Difference between revisions of "Intel VTune Tutorial/CPU Architecture"
Intel VTune Tutorial/CPU Architecture
Jump to navigation
Jump to search
(Created page with "Intel VTune Tutorials: CPU Architectures<nowiki /> {{DISPLAYTITLE:Intel VTune Tutorial: CPU Architectures}}<nowiki /> {{Syllabus Intel VTune Tutorial}}...") |
|||
| Line 4: | Line 4: | ||
__TOC__ | __TOC__ | ||
| − | The second tutorial of the Intel VTune | + | The second tutorial of the Intel VTune series serves as a small interlude, providing some context about CPU architecture concepts. |
It covers a brief description of the front-end and back-end of a CPU core. | It covers a brief description of the front-end and back-end of a CPU core. | ||
These concepts are a necessary basis to understand the language of VTune profiles. | These concepts are a necessary basis to understand the language of VTune profiles. | ||
| Line 20: | Line 20: | ||
{ | { | ||
|type="()"} | |type="()"} | ||
| − | + VTunes analysis results are often expressed in these terms | + | + VTunes analysis results are often expressed in these terms |
|| Correct. It is good to know what "Front End", "Back End", "Bad Speculation" or "Retired Instructions" refer to! | || Correct. It is good to know what "Front End", "Back End", "Bad Speculation" or "Retired Instructions" refer to! | ||
| − | - VTune measurements are always presented in a CPU diagram | + | - VTune measurements are always presented in a CPU diagram |
|| Not correct. The data is presented in different ways. This might be a useful gimmick though! | || Not correct. The data is presented in different ways. This might be a useful gimmick though! | ||
- It is impossible to use VTune without a thorough understanding of CPUs! | - It is impossible to use VTune without a thorough understanding of CPUs! | ||
| Line 59: | Line 59: | ||
{ | { | ||
|type="()"} | |type="()"} | ||
| − | - All metrics describe how efficiently data is moved from memory to the CPU | + | - All metrics describe how efficiently data is moved from memory to the CPU |
|| Not correct. This would not be very useful for programs performing complex computations on few data. | || Not correct. This would not be very useful for programs performing complex computations on few data. | ||
| − | - Bad results always highlight inefficiencies in computation | + | - Bad results always highlight inefficiencies in computation |
|| Not correct. This would not cover performance in moving data, or instruction decoding and speculation in the front-end. | || Not correct. This would not cover performance in moving data, or instruction decoding and speculation in the front-end. | ||
| − | + Each metric | + | + Each metric focuses on a different aspect of the architecture |
|| Correct. Each metric is affected by its own class of problems. The interpretation and approaches to improvement can be very different between each metric. | || Correct. Each metric is affected by its own class of problems. The interpretation and approaches to improvement can be very different between each metric. | ||
</quiz> | </quiz> | ||
| Line 72: | Line 72: | ||
{ | { | ||
|type="()"} | |type="()"} | ||
| − | - A source of data from network connections | + | - A source of data from network connections |
|| Not correct. A Network connection is a concept that mostly lives outside of the CPU architecture. | || Not correct. A Network connection is a concept that mostly lives outside of the CPU architecture. | ||
| − | + A channel to read/write data from memory or use co-processors | + | + A channel to read/write data from memory or use co-processors |
| − | || Correct. The | + | || Correct. The execution engine moving data from the L1 cache through ports, for example. |
| − | - CPUs have only a single port, connecting the front-end to the back-end | + | - CPUs have only a single port, connecting the front-end to the back-end |
|| Not correct. There are multiple ports. | || Not correct. There are multiple ports. | ||
</quiz> | </quiz> | ||
{{hidden end}} | {{hidden end}} | ||
Latest revision as of 09:40, 14 June 2022
| Tutorial | |
|---|---|
| Title: | Intel VTune Tutorial |
| Provider: | HPC.NRW
|
| Contact: | tutorials@hpc.nrw |
| Type: | Multi-part video |
| Topic Area: | Performance analysis |
| License: | CC-BY-SA |
| Syllabus
| |
| 1. Introduction | |
| 2. CPU Architecture | |
| 3. Analysis Types | |
| 4. Useful Tips | |
The second tutorial of the Intel VTune series serves as a small interlude, providing some context about CPU architecture concepts. It covers a brief description of the front-end and back-end of a CPU core. These concepts are a necessary basis to understand the language of VTune profiles.
Video