topic Re:Several questions about vtune memory access analysis in Analyzers

Several questions about vtune memory access analysis

armorhu — Thu, 06 Jan 2022 07:40:38 GMT

1 What is the term "Socket" in the Platform Diagram mean?

2 What is the term "Local DRAM" and "Remote DRAM" mean?

3 What is the term "Package_0/1" and "Channel 0-4" mean?

4 What is the difference between "DRAM Bandwidth" and "UPI Bandwidth" ?

5 after I try use memory access , I found that there is no way to get memory access profile data of a certain function , I need a top down call stack organization data .

6 In other words , is there any better tools for profile NUMA problem,which can be accurate to the specific code line and data structure, or draw all the memory addresses accessed by a piece of code , similar to this , so i can intuitivley see the problem in the code and slove it

Re:Several questions about vtune memory access analysis

JananiC_Intel — Thu, 06 Jan 2022 12:36:47 GMT

Hi,

Thanks for posting in Intel forums.

We are checking on this. We will get back to you soon.

Regards,

Janani Chandran

Re: Several questions about vtune memory access analysis

JananiC_Intel — Wed, 12 Jan 2022 06:03:49 GMT

Hi,

Sorry for the delay. Please find the below responses and let us know if you have any doubts.

1)What is the term "Socket" in the Platform Diagram mean?

Here Socket refers to cpu socket. Cpu socket is used to hold cpu in motherboard.

2)What is the term "Local DRAM" and "Remote DRAM" mean?

>Local DRAM metric that shows how often the CPU was stalled on loads from the local memory.

>Remote DRAM metric that shows how often the CPU was stalled on loads from the remote memory.

You can find this below documentation.

https://www.intel.com/content/www/us/en/develop/documentation/vtune-help/top/analyze-performance/microarchitecture-analysis-group/memory-access-analysis.html

3)What is the term "Package_0/1" and "Channel 0-4" mean?

Package0/1 refers to socket here and the channel refers to memory channels on the system. Typically there will be many memory channels for a socket or cpu. Some configs will have 4 and some will have 8 or 16.

4)What is the difference between "DRAM Bandwidth" and "UPI Bandwidth" ?

dram bandwidth->They must be doubled or quadrupled to calculate the total memory bandwidth of a multiprocessor workstation or server.

upi bandwidth->There are typically two to three UPI links between CPU sockets, but this will vary by platform.

Based on processor, transfer speed will vary from model to model.

5)after I try use memory access , I found that there is no way to get memory access profile data of a certain function , I need a top down call stack organization data .

To enable call stack build your application with debug and then enable call stack option by clicking the "collect stack" check box in vtune which will be available in how pane under details section.

6)In other words , is there any better tools for profile NUMA problem, which can be accurate to the specific code line and data structure, or draw all the memory addresses accessed by a piece of code , similar to this , so i can intuitivley see the problem in the code and solve it

Vtune is the tool provided by Intel to profile NUMA problem. Refer the below link to know about it in detail.

https://www.intel.com/content/www/us/en/developer/videos/how-numa-affects-your-workloads-intel-vtune-amplifier.html

Regards,

Janani Chandran

Re: Several questions about vtune memory access analysis

armorhu — Wed, 12 Jan 2022 13:23:46 GMT

Thanks for your Reply.

5)after I try use memory access , I found that there is no way to get memory access profile data of a certain function , I need a top down call stack organization data .

To enable call stack build your application with debug and then enable call stack option by clicking the "collect stack" check box in vtune which will be available in how pane under details section.

----

can you give me the commandline ? we are work in linux and no gui .

6) open the link i get this:

Re:Several questions about vtune memory access analysis

JananiC_Intel — Fri, 14 Jan 2022 11:42:56 GMT

Hi,

Thanks for the update.

1)To compile code with debug symbols use the below command line.

icc -g <source.cxx>

icc is Intel c++ compiler. You can replace it with your compiler.

2)Regarding your video url issue, we suggest you to use google chrome to open the link as we opened from chrome and it was working.

Hope this helps!

Regards,

Janani Chandran

Re:Several questions about vtune memory access analysis

JananiC_Intel — Tue, 25 Jan 2022 05:40:17 GMT

Hi,

Is your issue resolved? Do you have any update?

Regards,

Janani Chandran

Re:Several questions about vtune memory access analysis

JananiC_Intel — Mon, 31 Jan 2022 13:17:23 GMT

Hi,

We assume that your issue is resolved. If you need any additional information, please post a new question as this thread will no longer be monitored by Intel.

Regards,

Janani Chandran