-
Supercomputing in Plain EnglishPart II: The Tyranny ofthe
Storage HierarchyHenry Neeman, DirectorOU Supercomputing Center for
Education & ResearchUniversity of OklahomaWednesday September 5
2007
OU Supercomputing Center for Education & Research
-
This is an experiment!Its the nature of these kinds of
videoconferences that failures are guaranteed to happen!NO
PROMISES!So, please bear with us. Hopefully everything will work
out well enough.
OU Supercomputing Center for Education & Research
-
Access Grid/VRVSIf youre connecting via the Access Grid or VRVS,
the venue is:NCSA Venue WalkaboutIts available Wed Sep 5 2007
1:00-4:30pm Central Time, but the workshop starts at 3:00pm Central
Time.Many thanks to John Chapman of U Arkansas for setting this up
for us.
OU Supercomputing Center for Education & Research
-
iLincWe only have about 40-45 simultaneous iLinc connections
available.Therefore, each institution has at most one iLinc person
designated.If youre the iLinc person for your institution, youve
already gotten e-mail about it, so please follow the
instructions.If you arent your institutions iLinc person, then you
cant become it, because were completely out of iLinc
connections.Many thanks to Katherine Kantardjieff of California
State U Fullerton for setting this up for us.
OU Supercomputing Center for Education & Research
-
QuickTime BroadcastIf you dont have iLinc, you can connect via
QuickTime:rtsp://129.15.254.141/neeman_02.sdpWe strongly recommend
using QuickTime player, since weve seen it work.When you run it,
traverse the menusFile -> Open URLThen paste in the rstp URL the
Movie URL space, and click OK.Many thanks to Kevin Blake of OU for
setting this up.
OU Supercomputing Center for Education & Research
-
Phone BridgeIf all else fails, you can call into our phone
bridge:1-866-285-7778, access code 6483137#Please mute yourself and
use the phone to listen.Dont worry, Ill call out slide numbers as
we go.To ask questions, please use Google Talk or Gmail.Many thanks
to Amy Apon of U Arkansas for setting this up for us, and to U
Arkansas for absorbing the costs.
OU Supercomputing Center for Education & Research
-
Google TalkTo ask questions, please use our Google Talk group
chat session (text only).You need to have (or create) a gmail.com
account to use Google Talk.Once youve logged in to your gmail.com
account, go to:http://www.google.com/talk/and then contact the user
named:oscer.sipeAlternatively, you can send your questions by
e-mail to [email protected].
OU Supercomputing Center for Education & Research
-
This is an experiment!REMINDER:Its the nature of these kinds of
videoconferences that failures are guaranteed to happen!NO
PROMISES!So, please bear with us. Hopefully everything will work
out well enough.
OU Supercomputing Center for Education & Research
-
Okla. Supercomputing Symposium2006 Keynote:Dan AtkinsHead of
NSFsOffice ofCyberinfrastructure2004 Keynote:Sangtae KimNSF
SharedCyberinfrastructureDivision Director2003 Keynote:Peter
FreemanNSFComputer & InformationScience &
EngineeringAssistant Director2005 Keynote:Walt BrooksNASA
AdvancedSupercomputingDivision
Directorhttp://symposium2007.oscer.ou.edu/2007 Keynote:Jay
BoisseauDirectorTexas AdvancedComputing CenterUniv Texas AustinWed
Oct 3 2007 @ OUOver 280 registrations already!Free MPI workshop Tue
Oct 2!FREE Symposium! FREE Food!
OU Supercomputing Center for Education & Research
-
OutlineWhat is the storage hierarchy?RegistersCacheMain Memory
(RAM)The Relationship Between RAM and CacheThe Importance of Being
LocalHard DiskVirtual Memory
OU Supercomputing Center for Education & Research
-
What is the Storage Hierarchy?RegistersCache memoryMain memory
(RAM)Hard diskRemovable media (e.g., CDROM)InternetFast, expensive,
fewSlow, cheap, a lot[1]
OU Supercomputing Center for Education & Research
-
Henrys LaptopPentium 4 Core Duo T2400 1.83 GHz w/2 MB L2 Cache2
GB (2048 MB) 667 MHz DDR2 SDRAM100 GB 7200 RPM SATA Hard
DriveDVD+RW/CD-RW Drive (8x)1 Gbps Ethernet Adapter56 Kbps Phone
ModemDell Latitude D620[3]
OU Supercomputing Center for Education & Research
-
Storage Speed, Size, Cost* MFLOP/s: millions of floating point
operations per second** 8 64-bit integer registers, 8 80-bit
floating point registers, 16 128-bit floating point XMM
registers
OU Supercomputing Center for Education & Research
-
Registers[25]
OU Supercomputing Center for Education & Research
-
What Are Registers?Registers are memory-like locations inside
the Central Processing Unit that hold data that are being used
right now in operations.Arithmetic/Logic UnitControl
UnitRegistersFetch Next
InstructionAddSubMultDivAndOrNotIntegerFloating PointFetch
DataStore DataIncrement Instruction PtrExecute InstructionCPU
OU Supercomputing Center for Education & Research
-
How Registers Are UsedEvery arithmetic or logical operation has
one or more operands and one result.Operands are contained in
source registers.A black box of circuits performs the operation.The
result goes into a destination register.Example:addend in R0augend
in R1ADDsum in R25712Register RiRegister RjRegister
RkoperandoperandresultOperation circuitry
OU Supercomputing Center for Education & Research
-
How Many Registers?Typically, a CPU has less than 4 KB (4096
bytes) of registers, usually split into registers for holding
integer values and registers for holding floating point (real)
values, plus a few special purpose registers.Examples:IBM POWER5+
(found in IBM p-Series supercomputers): 80 64-bit integer registers
and 72 64-bit floating point registers (1,216 bytes) [12]Intel
Pentium4 EM64T: 8 64-bit integer registers, 8 80-bit floating point
registers, 16 128-bit floating point vector registers (400 bytes)
[4]Intel Itanium2: 128 64-bit integer registers, 128 82-bit
floating point registers (2304 bytes) [23]
OU Supercomputing Center for Education & Research
-
Cache[4]
OU Supercomputing Center for Education & Research
-
What is Cache?A special kind of memory where data reside that
are about to be used or have just been used.Very fast => very
expensive => very small (typically 100 to 10,000 times as
expensive as RAM per byte)Data in cache can be loaded into or
stored from registers at speeds comparable to the speed of
performing computations.Data that are not in cache (but that are in
Main Memory) take much longer to load or store.Cache is near the
CPU: either inside the CPU or on the motherboard that the CPU sits
on.
OU Supercomputing Center for Education & Research
-
From Cache to the CPUTypically, data move between cache and the
CPU at speeds relatively near to that of the CPU performing
calculations.351 GB/sec[7]
OU Supercomputing Center for Education & Research
-
Multiple Levels of CacheMost contemporary CPUs have more than
one level of cache. For example:Intel Pentium4 EM64T (Yonah)
[??]Level 1 caches: 32 KB instruction, 32 KB dataLevel 2 cache:
2048 KB unified (instruction+data)IBM POWER4 [12]Level 1 cache: 64
KB instruction, 32 KB dataLevel 2 cache: 1440 KB unified for each 2
CPUsLevel 3 cache: 32 MB unified for each 2 CPUS
OU Supercomputing Center for Education & Research
-
Why Multiple Levels of Cache?The lower the level of cache:the
faster the cache can transfer data to the CPU;the smaller that
level of cache is, becausefaster => more expensive =>
smaller.Example: IBM POWER4 latency to the CPU [12]L1 cache: 4
cycles = 3.6 ns for 1.1 GHz CPUL2 cache: 14 cycles = 12.7 ns for
1.1 GHz CPUExample: Intel Itanium2 latency to the CPU [19]L1 cache:
1 cycle = 1.0 ns for 1.0 GHz CPUL2 cache: 5 cycles = 5.0 ns for 1.0
GHz CPUL3 cache: 12-15 cycles = 12 15 ns for 1.0 GHz CPUExample:
Intel Pentium4 (Yonah) [??]L1 cache: 3 cycles = 1.64 ns for a 1.83
GHz CPU = 12 calculationsL2 cache: 14 cycles = 7.65 ns for a 1.83
GHz CPU = 56 calculationsRAM: 48 cycles = 26.2 ns for a 1.83 GHz
CPU = 192 calculations
OU Supercomputing Center for Education & Research
-
Cache & RAM LatenciesBetter[26]
OU Supercomputing Center for Education & Research
MemLat
3
3.04
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3.02
3
3
3
3
3
3
3
3.02
3
3
3
3
3
3
3
3.02
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3.02
3
3
3
3
3.03
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3.02
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3.6
6.8
9.77
12.75
14.14
14.15
14.15
14.14
14.14
14.18
14.15
14.14
14.15
14.15
14.14
14.18
14.16
14.16
14.14
14.19
14.14
14.14
14.13
14.14
14.14
14.13
14.17
14.15
14.14
14.16
14.14
14.16
14.15
14.16
14.14
14.16
14.14
14.18
14.14
14.16
14.16
14.14
14.15
14.14
14.22
14.16
14.14
14.17
14.14
14.14
14.14
14.17
14.15
14.14
14.16
14.17
14.14
14.14
14.14
14.14
14.14
14.17
14.18
14.16
14.15
14.16
14.15
14.14
14.14
14.16
14.14
14.19
14.15
14.16
14.13
14.14
14.15
14.14
14.16
14.18
14.16
14.14
14.14
14.15
14.14
14.17
14.17
14.17
14.16
14.15
14.2
14.19
14.19
14.21
14.25
14.23
14.26
14.26
14.24
14.25
14.26
14.24
14.27
14.25
14.27
14.26
14.41
14.24
14.28
14.29
14.26
14.28
14.25
14.34
14.3
14.51
14.31
14.37
14.28
14.38
14.54
14.44
14.43
15.08
14.51
14.69
15.33
14.95
16.21
16.72
16.65
18.42
19.44
20.33
26.06
31.63
37.64
42.81
44.53
45.52
45.24
45.73
46.74
46.48
46.53
46.75
46.49
46.62
46.89
47.01
46.86
46.93
47.05
46.89
47.3
47.1
46.94
46.81
47.23
46.9
47.31
47.52
47.12
47.63
46.96
47.11
47.25
47.11
47.36
47.81
47.54
47.12
47.38
47.24
47.39
47.55
47.35
47.4
47.24
47.19
47.51
47.52
47.53
47.29
47.64
48.85
47.66
3 cycles
14 cycles
47 cycles
Memory Latency
Array Size (bytes)
Latency (clock cycles)
Cache & RAM Latency: Intel T2400 (1.83 GHz)
Sheet1
10243
10883.04
11523
12163
12803
13443
14083
14723
15363
16003
16643
17283
17923
18563
19203
19843
20483
21123
22403.02
23683
24963
26243
27523
28803
30083
31363
32643.02
33923
35203
36483
37763
39043
40323
41603
43523.02
45443
47363
49283
51203
53123
55043
56963
58883
60803
62723
65283
67843
70403
72963
75523
78083.02
80643
83203
86403
89603
92803.03
96003
99203
102403
105603
109443
113283
117123
120963
124803
129283
133763
138243
142723
147203
152323
157443
162563
167683
173443
179203
184963
191363.02
197763
204163
210563
217603
224643
231683
239363
247043
255363
263683
272003
280963
289923
299523
309123
319363
329603.6
340486.8
351369.77
3628812.75
3744014.14
3865614.15
3987214.15
4115214.14
4249614.14
4384014.18
4524814.15
4672014.14
4819214.15
4972814.15
5132814.14
5299214.18
5465614.16
5638414.16
5817614.14
6003214.19
6195214.14
6393614.14
6598414.13
6809614.14
7027214.14
7251214.13
7481614.17
7718414.15
7961614.14
8211214.16
8473614.14
8742414.16
9017614.15
9305614.16
9600014.14
9900814.16
10214414.14
10534414.18
10867214.14
11212814.16
11564814.16
11929614.14
12307214.15
12697614.14
13094414.22
13504014.16
13926414.14
14361614.17
14816014.14
15283214.14
15763214.14
16256014.17
16768014.15
17292814.14
17836814.16
18400014.17
18976014.14
19571214.14
20185614.14
20819214.14
21472014.14
22144014.17
22841614.18
23558414.16
24300814.15
25062414.16
25849614.15
26662414.14
27500814.14
28364814.16
29254414.14
30169614.19
31116814.15
32089614.16
33094414.13
34131214.14
35200014.15
36300814.14
37440014.16
38611214.18
39820814.16
41068814.14
42355214.14
43680014.15
45049614.14
46457614.17
47910414.17
49408014.17
50956814.16
52550414.15
54195214.2
55891214.19
57638414.19
59443214.21
61305614.25
63225614.23
65203214.26
67244814.26
69350414.24
71520014.25
73760014.26
76070414.24
78451214.27
80908814.25
83443214.27
86054414.26
88748814.41
91526414.24
94387214.28
97337614.29
100384014.26
103526414.28
106764814.25
110105614.34
113548814.3
117100814.51
120761614.31
124537614.37
128435214.28
132454414.38
136595214.54
140864014.44
145267214.43
149811215.08
154496014.51
159328014.69
164307215.33
169446414.95
174745616.21
180211216.72
185843216.65
191654418.42
197644819.44
203827220.33
210201626.06
216774431.63
223552037.64
230540842.81
237747244.53
245177645.52
252844845.24
260748845.73
268902446.74
277305646.48
285977646.53
294918446.75
304140846.49
313651246.62
323456046.89
333568047.01
343993646.86
354745646.93
365836847.05
377273646.89
389068847.3
401228847.1
413772846.94
426707246.81
440044847.23
453798446.9
467980847.31
482611247.52
497696047.12
513254447.63
529299246.96
545843247.11
562905647.25
580499247.11
598643247.36
617356847.81
636652847.54
656550447.12
677068847.38
698227247.24
720051247.39
742553647.55
765760047.35
789696047.4
814374447.24
839827247.19
866073647.51
893139247.52
921056047.53
949843247.29
979526447.64
1010137648.85
1041708847.66
rmma_intelt2400_memfwdlat_20070
RIGHTMARK MEMORY ANALYZER V3.72 TEST RESULTS
============================================
CPU ModelGenuine Intel(R) Core(TM) Duo (Yonah) 1828.7 MHz
L1 Cache Line Size64 bytes
L2 Cache Line Size128 bytes
Test TypeD-Cache Latency
Test StatusCompleted successfully
Selected Tests120
Set Size134217728 bytes
Memory Allocation1
Thread Lock0
Min Block Size1024 bytes
Max Block Size10485760 bytes
Stride Size64 bytes
Segments Count1
NOP Count0
Measurement Mode0
NOP Latency0
D-CACHE FORWARD READ LATENCY TEST
=================================
Size(bytes)Latency(cycles)Latency(ns)
102431.64
10883.041.66
115231.64
121631.64
128031.64
134431.64
140831.64
147231.64
153631.64
160031.64
166431.64
172831.64
179231.64
185631.64
192031.64
198431.64
204831.64
211231.64
22403.021.65
236831.64
249631.64
262431.64
275231.64
288031.64
300831.64
313631.64
32643.021.65
339231.64
352031.64
364831.64
377631.64
390431.64
403231.64
416031.64
43523.021.65
454431.64
473631.64
492831.64
512031.64
531231.64
550431.64
569631.64
588831.64
608031.64
627231.64
652831.64
678431.64
704031.64
729631.64
755231.64
78083.021.65
806431.64
832031.64
864031.64
896031.64
92803.031.65
960031.64
992031.64
1024031.64
1056031.64
1094431.64
1132831.64
1171231.64
1209631.64
1248031.64
1292831.64
1337631.64
1382431.64
1427231.64
1472031.64
1523231.64
1574431.64
1625631.64
1676831.64
1734431.64
1792031.64
1849631.64
191363.021.65
1977631.64
2041631.64
2105631.64
2176031.64
2246431.64
2316831.64
2393631.64
2470431.64
2553631.64
2636831.64
2720031.64
2809631.64
2899231.64
2995231.64
3091231.64
3193631.64
329603.61.97
340486.83.72
351369.775.34
3628812.756.97
3744014.147.73
3865614.157.74
3987214.157.74
4115214.147.73
4249614.147.73
4384014.187.76
4524814.157.74
4672014.147.73
4819214.157.74
4972814.157.74
5132814.147.73
5299214.187.76
5465614.167.75
5638414.167.74
5817614.147.73
6003214.197.76
6195214.147.73
6393614.147.73
6598414.137.73
6809614.147.73
7027214.147.73
7251214.137.73
7481614.177.75
7718414.157.74
7961614.147.73
8211214.167.74
8473614.147.73
8742414.167.74
9017614.157.74
9305614.167.74
9600014.147.73
9900814.167.74
10214414.147.73
10534414.187.76
10867214.147.73
11212814.167.75
11564814.167.75
11929614.147.73
12307214.157.74
12697614.147.73
13094414.227.77
13504014.167.74
13926414.147.73
14361614.177.75
14816014.147.73
15283214.147.73
15763214.147.73
16256014.177.75
16768014.157.74
17292814.147.73
17836814.167.74
18400014.177.75
18976014.147.73
19571214.147.73
20185614.147.73
20819214.147.73
21472014.147.73
22144014.177.75
22841614.187.75
23558414.167.74
24300814.157.74
25062414.167.74
25849614.157.74
26662414.147.73
27500814.147.73
28364814.167.74
29254414.147.73
30169614.197.76
31116814.157.74
32089614.167.74
33094414.137.73
34131214.147.73
35200014.157.74
36300814.147.73
37440014.167.74
38611214.187.76
39820814.167.74
41068814.147.73
42355214.147.73
43680014.157.74
45049614.147.73
46457614.177.75
47910414.177.75
49408014.177.75
50956814.167.75
52550414.157.74
54195214.27.76
55891214.197.76
57638414.197.76
59443214.217.77
61305614.257.79
63225614.237.78
65203214.267.8
67244814.267.8
69350414.247.79
71520014.257.79
73760014.267.8
76070414.247.79
78451214.277.81
80908814.257.79
83443214.277.8
86054414.267.8
88748814.417.88
91526414.247.79
94387214.287.81
97337614.297.82
100384014.267.8
103526414.287.81
106764814.257.79
110105614.347.84
113548814.37.82
117100814.517.93
120761614.317.82
124537614.377.86
128435214.287.81
132454414.387.87
136595214.547.95
140864014.447.89
145267214.437.89
149811215.088.25
154496014.517.94
159328014.698.03
164307215.338.38
169446414.958.18
174745616.218.86
180211216.729.14
185843216.659.1
191654418.4210.07
197644819.4410.63
203827220.3311.12
210201626.0614.25
216774431.6317.3
223552037.6420.58
230540842.8123.41
237747244.5324.35
245177645.5224.89
252844845.2424.74
260748845.7325.01
268902446.7425.56
277305646.4825.42
285977646.5325.45
294918446.7525.56
304140846.4925.42
313651246.6225.49
323456046.8925.64
333568047.0125.71
343993646.8625.62
354745646.9325.66
365836847.0525.73
377273646.8925.64
389068847.325.86
401228847.125.75
413772846.9425.67
426707246.8125.6
440044847.2325.83
453798446.925.64
467980847.3125.87
482611247.5225.99
497696047.1225.77
513254447.6326.05
529299246.9625.68
545843247.1125.76
562905647.2525.84
580499247.1125.76
598643247.3625.9
617356847.8126.14
636652847.5426
656550447.1225.77
677068847.3825.91
698227247.2425.83
720051247.3925.91
742553647.5526
765760047.3525.89
789696047.425.92
814374447.2425.83
839827247.1925.81
866073647.5125.98
893139247.5225.98
921056047.5325.99
949843247.2925.86
979526447.6426.05
1010137648.8526.71
1041708847.6626.06
D-CACHE BACKWARD READ LATENCY TEST
==================================
Size(bytes)Latency(cycles)Latency(ns)
102431.64
10883.021.65
115231.64
121631.64
128031.64
134431.64
140831.64
147231.64
153631.64
160031.64
166431.64
172831.64
179231.64
185631.64
192031.64
198431.64
204831.64
211231.64
224031.64
236831.64
249631.64
262431.64
275231.64
288031.64
300831.64
313631.64
326431.64
339231.64
352031.64
364831.64
377631.64
390431.64
403231.64
416031.64
435231.64
454431.64
473631.64
492831.64
512031.64
531231.64
550431.64
569631.64
588831.64
608031.64
627231.64
652831.64
678431.64
704031.64
729631.64
755231.64
780831.64
806431.64
832031.64
864031.64
896031.64
928031.64
960031.64
992031.64
1024031.64
1056031.64
1094431.64
1132831.64
1171231.64
1209631.64
1248031.64
1292831.64
1337631.64
1382431.64
1427231.64
1472031.64
1523231.64
1574431.64
1625631.64
1676831.64
1734431.64
1792031.64
1849631.64
1913631.64
1977631.64
2041631.64
2105631.64
2176031.64
2246431.64
2316831.64
2393631.64
2470431.64
2553631.64
2636831.64
2720031.64
2809631.64
2899231.64
2995231.64
3091231.64
3193631.64
329603.61.97
340486.813.72
351369.745.33
3628812.756.97
3744014.137.73
3865614.197.76
3987214.167.74
4115214.147.73
4249614.147.73
4384014.147.73
4524814.167.74
4672014.157.74
4819214.167.74
4972814.157.74
5132814.147.73
5299214.167.74
5465614.147.73
5638414.137.73
5817614.147.73
6003214.157.74
6195214.167.74
6393614.187.75
6598414.147.73
6809614.137.73
7027214.157.74
7251214.147.73
7481614.167.74
7718414.167.75
7961614.157.74
8211214.147.73
8473614.147.73
8742414.167.74
9017614.147.73
9305614.137.73
9600014.147.73
9900814.147.73
10214414.157.74
10534414.197.76
10867214.137.73
11212814.157.74
11564814.157.74
11929614.137.73
12307214.167.74
12697614.147.73
13094414.157.74
13504014.27.77
13926414.157.74
14361614.167.75
14816014.147.73
15283214.137.73
15763214.147.73
16256014.157.74
16768014.157.74
17292814.167.74
17836814.147.73
18400014.147.73
18976014.157.74
19571214.187.75
20185614.157.74
20819214.147.73
21472014.167.74
22144014.167.74
22841614.167.74
23558414.147.73
24300814.147.73
25062414.137.73
25849614.197.76
26662414.167.74
27500814.147.73
28364814.157.74
29254414.167.74
30169614.167.74
31116814.157.74
32089614.177.75
33094414.177.75
34131214.157.74
35200014.147.73
36300814.187.75
37440014.157.74
38611214.157.74
39820814.147.73
41068814.187.76
42355214.177.75
43680014.147.73
45049614.147.73
46457614.147.73
47910414.147.73
49408014.157.74
50956814.157.74
52550414.167.74
54195214.197.76
55891214.187.76
57638414.217.77
59443214.37.82
61305614.327.83
63225614.277.8
65203214.287.81
67244814.297.81
69350414.337.84
71520014.257.79
73760014.617.99
76070414.337.84
78451214.387.86
80908814.337.84
83443214.357.85
86054414.437.89
88748814.367.85
91526414.267.8
94387214.397.87
97337614.287.81
100384014.317.82
103526414.387.87
106764814.347.84
110105614.437.89
113548814.417.88
117100814.517.93
120761614.347.84
124537614.47.87
128435214.397.87
132454414.517.94
136595214.357.85
140864014.668.02
145267215.348.39
149811214.778.08
154496016.59.02
159328015.198.31
164307215.788.63
169446415.278.35
174745615.28.31
180211215.788.63
185843217.589.61
191654418.5510.14
197644820.0510.97
203827221.611.81
210201626.2514.36
216774432.2617.64
223552039.621.65
230540842.9623.49
237747245.6224.95
245177645.9725.14
252844846.6125.49
260748846.8225.6
268902446.1825.25
277305647.2125.82
285977646.8725.63
294918446.9525.67
304140847.9326.21
313651247.2425.83
323456046.6425.5
333568047.2325.82
343993647.3425.89
354745646.9725.68
365836847.2825.86
377273647.4425.94
389068847.6426.05
401228846.9825.69
413772847.5225.99
426707247.4225.93
440044847.3725.9
453798447.4925.97
467980847.7126.09
482611247.9926.24
497696047.6726.07
513254447.425.92
529299247.6626.06
545843247.8526.17
562905647.6826.08
58049924826.25
598643247.4925.97
617356847.726.08
636652847.6326.05
656550448.0726.29
677068847.4925.97
698227247.7126.09
720051248.2626.39
742553647.3425.89
765760047.2825.86
789696047.6426.05
814374447.8826.18
839827248.0526.28
866073647.7926.13
893139247.8526.17
921056047.626.03
949843247.4125.92
979526447.8626.17
1010137648.8926.73
1041708847.4625.95
D-CACHE RANDOM READ LATENCY TEST
================================
Size(bytes)Latency(cycles)Latency(ns)
10243.031.65
108831.64
115231.64
121631.64
128031.64
13443.031.65
140831.64
147231.64
153631.64
160031.64
166431.64
172831.64
179231.64
185631.64
192031.64
19843.011.65
204831.64
211231.64
224031.64
236831.64
249631.64
26243.031.65
275231.64
288031.64
300831.64
31363.021.65
326431.64
339231.64
352031.64
364831.64
37763.021.65
390431.64
403231.64
416031.64
435231.64
45443.031.65
473631.64
492831.64
512031.64
531231.64
55043.021.65
569631.64
588831.64
608031.64
627231.64
65283.021.65
678431.64
704031.64
729631.64
755231.64
78083.021.65
806431.64
832031.64
864031.64
896031.64
928031.64
960031.64
992031.64
1024031.64
105603.021.65
1094431.64
1132831.64
1171231.64
1209631.64
1248031.64
1292831.64
1337631.64
1382431.64
1427231.64
147203.031.65
1523231.64
1574431.64
1625631.64
1676831.64
173443.031.66
1792031.64
1849631.64
1913631.64
197763.021.65
2041631.64
2105631.64
2176031.64
2246431.64
2316831.64
239363.021.65
2470431.64
2553631.64
2636831.64
272003.031.66
2809631.64
2899231.64
2995231.64
3091231.64
319363.031.66
329603.61.97
340486.653.64
351369.575.24
3628812.556.86
3744013.987.65
3865614.017.66
3987214.157.74
4115214.177.75
4249614.217.77
4384014.147.73
4524814.167.74
4672014.147.73
4819214.157.74
4972814.147.73
5132814.147.73
5299214.177.75
5465614.157.74
5638414.187.75
5817614.157.74
6003214.137.73
6195214.167.74
6393614.157.74
6598414.137.73
6809614.187.75
7027214.177.75
7251214.167.74
7481614.177.75
7718414.167.74
7961614.167.74
8211214.167.74
8473614.157.74
8742414.147.73
9017614.167.74
9305614.147.73
9600014.137.73
9900814.167.75
10214414.157.74
10534414.137.73
10867214.157.74
11212814.157.74
11564814.137.73
11929614.147.73
12307214.147.73
12697614.147.73
13094414.187.75
13504014.147.73
13926414.147.73
14361614.167.74
14816014.167.74
15283214.147.73
15763214.157.74
16256014.147.73
16768014.147.73
17292814.157.74
17836814.157.74
18400014.137.73
18976014.167.74
19571214.167.74
20185614.147.73
20819214.147.73
21472014.157.74
22144014.157.74
22841614.167.75
23558414.157.74
24300814.157.74
25062414.137.73
25849614.147.73
26662414.157.73
27500814.147.73
28364814.177.75
29254414.187.76
30169614.147.73
31116814.167.75
32089614.167.74
33094414.167.74
34131214.167.75
35200014.167.74
36300814.157.74
37440014.187.75
38611214.157.74
39820814.147.73
41068814.157.74
42355214.147.73
43680014.177.75
45049614.157.74
46457614.177.75
47910414.267.8
49408014.167.74
50956814.157.74
52550414.197.76
54195214.367.85
55891214.487.92
57638414.67.98
59443214.748.06
61305614.868.12
63225615.038.22
65203215.18.25
67244815.218.32
69350415.278.35
71520015.378.4
73760015.488.46
76070415.648.55
78451215.698.58
80908815.778.62
83443215.958.72
86054415.958.72
88748816.048.77
91526416.18.81
94387216.258.88
97337616.318.92
100384016.378.95
103526416.519.03
106764816.839.21
110105616.659.1
113548816.889.23
117100816.799.18
120761616.849.21
124537616.959.27
1284352179.3
132454417.159.38
136595217.319.47
140864017.49.52
145267217.549.59
149811217.439.53
154496017.69.62
159328018.4110.07
164307219.0810.44
169446418.3510.03
174745619.0710.43
180211219.9110.89
185843221.9812.02
191654423.4112.8
197644836.4619.93
203827242.323.13
210201649.3426.98
216774475.0541.04
223552094.0151.41
2305408101.1755.32
2377472108.3159.23
2451776111.6261.04
2528448114.4262.57
2607488116.7263.83
2689024116.0363.45
2773056116.7363.83
2859776117.3464.16
2949184118.0364.54
3041408120.8266.07
3136512121.8866.65
3234560124.1767.9
3335680127.0569.47
3439936127.3669.64
3547456129.8771.02
3658368131.2571.77
3772736136.5474.67
3890688135.9674.35
4012288140.1476.63
413772814177.1
4267072142.6778.01
4400448145.4779.55
4537984146.5480.13
4679808149.4881.74
4826112151.6682.93
4976960151.682.9
5132544155.9185.26
5292992156.2785.45
5458432157.4286.08
5629056158.6586.76
5804992159.0386.96
5986432161.0788.08
6173568162.0688.62
6366528163.589.4
6565504165.1490.3
6770688167.2891.47
6982272167.9991.86
7200512169.9592.93
7425536170.8993.45
7657600171.9894.04
7896960173.1394.67
8143744174.5195.43
8398272175.3895.9
8660736176.1496.32
8931392177.4697.04
9210560178.8697.81
9498432179.698.21
9795264180.4198.65
10101376179.5398.17
10417088180.2598.56
D-CACHE PSEUDO-RANDOM READ LATENCY TEST
=======================================
Size(bytes)Latency(cycles)Latency(ns)
10243.021.65
108831.64
115231.64
121631.64
12803.031.65
134431.64
140831.64
147231.64
153631.64
16003.031.65
166431.64
172831.64
179231.64
18563.031.65
192031.64
198431.64
204831.64
211231.64
224031.64
236831.64
249631.64
262431.64
27523.031.65
288031.64
300831.64
313631.64
32643.021.65
339231.64
352031.64
364831.64
377631.64
39043.031.66
403231.64
416031.64
435231.64
454431.64
473631.64
492831.64
512031.64
53123.031.66
550431.64
569631.64
588831.64
60803.021.65
627231.64
652831.64
678431.64
70403.021.65
729631.64
755231.64
780831.64
806431.64
83203.021.65
864031.64
896031.64
928031.64
96003.031.65
992031.64
1024031.64
1056031.64
109443.021.65
1132831.64
1171231.64
1209631.64
1248031.64
1292831.64
1337631.64
1382431.64
1427231.64
1472031.64
1523231.64
1574431.64
1625631.64
167683.031.65
1734431.64
1792031.64
1849631.64
1913631.64
1977631.64
2041631.64
2105631.64
217603.031.66
2246431.64
2316831.64
2393631.64
2470431.64
255363.031.66
2636831.64
2720031.64
2809631.64
289923.021.65
2995231.64
3091231.64
3193631.64
329603.611.97
340486.83.72
351369.85.36
3628812.756.97
3744014.167.74
3865614.187.75
3987214.177.75
4115214.167.74
4249614.157.74
4384014.157.74
4524814.167.74
4672014.157.74
4819214.147.73
4972814.157.74
5132814.147.73
5299214.177.75
5465614.197.76
5638414.197.76
5817614.157.74
6003214.177.75
6195214.167.74
6393614.167.75
6598414.187.75
6809614.167.74
7027214.167.74
7251214.217.77
7481614.157.74
7718414.187.75
7961614.217.77
8211214.187.75
8473614.197.76
8742414.217.77
9017614.137.73
9305614.177.75
9600014.167.74
9900814.197.76
10214414.147.73
10534414.157.74
10867214.167.74
11212814.217.77
11564814.167.74
11929614.167.75
12307214.157.74
12697614.167.74
13094414.167.75
13504014.157.74
13926414.187.75
14361614.177.75
14816014.167.74
15283214.177.75
15763214.217.77
16256014.187.75
16768014.197.76
17292814.177.75
17836814.187.75
18400014.27.77
18976014.187.76
19571214.187.75
20185614.157.74
20819214.187.76
21472014.167.74
22144014.177.75
22841614.177.75
23558414.197.76
24300814.187.76
25062414.167.74
25849614.187.75
26662414.177.75
27500814.197.76
28364814.167.75
29254414.217.77
30169614.187.75
31116814.177.75
32089614.187.76
33094414.187.75
34131214.157.74
35200014.197.76
36300814.177.75
37440014.177.75
38611214.177.75
39820814.167.75
41068814.27.77
42355214.177.75
43680014.187.76
45049614.187.76
46457614.167.74
47910414.167.74
49408014.187.75
50956814.157.74
52550414.177.75
54195214.187.76
55891214.237.78
57638414.227.78
59443214.237.78
61305614.267.8
63225614.277.8
65203214.297.82
67244814.297.82
69350414.267.8
71520014.277.8
73760014.337.84
76070414.317.83
78451214.357.84
80908814.297.81
83443214.297.81
86054414.337.84
88748814.297.81
91526414.277.81
94387214.277.8
97337614.437.89
100384014.317.82
103526414.277.8
106764814.277.8
110105614.347.84
113548814.337.83
117100814.327.83
120761614.437.89
124537614.347.84
128435214.37.82
132454414.357.85
136595214.698.03
140864014.397.87
145267214.387.86
149811214.638
154496014.597.98
159328014.78.04
164307214.968.18
169446415.238.33
174745615.218.31
180211216.058.78
185843218.219.96
191654419.4610.64
197644823.3612.77
203827229.7916.29
210201633.9618.57
216774461.8633.82
223552077.5542.41
230540889.1448.74
237747294.651.73
245177694.9251.9
252844895.4952.21
260748896.8252.95
268902497.5753.36
277305697.4453.28
285977698.153.65
294918498.2353.71
304140898.2653.73
313651298.3953.8
323456098.6753.95
333568099.254.25
343993699.1454.21
354745698.9354.1
365836899.1854.23
377273699.2954.3
389068899.8654.6
401228899.4154.36
413772899.3154.31
426707299.7254.53
440044899.6854.51
453798499.3654.33
467980899.1454.21
482611299.5554.44
497696099.6754.5
513254499.3154.31
529299299.2154.25
545843299.4854.4
5629056100.1254.75
580499299.4354.37
598643299.854.58
617356899.4354.37
636652899.5154.42
656550499.3954.35
677068899.6754.5
698227299.4954.4
720051299.254.25
742553699.9854.67
765760099.3854.34
789696099.8654.61
814374499.8254.59
839827299.5854.45
8660736100.2454.82
8931392100.2954.84
9210560100.1554.76
9498432100.0254.69
979526499.8754.61
10101376100.4554.93
10417088100.1654.77
-
Main Memory[13]
OU Supercomputing Center for Education & Research
-
What is Main Memory?Where data reside for a program that is
currently runningSometimes called RAM (Random Access Memory): you
can load from or store into any main memory location at any
timeSometimes called core (from magnetic cores that some memories
used, many years ago)Much slower => much cheaper => much
bigger
OU Supercomputing Center for Education & Research
-
What Main Memory Looks Like012345678910536,870,911You can think
of main memory as a big long 1D array of bytes.
OU Supercomputing Center for Education & Research
-
The Relationship BetweenMain Memory & Cache
OU Supercomputing Center for Education & Research
-
RAM is SlowCPU351 GB/sec[7]3.5 GB/sec[26] (1%)BottleneckThe
speed of data transferbetween Main Memory and theCPU is much slower
than thespeed of calculating, so the CPUspends most of its time
waitingfor data to come in or go out.
OU Supercomputing Center for Education & Research
-
Why Have Cache?CPUCache is nearly the same speedas the CPU, so
the CPU doesnthave to wait nearly as long forstuff thats already in
cache:it can do moreoperations per second!351 GB/sec[7]3.5
GB/sec[26] (1%)29.3 GB/sec[26] (8%)
OU Supercomputing Center for Education & Research
-
Cache & RAM Bandwidths[26]
OU Supercomputing Center for Education & Research
Cache & RAM Bandwidth
14247.1814210.38
14386.7714180.11
14435.2214165.56
14457.6514182.99
14466.7114185.19
14476.2814174.7
14485.114198.12
14498.3414169.3
14483.5114182.34
14498.0114181.44
14498.7414202.71
14508.1114139.8
14507.5814181.87
14513.114194.58
14500.0414158.15
14513.5114186.06
14516.7714194.11
14513.8114155.92
14509.3614188.26
14516.2714178.59
14518.3114180.03
14508.4714184.54
14511.1814191.08
14519.2214189.59
14515.1614190.39
14509.7714188.29
14507.1114183.02
14515.5914192.23
14508.5314186.89
14518.0514168.5
14203.114024.38
14196.0914044.27
11519.7211755.13
8650.69079.46
7843.678249.91
7844.668255.21
7848.98237.1
7856.328259.32
7841.438262.92
7856.788259.96
7847.188246.82
7853.258255.26
7848.058246.77
7838.118250.94
7853.978243.19
7853.788261.14
7849.578248.73
7854.248260.4
7853.68231.9
7847.248253.33
7857.718177.5
7861.58247.45
7857.648257.7
7860.948253.15
7837.248255.14
7853.538243.88
7850.788254.16
7860.198240.45
7862.658235.63
7860.98257.26
7865.088261.5
7865.778247.68
7858.748247.77
7857.178205.32
7863.588238.6
7861.458249.15
7863.958235.33
7865.718255.3
7855.078243.98
7868.618258.39
7853.498236.09
7860.658210.25
7859.688254
7867.578253.59
7867.228259.73
7851.688246.81
7866.578256.6
7861.838241.05
7836.388247.16
7867.898243.68
7867.728261.13
7865.678240.21
7867.298224.45
7860.898257.53
7866.978248.44
7857.758256.1
7864.658246.46
7866.928256.44
7864.358240.2
7868.738256.85
7857.198257.51
7850.648260.73
7865.468251.95
7858.488259.42
7860.478259.32
7864.988261.1
7872.738245.37
7867.558259.95
7861.678224.83
7857.068237.93
7860.468199.36
7849.258257.12
7870.998189.58
7865.28215.24
7862.68237.23
7852.58248.6
7852.088245.84
7866.578258.02
7856.98250.85
7853.578230.01
7840.18190.94
7820.758177.06
7806.388128.59
7777.878140.93
7779.198088.6
7778.958101.97
7774.668060.33
7752.698101
7770.038094.78
7776.568053.9
7758.878101.12
7772.358068.58
7772.168095.8
7763.728074.29
7773.928083.87
7777.268070.48
7771.848099.73
7719.598075.81
7768.828096.28
7775.098056.91
7771.168086.48
7755.968096.3
7760.128078.07
7759.88068.23
7631.788054.83
7618.28047.63
7732.027956.87
7655.858042.72
7725.87996.28
7714.618056.72
7515.067858
7624.127895.5
7637.937741.66
7636.537945.8
7341.237853.66
7515.287781.19
7237.047747.09
7461.547250.46
7256.97378.14
6696.777106.15
6737.496343.97
6547.116249.16
5991.115018.53
5836.484482.27
4959.793418.65
4657.392772.85
4336.382547.29
3988.922312.31
40162166.92
3931.62095.44
3810.271995.24
3820.211954.69
3848.281899.1
3799.91840.8
3772.091796.24
3765.111742.41
3750.181723.53
3743.71703.83
3682.081668.98
3664.431642.6
3704.841618.6
3619.751592.33
3603.381585.18
3629.121563.99
3661.951535.88
3667.141531.17
3665.121524.09
3658.151495.36
3603.331507.28
3649.911488.69
3607.021469.99
3496.531455.48
3640.171464.54
3644.711452.88
3596.211437.13
3635.891446.52
3537.251427.81
3623.651422.92
3597.521420.96
3601.781442.96
3568.331423.41
3584.591443.29
3608.711427.09
3626.661400.76
3584.251398.39
3566.341386.02
3616.271407.95
3500.151398.48
32 KB (L1 cache size)
2 MB (L2 cache size)
7.7 GB/sec
14.2 GB/sec
3.5 GB/sec
1.4 GB/sec
Read BW
Write BW
Array Size (bytes)
Bandwidth (MB/sec)
Cache & RAM Bandwidth: Intel T2400 (1.83 GHz)
Sheet1
102414247.1814210.38
204814386.7714180.11
307214435.2214165.56
409614457.6514182.99
512014466.7114185.19
614414476.2814174.7
716814485.114198.12
819214498.3414169.3
921614483.5114182.34
1024014498.0114181.44
1126414498.7414202.71
1228814508.1114139.8
1331214507.5814181.87
1433614513.114194.58
1536014500.0414158.15
1638414513.5114186.06
1740814516.7714194.11
1843214513.8114155.92
1945614509.3614188.26
2048014516.2714178.59
2150414518.3114180.03
2252814508.4714184.54
2355214511.1814191.08
2457614519.2214189.59
2560014515.1614190.39
2662414509.7714188.29
2764814507.1114183.02
2867214515.5914192.23
2969614508.5314186.89
3072014518.0514168.5
3174414203.114024.38
3276814196.0914044.27
3379211519.7211755.13
358408650.69079.46
378887843.678249.91
399367844.668255.21
419847848.98237.1
440327856.328259.32
460807841.438262.92
481287856.788259.96
501767847.188246.82
522247853.258255.26
542727848.058246.77
563207838.118250.94
583687853.978243.19
604167853.788261.14
624647849.578248.73
645127854.248260.4
665607853.68231.9
696327847.248253.33
727047857.718177.5
757767861.58247.45
788487857.648257.7
819207860.948253.15
849927837.248255.14
880647853.538243.88
911367850.788254.16
942087860.198240.45
972807862.658235.63
1003527860.98257.26
1044487865.088261.5
1085447865.778247.68
1126407858.748247.77
1167367857.178205.32
1208327863.588238.6
1249287861.458249.15
1290247863.958235.33
1331207865.718255.3
1382407855.078243.98
1433607868.618258.39
1484807853.498236.09
1536007860.658210.25
1587207859.688254
1638407867.578253.59
1689607867.228259.73
1751047851.688246.81
1812487866.578256.6
1873927861.838241.05
1935367836.388247.16
1996807867.898243.68
2068487867.728261.13
2140167865.678240.21
2211847867.298224.45
2283527860.898257.53
2355207866.978248.44
2437127857.758256.1
2519047864.658246.46
2600967866.928256.44
2682887864.358240.2
2775047868.738256.85
2867207857.198257.51
2959367850.648260.73
3061767865.468251.95
3164167858.488259.42
3266567860.478259.32
3368967864.988261.1
3481607872.738245.37
3594247867.558259.95
3706887861.678224.83
3829767857.068237.93
3952647860.468199.36
4085767849.258257.12
4218887870.998189.58
4352007865.28215.24
4495367862.68237.23
4638727852.58248.6
4792327852.088245.84
4945927866.578258.02
5109767856.98250.85
5273607853.578230.01
5447687840.18190.94
5621767820.758177.06
5806087806.388128.59
5990407777.878140.93
6184967779.198088.6
6379527778.958101.97
6584327774.668060.33
6799367752.698101
7014407770.038094.78
7239687776.568053.9
7475207758.878101.12
7710727772.358068.58
7956487772.168095.8
8212487763.728074.29
8478727773.928083.87
8744967777.268070.48
9021447771.848099.73
9308167719.598075.81
9605127768.828096.28
9912327775.098056.91
10229767771.168086.48
10557447755.968096.3
10895367760.128078.07
11243527759.88068.23
11601927631.788054.83
11970567618.28047.63
12349447732.027956.87
12738567655.858042.72
13137927725.87996.28
13557767714.618056.72
13987847515.067858
14428167624.127895.5
14888967637.937741.66
15360007636.537945.8
15841287341.237853.66
16343047515.287781.19
16855047237.047747.09
17387527461.547250.46
17940487256.97378.14
18503686696.777106.15
19087366737.496343.97
19691526547.116249.16
20316165991.115018.53
20951045836.484482.27
21606404959.793418.65
22282244657.392772.85
22978564336.382547.29
23705603988.922312.31
244531240162166.92
25221123931.62095.44
26009603810.271995.24
26828803820.211954.69
27668483848.281899.1
28538883799.91840.8
29440003772.091796.24
30361603765.111742.41
31313923750.181723.53
32296963743.71703.83
33310723682.081668.98
34355203664.431642.6
35430403704.841618.6
36546563619.751592.33
37693443603.381585.18
38881283629.121563.99
40099843661.951535.88
41359363667.141531.17
42659843665.121524.09
44001283658.151495.36
45383683603.331507.28
46807043649.911488.69
48271363607.021469.99
49786883496.531455.48
51343363640.171464.54
52951043644.711452.88
54609923596.211437.13
56320003635.891446.52
58081283537.251427.81
59904003623.651422.92
61777923597.521420.96
63713283601.781442.96
65710083568.331423.41
67768323584.591443.29
69888003608.711427.09
72079363626.661400.76
74332163584.251398.39
76656643566.341386.02
79052803616.271407.95
81530883500.151398.48
rmma_intelt2400_membw_20070904
RIGHTMARK MEMORY ANALYZER V3.72 TEST RESULTS
============================================
CPU ModelGenuine Intel(R) Core(TM) Duo (Yonah) 1828.7 MHz
L1 Cache Line Size64 bytes
L2 Cache Line Size128 bytes
Test TypeMemory Bandwidth
Test StatusCompleted successfully
Selected Tests7
Set Size134217728 bytes
Memory Allocation1
Thread Lock0
Min Block Size1024 bytes
Max Block Size8388608 bytes
CPU Register Usage0
Read Prefetch Type0
Prefetch Distance0 bytes
Block Prefetch Size1024 bytes
Stride Size64 bytes
Non-Temporal Store0
Copy-to-Self Mode0
MEMORY READ BANDWIDTH TEST
==========================
Size(bytes)Bandwidth(bytes/cycle)Bandwidth(MB/s)
10247.7914247.18
20487.8714386.77
30727.8914435.22
40967.9114457.65
51207.9114466.71
61447.9214476.28
71687.9214485.1
81927.9314498.34
92167.9214483.51
102407.9314498.01
112647.9314498.74
122887.9314508.11
133127.9314507.58
143367.9414513.1
153607.9314500.04
163847.9414513.51
174087.9414516.77
184327.9414513.81
194567.9314509.36
204807.9414516.27
215047.9414518.31
225287.9314508.47
235527.9414511.18
245767.9414519.22
256007.9414515.16
266247.9314509.77
276487.9314507.11
286727.9414515.59
296967.9314508.53
307207.9414518.05
317447.7714203.1
327687.7614196.09
337926.311519.72
358404.738650.6
378884.297843.67
399364.297844.66
419844.297848.9
440324.37856.32
460804.297841.43
481284.37856.78
501764.297847.18
522244.297853.25
542724.297848.05
563204.297838.11
583684.297853.97
604164.297853.78
624644.297849.57
645124.297854.24
665604.297853.6
696324.297847.24
727044.37857.71
757764.37861.5
788484.37857.64
819204.37860.94
849924.297837.24
880644.297853.53
911364.297850.78
942084.37860.19
972804.37862.65
1003524.37860.9
1044484.37865.08
1085444.37865.77
1126404.37858.74
1167364.37857.17
1208324.37863.58
1249284.37861.45
1290244.37863.95
1331204.37865.71
1382404.37855.07
1433604.37868.61
1484804.297853.49
1536004.37860.65
1587204.37859.68
1638404.37867.57
1689604.37867.22
1751044.297851.68
1812484.37866.57
1873924.37861.83
1935364.297836.38
1996804.37867.89
2068484.37867.72
2140164.37865.67
2211844.37867.29
2283524.37860.89
2355204.37866.97
2437124.37857.75
2519044.37864.65
2600964.37866.92
2682884.37864.35
2775044.37868.73
2867204.37857.19
2959364.297850.64
3061764.37865.46
3164164.37858.48
3266564.37860.47
3368964.37864.98
3481604.37872.73
3594244.37867.55
3706884.37861.67
3829764.37857.06
3952644.37860.46
4085764.297849.25
4218884.37870.99
4352004.37865.2
4495364.37862.6
4638724.297852.5
4792324.297852.08
4945924.37866.57
5109764.37856.9
5273604.297853.57
5447684.297840.1
5621764.287820.75
5806084.277806.38
5990404.257777.87
6184964.257779.19
6379524.257778.95
6584324.257774.66
6799364.247752.69
7014404.257770.03
7239684.257776.56
7475204.247758.87
7710724.257772.35
7956484.257772.16
8212484.257763.72
8478724.257773.92
8744964.257777.26
9021444.257771.84
9308164.227719.59
9605124.257768.82
9912324.257775.09
10229764.257771.16
10557444.247755.96
10895364.247760.12
11243524.247759.8
11601924.177631.78
11970564.177618.2
12349444.237732.02
12738564.197655.85
13137924.227725.8
13557764.227714.61
13987844.117515.06
14428164.177624.12
14888964.187637.93
15360004.187636.53
15841284.017341.23
16343044.117515.28
16855043.967237.04
17387524.087461.54
17940483.977256.9
18503683.666696.77
19087363.686737.49
19691523.586547.11
20316163.285991.11
20951043.195836.48
21606402.714959.79
22282242.554657.39
22978562.374336.38
23705602.183988.92
24453122.24016
25221122.153931.6
26009602.083810.27
26828802.093820.21
27668482.13848.28
28538882.083799.9
29440002.063772.09
30361602.063765.11
31313922.053750.18
32296962.053743.7
33310722.013682.08
343552023664.43
35430402.033704.84
36546561.983619.75
37693441.973603.38
38881281.983629.12
400998423661.95
41359362.013667.14
426598423665.12
440012823658.15
45383681.973603.33
468070423649.91
48271361.973607.02
49786881.913496.53
51343361.993640.17
52951041.993644.71
54609921.973596.21
56320001.993635.89
58081281.933537.25
59904001.983623.65
61777921.973597.52
63713281.973601.78
65710081.953568.33
67768321.963584.59
69888001.973608.71
72079361.983626.66
74332161.963584.25
76656641.953566.34
79052801.983616.27
81530881.913500.15
MAXIMAL READ BANDWIDTH
245767.9414519.22
MEMORY WRITE BANDWIDTH TEST
===========================
Size(bytes)Bandwidth(bytes/cycle)Bandwidth(MB/s)
10247.7714210.38
20487.7514180.11
30727.7514165.56
40967.7614182.99
51207.7614185.19
61447.7514174.7
71687.7614198.12
81927.7514169.3
92167.7614182.34
102407.7514181.44
112647.7714202.71
122887.7314139.8
133127.7514181.87
143367.7614194.58
153607.7414158.15
163847.7614186.06
174087.7614194.11
184327.7414155.92
194567.7614188.26
204807.7514178.59
215047.7514180.03
225287.7614184.54
235527.7614191.08
245767.7614189.59
256007.7614190.39
266247.7614188.29
276487.7614183.02
286727.7614192.23
296967.7614186.89
307207.7514168.5
317447.6714024.38
327687.6814044.27
337926.4311755.13
358404.969079.46
378884.518249.91
399364.518255.21
419844.58237.1
440324.528259.32
460804.528262.92
481284.528259.96
501764.518246.82
522244.518255.26
542724.518246.77
563204.518250.94
583684.518243.19
604164.528261.14
624644.518248.73
645124.528260.4
665604.58231.9
696324.518253.33
727044.478177.5
757764.518247.45
788484.528257.7
819204.518253.15
849924.518255.14
880644.518243.88
911364.518254.16
942084.518240.45
972804.58235.63
1003524.528257.26
1044484.528261.5
1085444.518247.68
1126404.518247.77
1167364.498205.32
1208324.518238.6
1249284.518249.15
1290244.58235.33
1331204.518255.3
1382404.518243.98
1433604.528258.39
1484804.58236.09
1536004.498210.25
1587204.518254
1638404.518253.59
1689604.528259.73
1751044.518246.81
1812484.518256.6
1873924.518241.05
1935364.518247.16
1996804.518243.68
2068484.528261.13
2140164.518240.21
2211844.58224.45
2283524.528257.53
2355204.518248.44
2437124.518256.1
2519044.518246.46
2600964.518256.44
2682884.518240.2
2775044.528256.85
2867204.528257.51
2959364.528260.73
3061764.518251.95
3164164.528259.42
3266564.528259.32
3368964.528261.1
3481604.518245.37
3594244.528259.95
3706884.58224.83
3829764.58237.93
3952644.488199.36
4085764.528257.12
4218884.488189.58
4352004.498215.24
4495364.58237.23
4638724.518248.6
4792324.518245.84
4945924.528258.02
5109764.518250.85
5273604.58230.01
5447684.488190.94
5621764.478177.06
5806084.448128.59
5990404.458140.93
6184964.428088.6
6379524.438101.97
6584324.418060.33
6799364.438101
7014404.438094.78
7239684.48053.9
7475204.438101.12
7710724.418068.58
7956484.438095.8
8212484.428074.29
8478724.428083.87
8744964.418070.48
9021444.438099.73
9308164.428075.81
9605124.438096.28
9912324.418056.91
10229764.428086.48
10557444.438096.3
10895364.428078.07
11243524.418068.23
11601924.48054.83
11970564.48047.63
12349444.357956.87
12738564.48042.72
13137924.377996.28
13557764.418056.72
13987844.37858
14428164.327895.5
14888964.237741.66
15360004.347945.8
15841284.297853.66
16343044.257781.19
16855044.247747.09
17387523.967250.46
17940484.037378.14
18503683.897106.15
19087363.476343.97
19691523.426249.16
20316162.745018.53
20951042.454482.27
21606401.873418.65
22282241.522772.85
22978561.392547.29
23705601.262312.31
24453121.182166.92
25221121.152095.44
26009601.091995.24
26828801.071954.69
27668481.041899.1
28538881.011840.8
29440000.981796.24
30361600.951742.41
31313920.941723.53
32296960.931703.83
33310720.911668.98
34355200.91642.6
35430400.891618.6
36546560.871592.33
37693440.871585.18
38881280.861563.99
40099840.841535.88
41359360.841531.17
42659840.831524.09
44001280.821495.36
45383680.821507.28
46807040.811488.69
48271360.81469.99
49786880.81455.48
51343360.81464.54
52951040.791452.88
54609920.791437.13
56320000.791446.52
58081280.781427.81
59904000.781422.92
61777920.781420.96
63713280.791442.96
65710080.781423.41
67768320.791443.29
69888000.781427.09
72079360.771400.76
74332160.761398.39
76656640.761386.02
79052800.771407.95
81530880.761398.48
MAXIMAL WRITE BANDWIDTH
10247.7714210.38
MEMORY COPY BANDWIDTH TEST
==========================
Size(bytes)Bandwidth(bytes/cycle)Bandwidth(MB/s)
10245.8210648.75
20486.2511431.17
30726.1811309.64
40966.3311583.39
51206.2511429.84
61446.3511607.8
71686.2911510.59
81926.3411599.43
92166.3111531.02
102406.3611622.61
112646.3211551.17
122886.3411599.04
133126.3211555.78
143366.3511620.62
153606.1711287.61
163846.1211196.41
174083.957228.88
184322.915326.2
194562.955396.99
2048035491.1
215042.965417.67
225282.935357.97
235522.965414.54
2457635491.85
256002.975432.51
266242.945384.95
276482.975435.02
286723.015498.14
296962.975440.47
307202.955403.39
317442.985447.28
3276835494.46
337922.985453.21
358402.985457.75
378882.985457.63
399362.985452.18
419842.995472.63
440322.995468.92
460802.995463.01
481282.985455.08
5017635481.14
5222435484.35
542722.995470.94
5632035484.3
583682.995463.75
6041635485.67
624642.995474.95
6451235489.08
6656035481.57
696323.015505.31
7270435491.62
757762.995472.22
7884835485.58
8192035485.19
849923.015498.06
880642.995465.31
911362.995476.62
942083.015502.87
972803.015499.31
10035235490.08
10444835491.96
1085443.015496.4
11264035490.72
1167363.015496.95
1208323.015502.93
12492835479.51
12902435488.61
13312035486.26
1382403.015509.75
14336035494.93
1484803.015502.54
1536003.015498.09
1587203.015495.8
1638403.015507.81
16896035495.35
1751043.015511.85
1812483.015503.42
1873923.015507.25
1935363.015502.67
1996803.015510.99
2068483.025513.68
21401635481.46
2211843.015503.89
2283523.025517.06
2355203.015500.83
2437123.025516.56
2519043.015496.48
2600963.015502.12
2682883.015497.64
27750435489.9
2867202.975435.19
2959362.995467.49
3061762.995471.84
3164162.985450.19
3266562.985458.68
3368962.985445.75
3481602.945380.86
3594242.995462.88
3706882.995467.74
3829762.985441.7
3952642.985454.69
4085762.995465.03
4218882.975434.2
4352002.985452.61
4495362.985447.14
4638722.965418.31
4792322.975433.75
4945922.985449.67
5109762.995461.12
5273602.985446.15
5447682.965420.68
5621762.985445.81
5806082.975436.67
5990402.95305.18
6184962.965412.74
6379522.975426.45
6584322.935359.76
6799362.965415.67
7014402.975426.47
7239682.935362.93
7475202.965412.62
7710722.925334.95
7956482.895288.48
8212482.815132.24
8478722.845188.02
8744962.734989.11
9021442.745004.81
9308162.624794.01
9605122.474511.26
9912322.314223.45
10229762.163958.48
10557441.993644.64
10895361.522771.07
11243521.332424.51
11601921.182149.8
11970561.12009.58
12349441.031874.84
12738561.041896.16
13137921.011847.3
13557760.991815.14
13987840.951733.94
14428160.921690.97
14888960.911662.4
15360000.891631.91
15841280.871592.98
16343040.861576.36
16855040.851555.82
17387520.841535.8
17940480.821501.29
18503680.811484.7
19087360.81459.02
19691520.791442.87
20316160.791444.34
20951040.781422.38
21606400.771404.3
22282240.761390.2
22978560.751379.78
23705600.741360.23
24453120.741358.37
25221120.731336.56
26009600.731332.53
26828800.721309.32
27668480.711306.57
28538880.71287.96
29440000.711294.49
30361600.71274.23
31313920.691265.65
32296960.691258.17
33310720.681247.35
34355200.681248.1
35430400.671232.97
36546560.671221.7
37693440.671216.35
38881280.661212.83
40099840.661206.05
41359360.651189.26
42659840.651183.24
44001280.641165.64
45383680.651184.89
46807040.641164.57
48271360.651190.02
49786880.631155.19
51343360.641163.24
52951040.641162.7
54609920.641168.97
56320000.651183.23
58081280.631160.75
59904000.631151.1
61777920.631158.05
63713280.641171.36
65710080.641170.73
67768320.631151.42
69888000.641168.89
72079360.631157.37
74332160.641167.04
76656640.641171.26
79052800.631153.88
81530880.641176.82
MAXIMAL COPY BANDWIDTH
102406.3611622.61
-
Cache Use JargonCache Hit: the data that the CPU needs right now
are already in cache.Cache Miss: the data that the CPU needs right
now are not currently in cache.If all of your data are small enough
to fit in cache, then when you run your program, youll get almost
all cache hits (except at the very beginning), which means that
your performance could be excellent!Sadly, this rarely happens in
real life: most problems of scientific or engineering interest are
bigger than just a few MB.
OU Supercomputing Center for Education & Research
-
Cache LinesA cache line is a small, contiguous region in cache,
corresponding to a contiguous region in RAM of the same size, that
is loaded all at once.Typical size: 32 to 1024 bytesExamplesPentium
4 (Yonah) [26]L1 data cache: 64 bytes per lineL2 cache: 128 bytes
per linePOWER4 [12]L1 instruction cache: 128 bytes per lineL1 data
cache: 128 bytes per lineL2 cache: 128 bytes per lineL3 cache: 512
bytes per line
OU Supercomputing Center for Education & Research
-
How Cache WorksWhen you request data from a particular address
in Main Memory, heres what happens:The hardware checks whether the
data for that address is already in cache. If so, it uses
it.Otherwise, it loads from Main Memory the entire cache line that
contains the address.For example, on a 1.83 GHz Pentium4 Core Duo
(Yonah), a cache miss makes the program stall (wait) at least 48
cycles (26.2 nanoseconds) for the next cache line to load time that
could have been spent performing up to 192 calculations! [26]
OU Supercomputing Center for Education & Research
-
If Its in Cache, Its Also in RAMIf a particular memory address
is currently in cache, then its also in Main Memory (RAM).That is,
all of a programs data are in Main Memory, but some are also in
cache.Well revisit this point shortly.
OU Supercomputing Center for Education & Research
-
Mapping Cache Lines to RAMMain memory typically maps into cache
in one of three ways:Direct mapped (occasionally)Fully associative
(very rare these days)Set associative (common)DONTPANIC!
OU Supercomputing Center for Education & Research
-
Direct Mapped CacheDirect Mapped Cache is a scheme in which each
location in main memory corresponds to exactly one location in
cache (but not the reverse, since cache is much smaller than main
memory).Typically, if a cache address is represented by c bits, and
a main memory address is represented by m bits, then the cache
location associated with main memory address A is MOD(A,2c); that
is, the lowest c bits of A.Example: POWER4 L1 instruction cache
OU Supercomputing Center for Education & Research
-
Direct Mapped Cache IllustrationMust go intocache
address11100101Main Memory Address0100101011100101Notice that
11100101 is the low 8 bits of 0100101011100101.
OU Supercomputing Center for Education & Research
-
Jargon: Cache ConflictSuppose that the cache address 11100101
currently contains RAM address 0100101011100101.But, we now need to
load RAM address 1100101011100101, which maps to the same cache
address as 0100101011100101.This is called a cache conflict : the
CPU needs a RAM location that maps to a cache line already in
use.In the case of direct mapped cache, every cache conflict leads
to the new cache line clobbering the old cache line.This can lead
to serious performance problems.
OU Supercomputing Center for Education & Research
-
Problem with Direct MappedIf you have two arrays that start in
the same place relative to cache, then they might clobber each
other all the time: no cache
hits!REAL,DIMENSION(multiple_of_cache_size) :: a, b, cINTEGER ::
index
DO index = 1, multiple_of_cache_size a(index) = b(index) +
c(index)END DO !! index = 1, multiple_of_cache_sizeIn this example,
a(index), b(index) and c(index) all map to the same cache line, so
loading c(index) clobbers b(index) no cache reuse!
OU Supercomputing Center for Education & Research
-
Fully Associative CacheFully Associative Cache can put any line
of main memory into any cache line.Typically, the cache management
system will put the newly loaded data into the Least Recently Used
cache line, though other strategies are possible (e.g., Random,
First In First Out, Round Robin, Least Recently Modified).So, this
can solve, or at least reduce, the cache conflict problem.But,
fully associative cache tends to be expensive, so its pretty rare:
you need Ncache. NRAM connections!
OU Supercomputing Center for Education & Research
-
Fully Associative IllustrationCould go intoany cache lineMain
Memory Address0100101011100101
OU Supercomputing Center for Education & Research
-
Set Associative CacheSet Associative Cache is a compromise
between direct mapped and fully associative. A line in main memory
can map to any of a fixed number of cache lines.For example, 2-way
Set Associative Cache can map each main memory line to either of 2
cache lines (e.g., to the Least Recently Used), 3-way maps to any
of 3 cache lines, 4-way to 4 lines, and so on.Set Associative cache
is cheaper than fully associative you need K . NRAM connections but
more robust than direct mapped.
OU Supercomputing Center for Education & Research
-
2-Way Set Associative IllustrationCould go into cache
address11100101Main Memory Address0100101011100101Could go
intocache address01100101OR
OU Supercomputing Center for Education & Research
-
Cache Associativity ExamplesPentium 4 EM64T (Yonah) [26]L1 data
cache: 8-way set associativeL2 cache: 8-way set associativePOWER4
[12]L1 instruction cache: direct mappedL1 data cache: 2-way set
associativeL2 cache: 8-way set associativeL3 cache: 8-way set
associative
OU Supercomputing Center for Education & Research
-
If Its in Cache, Its Also in RAMAs we saw earlier:If a
particular memory address is currently in cache, then its also in
Main Memory (RAM).That is, all of a programs data are in Main
Memory, but some are also in cache.
OU Supercomputing Center for Education & Research
-
Changing a Value Thats in CacheSuppose that you have in cache a
particular line of main memory (RAM).If you dont change the
contents of any of that lines bytes while its in cache, then when
it gets clobbered by another main memory line coming into cache,
theres no loss of information.But, if you change the contents of
any byte while its in cache, then you need to store it back out to
main memory before clobbering it.
OU Supercomputing Center for Education & Research
-
Cache Store StrategiesTypically, there are two possible cache
store strategies:Write-through: every single time that a value in
cache is changed, that value is also stored back into main memory
(RAM).Write-back: every single time that a value in cache is
changed, the cache line containing that cache location gets marked
as dirty. When a cache line gets clobbered, then if it has been
marked as dirty, then it is stored back into main memory (RAM).
[14]
OU Supercomputing Center for Education & Research
-
The Importance of Being Local[15]
OU Supercomputing Center for Education & Research
-
More Data Than CacheLets say that you have 1000 times more data
than cache. Then wont most of your data be outside the cache?
YES!
Okay, so how does cache help?
OU Supercomputing Center for Education & Research
-
Improving Your Cache Hit RateMany scientific codes use a lot
more data than can fit in cache all at once.Therefore, you need to
ensure a high cache hit rate even though youve got much more data
than cache.So, how can you improve your cache hit rate?Use the same
solution as in Real Estate:Location, Location, Location!
OU Supercomputing Center for Education & Research
-
Data LocalityData locality is the principle that, if you use
data in a particular memory address, then very soon youll use
either the same address or a nearby address.Temporal locality: if
youre using address A now, then youll probably soon use address A
again.Spatial locality: if youre using address A now, then youll
probably soon use addresses between A-k and A+k, where k is
small.Note that this principle works well for sufficiently small
values of soon.Cache is designed to exploit locality, which is why
a cache miss causes a whole line to be loaded.
OU Supercomputing Center for Education & Research
-
Data Locality Is EmpiricalData locality has been observed
empirically in many, many programs.void ordered_fill (int* array,
int array_length){ /* ordered_fill */ int index;
for (index = 0; index < array_length; index++) { array[index]
= index; } /* for index */} /* ordered_fill */
OU Supercomputing Center for Education & Research
-
No Locality ExampleIn principle, you could write a program that
exhibited absolutely no data locality at all:void random_fill (int*
array, int* random_permutation_index, int array_length){ /*
random_fill */ int index;
for (index = 0; index < array_length; index++) {
array[random_permutation_index[index]] = index; } /* for index */}
/* random_fill */
OU Supercomputing Center for Education & Research
-
Permuted vs. OrderedIn a simple array fill, locality provides a
factor of 8 to 20 speedup over a randomly ordered fill on a
Pentium4.
OU Supercomputing Center for Education & Research
Permuted vs Ordered
27.251.379997
11.070.76
4.050.36
1.640.18
0.710.09
0.350.04
0.180.02
0.080.01
Random
Ordered
Array size (log2 bytes)
CPU seconds
Sheet1
Sizelog(size)RandomOrdered
1342177282727.251.379997
671088642611.070.76
33554432254.050.36
16777216241.640.18
8388608230.710.09
4194304220.350.04
2097152210.180.02
1048576200.080.01
Sheet2
Sheet3
-
Exploiting Data LocalityIf you know that your code is capable of
operating with a decent amount of data locality, then you can get
speedup by focusing your energy on improving the locality of the
codes behavior.This will substantially increase your cache
reuse.
OU Supercomputing Center for Education & Research
-
A Sample ApplicationMatrix-Matrix MultiplyLet A, B and C be
matrices of sizesnr nc, nr nk and nk nc, respectively:The
definition of A = B C isfor r {1, nr}, c {1, nc}.
OU Supercomputing Center for Education & Research
-
Matrix Multiply: Nave VersionSUBROUTINE
matrix_matrix_mult_by_naive (dst, src1, src2, & & nr, nc,
nq) IMPLICIT NONE INTEGER,INTENT(IN) :: nr, nc, nq
REAL,DIMENSION(nr,nc),INTENT(OUT) :: dst
REAL,DIMENSION(nr,nq),INTENT(IN) :: src1
REAL,DIMENSION(nq,nc),INTENT(IN) :: src2
INTEGER :: r, c, q
CALL matrix_set_to_scalar(dst, nr, nc, 1, nr, 1, nc, 0.0) DO c =
1, nc DO r = 1, nr DO q = 1, nq dst(r,c) = dst(r,c) + src1(r,q) *
src2(q,c) END DO !! q = 1, nq END DO !! r = 1, nr END DO !! c = 1,
ncEND SUBROUTINE matrix_matrix_mult_by_naive
OU Supercomputing Center for Education & Research
-
Matrix Multiply w/InitializationSUBROUTINE
matrix_matrix_mult_by_init (dst, src1, src2, & & nr, nc,
nq) IMPLICIT NONE INTEGER,INTENT(IN) :: nr, nc, nq
REAL,DIMENSION(nr,nc),INTENT(OUT) :: dst
REAL,DIMENSION(nr,nq),INTENT(IN) :: src1
REAL,DIMENSION(nq,nc),INTENT(IN) :: src2
INTEGER :: r, c, q
DO c = 1, nc DO r = 1, nr dst(r,c) = 0.0 DO q = 1, nq dst(r,c) =
dst(r,c) + src1(r,q) * src2(q,c) END DO !! q = 1, nq END DO !! r =
1, nr END DO !! c = 1, ncEND SUBROUTINE
matrix_matrix_mult_by_init
OU Supercomputing Center for Education & Research
-
Matrix Multiply Via IntrinsicSUBROUTINE
matrix_matrix_mult_by_intrinsic ( & & dst, src1, src2, nr,
nc, nq) IMPLICIT NONE INTEGER,INTENT(IN) :: nr, nc, nq
REAL,DIMENSION(nr,nc),INTENT(OUT) :: dst
REAL,DIMENSION(nr,nq),INTENT(IN) :: src1
REAL,DIMENSION(nq,nc),INTENT(IN) :: src2
dst = MATMUL(src1, src2)END SUBROUTINE
matrix_matrix_mult_by_intrinsic
OU Supercomputing Center for Education & Research
-
Matrix Multiply BehaviorIf the matrix is big, then each sweep of
a row will clobber nearby values in cache.
OU Supercomputing Center for Education & Research
-
Performance of Matrix Multiply
OU Supercomputing Center for Education & Research
Chart1
0.020.010.01
0.050.050.03
0.10.110.08
0.380.760.28
1.782.180.58
8.979.592.4
17.920.984.72
72.881.4618.76
156.6161.4437.41
723.9690.98151.55
Naive
Init
Intrinsic
Total Problem Size in bytes (nr*nc+nr*nq+nq*nc)
CPU sec
Matrix-Matrix Multiply
Sheet1
819200.020.010.01
1966080.050.050.03
3276800.10.110.08
7864320.380.760.28
13107201.782.180.58
31457288.979.592.4
524288017.920.984.72
1258291272.881.4618.76
20971520156.6161.4437.41
50331648723.9690.98151.55
Sheet2
Sheet3
-
Tiling
OU Supercomputing Center for Education & Research
-
TilingTile: a small rectangular subdomain of a problem domain.
Sometimes called a block or a chunk.Tiling: breaking the domain
into tiles.Tiling strategy: operate on each tile to completion,
then move to the next tile.Tile size can be set at runtime,
according to whats best for the machine that youre running on.
OU Supercomputing Center for Education & Research
-
Tiling CodeSUBROUTINE matrix_matrix_mult_by_tiling (dst, src1,
src2, nr, nc, nq, & & rtilesize, ctilesize, qtilesize)
IMPLICIT NONE INTEGER,INTENT(IN) :: nr, nc, nq
REAL,DIMENSION(nr,nc),INTENT(OUT) :: dst
REAL,DIMENSION(nr,nq),INTENT(IN) :: src1
REAL,DIMENSION(nq,nc),INTENT(IN) :: src2 INTEGER,INTENT(IN) ::
rtilesize, ctilesize, qtilesize
INTEGER :: rstart, rend, cstart, cend, qstart, qend
DO cstart = 1, nc, ctilesize cend = cstart + ctilesize - 1 IF
(cend > nc) cend = nc DO rstart = 1, nr, rtilesize rend = rstart
+ rtilesize - 1 IF (rend > nr) rend = nr DO qstart = 1, nq,
qtilesize qend = qstart + qtilesize - 1 IF (qend > nq) qend = nq
CALL matrix_matrix_mult_tile(dst, src1, src2, nr, nc, nq, &
& rstart, rend, cstart, cend, qstart, qend) END DO !! qstart =
1, nq, qtilesize END DO !! rstart = 1, nr, rtilesize END DO !!
cstart = 1, nc, ctilesizeEND SUBROUTINE
matrix_matrix_mult_by_tiling
OU Supercomputing Center for Education & Research
-
Multiplying Within a TileSUBROUTINE matrix_matrix_mult_tile
(dst, src1, src2, nr, nc, nq, & & rstart, rend, cstart,
cend, qstart, qend) IMPLICIT NONE INTEGER,INTENT(IN) :: nr, nc, nq
REAL,DIMENSION(nr,nc),INTENT(OUT) :: dst
REAL,DIMENSION(nr,nq),INTENT(IN) :: src1
REAL,DIMENSION(nq,nc),INTENT(IN) :: src2 INTEGER,INTENT(IN) ::
rstart, rend, cstart, cend, qstart, qend
INTEGER :: r, c, q
DO c = cstart, cend DO r = rstart, rend IF (qstart == 1)
dst(r,c) = 0.0 DO q = qstart, qend dst(r,c) = dst(r,c) + src1(r,q)
* src2(q,c) END DO !! q = qstart, qend END DO !! r = rstart, rend
END DO !! c = cstart, cendEND SUBROUTINE
matrix_matrix_mult_tile
OU Supercomputing Center for Education & Research
-
Performance with Tiling
OU Supercomputing Center for Education & Research
-
The Advantages of TilingIt allows your code to exploit data
locality better, to get much more cache reuse: your code runs
faster!Its a relatively modest amount of extra coding (typically a
few wrapper functions and some changes to loop bounds).If you dont
need tiling because of the hardware, the compiler or the problem
size then you can turn it off by simply setting the tile size equal
to the problem size.
OU Supercomputing Center for Education & Research
-
Will Tiling Always Work?Tiling WONT always work. Why?Well,
tiling works well when:the order in which calculations occur doesnt
matter much, ANDthere are lots and lots of calculations to do for
each memory movement.If either condition is absent, then tiling
wont help.
OU Supercomputing Center for Education & Research
-
Hard Disk
OU Supercomputing Center for Education & Research
-
Why Is Hard Disk Slow?Your hard disk is much much slower than
main memory (factor of 10-1000). Why?Well, accessing data on the
hard disk involves physically moving:the disk platterthe read/write
headIn other words, hard disk is slow because objects move much
slower than electrons: Newtonian speeds are much slower than
Einsteinian speeds.
OU Supercomputing Center for Education & Research
-
I/O StrategiesRead and write the absolute minimum amount.Dont
reread the same data if you can keep it in memory.Write binary
instead of characters.Use optimized I/O libraries like NetCDF [17]
and HDF [18].
OU Supercomputing Center for Education & Research
-
Avoid Redundant I/OAn actual piece of code seen at OU:for (thing
= 0; thing < number_of_things; thing++) { for (time = 0; time
< number_of_timesteps; time++) { read(file[time]);
do_stuff(thing, time); } /* for time */} /* for thing */Improved
version:for (time = 0; time < number_of_timesteps; time++) {
read(file[time]); for (thing = 0; thing < number_of_things;
thing++) { do_stuff(thing, time); } /* for thing */} /* for time
*/Savings (in real life): factor of 500!
OU Supercomputing Center for Education & Research
-
Write Binary, Not ASCIIWhen you write binary data to a file,
youre writing (typically) 4 bytes per value.When you write ASCII
(character) data, youre writing (typically) 8-16 bytes per value.So
binary saves a factor of 2 to 4 (typically).
OU Supercomputing Center for Education & Research
-
Problem with Binary I/OThere are many ways to represent data
inside a computer, especially floating point (real) data.Often, the
way that one kind of computer (e.g., a Pentium4) saves binary data
is different from another kind of computer (e.g., a POWER5).So, a
file written on a Pentium4 machine may not be readable on a
POWER5.
OU Supercomputing Center for Education & Research
-
Portable I/O LibrariesNetCDF and HDF are the two most commonly
used I/O libraries for scientific computing.Each has its own
internal way of representing numerical data. When you write a file
using, say, HDF, it can be read by a HDF on any kind of
computer.Plus, these libraries are optimized to make the I/O very
fast.
OU Supercomputing Center for Education & Research
-
Virtual Memory
OU Supercomputing Center for Education & Research
-
Virtual MemoryTypically, the amount of main memory (RAM) that a
CPU can address is larger than the amount of data physically
present in the computer.For example, Henrys laptop can address 32
GB of main memory (roughly 32 billion bytes), but only contains 2
GB (roughly 2 billion bytes).
OU Supercomputing Center for Education & Research
-
Virtual Memory (contd)Locality: most programs dont jump all over
the memory that they use; instead, they work in a particular area
of memory for a while, then move to another area.So, you can
offload onto hard disk much of the memory image of a program thats
running.
OU Supercomputing Center for Education & Research
-
Virtual Memory (contd)Memory is chopped up into many pages of
modest size (e.g., 1 KB 32 KB; typically 4 KB).Only pages that have
been recently used actually reside in memory; the rest are stored
on hard disk.Hard disk is 10 to 1,000 times slower than main
memory, so you get better performance if you rarely get a page
fault, which forces a read from (and maybe a write to) hard disk:
exploit data locality!
OU Supercomputing Center for Education & Research
-
Cache vs. Virtual MemoryLines (cache) vs. pages (VM)Cache faster
than RAM (cache) vs. RAM faster than disk (VM)
OU Supercomputing Center for Education & Research
-
Storage Use StrategiesRegister reuse: do a lot of work on the
same data before working on new data.Cache reuse: the program is
much more efficient if all of the data and instructions fit in
cache; if not, try to use whats in cache a lot before using
anything that isnt in cache (e.g., tiling).Data locality: try to
access data that are near each other in memory before data that are
far.I/O efficiency: do a bunch of I/O all at once rather than a
little bit at a time; dont mix calculations and I/O.
OU Supercomputing Center for Education & Research
-
To Learn More
Supercomputinghttp://www.oscer.ou.edu/education.phphttp://symposium2007.oscer.ou.edu/
OU Supercomputing Center for Education & Research
-
Thanks for your attention!
Questions?
OU Supercomputing Center for Education & Research
-
References[1]
http://graphics8.nytimes.com/images/2007/07/13/sports/auto600.gif[2]
http://www.vw.com/newbeetle/[3]
http://img.dell.com/images/global/products/resultgrid/sm/latit_d630.jpg[4]
http://en.wikipedia.org/wiki/X64 [5] Richard Gerber, The Software
Optimization Cookbook: High-performance Recipes for the Intel
Architecture. Intel Press, 2002, pp. 161-168.[6]
http://www.anandtech.com/showdoc.html?i=1460&p=2[8]
http://www.toshiba.com/taecdpd/products/features/MK2018gas-Over.shtml[9]
http://www.toshiba.com/taecdpd/techdocs/sdr2002/2002spec.shtml[10]
ftp://download.intel.com/design/Pentium4/manuals/24896606.pdf[11]
http://www.pricewatch.com/[12] S. Behling, R. Bell, P. Farrell, H.
Holthoff, F. OConnell and W. Weir, The POWER4 Processor
Introduction and Tuning Guide. IBM Redbooks, 2001.[13] http