Wikipedia:Reference desk/Archives/Computing/2018 May 5

= May 5 =

Is there a way to see a "real" Task Manager on Windows 10?
A nearly new laptop claims it is running at "2% disk" or "0.1 MB/s" on Task Manager, and only one user is shown on that program's Users screen, but the disk can be heard running continuously for several minutes. Is there any way to tell what is really running the disk? Wnt (talk) 06:49, 5 May 2018 (UTC)
 * AFAIK Windows 10 Task Manager can show all the details visible on previous versions of Windows, and more so your question is fairly unclear. I suggest make sure you click on 'more details' and also check out the details tab. If you just want to look at disk usage, 'Resource Monitor' which can also be opened view the performance tab of Task Manager is likely a better choice. Note that if you are not an administrator, it's unlikely you will be able to see details for other users and in some cases you may need UAC elevation. If you can actually 'hear' the disk, this likely means that something is doing a lot of seeking, so you should expect a fairly low rate although 0.1 MB/s does sound a little low. (In other words, hearing the disk doesn't mean the disk is going very fast, it actually means the disk is going to be performing quite slowly.) Of course it could have nothing to do with the OS, especially if you have an SSHD or some other fancy disk. Nil Einne (talk) 09:11, 5 May 2018 (UTC)
 * In case my response above is unclear, what I mean is that with modern hard disks, while not as quiet as when powered off, a hard drive that is simply doing a continuous full sequential speed read or write is actually fairly silent. If the hard drive itself is noisy (as opposed to some fan or something else) this would generally indicate a fair amount of seeking or stopping and starting or something similar. So if you're thinking that 'hearing the hard disk' means 'very high transfer speeds', you're mistaken. The highest transfers speeds are when you can't really hear the HD. Also while I'm not sure how the % disk active time is determined, I wouldn't trust it to be a greatly meaningful metric. If you really want to analyse disk performance over time, Performance Monitor is probably a better tool although you would also need to have some understanding of it and remember it isn't going to pick up stuff not happening at the OS level. (Again think of SSHDs and probably some SMR HDs.) People tend to panic themselves over nothing when they don't understand stuff e.g. committed RAM or whatever it was in XP that people used to panic over. Nil Einne (talk) 10:22, 5 May 2018 (UTC)
 * You can open the Resource Monitor. Don't know about Windows 10, but in Windows 7 you can access it from the Performance tab of the Task Manager. There you can find a list of all files being accessed, as well as total real I/O speed (as opposed to a percentage which I don't know what it is supposed to convey). 93.136.56.19 (talk) 14:35, 11 May 2018 (UTC)

MULTIPROCESSORS
MULTIPROCESSORS

A year ago (June 18th 2017) I posted on the issue and got helpful advice. Now I want to ask more questions. There were hardware suggestions and I wonder if more advanced things are available now. First I want to describe what I want to do.

I need a GPU with a CPU or a number of CPU's like dual, or quadruple arrangement. I expect them to control the multitude of GPU's, perhaps 3,000. I will have a webcam and the images from the webcam will be read at a rate of perhaps 10 a second. Each flat image will be converted into an image on a virtual hemispherical surface by the CPU's. Then the image on the hemisphere will be divided (fragmented) into 3,000 or so portions to provide parallel treatment of the fragments. I expect that the GPU's will do that job. The treatment in the small fragments will amount to numerical integration. Then the CPU's will add calculated numbers and provide a multitude of combinations.

I've read that EVGA GeForce GTX TITAN has up to 7,000 gpu channels. I found some on Amazon but they are not new, used. Is it safe to buy such? There is another model, though:. But it has half of the CUDA channels than the previous model and is twice as expensive. Are there any better models now? AboutFace 22 (talk) 21:20, 5 May 2018 (UTC)


 * I'm confused. Have you actually worked out whether what you are trying to do is doable with OpenCL or DirectCompute or CUDA or whatever you want to use? Have you also tried it on a lower end GPU or alternatively I guess some cloud computing platform? Since you're asking here, I take it you are not the child of Bill Gates or Jeff Bezos or something so may want to waste money on something you're not likely to be able to use, so verifying the GPU is likely to be useful seems a basic first step before worrying about what the best choice is or whether to choose a used device. I also don't understand the CPU point. GPUs don't generally come with CPUs at least by normal definitions of those words for high end computing. (You can get SOCs or APUs or whatever you want to call them with CPUs and GPUs but they aren't at the level you seem to be aiming for and that's more like a CPU with a GPU than a GPU with a CPU.) The computer you plan to put the GPU on will have at least one multicore CPU. Nil Einne (talk) 22:41, 5 May 2018 (UTC)

I haven't tried OpenCL. I program in C & C++, C#, FORTRAN and they are the languages for Nvidia, correct? Also I've read  that NVidia systems are the most popular. So, I assume, if I somehow make sure a GPU I will purchase will fit into my hardware, I won't have any problems with programming. Currently all my software's in C++. Yes, right now I have essentially one channel. My hardware now is HP Pavilion, a relatively new system, 2 years old perhaps. It is an image recognition task. It takes a few seconds to finish all computations although the algorithm is highly optimized. A few seconds for a small image is not acceptable. iCloud - I don't want to try it, too much trouble to learn how it works, maybe in the future. I also want my system to be portable, to be able to move it around geographically. I went to a Best Buy yesterday explained to a man from the Geek Squad what I need to do, he said NVidia GPU's are for you. He talked about the CUDA. "that's more like a CPU with a GPU than a GPU with a CPU" - I don't know, it's always been my understanding that it is a CPU with a GPU, I add a GPU to an existing CPU, but who cares. AboutFace 22 (talk) 14:16, 6 May 2018 (UTC)
 * OpenCL (or whatever API) and the languages you use are largely distinct issues. As was explained last time you asked, there's is no guarantee what you are trying to do is going to work particularly well on a GPU. If everything worked well on a GPU and it was trivial to port it, everything would be run on the GPU. But clearly, despite Nvidia, and to a lesser extent AMD's desires, this is very far from happening, the GPGPU revolution has only happened to a limited extent. The person at Best Buy is clearly talking nonsense and probably trying to say something like a CUDA GPU is like a CPU, except it's still very far from true. You don't add an external GPU to an existing CPU, you add it to a motherboard where it is somehow connected to the CPU also on the motherboard. I have no idea why you brought up iCloud. I never said anything about iCloud but simply cloud service with GPUs available. Does iCloud even have user accessible cloud GPU services? In any case, considering you have never apparently programmed anything for a GPU, if you can't work out how to get your program working on a cloud computing service, I wouldn't give you much chance of get it working on a local GPU either. The work required to get it working on a GPU, is going to be far more difficult than the differences of working on a cloud GPU instead of a local GPU. BTW, you were told most of this last time you asked the question, so I'm sure why it's still an issue of confusion. I find it very unlikely you don't have a current GPU even if it's just an Intel iGPU so if you aren't planning to use CUDA (and if you are, why?), it would probably be best to at least try to get it work on your existing GPU with OpenCL or DirectCompute before you buy anything. While there are plenty of differences between GPUs, if things are really as simple as you seem to think, it shouldn't matter. Let me repeat for the last time, that having the GPU doesn't mean you code is somehow going to magically start working on the GPU. And whether you want to use C & C++, C#, FORTRAN, you're still going to use some API to use the GPU, probably either OpenCL, DirectComputer or CUDA. I mean even if you don't port your code, you should at least try some code to get an idea of what you're doing. Nil Einne (talk) 03:58, 7 May 2018 (UTC)

@Einne, thanks for a useful cold shower. To try it on an existing GPU, it means I should try to do it on my HP pavilion. The old computer I've planned to use is Dell R5400, rack mount. It does have a video card inside, I think. I actually focused, or narrowed my choice down to NVIDIA Tesla V100. that is about 10 grand, perhaps even 20. Instead, after reading your message, which I definitely appreciate, I may buy a smaller, less advanced GPU, perhaps refurbished and try to work it out. This is a data sheet for Tesla V100. What I need is special functions, numerical integration, etc, trivial things. What is the term they are using SXM2? or another one: PCIe? Sometimes they print it like "PCle," I mean it looks like it is an "L" letter in the third position, hard to figure out.

There are quite a few tutorials on the web that explain how to code for CUDA, this one for tensor cores. It is a trivial thing for me to master. AboutFace 22 (talk) 15:00, 7 May 2018 (UTC)

Personal identity and WHOIS
Can WHOIS reveal one's real-life identity? — Preceding unsigned comment added by 97.80.134.208 (talk) 21:46, 5 May 2018 (UTC)
 * It is only going to reveal the information stored in the whois database, which may be company and email address and name. See also whois. Graeme Bartlett (talk) 02:43, 6 May 2018 (UTC)