of accomplishing the same functionality? What are some resources that you could point me to so I can begin understanding what is going on in the driver level when adjusting the voltage of a CPU?
The Problem is, i7 Cpus have no possibility to manage the voltage via software. The reason is simple. Intel has cut the external register address for voltage management, only AMD cpus has the possibility to manage voltage via software, or older core duo intel cpus. Coolbook wors but not very well on OSX Lion on older Intel CPUs, there is no update for Lion at this time. I hope that help.
I have used this on my Android phones. Not sure how it would work on a mbp but they are both the linux environment:
echo> 0 /sys/devices/system/cpu/cpu1/online
On my phone it stops the phone from using the second core and saves a little battery life. Hope it helps to reverse it. Just replace the 0 with a 1 (you must be a superuser for this to work).
We are preparing a solution to multicast the teacher's screen to 40 students' pcs.
Teacher and students pcs can be both Ubuntu and Windows.
Some solutions were tested :
iTalc ... not stable yet.
multicast a "vnc -viewonly" ... no solution found
capture the screen with VLC and multicast it.
That latest seems to work ... except that with a resolution like 1920x1200 it is just too much CPU intensive.
One idea would be to capture only the 4th of the screen. CPU is not saturated anymore but everything becomes really slow and the surface is too small anyway.
A second idea is to buy a PCI card (or something) which will be dedicated to real-time video encoding.
Anyone has an experience/knowledge on it?
Thanks!
Try the TightVNC project's TightProjector.
TightProjector is a program that can transmit the screen of a
particular Windows computer to other computers in the same local-area
network. The data is transmitted continuously, in real time.