- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
In the above screenshot, is the 56.2% of clockticks for DRAM Bound 56.2% of all clockticks that occurred during the collection?
If that is the case, the 84.2% of clockticks for Local Memory can't be the percentage of all clockticks, right, because that couldn't be larger than 56.2. So is the 84.2% number the percentage of the 56.2% of DRAM clockticks?
Finally, is the percentage of total clockticks proportional to the percentage of elapsed time? For example, if I understand correctly that we have 54.3% of all clockticks spent reading DRAM, and since the elapsed time is 49.914 seconds, is the amount of time spent accessing DRAM 0.543*49.914 = 27.103302 seconds?
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The uarch-exploration is based on TMA, all the CPU metrics are defined in the file below, you can refer:
https://github.com/intel/perfmon/blob/main/TMA_Metrics-full.xlsx
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
The MEM_Bandwidth row has no entries except for SNB (Sandy Bridge if I understand correctly). My machine is Skylake. Should I assume the metric is calculated the same way as for SNB?
Also, do you think you could answer this question directly: is the percentage of total clockticks proportional to the percentage of elapsed time?
If you could answer my first two questions directly as well, that would also be helpful.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Yes, the SKL uses the same formula as SNB.
From the formula, the percentage of total clockticks should be proportional to the percentage of elapsed time.

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page