Crafty 23.1 scaling problem on Nehalem octa

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

User avatar
Peter Skinner
Posts: 1763
Joined: Sun Feb 26, 2006 1:49 pm
Location: Edmonton, Alberta, Canada
Full name: Peter Skinner

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by Peter Skinner »

zullil wrote:
Peter Skinner wrote:
Hugo wrote: nps=20.2M
No, you are getting 20.2Mnps. Not 20000.

Peter
Didn't he write 20000K?
Yes, there is a difference between 20,000Knps and 20.2Mnps.

I would like to know where he is seeing the 20,000Knps number. The log clearly shows 20.2Mnps. Significant difference.

Peter
I was kicked out of Chapters because I moved all the Bibles to the fiction section.
bob
Posts: 20943
Joined: Mon Feb 27, 2006 7:30 pm
Location: Birmingham, AL

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by bob »

Peter Skinner wrote:
zullil wrote:
Peter Skinner wrote:
Hugo wrote: nps=20.2M
No, you are getting 20.2Mnps. Not 20000.

Peter
Didn't he write 20000K?
Yes, there is a difference between 20,000Knps and 20.2Mnps.

I would like to know where he is seeing the 20,000Knps number. The log clearly shows 20.2Mnps. Significant difference.

Peter
Think carefully, peter. :)

K = 1000.

So 20,000K == 20,000 x 1000 = 20,000,000

:)

Easy to make that mistake late at night.
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by zullil »

Peter Skinner wrote:
zullil wrote:
Peter Skinner wrote:
Hugo wrote: nps=20.2M
No, you are getting 20.2Mnps. Not 20000.

Peter
Didn't he write 20000K?
Yes, there is a difference between 20,000Knps and 20.2Mnps.

I would like to know where he is seeing the 20,000Knps number. The log clearly shows 20.2Mnps. Significant difference.

Peter
Well, now I'm confused too.

I assumed that "M" in the log meant 10^6, and that "K" for Clemens meant 10^3. So it's 20,200,000 vs. 20,000,000 nodes per second, a difference of 1%. Since that can't be considered "significant," I must be misunderstanding some notation here.

But even if that "M" means 2^20 and "K" just means 10^3, the difference between 20.2M and 20000K is about 6%.

What am I misunderstanding?
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by zullil »

Ignore my last post. Bob's reply, which must have arrived while I was typing mine, suffices.
Hugo
Posts: 782
Joined: Tue Dec 01, 2009 11:10 am

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by Hugo »

Hello Peter

here is your compile:

unable to open book file [./book.bin].
book is disabled
unable to open book file [./books.bin].

Initializing multiple threads.
System is NUMA. 2 nodes reported by Windows
Node 0 CPUs: 0 1 2 3
Node 1 CPUs: 4 5 6 7
Current ideal CPU is 4
Exchanging nodes 0 and 1

Crafty v23.1 (1 cpus)

White(1): mt 8
max threads set to 8.
White(1): bench
Running benchmark. . .
.Starting thread on node 1 CPU mask 15
Starting thread on node 0 CPU mask 240
Starting thread on node 1 CPU mask 15
Starting thread on node 0 CPU mask 240
Starting thread on node 1 CPU mask 15
Starting thread on node 0 CPU mask 240
Starting thread on node 1 CPU mask 15
.....
Total nodes: 154314273
Raw nodes per second: 12141170
Total elapsed time: 12.71
White(1):

here is Jims latest compile:

unable to open book file [./book.bin].
book is disabled
unable to open book file [./books.bin].

Initializing multiple threads.
System is NUMA. 2 nodes reported by Windows
Node 0 CPUs: 0 1 2 3
Node 1 CPUs: 4 5 6 7
Current ideal CPU is 7
Exchanging nodes 0 and 1

Crafty v23.1 (1 cpus)

White(1): mt 8
max threads set to 8.
White(1): bench
Running benchmark. . .
.Starting thread on node 1 CPU mask 15
Starting thread on node 0 CPU mask 240
Starting thread on node 1 CPU mask 15
Starting thread on node 0 CPU mask 240
Starting thread on node 1 CPU mask 15
Starting thread on node 0 CPU mask 240
Starting thread on node 1 CPU mask 15
.....
Total nodes: 173521932
Raw nodes per second: 22389927
Total elapsed time: 7.75
White(1):
regards, Clemens

PS.: about the nodes I wrote:
some gui show nodes always in Kilo-nodes. So when I run crafty with uci adapter, I get 20.000Kilonodes.
My understanding was that 20,2Mnodes means Meganodes
Kilo=1 thousand
Mega=1 milllion
think we both mean the sams :)
User avatar
Peter Skinner
Posts: 1763
Joined: Sun Feb 26, 2006 1:49 pm
Location: Edmonton, Alberta, Canada
Full name: Peter Skinner

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by Peter Skinner »

bob wrote: Think carefully, peter. :)

K = 1000.

So 20,000K == 20,000 x 1000 = 20,000,000

:)

Easy to make that mistake late at night.
The -40C conditions here in Edmonton have officially frozen my brain.

Sorry about that, obviously I wasn't thinking clearly.

With the windchill today, we are supposed to dip once again beyond the -40C (also -40F for those in farenheitland) mark... man I love when these arctic fronts move in :)

Peter
I was kicked out of Chapters because I moved all the Bibles to the fiction section.
User avatar
Peter Skinner
Posts: 1763
Joined: Sun Feb 26, 2006 1:49 pm
Location: Edmonton, Alberta, Canada
Full name: Peter Skinner

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by Peter Skinner »

I just compiled a new version for you Clemons with the EM64T instructions.

This is what I get on my Phenom 920 system:

D:\Chess\Compiling\crafty-23.1>crafty
unable to open book file [./book.bin].
book is disabled
unable to open book file [./books.bin].

Initializing multiple threads.
System is SMP, not NUMA.

Crafty v23.1 (1 cpus)

White(1): mt 4
max threads set to 4.
White(1): bench
Running benchmark. . .
......
Total nodes: 146566996
Raw nodes per second: 12368522
Total elapsed time: 11.85
White(1):

You should get significantly higher.
I was kicked out of Chapters because I moved all the Bibles to the fiction section.
Hugo
Posts: 782
Joined: Tue Dec 01, 2009 11:10 am

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by Hugo »

Hello Peter

thank you for the compile. It is the sam old "block" :

unable to open book file [./book.bin].
book is disabled
unable to open book file [./books.bin].

Initializing multiple threads.
System is NUMA. 2 nodes reported by Windows
Node 0 CPUs: 0 1 2 3
Node 1 CPUs: 4 5 6 7
Current ideal CPU is 0

Crafty v23.1 (1 cpus)

White(1): mt 8
max threads set to 8.
White(1): bench
Running benchmark. . .
.Starting thread on node 1 CPU mask 240
Starting thread on node 0 CPU mask 15
Starting thread on node 1 CPU mask 240
Starting thread on node 0 CPU mask 15
Starting thread on node 1 CPU mask 240
Starting thread on node 0 CPU mask 15
Starting thread on node 1 CPU mask 240
.....
Total nodes: 190647091
Raw nodes per second: 12135397
Total elapsed time: 15.71
White(1):

regards, Clemens
User avatar
Peter Skinner
Posts: 1763
Joined: Sun Feb 26, 2006 1:49 pm
Location: Edmonton, Alberta, Canada
Full name: Peter Skinner

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by Peter Skinner »

Hugo wrote:Hello Peter

thank you for the compile. It is the sam old "block" :

unable to open book file [./book.bin].
book is disabled
unable to open book file [./books.bin].

Initializing multiple threads.
System is NUMA. 2 nodes reported by Windows
Node 0 CPUs: 0 1 2 3
Node 1 CPUs: 4 5 6 7
Current ideal CPU is 0

Crafty v23.1 (1 cpus)

White(1): mt 8
max threads set to 8.
White(1): bench
Running benchmark. . .
.Starting thread on node 1 CPU mask 240
Starting thread on node 0 CPU mask 15
Starting thread on node 1 CPU mask 240
Starting thread on node 0 CPU mask 15
Starting thread on node 1 CPU mask 240
Starting thread on node 0 CPU mask 15
Starting thread on node 1 CPU mask 240
.....
Total nodes: 190647091
Raw nodes per second: 12135397
Total elapsed time: 15.71
White(1):

regards, Clemens
This is an Intel system correct? If so, why does it say it is NUMA? That is AMD only if I am correct.

Peter
I was kicked out of Chapters because I moved all the Bibles to the fiction section.
bob
Posts: 20943
Joined: Mon Feb 27, 2006 7:30 pm
Location: Birmingham, AL

Re: Crafty 23.1 scaling problem on Nehalem octa

Post by bob »

Hugo wrote:Hello Peter

thank you for the compile. It is the sam old "block" :

unable to open book file [./book.bin].
book is disabled
unable to open book file [./books.bin].

Initializing multiple threads.
System is NUMA. 2 nodes reported by Windows
Node 0 CPUs: 0 1 2 3
Node 1 CPUs: 4 5 6 7
Current ideal CPU is 0

Crafty v23.1 (1 cpus)

White(1): mt 8
max threads set to 8.
White(1): bench
Running benchmark. . .
.Starting thread on node 1 CPU mask 240
Starting thread on node 0 CPU mask 15
Starting thread on node 1 CPU mask 240
Starting thread on node 0 CPU mask 15
Starting thread on node 1 CPU mask 240
Starting thread on node 0 CPU mask 15
Starting thread on node 1 CPU mask 240
.....
Total nodes: 190647091
Raw nodes per second: 12135397
Total elapsed time: 15.71
White(1):

regards, Clemens
We need to discuss hardware. What is this "starting thread on node 1" stuff? That appears to be NUMA. Crafty says this system is NUMA. Which suggests that it is an AMD box, not a Core I7 box, or at least not one I have run on.

AMD is the only intel-compatible chip that has local memory for each CPU socket, at least as far as X86 boxes I have used. Itanium boxes are typically NUMA, as are some others, but not X86.

Can you provide explicit details of the hardware configuration (who made it, system name, model number, etc) and I will see what I can find out.

Another quick idea. Boot the system, and hit whatever (F2, F10, etc) you need to enter BIOS setup. Look for the memory options and see if you see something like "mode NUMA/SMP". If you do, switch it to NUMA. Otherwise the thing interleaves pages of memory between the two memory units which will hurt everywhere as Crafty knows how to acquire local memory, which is much more efficient on a real NUMA box.