Intel® Moderncode for Parallel Architectures
Support for developing parallel programming applications on Intel® Architecture.

Intel MLC(Memory Latency Checker) distrust results

HyunOck_P_
Beginner
656 Views

Recently, Our servers appears abnormal performance.

So, I tested performance with Intel MLC v2.3 after reboot.

First result is very good.

However, I returned strange result after 3~5 times repeat mlc.

I think last result is credible data.

Finally, I replaced memory and performance becomes good.

In question, Why mlc result is good when immediately reboot?

Why the origin result is show later?

Please help me.

 

Machine : Lenovo x3950 x6 4 computer book / 1 TB memory (each book = 256GB)

------- first test --------

Using buffer size of 200.000MB
Measuring idle latencies (in ns)...
        Memory node
Socket       0       1       2       3
     0   108.7   187.4   224.6   185.4
     1   187.7   108.5   185.7   224.4
     2   224.2   185.6   116.6   185.6
     3   185.0   224.6   189.8   108.8

Measuring Peak Memory Bandwidths for the system
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using traffic with the following read-write ratios
ALL Reads        :      202381.1
3:1 Reads-Writes :      174073.8
2:1 Reads-Writes :      167462.2
1:1 Reads-Writes :      158554.0
Stream-triad like:      159076.0

Measuring Memory Bandwidths between nodes within system
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using Read-only traffic type
        Memory node
 Socket      0       1       2       3
     0  50558.5 12117.5 12069.9 12712.5
     1  12351.6 50751.4 12728.7 12055.4
     2  12268.8 12710.1 50661.5 12173.9
     3  12732.0 12054.1 12104.0 50675.2

-------------End of first test ----------

---------Fourth result------

Using buffer size of 200.000MB
Measuring idle latencies (in ns)...
        Memory node
Socket       0       1       2       3
     0   147.3   261.6   272.1   244.0
     1   247.7   108.7   204.6   236.9
     2   270.2   202.6   108.5   183.8
     3   246.0   236.3   208.3   108.7

Measuring Peak Memory Bandwidths for the system
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using traffic with the following read-write ratios
ALL Reads        :      189662.5
3:1 Reads-Writes :      166338.1
2:1 Reads-Writes :      160815.7
1:1 Reads-Writes :      152592.7
Stream-triad like:      31839.4

Measuring Memory Bandwidths between nodes within system
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using Read-only traffic type
        Memory node
 Socket      0       1       2       3
     0  37788.7  7831.2 12066.2 12714.4
     1   9098.5 50648.0 12686.3  6726.6
     2   9153.9  8003.7 50749.6 12258.4
     3   9191.7  7923.7 12079.6 50702.9

 

0 Kudos
1 Reply
Gael_H_Intel
Moderator
656 Views

Hello,

This is off-topic for this particular forum.  I will see if I can move your question to a more appropriate place.

I see that when you go to the landing page and click on "Forum", it sends you to the Business Client Forum.  I am working on getting this resolved.

 

Gael

0 Kudos
Reply