Share
When installing Ubuntu 18.04, Canonical's installer will offer to send some statistics to them. Canonical have now released some of this. One thing to note, is that this data does not include Ubuntu Server, Ubuntu Core, cloud images or and any other Ubuntu derivatives that don't include the report in their own installer.
They've had some good results from it, with 66% of people sending them their data although they don't mention how many results this is actually from. It's a nice start, but I think they really need to do some separation of physical and virtual machines, since it seems they're merged together which will skew a bunch of the data I would imagine. If you're interested in seeing what data is sent on Ubuntu, it can be found in "/.cache/ubuntu-report/".
Their data shows that 98% of people are using the 64bit version on Ubuntu, which lines up with our own user survey. On top of that, a desktop resolution of 1920x1080 remains the most popular at 28% with 1366x768 being the next highest at 25%. What's interesting, is that higher resolutions have a pretty low use with 2560x1440 and 3840x2160 both only seeing 1% although that could easily be watered down due to virtual machines.
One thing that's quite odd is the CPU section under the "Number of CPUs" heading, which claims 27% of people have 4-6 CPUs. Something about that doesn't seem right. 27% of people have at least 4 CPUs in the computer they're installing Ubuntu on? I think they need to improve the wording on this quite a bit just so it's crystal clear on exactly what the statistic represents. Likely CPU cores. I've let them know about it to take a look.
Take a look here at their full statistics page if you're interested.
Article edited and re-posted due to a mess up with the text causing some confusion, mostly my own—apologies.
So it's both possible and easy to track each individual stats if one wants to. And yes there is a difference between two single threaded CPU:s and one dual core CPU if you are a programmer but that is possible even more beyond the topic of this thread.
The kernel separates all this for good reason and things like the scheduler threats them differently.
Maybe Liam is busy incorporating Proton in the stats? :-)
As in thread the lightweight process. Yep, there’s a lot of misnomers out there.
My opinion is that the Ubuntu statistics are not well defined. They are using CPU and GPU to refer to the physical hardware - and it is clear from reading the GOL comments that they didn't define this well leading to misinterpretation.
I don't like their World Usage Location Charts - The chart on the stats page looks like USA hasn't got a big user base, and then the description and chart on the Metrics page suggests that the USA does have a big user base.
They split the data into Physical and Virtual installs of Ubuntu, and then they don't provide stats for each of those, but then they are suggesting conclusions about Desktop Users when their stats could easily be referring to virtual "cloud" installs. For example, one reason that USA could have lots of Ubuntu "users" - is the fact that Amazon AWS Cloud has massive USA-based Data Centres hosting Virtual Machines.
These stats could have lots of value, but as Ubuntu have provided them to website visitors - not so much.
Long Opinion:
The actual information collected is quite sparse; I have included, as an example, the full information generated by ubuntu-report on the machine I am using at the end of this post. It does fully report processor geometry ( Sockets/Cores/HW Threads), but doesn't even include, for example, the clock speed the cores can run at.
Anyone that has looked at articles from the likes of Tom's Hardware, Anandtech and Phoronix will know that it is actually very hard to really compare CPU performance through simple metrics that take no account of chip architecture.
The physical CPU that you drop in a motherboard socket these days bears no relation to the physical CPU of 10-15 years ago, when 2-way and 4-way multi-socket motherboards were the only real way to get higher performing PCs. Those CPUs had a single core plus cache memory, etched on a single piece of silicon, and that's it.
The reason we have this CPU terminology problem today was the development of the MCM ( multi-chip module ), whereby several separately etched pieces of silicon, or one piece etched in several steps, are electrically integrated into a single physical package.
A modern physical CPU contains not just multiple copies of what used to be called a CPU, but also now contains the functions that used to be performed by separate physical chips on the motherboard ( the Northbridge and Southbridge ), and even GPU logic and memory. The MCM is also the reason why there are so many different physical CPUs to choose from, as Intel and AMD can assemble different components into a physical CPU for different purposes ( server, desktop, router etc ).
The state of the art for this integration is something like the upcoming Zen 2 range of processors from AMD, which will be able to ship a single physical server CPU with 64 cores, each with 2 hardware threads. Ubuntu-report would count this single physical package as "128 CPUs".
It is important to realize that putting multiple CPU cores into a single physical package will, under almost all circumstances, be higher performance and cheaper than the eqivalent number of 1-core CPUs in separate sockets. The only limitations to MCM integration is the number of input/output pins available in your socket design, and the ability of a cooling solution to remove heat from the package.
So there is often very little to be gained from even 2 physical CPU sockets on a motherboard today, and you are generally talking about eye-watering price points if you go down this route.
As each CPU core provides an independent ability to perform computations, I don't think it is unreasonable to count it as an independent CPU. I have more reservations with hardware threads; although you can use them to schedule more than one concurrent workload to a CPU core, each of those workloads will take longer, so hardware threads will not always improve performance; it depends on the workloads available.
====================================================================
Sample Ubuntu-Report Data Collected
====================================================================
{
"Version": "18.04",
"OEM": {
"Vendor": "To Be Filled By O.E.M.",
"Product": "To Be Filled By O.E.M."
},
"BIOS": {
"Vendor": "American Megatrends Inc.",
"Version": "P2.30"
},
"CPU": {
"OpMode": "32-bit, 64-bit",
"CPUs": "8",
"Threads": "2",
"Cores": "4",
"Sockets": "1",
"Vendor": "GenuineIntel",
"Family": "6",
"Model": "45",
"Stepping": "7",
"Name": "Intel(R) Core(TM) i7-3820 CPU @ 3.60GHz",
"Virtualization": "VT-x"
},
"Arch": "amd64",
"GPU": [
{
"Vendor": "10de",
"Model": "1b82"
}
],
"RAM": 32.9,
"Disks": [
128,
128,
500.1,
512.1
],
"Partitions": [
107.6,
0.1,
124.9,
500,
479.7
],
"Screens": [
{
"Size": "597mmx336mm",
"Resolution": "2560x1440",
"Frequency": "59.95"
}
],
"Autologin": false,
"LivePatch": true,
"Session": {
"DE": "ubuntu:GNOME",
"Name": "ubuntu",
"Type": "x11"
},
"Language": "en_GB",
"Timezone": "Europe/London"
Seriously though, who cares??
I wonder if Canonical will ever release overall desktop numbers, rather than just stats collected. I'd just like to know how many Ubuntu desktops exist (that aren't virual/test machines). That would be cool.
You are not crazy but you are definitely wrong. A central processing unit is a module that can run code autonomously. It doesn't matter how many of them are put into a chip, if a module can run code completely independent of the rest of the chip, it is a CPU.
Haha, I think the reason for the 50/50 split is because everyone is using a different sense of the word. There's the general public/product sense, then there's the technical definition.
For instance, while I hold to the fact that a core *is* a CPU, if I went to the store I would ask for a multi-core "CPU". Even though I'm well aware there are multiple CPUs in it, I will still call the monolithic chip a single CPU.
However, if you ask me if a core is a CPU I would say yes, because a CPU is a processing unit that contains an IO controller, a math processor and a logic processor. Thus, each core meets the exact definition of a CPU.
These stats aren't primarily for just whoever happens to read them, nor to be able to report about statistics for the sake of reporting them. These stats are for developers who want to know what they ought to be targeting. So, it actually makes perfect sense to just label it as "CPUs".
Funny thing is that if you talk to some older people they call the entire machine a CPU. If you said, "I have 8 CPUs", they would assume you're saying you have 8 entire desktop computers. My computer science teacher when I was in high school used to use that sense of the word.
Last edited by Natedawg at 19 October 2018 at 11:39 pm UTC
I remember that...
My first incursion with the computers was in the mid 1980's, when they were connected to a normal TV and they used audio cassettes for the software.. I remember using the program LOGO...
Anyway, in 1998 I got my first PC and in those days they called "CPU" to the assembled case (mobo, processor, RAM, disk, video and the optional audio card)