Main
Date: 21 Jun 2006 08:58:51
From: Thomas T. Veldhouse
Subject: Hyperthreading and Dual Core

I am wondering if any of the "Deep" multiprocess chess engines can properly
utilize hyperthreading? Since they are SMP capable programs, they should use
it, but do they use it efficiently like they would a true multip-processor
machine?

Also, is there any reason these same programs wouldn't run equally well on a
dual-core machine as opposed to multi-processor machine? Which leads to
another interesting question. How about those dual-core hyperthreaded chips
that Intel is putting out? Thus, an appearance of four CPUS, two physical and
two logical; will these programs utilize them?

How about crafty?

--
Thomas T. Veldhouse
Key Fingerprint: 2DB9 813F F510 82C2 E1AE 34D0 D69D 1EDC D5EC AED1





 
Date: 25 Jun 2006 08:33:21
From: Guy Macon
Subject: Re: Hyperthreading and Dual Core



AMD is rumoured to be about to announce a processor with
antihyperthreading; two cores that look like one faster core
to single-processor applications.




  
Date: 26 Jun 2006 08:14:42
From: Thomas T. Veldhouse
Subject: Re: Hyperthreading and Dual Core
Guy Macon <http://www.guymacon.com/ > wrote:
>
>
>
> AMD is rumoured to be about to announce a processor with
> antihyperthreading; two cores that look like one faster core
> to single-processor applications.
>

While this would be cool, I have my doubts about feasibility and thus, this
report (no offense intended). Certainly, you will not see anything like a AMD
Athlon X2 4800+ looking like a Athlon 9800+. I am curious what the effective
rating would be and how stable it would be as a single CPU core.

--
Thomas T. Veldhouse
Key Fingerprint: 2DB9 813F F510 82C2 E1AE 34D0 D69D 1EDC D5EC AED1



 
Date: 21 Jun 2006 17:38:14
From: [email protected]
Subject: Re: Hyperthreading and Dual Core
I have a Pentium D 920 (dual-core).

I've used Rybka 2.0 beta and Deep Shredder 10.

Both chess programs utilized BOTH cores at 100%.

Unfortunately, the 920 doesn't have hypertheading.




--
--------------------------------- --- -- -
Posted with NewsLeecher v3.7 Final
Web @ http://www.newsleecher.com/?usenet
------------------- ----- ---- -- -



 
Date: 21 Jun 2006 15:51:58
From: Gian-Carlo Pascutto
Subject: Re: Hyperthreading and Dual Core
Thomas T. Veldhouse wrote:
> I am wondering if any of the "Deep" multiprocess chess engines can properly
> utilize hyperthreading? Since they are SMP capable programs, they should use
> it, but do they use it efficiently like they would a true multip-processor
> machine?
>
> Also, is there any reason these same programs wouldn't run equally well on a
> dual-core machine as opposed to multi-processor machine? Which leads to
> another interesting question. How about those dual-core hyperthreaded chips
> that Intel is putting out? Thus, an appearance of four CPUS, two physical and
> two logical; will these programs utilize them?
>
> How about crafty?

The "Deep" programs will utilize them as they would utilize a true
multiprocessor system. For dual-core, the speedups will be similar to a
true multiprocessor. For hyperthreading, the speedups are basically
nonexistant (which isn't surprising, since you really *dont* have extra
processing units).

--
GCP


  
Date: 21 Jun 2006 14:04:40
From: Jud McCranie
Subject: Re: Hyperthreading and Dual Core
On Wed, 21 Jun 2006 15:51:58 GMT, Gian-Carlo Pascutto
<[email protected] > wrote:

>true multiprocessor. For hyperthreading, the speedups are basically
>nonexistant (which isn't surprising, since you really *dont* have extra
>processing units).

In tests or other programs, I thin khyperthreading gives a 10-15%
boost.
---
Replace you know what by j to email


   
Date: 22 Jun 2006 06:15:36
From: Gian-Carlo Pascutto
Subject: Re: Hyperthreading and Dual Core
Jud McCranie wrote:
> On Wed, 21 Jun 2006 15:51:58 GMT, Gian-Carlo Pascutto
> <[email protected]> wrote:
>
>
>>true multiprocessor. For hyperthreading, the speedups are basically
>>nonexistant (which isn't surprising, since you really *dont* have extra
>>processing units).
>
>
> In tests or other programs, I thin khyperthreading gives a 10-15%
> boost.

In terms of nodes-per-second, yes. But no speedup in terms of actually
finding the right moves faster.

--
GCP


    
Date: 22 Jun 2006 16:09:49
From: Alexander Belov
Subject: Re: Hyperthreading and Dual Core
"Gian-Carlo Pascutto" <[email protected] > wrote in message
news:[email protected]...
> Jud McCranie wrote:
> > On Wed, 21 Jun 2006 15:51:58 GMT, Gian-Carlo Pascutto
> > In tests or other programs, I thin khyperthreading gives a 10-15%
> > boost.
>
> In terms of nodes-per-second, yes. But no speedup in terms of actually
> finding the right moves faster.

I've got up to 13% boost in time of finding the right moves faster with
HyperThreading, while less than 2% increase in nodes-per-second. It seems
search threading in case of HT is able to fix some small and unpredictable
ordering problems and thus improve search time.




     
Date: 22 Jun 2006 17:10:09
From: Gian-Carlo Pascutto
Subject: Re: Hyperthreading and Dual Core
Alexander Belov wrote:
> "Gian-Carlo Pascutto" <[email protected]> wrote in message
> news:[email protected]...
>
>>Jud McCranie wrote:
>>
>>>On Wed, 21 Jun 2006 15:51:58 GMT, Gian-Carlo Pascutto
>>>In tests or other programs, I thin khyperthreading gives a 10-15%
>>>boost.
>>
>>In terms of nodes-per-second, yes. But no speedup in terms of actually
>>finding the right moves faster.
>
>
> I've got up to 13% boost in time of finding the right moves faster with
> HyperThreading, while less than 2% increase in nodes-per-second. It seems
> search threading in case of HT is able to fix some small and unpredictable
> ordering problems and thus improve search time.

In other words, the program you tested is buggy in single-CPU mode :-P

--
GCP


      
Date: 23 Jun 2006 12:10:32
From: Alexander Belov
Subject: Re: Hyperthreading and Dual Core

"Gian-Carlo Pascutto" <[email protected] > wrote in message
news:[email protected]...
> Alexander Belov wrote:
> > I've got up to 13% boost in time of finding the right moves faster with
> > HyperThreading, while less than 2% increase in nodes-per-second. It
seems
> > search threading in case of HT is able to fix some small and
unpredictable
> > ordering problems and thus improve search time.
>
> In other words, the program you tested is buggy in single-CPU mode :-P

I don't know whether it is buggy or not in single-CPU mode. I can only say
that the program could draw the game with Crafty 19.03 having in average
x1.5 time less depth of search. When it was buggy it couldn't do it. It is
just one case (in some other positions depth of search still rules), but as
I know buggy programs cannot keep up with stable GM programs.
Concerning chess multithreading efficiency - it is still wide field for
investigation, and some solutions can be better than the other.




    
Date: 22 Jun 2006 12:49:48
From: HD
Subject: Re: Hyperthreading and Dual Core
Gian-Carlo Pascutto wrote:
> Jud McCranie wrote:
>> On Wed, 21 Jun 2006 15:51:58 GMT, Gian-Carlo Pascutto
>> <[email protected]> wrote:
>>
>>
>>> true multiprocessor. For hyperthreading, the speedups are basically
>>> nonexistant (which isn't surprising, since you really *dont* have extra
>>> processing units).
>>
>> In tests or other programs, I thin khyperthreading gives a 10-15%
>> boost.
>
> In terms of nodes-per-second, yes. But no speedup in terms of actually
> finding the right moves faster.
>

Agree.
In fact my tests showed that HyperTreading can be damaging for the play
of some engines in some situations. Therefore I started to switch off
the 2. tread in all engines smp-capable when playing them on that machine.