CentOS - Mar 2007 - DDR2 compile times (was Re: segfaults with 8 gig of ram)

Hello All.
Below are some compile times for 2.6.20 on an fx-62
running Centos64 with various 2 gig sticks of
dual channel ram (See previous posts with subject
of 'segfaults with 8gig of ram' for more details.)

It seems dual channel will compile a kernel faster,
but only gained around 30 seconds over 20 minutes.
Below are the shortest/longest compile times of
several compiles done over the weekend.

Compile consisted of make menuconfig (save .config)
time make dep clean bzImage modules


2 sticks dual channel @800mhz
10 samples
===============================real 791.46
user 664.10
sys 93.85

real 799.88
user 665.30
sys 93.87


4 sticks dual channel @400mhz
5 samples (before it hurled booting)
===============================real 876.19
user 712.98
sys 116.18

real 846.06
user 710.16
sys 115.42



3 sticks single channel @800mhz
6 samples
===============================real 815.74
user 673.73
sys 108.30

real 822.95
user 674.32
sys 108.62


-- 

Flambeau Inc. Technology Center - Baraboo, WI
Email    : tblader at flambeau.com
Keyserver: http://pgp.mit.edu KeyID: 0x00E9EC2C

tblader wrote:
> Hello All.
> Below are some compile times for 2.6.20 on an fx-62
> running Centos64 with various 2 gig sticks of
> dual channel ram (See previous posts with subject
> of 'segfaults with 8gig of ram' for more details.)
> 
> It seems dual channel will compile a kernel faster,
> but only gained around 30 seconds over 20 minutes.
> Below are the shortest/longest compile times of
> several compiles done over the weekend.
> 
> Compile consisted of make menuconfig (save .config)
> time make dep clean bzImage modules
> 
> 
> 2 sticks dual channel @800mhz
> 10 samples
> ===============================> real 791.46
> user 664.10
> sys 93.85
> 
> real 799.88
> user 665.30
> sys 93.87
> 
> 
> 4 sticks dual channel @400mhz
> 5 samples (before it hurled booting)
> ===============================> real 876.19
> user 712.98
> sys 116.18
> 
> real 846.06
> user 710.16
> sys 115.42
> 
> 
> 
> 3 sticks single channel @800mhz
> 6 samples
> ===============================> real 815.74
> user 673.73
> sys 108.30
> 
> real 822.95
> user 674.32
> sys 108.62
> 
> 
I am putting together a pretty high-end workstation and originally had 
ordered 2gb sticks. Here's my [main] hardware specs:

Abit AB9 QaudGT motherboard
Intel Core 2 Duo E6600
Seasonic M12 PSU
3 Seagate 250gb SATA II disks

Since my main goal is to run VMWare Workstation with host OS Centos 5 
(when it comes out), I was originally interested in using 8gb memory. 
But after posting the following thread on the Abit motherboard forums, I 
discovered that the Intel P965 Express chipset does not fully support 
PC6400 (800) DDR2 memory:

http://forum.abit-usa.com/showthread.php?t=121720

So I returned that memory and went with 4 1gb sticks of low latency 
(4-4-3-8) memory instead, trading memory capacity for performance.

tblader spake the following on 3/29/2007 6:19 AM:
> Hello All.
> Below are some compile times for 2.6.20 on an fx-62
> running Centos64 with various 2 gig sticks of
> dual channel ram (See previous posts with subject
> of 'segfaults with 8gig of ram' for more details.)
> 
> It seems dual channel will compile a kernel faster,
> but only gained around 30 seconds over 20 minutes.
> Below are the shortest/longest compile times of
> several compiles done over the weekend.
> 
> Compile consisted of make menuconfig (save .config)
> time make dep clean bzImage modules
> 
> 
> 2 sticks dual channel @800mhz
> 10 samples
> ===============================> real 791.46
> user 664.10
> sys 93.85
> 
> real 799.88
> user 665.30
> sys 93.87
> 
> 
> 4 sticks dual channel @400mhz
> 5 samples (before it hurled booting)
> ===============================> real 876.19
> user 712.98
> sys 116.18
> 
> real 846.06
> user 710.16
> sys 115.42
> 
> 
> 
> 3 sticks single channel @800mhz
> 6 samples
> ===============================> real 815.74
> user 673.73
> sys 108.30
> 
> real 822.95
> user 674.32
> sys 108.62
> 
> 
Dual channel at 400Mhz and single channel at 800mhz... Single channel at twice
the speed would be pretty close to equal. Now if you did dual channel at
800mhz you might see a difference.
Try a test with 2 sticks so you can get back to dual channel, put the memory
back up to its SPD detected speed and test that.

http://www.abit-usa.com/test-report/kn9.htm
shows the tested memory with that board.

-- 

MailScanner is like deodorant...
You hope everybody uses it, and
you notice quickly if they don't!!!!