ramsmp - RAM sebesség benchmarkolás - GANGED vs UNGANGED

Segédprogramok

Üdvözlet!

Gondoltam kell ennek egy topic, mert érdekes. Lehet. Valakinek.

Szóval van ez a ramsmp nevű program: http://alasir.com/software/ramspeed/
Amivel különböző benchmarkokat lehet a ramra ráengedni.

Kipróbáltam GANGED és UNGANGED módban is őket. (AMD Phenom II X4 955 BE C3 + 2x2GB DDR2 800-MHz CL5 KIT)
Az OS openSUSE 12.3 amd64, a kernel 3.7-es, AMD K8 kóddal fordítva, egyénileg. Más buhera nincs.

Én kézzel telepítettem a programot és megcsináltam, hogy a 

ramsmp

paranccsal lehessen hívni.

Aztán írtam hozzá egy kis scriptet: 

#!/bin/bash
# init

# This will prompt for the Enter key to be pressed.
echo ''
echo '***************************************************'
echo '*  This will benchmark your memory using ramsmp.  *'
echo '*  Output data will be written to ~/ramtest.log   *'
echo '***************************************************'
echo ''
read -p "Press [Enter] key to start benchmarking..."

#This will start benchmarking
echo ''
echo 'Executing the INTmark test...'
echo 'Executing the INTmark test...' >> ~/ramtest.log
ramsmp -r -b 3 >> ~/ramtest.log
echo 'Finished the INTmark test!' >> ~/ramtest.log
echo 'Executing the FLOATmem test...' >> ~/ramtest.log
echo 'Finished the INTmark test!'
echo 'Executing the FLOATmem test...'
ramsmp -r -b 6 >> ~/ramtest.log
echo 'Finished the FLOATmem test!' >> ~/ramtest.log
echo 'Executing the MMXmark test...' >> ~/ramtest.log
echo 'Finished the FLOATmem test!'
echo 'Executing the MMXmark test...'
ramsmp -r -b 9 >> ~/ramtest.log
echo 'Finished the MMXmark test!' >> ~/ramtest.log
echo 'Executing the SSEmem test...' >> ~/ramtest.log
echo 'Finished the MMXmark test!'
echo 'Executing the SSEmem test...'
ramsmp -r -b 12 >> ~/ramtest.log
echo 'Finished the SSEmem test!' >> ~/ramtest.log
echo 'Executing the MMXmark (nt) test...' >> ~/ramtest.log
echo 'Finished the SSEmem test!'
echo 'Executing the MMXmark (nt) test...'
ramsmp -r -b 15 >> ~/ramtest.log
echo 'Finished the MMXmark (nt) test!' >> ~/ramtest.log
echo 'Executing the SSEmem (nt) test...' >> ~/ramtest.log
echo 'Finished the MMXmark (nt) test!'
echo 'Executing the SSEmem (nt) test...'
ramsmp -r -b 18 >> ~/ramtest.log
echo 'Finished the SSEmem (nt) test!'
echo 'Finished the SSEmem (nt) test!'
>> ~/ramtest.log
echo ''
echo 'Finished all tests!'
echo 'Finished all tests!' >> ~/ramtest.log

Ez lefuttattva a ~/ -be tesz egy ramtest.log fájlt, az eredménnyel. Elég egyszerű dolog.

Szeretném megosztani az eredményeim is!
UNGANGED MÓD

Executing the INTmark test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

INTEGER   Copy:      5856.96 Mb/s
INTEGER   Scale:     5732.60 Mb/s
INTEGER   Add:       5313.90 Mb/s
INTEGER   Triad:     5275.50 Mb/s
---
INTEGER   AVERAGE:   5544.74 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished the INTmark test!
Executing the FLOATmem test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

FL-POINT  Copy:      5898.57 Mb/s
FL-POINT  Scale:     5862.29 Mb/s
FL-POINT  Add:       6459.27 Mb/s
FL-POINT  Triad:     6502.70 Mb/s
---
FL-POINT  AVERAGE:   6180.71 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished the FLOATmem test!
Executing the MMXmark test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

MMX       Copy:      5074.84 Mb/s
MMX       Scale:     5091.86 Mb/s
MMX       Add:       5393.58 Mb/s
MMX       Triad:     5290.56 Mb/s
---
MMX       AVERAGE:   5212.71 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished the MMXmark test!
Executing the SSEmem test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

SSE       Copy:      5320.47 Mb/s
SSE       Scale:     5285.97 Mb/s
SSE       Add:       5561.15 Mb/s
SSE       Triad:     5457.51 Mb/s
---
SSE       AVERAGE:   5406.28 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished the SSEmem test!
Executing the MMXmark (nt) test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

MMX (nt)  Copy:      9676.31 Mb/s  [NTA prefetch]
MMX (nt)  Scale:     9743.89 Mb/s  [NTA prefetch]
MMX (nt)  Add:       8637.64 Mb/s  [T0 prefetch]
MMX (nt)  Triad:     8973.19 Mb/s  [T0 prefetch]
---
MMX (nt)  AVERAGE:   9257.76 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished the MMXmark (nt) test!
Executing the SSEmem (nt) test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

SSE (nt)  Copy:      9517.12 Mb/s  [NTA prefetch]
SSE (nt)  Scale:     9506.87 Mb/s  [NTA prefetch]
SSE (nt)  Add:       8729.61 Mb/s  [T0 prefetch]
SSE (nt)  Triad:     8986.34 Mb/s  [T0 prefetch]
---
SSE (nt)  AVERAGE:   9184.99 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished all tests!

GANGED MÓD 

Executing the INTmark test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

INTEGER   Copy:      4824.60 Mb/s
INTEGER   Scale:     4811.52 Mb/s
INTEGER   Add:       4168.50 Mb/s
INTEGER   Triad:     4068.89 Mb/s
---
INTEGER   AVERAGE:   4468.38 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished the INTmark test!
Executing the FLOATmem test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

FL-POINT  Copy:      4724.16 Mb/s
FL-POINT  Scale:     4758.73 Mb/s
FL-POINT  Add:       5513.20 Mb/s
FL-POINT  Triad:     5545.72 Mb/s
---
FL-POINT  AVERAGE:   5135.45 Mb/s

Finished the FLOATmem test!
Executing the MMXmark test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

MMX       Copy:      4331.54 Mb/s
MMX       Scale:     4387.29 Mb/s
MMX       Add:       4689.86 Mb/s
MMX       Triad:     4632.00 Mb/s
---
MMX       AVERAGE:   4510.17 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished the MMXmark test!
Executing the SSEmem test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

SSE       Copy:      4427.82 Mb/s
SSE       Scale:     4511.41 Mb/s
SSE       Add:       4836.88 Mb/s
SSE       Triad:     4923.06 Mb/s
---
SSE       AVERAGE:   4674.79 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished the SSEmem test!
Executing the MMXmark (nt) test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

MMX (nt)  Copy:      9078.88 Mb/s  [NTA prefetch]
MMX (nt)  Scale:     9052.25 Mb/s  [NTA prefetch]
MMX (nt)  Add:       8313.64 Mb/s  [T0 prefetch]
MMX (nt)  Triad:     8087.18 Mb/s  [T0 prefetch]
---
MMX (nt)  AVERAGE:   8632.99 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished the MMXmark (nt) test!
Executing the SSEmem (nt) test...
RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

SSE (nt)  Copy:      9211.12 Mb/s  [NTA prefetch]
SSE (nt)  Scale:     8837.63 Mb/s  [NTA prefetch]
SSE (nt)  Add:       8135.74 Mb/s  [T0 prefetch]
SSE (nt)  Triad:     8325.29 Mb/s  [T0 prefetch]
---
SSE (nt)  AVERAGE:   8627.44 Mb/s

RAMspeed/SMP (Linux) v3.5.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

8Gb per pass mode, 2 processes

Finished all tests!

Az eredmények eléggé az UNGANGED mód mellett szólnak. Úgy tudom ilyenkor a 2 csatornás ram vezérlő 2x64bit szávszélességgel gazdálkodik, míg GANGED módban 1x128bit sávszéleségel.
Meg tudja mondani valaki, hogy miért jobb az UNGANGED?

Meg persze ha más is megcsinálja a tesztet, itt megoszthatja az eredményét. Csak mint érdekesség.

  • 7185 megtekintés

( uid_395 v | 2013. 10. 20., v – 20:41 )

"Dual-channel was originally conceived as a way to maximize memory throughput by combining two 64-bit buses into a single 128-bit bus. This is retrospectively called the "ganged" mode. However, due to lackluster performance gains in consumer applications as discussed above, more modern implementations of dual-channel use the "unganged" mode by default, which maintains two 64-bit memory buses but allows independent access to each channel, in support of multithreading with multi-core processors."

hirtelen ennyi :)

http://en.wikipedia.org/wiki/Multi-channel_memory_architecture#Ganged_v…