Most powerful machine you've put gentoo on?

[phsythax]

impressive how fast the technology is moving ahead.

[rbr28]

Finallly got around to upgrading my old P4. Upgraded for about $400, just adding the components below to my existing system.

Intel Quad Core Q9400 2.66Ghz (no overclock yet)
Super Talent PC1600 4GB 7-7-7-21
BFG Geforce 9600GT 512MB
Gigabyte GA-EP45T-UD3LR motherboard

Rebuilt my OS from scratch using GCC 4.3.3 so I could use -march=core2. I wasn't brave enough to try GCC 4.4 yet. Also enabled -msse4.1. Streamlined everything beyond belief. I put Ubuntu 9.04 on the system just before Gentoo to compare. Gentoo took way longer but was tons faster. I'm amazed at the responsiveness of KDE4.

[d2_racing]

Hi, are you sure that your Quad Core is compatible with -msse4.

I tought that it was for the CoreI7 :

http://en.gentoo-wiki.com/wiki/Safe_Cfl ... 2B.2FExxxx

Also, for my Intel Quad Core Q9550, I use this :

Code: Select all

CFLAGS="-O2 -march=native -pipe"
CHOST="x86_64-pc-linux-gnu"
CXXFLAGS="${CFLAGS}"
MAKEOPTS="-j5"

[rbr28]

Checked the link you pointed to and I guess that although they refer to the i7 as having sse 4.1, they don't necessarily state that the other cores don't have it. For some reason there's a lot of misinformation about the Q9400 out there. I've also seen mention in forums that this processor lacks vt extensions. I can say without a doubt that it has sse 4.1 and vt-x. The Intel product datasheet is always the most accurate source, sometimes even the information on the rest of their site is incomplete.

Also, the i7 added sse 4.2, but I haven't checked to see if those optimizations are in gcc 4.4, I didn't notice that they were present in gcc 4.3.3

[Yamakuzure]

you have wayy too much swap - the general rule nowadays is to not go over 512MB... of course, if you have that much ram, why do you need swap anyway?

There are reasons to have relatively large quantities of swap space.

And I have mounted /tmp on 1GB tmpfs, and /var/tmp on 10GB tmpfs. I will upgrade my RAM to 8GB soon, and the normal way for swap is RAM * 1.5, which is 12GB. (And with 11GB on tmpfs drives, I won't get memory problems with 12GB swap either, even when compiling OpenOffice.

btw:

Also, for my Intel Quad Core Q9550, I use this :

Code: Select all

CFLAGS="-O2 -march=native -pipe"
CHOST="x86_64-pc-linux-gnu"
CXXFLAGS="${CFLAGS}"
MAKEOPTS="-j5"

I have edited my flags according to what AMD is recommending for PhenomII. (But I am not brave enough to use *everything* they recommend, though) :

Code: Select all

# grep -i FLAG /etc/make.conf
## C-/LD-Flags Update for new Phenom II System, according to "Compiler Usage Guidelines for 64-Bit OS on AMD64"
CFLAGS="-march=amdfam10 -O2 -pipe -mno-push-args -mcx16 -msahf -funroll-loops -ftree-vectorize"
LDFLAGS="-s -Wl,-O1 -Wl,--as-needed -Wl,--sort-common -Wl,-z,now -Wl,--hash-style=gnu"
CXXFLAGS="${CFLAGS}"

Unforunately they completely forbid me to use "gold" USE Flag with binutils, as it neither knows --sort-common, nor -z,now...

[Mr. M]

So far, my desktop with a Core i7 2.66GHz CPU, 6GB of RAM. Quad core really helps when you are using Gentoo, I just upgraded from KDE 4.2.3 to KDE 4.2.4 in probably less than 1h.

I will soon be installing Gentoo on a small cluster consisting of 4 dell power edge servers with dual quad core Xeon E5570, 24GB RAM each

[Traffixxx]

Intel(R) Core(TM)2 Quad CPU Q9550 @ 2.83GHz
2 GB DDR2-800
Asus P5Q pro
discs: 300GB, 500GB, 2 x 1.5T and 2 x 160GB
ATI Technologies Inc RV770 [Radeon HD 4870]

[warrawarra]

Hi, are you sure that your Quad Core is compatible with -msse4.

I thought that it was for the CoreI7 :

http://en.gentoo-wiki.com/wiki/Safe_Cfl ... 2B.2FExxxx

Also, for my Intel Quad Core Q9550, I use this :

Code: Select all

CFLAGS="-O2 -march=native -pipe"
CHOST="x86_64-pc-linux-gnu"
CXXFLAGS="${CFLAGS}"
MAKEOPTS="-j5"

Some quads have sse4.1
http://processorfinder.intel.com/Default.aspx
http://www.cpu-world.com/CPUs/Core_2/In ... 50%29.html

The -j options was originally for 1x cpu at 16bit/32bit -j1+1 then came 64bit = 2x32bit then came hyper threading 3x32bit or 3x64bit=6x32bit then came the quad 6x32bit x 4 cores if hyper threading or 4x32bit x 4cores so a quad 64bit should be able to minm -j8 on mmx/sse1 if thinking of original pentuim cpu's.
Update that to today's sse4.1 64bit quads. Then load balance this with a ati stream or nv cuda card and you need to rebuild your kernel to 8+ cores / MPI .
Things are never as efficient as above math so (conservative to potential) dual core 32bit = -j8 to -j16 and 64bit quad -j32 to -j64/-j128 and 8cores to -j256/-j512.
640 ati stream 4770 = -j ??? , 800 ati stream 4850 = -j ??? , 1600 ati stream ati 4850x2 = -j ???

I have been using for 3 years -j16 on a intel t2600 core duo 32bit laptop and runs perfect. No idea why you are only -j5 , my laptop runs almost optimal and loves this to pieces with "-j16" (dell e1705 t2600 nv7800gtx) , used several -march= options same results.

The 8 core intels coming out sept - nov 2009 will have 128 threads so 16 per core and leave some for the host OS to work with 128 - 16 = 112 usable.

The i7c has hyper threading and VT so the OS thinks you have 8 cores and should at least be able to do "-j48 " or hyper threaded "-j64" or 3 to 4 times my laptop dual core 32bit.

My q9550 runs at -j32 and sometimes uses all 4 cores "no hyper threading" 2.8ghz 53C leaving 20C temp gap. Would still like to get optimal -j for the q9550 for minm. 95% load.

Have a look at "-j5 -s" or "-j16 -s" and what it does.

I have read somewhere you can 2nd level optimize, as example " -march=i686 -mtune=nocona" to have default i686 optimized for nocona ie: both i686 and nocona optimized instead of only nocona and other i686 support is dropped from the build. This both option is good for livecd's .
Have a look at the gcc 4.4.1 and beta 4.5.* docs for more info.
x86 versus i686 is more optimized for slower 486 pc's and outperforms i686 + less bugs / issues. Most problems has been fixed for x86 "486" versus i686. Using python 2.6.2 gcc 4.4.0 openrc5 and you will notice the speed difference on both x86 and i686 builds.

You guy's should try ati stream and running 3 to 8 vm on a ati video card like they do in ubuntu / amd developer forums using MPI minm 640 stream processors to 6400 (4x4850x2) then the cpu amd X4 PII 955 is actually a glorified memory controller.

Have fun.
The future is here it is just not widely distributed yet.

[poly_poly-man]

buy a ton of ram and make your makeopts "-j"

[sk8harddiefast]

in my desktop pc.
Intel Quad Core Q6600 + 2g ddr3 Ram + ~1,5 tera hdd

[cach0rr0]

Code: Select all

vendor_id	: AuthenticAMD
cpu family	: 16
model		: 2
model name	: AMD Phenom(tm) 9950 Quad-Core Processor
stepping	: 3
cpu MHz		: 2606.303
cache size	: 512 KB
physical id	: 0
siblings	: 4
core id		: 3
cpu cores	: 4
apicid		: 3
initial apicid	: 3
fpu		: yes
fpu_exception	: yes
cpuid level	: 5
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 pdpe1gb rdtscp lm 3dnowext 3dnow constant_tsc rep_good pni monitor cx16 lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs
bogomips	: 5213.26
TLB size	: 1024 4K pages
clflush size	: 64
cache_alignment	: 64
address sizes	: 48 bits physical, 48 bits virtual
power management: ts ttp tm stc 100mhzsteps hwpstate

(times 4 of course - meaning, a singular quad core Phenom 9950)

also

Code: Select all

gentoob0x insecure # free -m
             total       used       free     shared    buffers     cached
Mem:          7921       5297       2623          0        320       3464
-/+ buffers/cache:       1512       6408
Swap:          980          0        980

I should disabled the onboard vid card to be honest since this is my server, just havent done it yet as things are working beautifully and i despise reboots unless it's a kernel rebuild

[Malakai]

Allow me to share my specs! Runs gentoo and vista 64 ultimate. I <3 AMD, I got the cpu and mobo for 280 shipped from newegg back in say, march. Prices are the same even now months later. And got 8gb of the ddr2 for 110 shipped. amazing value, this thing compiles like a beast!


Phenom II 940 BE 3.6ghz(240x15)@1.44v, 2.16ghz NB, 1.92ghz HT | Sunbeam Core Contact 120
Gigabyte MA790GP-UD4H 790GX Mobo | Sapphire HD4830 @680/1050 | Thermaltake 550W PSU
8gb(4x2) G.Skill DDR2-1066C5@960mhz 5-5-5-15 2.0v | Logitech z-2100e 200W 2.1 Speakers
WD 640 Black & WD 1TB Green HDDs | Logitech MX518 Mouse | Antec 300 fully loaded with fans
BenQ G2400WD 24" Widescreen TN | Viewsonic VP191b 19" | Vista Ult 64-bit | Linksys WRT54GL

[hielvc]

Spec's for my old most powerfull machine and the new most powerfull machine. MB Fry's special $49 Gigabyte GA-MS1PME-S2P AM2+ AM3, fRY'S $30 speacial 2gigs of Kingston or Corsair slow ram, Geeks cheapies $54 each Seagate 250Gig SATA1 raid0 The processors: 1st AMD X2 BE2300 45W sse3 1.8ghz with 512 cache. 2nd $64 fRY'S special AMD ATHLON X2 7650 BE 2.8ghz OCable if the el cheapo MB could.
Video MB geforce 6150 that nvivia-drivers refuse to have anything to do with, Hay xorg "nv" runs pengine racer runs fine.

Not listed L3 2meg cache

• processor : 0
  vendor_id : AuthenticAMD
  cpu family : 16
  model : 2
  model name : AMD Athlon(tm) 7850 Dual-Core Processor
  stepping : 3
  cpu MHz : 2813.047
  cache size : 512 KB
  physical id : 0
  siblings : 2
  core id : 0
  cpu cores : 2
  apicid : 0
  initial apicid : 0
  fpu : yes
  fpu_exception : yes
  cpuid level : 5
  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 pdpe1gb rdtscp lm 3dnowext 3dnow constant_tsc rep_good nonstop_tsc extd_apicid pni monitor cx16 lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs
  bogomips : 5628.50
  TLB size : 1024 4K pages
  clflush size : 64
  cache_alignment : 64
  address sizes : 48 bits physical, 48 bits virtual
  power management: ts ttp tm stc 100mhzsteps hwpstate

Genlop listings show the BE2300 first then the 7850 second ( this a 1 month old AMD64 install which started with the BE2300). The BE2300 and the first 7850 time are with -j4. the last 3 7850's times are -j16, then -j8 and lastly -j6. Test conditions, Listening the the Goblet of Fire, 2 or 3 terminals running htop, genlop and nano make.conf, and occasional wiggling of the mouse to kill xscrensaver.

For the BE2300 CFLAGS="-march=k8-sse3 -O2 -pipe"
For the 7850 CFLAGS="-march=amdfam10 -O2 -pipe"

• genlop -t gcc
  ' * sys-devel/gcc
  
  Sat Jun 27 16:04:14 2009 >>> sys-devel/gcc-4.3.3-r2
  merge time: 58 minutes and 13 seconds.
  
  Mon Jul 6 17:02:23 2009 >>> sys-devel/gcc-4.3.3-r2
  merge time: 1 hour, 4 minutes and 25 seconds.
  
  Thu Jul 23 14:28:33 2009 >>> sys-devel/gcc-4.3.3-r2
  merge time: 23 minutes and 27 seconds.
  
  Fri Jul 31 00:54:22 2009 >>> sys-devel/gcc-4.3.3-r2
  merge time: 24 minutes and 26 seconds.
  
  Fri Jul 31 12:18:06 2009 >>> sys-devel/gcc-4.3.3-r2
  merge time: 23 minutes and 31 seconds.
  
  Fri Jul 31 13:56:42 2009 >>> sys-devel/gcc-4.3.3-r2
  merge time: 23 minutes and 24 seconds.

and

• * sys-libs/glibc
  
  Sat Jun 27 15:06:01 2009 >>> sys-libs/glibc-2.9_p20081201-r2
  merge time: 55 minutes and 32 seconds.
  
  Thu Jul 23 14:50:10 2009 >>> sys-libs/glibc-2.9_p20081201-r2
  merge time: 21 minutes and 37 seconds.
  
  Fri Jul 31 00:29:56 2009 >>> sys-libs/glibc-2.9_p20081201-r2
  merge time: 22 minutes and 5 seconds.
  
  Fri Jul 31 11:54:34 2009 >>> sys-libs/glibc-2.9_p20081201-r2
  merge time: 21 minutes and 10 seconds.
  
  Fri Jul 31 13:33:18 2009 >>> sys-libs/glibc-2.9_p20081201-r2
  merge time: 21 minutes and 15 seconds.

While some may consider this not quit , err , controlled enough it is interesting that build times for both builds seem to like the -j8 setting.

[durian]

New i7 920 machine with 6 GB of memory:

Code: Select all

margretetorp ~ # genlop -i gcc
 * sys-devel/gcc

   Total builds: 1
   Global build time: 10 minutes and 55 seconds.

   Info about currently installed ebuild:

   * sys-devel/gcc-4.3.2-r3
   Install date: Fri Aug 14 07:26:03 2009

-peter

[Thaidog]

I just got root access to a 128 core Altix supercomputer! Right now we are only testing our ICC kernel but we are also making an icc stage 3 tarball that I hope to install on it as well

[dol-sen]

@#$@$%, I finally get to build a new one (7 years since my last one) and it's already nearly you guy's cast offs

Intel Q8400 SLGT6 quad core @2.66Ghz.
GigaByte EP45-UD3R mobo
4Gig ram (slow 400's) - hope to get some nice fast ones.
LG BD-Rom/DVD & CD rewriter drive
WD 500G sata 300 HDD
MSI N9400GT-MD1G 1G ram w/HDMI out
Cooler Master Sileo500 case/ps <== nice and quiet
Asus VH242 - 23.6in. widescreen monitor 1900x1080 <== don't need a magnifying glass for these aging eyes

re-using keyboards/mice and Klipsh THX certified 4.1 speakers (local auction buy several years ago, good price )

I also want to say that things are working quite well too. Still some things to iron out, but heh, isn't that what progress is all about.

I also see I can optimize a little more too. lshw reports this cpu having sse4.1

[Raze]

2x Intel Xeon CPU E5530 @ 2.40GHz
24GB Ram
3ware 9690SA
2x FUJITSU MBA3147RC ( 136GB, 15K SAS )

Goes like hell

Two of these machines replaced 3 older, Netburst based Xeon. Querytime and systemload reduced by 50%

[poly_poly-man]

2x Intel Xeon CPU E5530 @ 2.40GHz
24GB Ram
3ware 9690SA
2x FUJITSU MBA3147RC ( 136GB, 15K SAS )

Goes like hell

Two of these machines replaced 3 older, Netburst based Xeon. Querytime and systemload reduced by 50%

please compile a kernel with make -j, and tell me how long it takes... I want to drool

[d2_racing]

In fact, also how long for GCC 4.4.1

[hielvc]

Yeah that kernel compile will be about 35 sec, duh BASSSSSSTUUUURD , in other words fish droppings gcc 5min or less.

[JanR]

Hi,

dual Xeon 5570 (2x2.93 GHz, turbo up to 3,33 GHz)... compiles Openoffice in 25 minutes and an optimized kernel with -j32 in less than 30 seconds. 48 GB RAM make this machine even nicer...

Greetings,

Jan

[poly_poly-man]

Hi,

dual Xeon 5570 (2x2.93 GHz, turbo up to 3,33 GHz)... compiles Openoffice in 25 minutes and an optimized kernel with -j32 in less than 30 seconds. 48 GB RAM make this machine even nicer...

Greetings,

Jan

can you try with -j instead of -j32?

[hielvc]

I tried your -j and it locked my box to point where I had to power off. Is that what you expected ? What is the point of "-j"

[poly_poly-man]

-j compiles every single job at once...

the kernel should only take about 16gb ram... did you try on a fresh boot?

[JanR]

Hi,

now my try with -j...

I compile at a ramdisk (tmpfs), make one warmup run and count the second run.

First with -j32:

(> /dev/null should avoid impact of console/ssh/whatever):

Code: Select all

make clean; /usr/bin/time make -j32 > /dev/null 

492.45user 37.89system 0:36.61elapsed 1448%CPU (0avgtext+0avgdata 0maxresident)k

This is the correct time... the time mentioned in the previous post was just from memory. I can post the config, it is mixture of the config of my i7 workstation and optimizations for that NUMA server.

Now -j (that machine has more than enough memory...):

Code: Select all

make clean ; /usr/bin/time make -j > /dev/null

496.01user 34.29system 0:37.02elapsed 1432%CPU (0avgtext+0avgdata 0maxresident)k

Worse, as I exepcted due to the scheduling overhead.

If we chose something smaller...

Code: Select all

 make clean ; /usr/bin/time make -j24 > /dev/null 

482.53user 37.17system 0:37.04elapsed 1402%CPU (0avgtext+0avgdata 0maxresident)k

...it also becomes slower.

Greetings,

Jan