Dedicated GPU Support

[CaFri]

Hello,

I am currently using Diadem 2012. I am handling rather large sets(before reduction length per channel around 12 Million , after reduction around 1,2 Mio per channel) of Data and often need analyze the data using VIEW. I also need to compare the expirements by stacking multiple channels on top of each other in the same sheet under VIEW.

My problem is that at some point DIAdem is starting to lag. Meaning he has to redraw all the lines (lets just say the 5 meassured forces from 5 experiments, so 5 channels, each 1,2 Mio meassured points)). This is obviously a bit annoying and a result form using large amounts of data.

The question is where is the bottleneck? Is the CPU(i5 3rd gen, 2,5ghz) to slow, is the (integreted) GPU to slow, is the SSD to slow, or is it simply to much for diadem? There must be a reason why DIAdem has to reload the graphs after sligthly moving them(maybe not enough graphics memory).

It is not an defastating issue, still it is annoying and waiting for a PC/Programm to perform is not ideal. We are planing to buy some new hardware and i would like to know what is more importent for diadem:

Do i need a dedicated GPU, is Diadem utilizing on that, faster CPU, fast SSD?

I would be thankfull for any recommendations, greetings,

CaFri

[MelanieE]

Hi CaFri,

since I don't know your system and the properties of the data you load, let me just give you a hint, how you can get closer to the bottleneck.

Windows brings some cool features, that lets you analyse runtime issues. So how to indicate the bottleneck:

To see, if all the data can be handled in memory, you can watch the page faults in Windows Task Manager. If the page faults are couting up while waiting for the data to be redrawn, you simply need more RAM, so all the data can be hosted in memory. A page fault indicates, that data has been written to the disk since there is no more space available in memory for it.

Please also keep in mind, that also the OS may limit the accessible memory of an application.

If there are no or nearly no page faults, there has to be another reason for the lag. If the CPU is the limitation, the CPU usage in Task Manager will rise.

If both indication do not show a significant change, the GPU or the OS is the limitation. Both need to make a lot of calculations to redraw.

If you do not use it already, I'd recommend a system with Windows 10 and Diadem 64 bit for the new machine, since 64 bit application can access more memory and Windows 10 has a better graphical performance, than Windows 7.

Best regards,

Melanie Eisfeld

Staff Applications Engineer

National Instruments

[CaFri]

Hi Melaniee,

thank you for your help. I checked the page faults, there seem to be none, so RAM is not an issue(one opened instance of Diadem uses only around 100MB) . I already knew that the CPU was using 100% of one of his cores(Dualcore CPU with HyperThread), so this seems to be the bottleneck. Also checked on the SSD, no issue there. Since it is an integrated GPU i cannot complete take that out of the equation, but it seems not be an issue.

But I am still wondering why Diadem cant handle to "remember" sections? As an example, I want to look closer into a certain area of my graphs. The complete graph has an range on the x-axis from 0-100 seconds. I want to look at 20-40secs, so I zoom in. Diadem ist than reloading every point from 0-100, but is only displaying the ones from 20-40secs. 

When i want to switch to the next section, 40-60secs, Diadem is again realoading every point from 0-100 and after that displays the section, 40-60secs.

The question for me was, why can't Diadem remember the section 40-60secs, while I am looking at 20-40secs? The programm has obviously loaded and scaled the section 40-60secs, while focusing 20-40secs.

Do you know if the 64Bit version of DIAdem is support multicore rendering/visualization. I already know that it supports multicore scripting, which is nice.

Thank you, greetings

CaFri

[MelanieE]

Hi CaFri,

can you provide a bit more information about how you load the data and from which file?

Which file type do you use?

Do you use a data plugin?

How large is one file?

Is the data distributed over multiple files?

Best,

Melanie

[CaFri]

Hi Melaniee,

my measurements are saved as tdm files(together with the tdx file). The measurements are done using a DAC Layout. Each measurement writes/saves 8 channels(each around 12 Mio. points long), when saved one file is around 1.2 GB. I then reduce the rawdata to 1/10 of the points, using a script(using the command "Call DataFileLoadRed(..)"). Afterwards the files are also around 1/10th in size, so around 100MB.

I dont use a plugin. After my Diadem-Scripts are done with the rawdata and I am actually starting to work with the measurment, I manualy load the files from the navigator(right click on the file and select "load data", or using drag&drop) into the "dataportal"(not sure about the translation("Datenportal"), i only have the German version, I am talking about the area, which is normaly on the right side in Diadem, showing you the groups and their channels).

For comparing different meassurments it do happens that I am selecting channels from different files(draging Channel D from file B, and droping it next to Channel C from file A), these "channel-comparisons" are then saved as/under one new file. This new file tends to be around 500MB, due to an increasing amount of measurements.#

As an example: The file I am currently working with is around 500MB, and contains 11 Groups, each having 6 channels. To compare them, I go to VIEW select a 2D-Axis-Diagramm and then compare one channel from each group, which sums up to a curve set of 11 curves in diagram(at the moment, the amount is going to rise).

I hope i could help you with these informations, thank you very much for your help so far. I am still new to using forums, I am working on providing better informations.

Regards,

Carl

[MelanieE]

Hi Carl,

thank you for the detailed description.

In general Diadem VIEW will always redraw the complete data set, that you have in the VIEW currently.

Methods, that can reduce the redraw time for you, can be resampling the channel, selectively load the data or load only the parts of the channels, you need to handle in the session.

The Resample Method:

The resampling function can be found under the Analysis tab >> Channel Functions >> Resampling. The purpose is to reduce the samples per channel, without manipulating the original channel data in the file. Detailed information can be found under the following link:

Resampling - DIAdem 2015 Help - National Instruments
http://zone.ni.com/reference/en-XX/help/370858M-01/exploff/examples/expl_resampling/

To load data selectively, use the NAVIGATOR tab and load your data through the context menu of the files.

One option is to select Data Reduction from the context menue. Here you can specify which data to load, for example through the number of intervals.

The third option is to select Open with.. from the context menue and use the Selective Opening. Here you can select frames of data to load into memory, so Diadem does not have to hold all data in memory and VIEW can not redraw all the data in the end.

Best,

Melanie