On Saturday 30 August 2003 14:11 pm, Ken Schneider wrote:

-----Original Message----- From: Bruce Marshall To: SLE Date: Sat, 30 Aug 2003 13:44:42 -0400 Subject: Re: [SLE] For or against ..Hyperthreading.

...

On Friday 29 August 2003 23:57 pm, Ben Rosenberg wrote:

...

I have partly a question and partly a survey. I just got a P4 2.4ghz (c) proc, mb w/ 512M DDR400 dropped in my lap at work so I could upgrade my workstation in the office. It does have hyperthreading. I've never user HT hardware and I've heard conflicting reports about it. I have it turned off right now and I haven't installed the SMP kernel since I'm not sure I'm going to bother. I guess my question is...for a workstation doing webbrowsing, documents in OO, Gaim, playing mp3's and s**tloads of xterms. Does HT give or take away

from

...

the enviroment as far as speed and all that. All opinions are

welcome.

...

:)

No one's taken this on so I'll give it a crack:

I'm going to assume that HT works in a similar manner to having two cpu in SMP formation... (although something tells me that it's not quite that efficient.)

1) Having one CPU that is twice (or close to twice) as fast as two cpu's is always the best way to go.

2) The reason for (1) is that most of your computing time is going to be spent on one task, and in an SMP situation you can only apply 1 cpu to that task at a time. Therefore, a 2Ghz cpu in that situation will do MUCH better than two cpu's of 1Ghz.

Wrong answer! The opposite of this is true. When was the last time you saw your computer running only one task, even DOS did more then one thing at a time. If you compile a kernel or any program and then go and play a game while it is running, 2 cpu's will out perform a single cpu system ALWAYS. I have an SMP system at home with 2 P3-750 cpus that outperforms the single cpu system at work (P41.6).

B-z-z-z-t-tttt!! Thanks for playing.... The answer to your question is "yes, there almost always will be more than one thread running, but are they doing anything? And even if there is more than one thread running, my guess would be that they aren't using much cpu time... ergo, they can't add much to 'speeding up' your system." Example: You want to crunch some numbers with SETI. And also, in the background, your system is going to be doing an updatedb or some other typical maintenance task. Said tasks don't take much computer power. (but use a lot of I/O) Which is going to get more work done for the SETI program? 1 2-Ghz cpu or 2 1-Ghz cpu? Answer: a flat-out no-brainer... the 2-Ghz cpu. And the same applies to your scenario above!! With a 2-Ghz cpu you *ALWAYS* have 2-Ghz of power to apply to a single task (which is usually the case when power is needed) With two cpu's you can *NEVER* apply more than 1-Ghz to a single task... And with the 2-Ghz cpu.. you have a good chance of apply that 2-Ghz over as many tasks as possible that need cpu power. I'll take a single cpu any day over multiple cpu's that DON'T ADD UP TO MORE MIPS than the single cpu. And that was my point.

...

3) Given that you have (two) cpus so-to-speak, there is not much harm in using HT. There might be a slight addition of system overhead in scheduling two cpu's but that should be gained by the 2nd cpu being able to handle other tasks such as spooling and printing, and all the other system type threads that occur while your doing your own tasks of word processing, etc.

4) On the other hand, since I don't know much about HT, if it takes a 2Ghz cpu and turns it into (mostly) two 1Ghz cpu's by sharing all of the resources in the chip, then it would be best not to use HT.

So in a nutshell, it shouldn't hurt, might help a little, you probably won't notice any difference, and besides, it's a neat conversation piece. :-)

Ken

-- +----------------------------------------------------------------------------+ + Bruce S. Marshall bmarsh@bmarsh.com Bellaire, MI 08/30/03 16:23 + +----------------------------------------------------------------------------+ "It's a small world, but I wouldn't want to paint it." - Steven Wright

On Sunday 31 August 2003 4:36 am, John Andersen wrote:

On Saturday 30 August 2003 12:32, Bruce Marshall wrote:

...

I'll take a single cpu any day over multiple cpu's that DON'T ADD UP TO MORE MIPS than the single cpu. And that was my point.

A very well reasond answer Bruce.

Unfortunatly it just proves you have never touched a Dual CPU machine running and SMP kernel in your life.

Wrong.... Both SMP PC's and IBM mainframes with up to 5 engines....

Because in spite of your reasoning, the facts are the opposite.

ANYONE who has used dualies would tell you two 500hmz CPUs easliy outperform a single 1Ghz cpu on the normal mix of applications you run on a typical linux machine, and they do it for LESS money. Usually enough less to afford the dual motherboard.

And that's got to be a crock.... So, for example, I'll get more SETI work done on the above SMP machine? I'd like to see you prove it. Define 'normal mix' for a single user machine.

CPU time is hardly ever the bottelneck in computers these days, except on the most compute intensive task.

You're right on that...

You assume in your SETI example that the CPU doing updatedb does nothing else except the updatedb. That's not true, it can run SETI while it is waiting on diskIO.

Uhhh... that was my point... With an SMP machine, you're going to have 1 cpu running SETI and 1 cpu doing the updatedb... Not always the same cpu doing either, but in general, 1 for each. Now if you want to run 2 SETI programs, then you would be applying 2 cpus to SETI, and both would keep the updatedb running with what little cpu power it needs. But where are you gaining in the SETI area? 2 cpu' at 1Ghz are not going to be any better than 1 cpu at 2Ghz. If you think so, please explain where the cycles are coming from.

Instruction fetch takes longer than instruction execution by several orders of magnitude. With two CPUs fetching data and instructions tasks are seldom ever backed up, and the machine remains responsive even under high load.

Wow, that's a stretch...

A load of 50 would burry most single processor machines yet I've seen that often on a busy dual processor machine and it just runs right thru it.

Pure perception if you think the above is the case.

In all but the most compute intensive tasks I'll take two half speed CPUs over a single full-speed one any day.

It's your choice.

I ran RC5 crunchers in several of my machines for several years. The dualies (running two curnchers) always outperformed Single CPU machines twice their speed even though they theoretically should not have had an advantage on such a compute intensive task.

If you're baseing your theory on Win2k or NT platforms you'd be right, Windows can only get 1.4 times a single processor performance with dual processors. Linux gets 2x.

And when you want to do an addtionial task theres always a cpu cycle available - never any sluggish preformance.

Don't theorize. Go get a dualie and bench it.

I've had one... not worth the extra expense......

-- _____________________________________ John Andersen

-- +----------------------------------------------------------------------------+ + Bruce S. Marshall bmarsh@bmarsh.com Bellaire, MI 08/31/03 11:47 + +----------------------------------------------------------------------------+ "He knows nothing: he thinks he knows everything- that clearly points to a political career." --George Bernard Shaw

On Sunday, August 31, 2003, at 03:41 AM, John Andersen wrote:

On Saturday 30 August 2003 10:12, will wrote:

...

Here's a link to a lengthy benchmark using win xp. Takes a while to get to the good parts. A lot of explanation of the hw.

Hold on there, Will, we are talking LINUX here. Windows platforms top out at 1.4 times a single processor thru-put when running 2 cpus.

I understand the difference. I also noted in my reply that I was not aware of the performance increases with an optimized linux kernel and would be interested to learn more.

They don't start breaking even till you get 4 CPUs.

Even at maximum tweaking any win os still manages to leverage minimal gains from any system over other alternatives. Thus the longevity of the linux turnover in hardware deployment. Linux can do more with less. We all know this. http://pgmeter.sourceforge.net/pgmeter.pdf

Linux will get 2X the performance out of dual processors with almost no performance penalty for the SMP task management.

Even under the best tweaking and optimization I have not run across any processor that runs double. Apple has tried this for years while only being able to show a 1.5 or so increase in processing. Except in certain situations with extremely optimized programs. Which people like to use for advertising. "Gaussian blur runs x times faster on our machine with optimized x under photoshop." How many people use photo shop to check their email? or even know what a Gaussian blur is? It all just sounds good. http://www-106.ibm.com/developerworks/linux/library/l-kperf/ http://eshop.macsales.com/Reviews/Framework.cfm?page=/benchmarks/ 071803/main071803.html

There are some tasks, like Vmware, which can only utilize a single processor, but most other real world job mixes multi-thread all over the place.

The posted benchmarking was to mark the difference in the HT v. non HT processors. It also gave various comparison to other factors which I felt might help Ben or at least give him a direction for more research. His question was directly related to the average desktop user. Not which processor is better for displaying and calculating the mutation factors in the human genome when exposed to "x" type of catalyst under "y" conditions. The xp benchmarks just showed that ht usually results in a performance hit, though minor, and relatively few performance increased. I couldn't find anything that related to ht on linux, though, as you mentioned is basically using 2 lower powered procs to attempt to gain the same output as a faster proc while being cheaper in hardware deployment costs and HT is just attempting to gain this performance out of a single chips. The average email checking, web browsing, game playing user will almost never see an increase in performance from a dual proc or HT. That doesn't mean there aren't user that don't need them and can't take full advantage of them. Those same people use clusters and distributed processing also while making the big bucks. Thanks for the opportunity to clarify myself, will Ps. I don't mean to seem hostile. Just getting hungry.

This has been the most interesting thread I've followed in a long time. Y'all please keep it going, I'm learning a bunch. Patrick Hooker Adalberto Castelo wrote:

There are two contexts you want to analyze this in: .....

On Sunday 31 August 2003 21:52, Filipe Joel Almeida wrote: <SNIP>

What are the major benefits of a 64 bit processor over a 32 bit processor?

Let's say for mail or Database server... would there be significant benefits from having a Dual Opteron over a Dual Xeon, and stuff like that?

Can anyone please shed some light on this subject?

SFAIUI, assuming you are running a 64bit OS with apps compiled for 64bit, then there are several ways that the apparent speed may be improved: - with 64-bit internal registers, number crunching can operate directly on larger values, so maths-intensive processes can obviously benefit. The Opteron and Athlon-64 also have a different memory access archetecture so fetch and put operations are (supposed to be) more efficient. - 64-bit Address registers means more memory is directly addressable, meaning that larger files (or portions of them) can be kept in RAM at a time. This meand a database (for example) can potentially have a 'live' copy of an entire (set of) table(s) im memory, rather than paging sections to disk, and that swap may become effectively redundant. Also, larger memory blocks are directly accessible, meaning that block operations can be performed on larger blocks of memory. This would also speed up databases, and especially graphics and video apps. - 64-bit instruction registers opens up a whole magnitude of space for new CPU instructions (in principle if not yet in practice) to allow the processor to do a wider range of operations as primitive instructions (for example, fetch, shift-left, put could be a single opcode instead of three) or provide for sub-instructions or parameterised instructions. All this means less instructions would need to be fetched to perform the same ammount of work. The AMD chips are backwards compatible, so they will run existing OS's (albeit in 32-bit state), and 32 (or even 16) -bit apps on 64bit OS's natively. The Intel offering is NOT backwards compatible in that it requires a 64 OS, but can run 32-bit apps (under a 64-bit OS) in a mode which emulates a 32-bit Intel chip. I'm sure there's other benefits as well... Dylan -- Sweet moderation Heart of this nation Desert us not We are between the wars - Billy Bragg

* Filipe Joel Almeida (Sun, Aug 31, 2003 at 09:52:22PM +0100)

But personally I'm also curious about something else. I have asked this lots of times, one of them in this list, but noone has been able to give me a good answer.

What are the major benefits of a 64 bit processor over a 32 bit processor?

The main reason (for us at least) is that you can address more then 2,3,3.5G RAM in a single application.

Let's say for mail or Database server... would there be significant benefits from having a Dual Opteron over a Dual Xeon, and stuff like that?

Mail server, I don;'t think so. Dqatabase server, yup, if the database is so big it needs more then 3.5G to fit completely into memory. (there's much more to it though then just what I wrote above ) Kind regards, -- Gerhard den Hollander Phone :+31-10.280.1515 ICT manager Direct:+31-10.280.1539 Jason Geosystems BV Fax :+31-10.280.1511 gdenhollander@Fugro-Jason.com POBox 1573 visit us at http://www.Fugro-Jason.com 3000 BN Rotterdam JASON.......#1 in Reservoir Characterization The Netherlands This e-mail and any attachment is/are intended solely for the named addressee(s) and may contain information that is confidential and privileged.