[Hardware] Dual-core and/or SMP + Win9x/ME

Vadim Tukaev shiz0rat at real.xakep.ru
Thu Oct 20 02:06:44 EDT 2005

Floris Dansen wrote:
> Hi,


> The distributed.net client can ONLY use processors/cores that are also
> supported by the operating system.

So, I'm wrong. I hoped, that dnetc have one`s own "intellect" for 
detecting number of processors and their capabilities. Am I talk rot?

> Windows 9x/me do NOT support multiple processors/cores,

I know that. It is precisely because I'm asked my primordial question.

> therefore the client will only use one.

I understand this yet, but it not seemsed to me as obvious conclusion.

> The least. Furthermore, there is a difference in performance with different
> operating systems; this has to do with how well the OS is programmed, if the
> OS efficiently uses new processor features etc.

I know that too, that is why I used word "approximately".

> If I remember correctly, the client runs fastest on Linux or Windows XP

I can flatly assert, that OS/2 not slower them and may be even faster.

> So, to apply that to your scenarios:
> Scenario 1: More or less, small differences exist

Of course, I implied maximal cleaned system with extirpated services.

> (or large if you count DOS as an OS ;))

Yes, I count DOS as an OS. Not ugliest OS, that I seen... Windows much 
more pathetic, because it want to appears as multitasking OS, although 
it must be correctly called "fewtasking". So, DOS is honest at any rate.

> Scenario 2: If the number of processors > 1, you will need to use an OS
> supporting that. If the OS supports it, you can more or less say that if you
> use processors doing 1Mkeys/s, the output will be number of processors times
> 1Mkeys/s. In your example: 1Mkeys/s because of Win9x/Me.


> Note that this is in most cases, where the average machine load is low. In
> specific situations with heavy loads, this doesn't apply.

That goes without saying!

> Scenario 3: Again, Win9x/ME do not support multiple processors/cores, you
> will get as much performance as with a single core processor with the same
> clockspeed.

Understanded yet.

> Scenario 4: See scenario 3. You WILL need an OS supporting the amount of
> CPU's in your system. Note that if you have 2 dual core processors, your OS
> will need to support 4 CPU's.

I suspected it. Thanks for this verification, it is extremely important.

> This is for example not the case with the standard Windows NT 4 machines.

You right, if "NT4.0 Workstation" was meant by "standart". NT4.0 Server 
support 4 CPU, but its OEM-modifications able to support up to 32 CPUs.

> Question: If you have the meens to put together a 4-core machine,

But I havn't it. ;) I have a love of knowledge, so just interested.

> why would you want to run something as awful as Win9x/me ?

1.I want separate computer for gaming ONLY. AFAIK, best OS for this 
purpose is Win98 (all my friends, who have skills in Windows, say thus).

2.I am fanat of distributed calculations, especially d-net.

3.I take the liberty of for buying dual-proc/dual-core machine instead 
of single-proc/single-core. Of course, this unnecessary expenditure 
dosn't affect speed and quality of games. In other words three cores in 
a four always give 100% of their power to dnetc. Only one core, which 
detected by Win98, process game and Win98 as such.

My train of thought:
For example, we have two computers: FirstKomp with M.A.D. UnoKore 3GHz 
and SecondKomp with two M.A.D. PairKore 3GHz. We play game, which 
require 2.7Ghz. Windows grab hipotetically 9% of CPU power, i.e. 270Mhz.

FirstKomp. Total requiments of game and Win - 99%. So, 1% for dnetc.
SecondKomp. One core process game and Win, as in FirstKomp. 1% for dnet. 
Other three cores process dnetc only. So, we have 301% of UniKore. =)

As you can see, we have same efficiency in games and much higher in 
dnetc. Unfortunately, it is impossible, as far as I understand. I must 
use one UniKore 3GHz with Win98 OR SMP/Dual-core with Windows 2000/XP.

ThirdKomp. Two M.A.D. PairKore 2.7GHz with any Windows NT 5.x, which 
grab hipotetically 20% of CPU power. One 2.7GHz-core process game, which 
require 2.7GHz, so 0% for dnetc. Another core process NT as such, so 80% 
for dnetc. Others cores give full power to dnetc, so summ is 280%. Of 
course, this is dumb scenario too, because expected smart OS with ideal 
allocation of recources. Windows never be so smart. But it's just model.

> regards,
> Floris Dansen
> (Floppus)
> -----Oorspronkelijk bericht-----
> Van: hardware-bounces at lists.distributed.net
> [mailto:hardware-bounces at lists.distributed.net]Namens Vadim Tukaev
> Verzonden: zondag 16 oktober 2005 7:00
> Aan: hardware at lists.distributed.net
> Onderwerp: [Hardware] Dual-core and/or SMP + Win9x/ME
> Hi/2!
> What happend, if I start distributed.net-client on SUBJ machine? Can
> dnetc utilize all processor power in this situation? In other words:
> Scenario1. Let us assume, that I have computer with processor, which
> produce 1M keys/sec. I run dnetc in any OS and get approximately 1M k/s.
> Scenario2. I install to my computer two processors, each of them equal
> to processor from Scenario1. I run dnetc on top of Win9x/ME, not
> NT4/2000/XP. How many keys/sec I can get practically? 1M or 2M?
> Scenario3. I install dual-core processor, each core of it equal to
> processor in Scenario1. I run dnetc on top of Win9x/ME. What I get?
> Scenario4. I install two Scenario3-processors (i.e. dual-core). They
> able to produce 4M k/s theoretically, but can I get such on Win9x/ME?
> --
> Open OS/2! (Members: 13; Rank: 1206; Blocks: 89020; Working: 331 days)
> http://stats.distributed.net/team/tmsummary.php?project_id=8&team=30911
> _______________________________________________
> Hardware mailing list
> Hardware at lists.distributed.net
> http://lists.distributed.net/mailman/listinfo/hardware
> _______________________________________________
> Hardware mailing list
> Hardware at lists.distributed.net
> http://lists.distributed.net/mailman/listinfo/hardware

Open OS/2! (Members: 13; Rank: 1192; Blocks: 91547; Working: 336 days)

More information about the Hardware mailing list