I have 2 computers I am using in a small cluster. Each has 2 Intel Xeon quad-cord processors.
I just wanted to verify that in my host file, I should specify 8 slots per host.
The tail of the /proc/cpu file looks like the following (with procs 0 - 6 before this):
.... processors 0-6 above ....
processor : 7
vendor_id : GenuineIntel
cpu family : 6
model : 23
model name : Intel(R) Xeon(R) CPU E5420 # 2.50GHz
stepping : 6
microcode : 0x606
cpu MHz : 1998.000
cache size : 6144 KB
physical id : 1
siblings : 4
core id : 3
cpu cores : 4
apicid : 7
initial apicid : 7
fpu : yes
fpu_exception : yes
cpuid level : 10
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx lm constant_tsc arch_perfmon pebs bts rep_good nopl aperfmperf pni dtes64 monitor ds_cpl vmx est tm2 ssse3 cx16 xtpr pdcm dca sse4_1 lahf_lm dtherm tpr_shadow vnmi flexpriority
bogomips : 4987.49
clflush size : 64
cache_alignment : 64
address sizes : 38 bits physical, 48 bits virtual
power management:
8 hosts per node would certainly be a valid way of doing things.
You could potentially also use fewer if you plan to do something like use MPI + threads. It just depends on your application. In general though, it is a safe way to go to say that you plan to use 1 rank per core.
Related
I am running Ubuntu 20.04.1 LTS/tf version 2.4.0, and I am not able to run the tensorflow library, because it always results in an error
This is the only line that I put in
import tensorflow as tf
This is the error it gives out
Illegal instruction (core dumped)
These are the processor specs
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 58
model name : Intel(R) Core(TM) i5-3437U CPU # 1.90GHz
stepping : 9
microcode : 0x21
cpu MHz : 842.451
cache size : 3072 KB
physical id : 0
siblings : 4
core id : 0
cpu cores : 2
apicid : 0
initial apicid : 0
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm cpuid_fault epb pti ssbd ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid fsgsbase smep erms xsaveopt dtherm ida arat pln pts md_clear flush_l1d
bugs : cpu_meltdown spectre_v1 spectre_v2 spec_store_bypass l1tf mds swapgs itlb_multihit srbds
bogomips : 4789.04
clflush size : 64
cache_alignment : 64
address sizes : 36 bits physical, 48 bits virtual
power management:
If more information will be needed I will provide it.
I have the same problem running the tf version 2.4.0 in Ubuntu 18.04.4 LTS. I have been looking for a solution but I didn't find it yet so, for now, I am using the previous version which works for me.
pip uninstall tensorflow
pip install tensorflow==2.3.1
It should be fixed now in 2.4.1
I am running v8's benchmark program, and I run the following command
./tools/cpu.sh fast
It prints out
Setting CPU frequency governor to "ondemand"
./tools/cpu.sh: line 13: /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor: no such file or directory
And I run
# ls /sys/devices/system/cpu/cpu0
cache crash_notes crash_notes_size microcode node0 power subsystem topology uevent
And find there is no "cpufreq"
After some searching, I find that I should install cpufrequtils, and I run
yum install cpufrequtils
After that, no thing works. So I wonder what is wrong here.
My system is
# lsb_release -a
LSB Version: :core-4.1-amd64:core-4.1-noarch
Distributor ID: CentOS
Description: CentOS Linux release 7.2 (Final)
Release: 7.2
Codename: Final
And my cpuinfo is
cat /proc/cpuinfo
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 94
model name : Intel(R) Xeon(R) Gold 61xx CPU
stepping : 3
microcode : 0x1
cpu MHz : 2499.998
cache size : 4096 KB
physical id : 0
siblings : 1
core id : 0
cpu cores : 1
apicid : 0
initial apicid : 0
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl eagerfpu pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch fsgsbase bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx rdseed adx smap xsaveopt xsavec xgetbv1 arat
bogomips : 4999.99
clflush size : 64
cache_alignment : 64
address sizes : 46 bits physical, 48 bits virtual
power management:
It depends on your kernel configuration whether the governor interface is exposed at all, and which governors are available. I don't know any specifics about CentOS. On Debian/Ubuntu the governors should be available by default (last I checked, they were).
Maybe https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/power_management_guide/cpufreq_governors helps?
I would like to know the following L3 cache parameters. But not sure how to get them, I also pasted the /proc/cpuinfo output (4 processors, only pasted the first one, the others are repetitive.)
CACHE_SIZE
LINE_SIZE
Associativity
$ cat /proc/cpuinfo
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 58
model name : Intel(R) Core(TM) i7-3520M CPU # 2.90GHz
stepping : 9
microcode : 0x15
cpu MHz : 1200.000
cache size : 4096 KB
physical id : 0
siblings : 4
core id : 0
cpu cores : 2
apicid : 0
initial apicid : 0
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm ida arat epb xsaveopt pln pts dtherm tpr_shadow vnmi flexpriority ept vpid fsgsbase smep erms
bogomips : 5786.68
clflush size : 64
cache_alignment : 64
address sizes : 36 bits physical, 48 bits virtual
power management:
UPDATE 1: It seems that someone posted the cache size and associativity here:
http://www.cpu-world.com/CPUs/Core_i7/Intel-Core%20i7-3520M%20%28PGA%29%20Mobile%20processor.html
But still I dont know the line size.
Hwloc / lstopo
Hwloc (Portable hardware locality) is a small utility that reports the structure of the processor in a neat visual diagram. The diagram shows the number of cores, hyperthreads and cache size. A single diagram tells it all.
$ sudo apt-get install hwloc
$ hwloc
I'm trying to get the size of different cache level in my system.
I tried two techniques.
a) Using information from /sys/device. Here is the output.
$ cat /sys/devices/system/cpu/cpu0/cache/index1/size
32K
$ cat /sys/devices/system/cpu/cpu0/cache/index2/size
256K
$ cat /sys/devices/system/cpu/cpu0/cache/index3/size
8192K
b) Using information from dmidecode
$ sudo dmidecode -t cache
Cache Information
Socket Designation: CPU Internal L1
Configuration: Enabled, Not Socketed, Level 1
Operational Mode: Write Through
Location: Internal
Installed Size: 256 KB
Maximum Size: 256 KB
< .... >
Cache Information
Socket Designation: CPU Internal L2
Configuration: Enabled, Not Socketed, Level 2
Operational Mode: Write Through
Location: Internal
Installed Size: 1024 KB
Maximum Size: 1024 KB
< .... >
Cache Information
Socket Designation: CPU Internal L3
Configuration: Enabled, Not Socketed, Level 3
Operational Mode: Write Back
Location: Internal
Installed Size: 8192 KB
Maximum Size: 8192 KB
< .... >
The size reported for L2 and L3 cache is different. Any ideas as to a) why this discrepancy? b) Which method gives the correct value?
Other related information:
$uname -a
Linux 3.0.0-14-generic #23somerville3-Ubuntu SMP Mon Dec 12 09:20:18 UTC 2011 x86_64 x86_64 x86_64 GNU/Linux
cat /proc/cpuinfo
processor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 58
model name : Intel(R) Core(TM) i7-3770 CPU # 3.40GHz
stepping : 9
cpu MHz : 2400.000
cache size : 8192 KB
physical id : 0
siblings : 8
core id : 0
cpu cores : 4
apicid : 0
initial apicid : 0
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm sse4_1 sse4_2 x2apic popcnt aes xsave avx f16c rdrand lahf_lm ida arat epb xsaveopt pln pts dts tpr_shadow vnmi flexpriority ept vpid fsgsbase smep erms
bogomips : 6784.23
clflush size : 64
cache_alignment : 64
address sizes : 36 bits physical, 48 bits virtual
power management:
< ... >
A few things:
You have a quad-core CPU
The index<n> name in /sys/devices/system/cpu/cpu<n>/cache does not correspond to L1/L2/L3 etc. There is a .../index<n>/level file that will tell you the level of the cache.
Your L1 cache is split into two caches (likely index0 and index1), one for data, and one for instructions (see .../index<n>/type), per core. 4 cores * 2 halves * 32K matches the 256K that dmidecode reports.
The L2 cache is split per-core. 4 cores * 256K (from index2) = 1024K, which matches dmidecodes L2 number.
I am doing performance analysis on linux for large scale programs which is memory driven(tens of Gigabytes memory).
I am thinking if it's possible to config linux/hardware to be more suitable to run such kind of large programs. But I am not familiar with this side.
Anybody have points about how to config
memory allocation strategy of OS
cache config for CPU
else...
Any comment is appreciated..
This is the typical CPU model (4 Opteron processors each has dual core):
processor : 3
vendor_id : AuthenticAMD
cpu family : 15
model : 65
model name : Dual-Core AMD Opteron(tm) Processor 2218
stepping : 2
cpu MHz : 2600.000
cache size : 1024 KB
physical id : 1
siblings : 2
core id : 1
cpu cores : 2
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt rdtscp lm 3dnowext 3dnow pni cx16 lahf_lm cmp_legacy svm extapic cr8_legacy
bogomips : 5200.09
TLB size : 1088 4K pages
clflush size : 64
cache_alignment : 64
address sizes : 40 bits physical, 48 bits virtual
power management: ts fid vid ttp tm stc
Useful for investigating memory / caching on a multi-socket system:
hwloc's lstopo (example):
lstopo
numactl / libnuma (but only if it really is a NUMA system)
numactl --hardware
numactl --show
sysfs, procfs:
sudo grep . /sys/devices/system/cpu/cpu*/cpufreq/*
grep . /sys/devices/system/cpu/cpu*/topology/physical_package_id
sudo grep . /proc/irq/*/smp_affinity # compare w/ /proc/interrupts