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. :) -- Ben Rosenberg ---===---===---===--- mailto:ben@whack.org ----- If two men agree on everything, you can be sure that only one of them is doing the thinking.
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. 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. :-) -- +----------------------------------------------------------------------------+ + Bruce S. Marshall bmarsh@bmarsh.com Bellaire, MI 08/30/03 13:36 + +----------------------------------------------------------------------------+ "Common sense is instinct. Enough of it is Genius." - George Bernard Shaw
-----Original Message----- From: Bruce Marshall <bmarsh@bmarsh.com> To: SLE <suse-linux-e@suse.com> 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).
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
On Saturday 30 August 2003 14:11 pm, Ken Schneider wrote:
-----Original Message----- From: Bruce Marshall <bmarsh@bmarsh.com> To: SLE <suse-linux-e@suse.com> 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 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. 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. CPU time is hardly ever the bottelneck in computers these days, except on the most compute intensive task. 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. 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. 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. In all but the most compute intensive tasks I'll take two half speed CPUs over a single full-speed one any day. 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. -- _____________________________________ John Andersen
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
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Alle 19:44, sabato 30 agosto 2003, Bruce Marshall ha scritto:
On Friday 29 August 2003 23:57 pm, Ben Rosenberg wrote:
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. :-)
http://www.arstechnica.com/paedia/h/hyperthreading/hyperthreading-1.html I do not know if the link has already been posted as I have not being following this thread closely, but that link explains the argument quite well. Hyperthreading should not hurt at all, as the cpu works in the usual way. The only difference is that HT allows the idle parts of the CPU to be used by another thread, so it should be faster than a non-HT machine, but slower than a SMP one. Praise -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.2-rc1-SuSE (GNU/Linux) iD8DBQE/UO296v3ZTabyE8kRAmgTAKDNGpwhkx9In6aJfJejSBGxTewaDgCgmh1Z xg1hvzaawUg9aEa91jIAnXk= =8YfP -----END PGP SIGNATURE-----
On Friday, August 29, 2003, at 10:57 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. :)
I don't have too much experience in dual/ht processors, but all the information I've read leads me to the conclusion that duals are slower for common tasks. Unless your doing a lot of Photoshop type work, video editing, or audio editing then you really have nothing to gain. 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. http://www17.tomshardware.com/cpu/20021114/ I recommend the average user like me :-) just get a faster single and not pay for an additional processor that will rarely get used. I have a dual pIII board that I experimented with a while. Used 2kpro, 2kserver, SuSE 7.3 and mandrake. Couldn't tell the difference with any of them so I just made 2 boxes and sold one to pay for the other. I would think you take a performance hit using ht. Tom's test seem to show that also. Of course I could be wrong. Linux might take better advantage of the architecture with a custom kernel. I would be interested in seeing more info. For what it's worth. will Ps. I recently did a lot of research on mac duals for a friend. I can dig those up if you want some RISC processor info also. Pretty much the same as P4. No substantial gains for average user, only mathematically intensive apps and a/v rendering. Still seems performance is more related to bus speeds, vid processors, and ram than proc speeds and how many you have. Quality and not quantity still seems to rule the day. Pss. My apologies for the ramble. The more I proof read this thing, the more involved I become in it. I'm also a bit embarrassed replying to one of the list guru's technical questions.
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. They don't start breaking even till you get 4 CPUs. Linux will get 2X the performance out of dual processors with almost no performance penalty for the SMP task management. 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. -- _____________________________________ John Andersen
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.
There are two contexts you want to analyze this in: scientific applications and non-scientific apps. SA are usually compute intensive algorithms that take a task and divide it in smaller parts -- the term SIMD, tho it usually refers to computer architectures, could also be used here (Single Instruction stream, Multiple Data streams). So, in a common parallel SA, all pieces are executing the same instructions over and over again, but over diff parts of the data. This is by far the most common kind of high demand sci app you'll see: weather models, finite element analysis, protein folding, pattern discovery, etc. There are other types of algorithms (I described the divide-and-conquer type), but they are far less common. NonSA is a context where you have lots of different apps doing different things on different data (most common) or one app doing different things over the same data. Now, to answer your question: if you'll run on a SA context, forget about HT and go get a real SMP machine. HT will slow you down. Reason: HT is just a mechanism that facilitates parallel access to the computing units inside the processor (integer units, floating point units, etc.). The clock speed is not affected (theoretically). So, if all your threads are trying to use THE SAME units, there'll be contention, and delay while a thread wants for a unit to clear. These delays involve overheads, that will make the app slower than in a single thread mode. If you'll run on the NonSA context, then HT will perhaps help you (true SMP will always be better tho). Reason: mix of apps (integer based, floating point based, etc.) may be using different processor units, effectively allowing HT to seem like two different processors. It will of course depend on the mis of apps. My gut feeling is that it would help. I hope that helps. Adalberto On Friday 29 August 2003 23:57, 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. :)
-- Ben Rosenberg ---===---===---===--- mailto:ben@whack.org ----- If two men agree on everything, you can be sure that only one of them is doing the thinking.
On Sun, 2003-08-31 at 15:45, Adalberto Castelo wrote:
There are two contexts you want to analyze this in: scientific applications and non-scientific apps. SA are usually compute intensive algorithms that take a task and divide it in smaller parts -- the term SIMD, tho it usually refers to computer architectures, could also be used here (Single Instruction stream, Multiple Data streams). So, in a common parallel SA, all pieces are executing the same instructions over and over again, but over diff parts of the data. This is by far the most common kind of high demand sci app you'll see: weather models, finite element analysis, protein folding, pattern discovery, etc. There are other types of algorithms (I described the divide-and-conquer type), but they are far less common.
NonSA is a context where you have lots of different apps doing different things on different data (most common) or one app doing different things over the same data.
Now, to answer your question: if you'll run on a SA context, forget about HT and go get a real SMP machine. HT will slow you down. Reason: HT is just a mechanism that facilitates parallel access to the computing units inside the processor (integer units, floating point units, etc.). The clock speed is not affected (theoretically). So, if all your threads are trying to use THE SAME units, there'll be contention, and delay while a thread wants for a unit to clear. These delays involve overheads, that will make the app slower than in a single thread mode.
If you'll run on the NonSA context, then HT will perhaps help you (true SMP will always be better tho). Reason: mix of apps (integer based, floating point based, etc.) may be using different processor units, effectively allowing HT to seem like two different processors. It will of course depend on the mis of apps. My gut feeling is that it would help.
I hope that helps.
Adalberto
Well Adalberto, that was definitively a pretty good explanation! 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? 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?
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
On Sunday 31 August 2003 16:52, Filipe Joel Almeida wrote:
Well Adalberto, that was definitively a pretty good explanation!
Thanks :)
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?
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?
I think the great advantage is the increase in memory address space. Right now, linux running in x86 has a limit of 2GB (31 bits) of memory per process (no matter how much system memory you have). With substantial kernel meddling (don't have the refs at hand) apparently you can get to 3.5 GB per process. It's very easy to hit that barrier for some applications (big databases probably being one case). Speed wise, I don't see much benefit in the case at hand. I confess I'm not familiar with the opteron's architecture, but I assume it's little more than a wider athlon (as opposed to the itanium series, which is actually a very very different beast than the pentium - but even there the speed gains are hard to measure, since the performance depends a lot on compiler intelligence) which would explain why it can have a 32 bit 'mode' (according to a recent article in The Register, AMD is even going to start shipping 32 bit locked opterons under the Athlon XP label) where it is binary compatible with x86. So, no substantial speed gains should be expected there, besides the ones derived from Moore's law.
* Filipe Joel Almeida <fijo.lists@netcabo.pt> (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.
* Gerhard den Hollander (gerhard@fugro-jason.com) [030901 01:04]:
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.
I'd have to agree with Gerhard on the mail server answer. We just put 5 v210 1U Sun's (dual 1.4ghz UltraIII's) as new mail servers into our main cluster last week. The first of them had 10k SCSI drives because that's what it came with and this was a test box at first. It seemed to choke on quite a bit less work then a machine (Sunblade 2k) that had cpu's 500mhz slower..the difference between the two was the machine with the slower cpu's had 15k SCSI drives in it. I rebuilt the v210 with new 15k drives and it blasted. It's because most mail servers need io more then they need cpu power..tons of read/writes. -- Ben Rosenberg ---===---===---===--- mailto:ben@whack.org ----- If two men agree on everything, you can be sure that only one of them is doing the thinking.
participants (11)
-
Adalberto Castelo -
Ben Rosenberg -
Bruce Marshall -
Dylan -
Filipe Joel Almeida -
Gerhard den Hollander -
John Andersen -
Ken Schneider -
Patrick Hooker -
Praise -
will