What are the stats on hyperthreading? I may need a Mobo.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 I'm trying to decide if buying a new mother board is worth the hassle. I learned earlier today that the Intel motherboard that my Intel CPU came on is one rev off from supporting hyperthreading. All I need is support for 2 512MB dimms, IDE, AGP, USB, PCI, and, of course, hyperthreading. I don't want on-board sound or ethernet. It just gets in the way. How much difference with HT make? Anybody have any benchmark info? I know, I just said yesterday on the programming list that I don't trust benchmarks. But they might at least give a hint. I do sometimes use stuff like Mathematica, Java3D, POVRay, etc., which are likely to be affected by hyperthreading. I also suspect my compiles will go faster. I'm just wondering how much. Everytime I pull the chip of the motherboard, or do anything else with the hardware, I run the risk of destroying it. I've been working with electronics since 1974. I know what I'm doing, but that doesn't mean I can't break something. Does anybody know if, and how outer CPU bus speed is reported by hwinfo? What's it called, if it is in the hwinfo? I don't see it there. Is there another method of probing the chip to get different info? STH -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.2 (GNU/Linux) iD8DBQFAR6HywX61+IL0QsMRArPnAKCPpQuCjIGwCHIcEYfTc3+JC9QmxgCbB+qh Bea4xzppG6pD39eFQFTJZ3g= =NXVr -----END PGP SIGNATURE-----
On Thursday 04 March 2004 04:38 pm, Steven T. Hatton wrote:
I'm trying to decide if buying a new mother board is worth the hassle. I learned earlier today that the Intel motherboard that my Intel CPU came on is one rev off from supporting hyperthreading. All I need is support for 2 512MB dimms, IDE, AGP, USB, PCI, and, of course, hyperthreading. I don't want on-board sound or ethernet. It just gets in the way.
How much difference with HT make? Anybody have any benchmark info? I know, I just said yesterday on the programming list that I don't trust benchmarks. But they might at least give a hint. I do sometimes use stuff like Mathematica, Java3D, POVRay, etc., which are likely to be affected by hyperthreading. I also suspect my compiles will go faster. I'm just wondering how much. Everytime I pull the chip of the motherboard, or do anything else with the hardware, I run the risk of destroying it. I've been working with electronics since 1974. I know what I'm doing, but that doesn't mean I can't break something.
Does anybody know if, and how outer CPU bus speed is reported by hwinfo? What's it called, if it is in the hwinfo? I don't see it there. Is there another method of probing the chip to get different info?
STH
Steven, If you are wanting better performance, actually doing two things at once, then go for a dual cpu motherboard. If you are just wanting speedups, then go for the Opteron/Athlon64 cpu. Hyperthreading was just another gimmick Intel tried to use for an already bad cpu, the Pentium IV. Lee -- --- KMail v1.6 --- SuSE Linux Pro v9.0 --- Registered Linux User #225206 On any other day, that might seem strange...
fredag 05 mars 2004 03:36 skrev BandiPat:
Steven, If you are wanting better performance, actually doing two things at once, then go for a dual cpu motherboard. If you are just wanting speedups, then go for the Opteron/Athlon64 cpu. Hyperthreading was just another gimmick Intel tried to use for an already bad cpu, the Pentium IV.
Which brings me to the question, of how does Athlon64 perform. Is it worth upgrading to it, there are numerous different reports out there.
On Thursday 04 March 2004 10:22 pm, Örn Hansen wrote:
fredag 05 mars 2004 03:36 skrev BandiPat:
Steven, If you are wanting better performance, actually doing two things at once, then go for a dual cpu motherboard. If you are just wanting speedups, then go for the Opteron/Athlon64 cpu. Hyperthreading was just another gimmick Intel tried to use for an already bad cpu, the Pentium IV.
Which brings me to the question, of how does Athlon64 perform. Is it worth upgrading to it, there are numerous different reports out there. =============
The Athlon64 is the lower cost version of the Opteron cpu. I don't remember what it is missing that the Opteron has, maybe someone that has done a bit more research can volunteer that information. I have not experienced the chip myself, but have spoken to a couple people that have and they seemed pleased. I believe there are a couple of folks on this list that have it. Lee -- --- KMail v1.6 --- SuSE Linux Pro v9.0 --- Registered Linux User #225206 On any other day, that might seem strange...
I did a dumb one, from root issued "crontab -r" :( Had too many ssh windows up and got confused. This was a recent install of SLES 8 for AMD64. I had made not crontab entries for root myself. I have configured Apache, email and so on. Did I lose anything? If so how can I recover? Here's what I think is relevant listing: kingfish:/etc # crontab -l no crontab for root kingfish:/etc # ls cron* crontab crontab.old cron.d: . .. evlogmgr seccheck sysstat cron.daily: . logrotate suse.de-clean-tmp .. suse.de-backup-rc.config suse.de-clean-vi clean_catman suse.de-backup-rpmdb suse.de-cron-local do_mandb suse.de-check-battery cron.hourly: . .. cron.monthly: . .. cron.weekly: . .. kingfish:/etc # Thank you - Richard
On Thu, 4 Mar 2004 21:37:43 -0700
"Richard Mixon (qwest)"
Did I lose anything? If so how can I recover?
Browse the cron rpm with mc and copy /etc/crontab to your system. Charles -- "A word to the wise: a credentials dicksize war is usually a bad idea on the net." (David Parsons in c.o.l.development.system, about coding in C.)
On Thu, 4 Mar 2004 21:37:43 -0700
"Richard Mixon (qwest)"
kingfish:/etc # crontab -l no crontab for root
This is fine since you don't have a crontab for root to begin with.
kingfish:/etc # ls cron* crontab crontab.old ...
Everything is fine, You didn't loose anything. Charles -- linux: because a PC is a terrible thing to waste (ksh@cis.ufl.edu put this on Tshirts in '93)
The 2004-03-04 at 21:37 -0700, Richard Mixon (qwest) wrote:
This was a recent install of SLES 8 for AMD64. I had made not crontab entries for root myself. I have configured Apache, email and so on.
Did I lose anything? If so how can I recover?
I don't know your version of suse, but customarily suse uses /etc/crontab, not root's crontab. You should have lost nothing. -- Cheers, Carlos Robinson
BandiPat
The Athlon64 is the lower cost version of the Opteron cpu. I don't remember what it is missing that the Opteron has, maybe someone that has done a bit more research can volunteer that information.
It's missing the second RAM channel and most of all it's missing two Hypertransport channels so it can't be used to build multiprocessor systems, not even dual ones. Philipp
Philipp Thomas wrote:
BandiPat
[Thu, 4 Mar 2004 23:07:54 -0500]: The Athlon64 is the lower cost version of the Opteron cpu. I don't remember what it is missing that the Opteron has, maybe someone that has done a bit more research can volunteer that information.
It's missing the second RAM channel and most of all it's missing two Hypertransport channels so it can't be used to build multiprocessor systems, not even dual ones.
Philipp
I'm sure I've seen dual Athlon 64 mobo's on sale and a friend and I have discussed watching the price drops on them, it could be they didn't make a distinction between Athlon64 and Opteron, caveat emptor perhaps? -- i.e you could buy one of those boards and 2 Athlon 64 CPU's and discover you have a spare CPU you can't use. My earlier mail got bounced ..... I'm expecting delivery of an Acer Aspire 1501LCe laptop tomorrow, Athlon64 3000+, 512M DDR333, 10/100/1000 Nic, CD RW/DVD etc. and I shall install gentoo 64-bit on the 40G drive and SuSE 64-bit on the spare 20G for comparison. All the comparisons I've seen so far have been between 32-bit operating systems on P4 vs Athlon64, but I'm interested only in 64-bit performance. Regards Sid. -- Sid Boyce .... Hamradio G3VBV and keen Flyer Linux Only Shop.
I'm sure I've seen dual Athlon 64 mobo's on sale and a friend and I have discussed watching the price drops on them, it could be they didn't make a distinction between Athlon64 and Opteron, caveat emptor perhaps? -- i.e you could buy one of those boards and 2 Athlon 64 CPU's and discover you have a spare CPU you can't use.
nope. the athlon 64 uses a 764-pin connection to the motherboard while the athlon 64 fx and opteron use a 940-pin connection. pretty soon the athlon 64 fx will switch to a 939-pin so it won't require use of registered ram. those extra pins provide the second memory channel and extra hypertransport channels that philipp was talking about. -- trey
Trey Gruel wrote:
I'm sure I've seen dual Athlon 64 mobo's on sale and a friend and I have discussed watching the price drops on them, it could be they didn't make a distinction between Athlon64 and Opteron, caveat emptor perhaps? -- i.e you could buy one of those boards and 2 Athlon 64 CPU's and discover you have a spare CPU you can't use.
nope. the athlon 64 uses a 764-pin connection to the motherboard while the athlon 64 fx and opteron use a 940-pin connection. pretty soon the athlon 64 fx will switch to a 939-pin so it won't require use of registered ram. those extra pins provide the second memory channel and extra hypertransport channels that philipp was talking about.
And to think that the first microprocessor (Intel 4004) used a 16 pin package!
Örn Hansen wrote:
fredag 05 mars 2004 03:36 skrev BandiPat:
Steven, If you are wanting better performance, actually doing two things at once, then go for a dual cpu motherboard. If you are just wanting speedups, then go for the Opteron/Athlon64 cpu. Hyperthreading was just another gimmick Intel tried to use for an already bad cpu, the Pentium IV.
Which brings me to the question, of how does Athlon64 perform. Is it worth upgrading to it, there are numerous different reports out there.
Most of the reviews I've seen of the Athlon64 have been comparing it on 32-bit operating systems, e.g against P4 which tells me nothing about 64-bit, however, I shall let the list know my experience when I have a 64-bit operating system installed on the laptop I expect to land on my doorstep on Monday. I'm preparing to install gentoo 64-bit on it at first and later SuSE 64-bit on another hard drive. Regards Sid. -- Sid Boyce .... Hamradio G3VBV and keen Flyer Linux Only Shop.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Thursday 04 March 2004 09:36 pm, BandiPat wrote:
On Thursday 04 March 2004 04:38 pm, Steven T. Hatton wrote: ***************
Steven, If you are wanting better performance, actually doing two things at once, then go for a dual cpu motherboard. If you are just wanting speedups, then go for the Opteron/Athlon64 cpu. Hyperthreading was just another gimmick Intel tried to use for an already bad cpu, the Pentium IV.
Lee
AMD does hyperthreading as well. The difference is the AMD chip emulates three processors, whereas the Intel emulates two. I already have a Pentium chip with hyperthreading. It costs far more than the motherboard which can be had at less that $100 US. I just want to know if it is worth the hassle, and _risk_, of swapping it out. I'm interested in demonstrable metrics. I would more than likely go with a 64bit chip before I went to two processors. The inter-processor protocols, the physical parameters introduced by the finite propagation speed of electromagnetic impulses, and the need for software to support (external) SMP make it unattractive.
-- --- KMail v1.6 --- SuSE Linux Pro v9.0 ---
STH -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.2 (GNU/Linux) iD8DBQFASCXwwX61+IL0QsMRAiUKAJ9BiMT0H9VndbBCq8g3pSRAr0SJwwCeMRoM aVZia2NDslSOnbFQoXe5cvo= =iAzV -----END PGP SIGNATURE-----
On Friday 05 March 2004 02:02 am, Steven T. Hatton wrote:
I would more than likely go with a 64bit chip before I went to two processors. The inter-processor protocols, the physical parameters introduced by the finite propagation speed of electromagnetic impulses, and the need for software to support (external) SMP make it unattractive.
I have a P4 2.8 running quite nicely here with HT. Can we gin up some simple test to see what it does? -- +----------------------------------------------------------------------------+ + Bruce S. Marshall bmarsh@bmarsh.com Bellaire, MI 03/05/04 07:25 + +----------------------------------------------------------------------------+ "When money talks, it usually says "bend over".
On Friday 05 March 2004 07:27 am, Bruce Marshall wrote:
On Friday 05 March 2004 02:02 am, Steven T. Hatton wrote:
I would more than likely go with a 64bit chip before I went to two processors. The inter-processor protocols, the physical parameters introduced by the finite propagation speed of electromagnetic impulses, and the need for software to support (external) SMP make it unattractive.
I have a P4 2.8 running quite nicely here with HT. Can we gin up some simple test to see what it does?
I think I have an answer for you and Lee is right. I wrote a simple looping program that would show the elapsed time for nnnn passes through the loop. Run by itself, a valid time would be 1.55 secs for one pass. Start up another copy of the same program, and the time for each became 2.9. Close enough to twice the time that it's obvious the HT is a sham. -- +----------------------------------------------------------------------------+ + Bruce S. Marshall bmarsh@bmarsh.com Bellaire, MI 03/05/04 07:55 + +----------------------------------------------------------------------------+ "Why are there Braille signs on drive-up ATM's?"
On Friday 05 March 2004 13.57, Bruce Marshall wrote:
On Friday 05 March 2004 07:27 am, Bruce Marshall wrote:
On Friday 05 March 2004 02:02 am, Steven T. Hatton wrote:
I would more than likely go with a 64bit chip before I went to two processors. The inter-processor protocols, the physical parameters introduced by the finite propagation speed of electromagnetic impulses, and the need for software to support (external) SMP make it unattractive.
I have a P4 2.8 running quite nicely here with HT. Can we gin up some simple test to see what it does?
I think I have an answer for you and Lee is right.
I wrote a simple looping program that would show the elapsed time for nnnn passes through the loop. Run by itself, a valid time would be 1.55 secs for one pass. Start up another copy of the same program, and the time for each became 2.9. Close enough to twice the time that it's obvious the HT is a sham.
I think you miss the point of HT. HT lets the CPU do other things while it would otherwise be idle, such as for external IO and data fetching. It's an idea to avoid pipeline stalls. In real world settings, most benchmarks show something like a 30% improvement in performance. What exactly does your loop do, other than just loop that is?
On Friday 05 March 2004 08:01 am, Anders Johansson wrote:
On Friday 05 March 2004 13.57, Bruce Marshall wrote:
On Friday 05 March 2004 07:27 am, Bruce Marshall wrote:
On Friday 05 March 2004 02:02 am, Steven T. Hatton wrote:
I would more than likely go with a 64bit chip before I went to two processors. The inter-processor protocols, the physical parameters introduced by the finite propagation speed of electromagnetic impulses, and the need for software to support (external) SMP make it unattractive.
I have a P4 2.8 running quite nicely here with HT. Can we gin up some simple test to see what it does?
I think I have an answer for you and Lee is right.
I wrote a simple looping program that would show the elapsed time for nnnn passes through the loop. Run by itself, a valid time would be 1.55 secs for one pass. Start up another copy of the same program, and the time for each became 2.9. Close enough to twice the time that it's obvious the HT is a sham.
I think you miss the point of HT. HT lets the CPU do other things while it would otherwise be idle, such as for external IO and data fetching. It's an idea to avoid pipeline stalls. In real world settings, most benchmarks show something like a 30% improvement in performance.
What exactly does your loop do, other than just loop that is?
No, I didn't miss that point..... It's true that in some instances two cpu's (even of 1/2 the power of one) can be a better deal, but not very often. However, comparing a 2.8 P4 without HT to a dual-cpu 1.4 setup (a fair comparison), there is one advantage: when there's only one task to perform, it gets a cpu of 2.8ghz instead of just 1.4ghz. For the money, the HT doesn't really have any disadvantages that I can see, it just doesn't have all the advantages of two P4 2.8's. (which would cost a lot more money) -- +----------------------------------------------------------------------------+ + Bruce S. Marshall bmarsh@bmarsh.com Bellaire, MI 03/05/04 08:08 + +----------------------------------------------------------------------------+ "The Information Highway: 500 channels and not a thing to watch."
* Bruce Marshall
I would more than likely go with a 64bit chip before I went to two processors. The inter-processor protocols, the physical parameters introduced by the finite propagation speed of electromagnetic impulses, and the need for software to support (external) SMP make it unattractive.
I have a P4 2.8 running quite nicely here with HT. Can we gin up some simple test to see what it does?
I think I have an answer for you and Lee is right.
I wrote a simple looping program that would show the elapsed time for nnnn passes through the loop. Run by itself, a valid time would be 1.55 secs for one pass. Start up another copy of the same program, and the time for each became 2.9. Close enough to twice the time that it's obvious the HT is a sham.
``obvious that HT is a sham'' and that is based on one simple looping program. Please read up onw hat HT is, what it does and under which circumstances it can be benificial. Or give it a try on your machien, and check the performance of a bunch of common programs with and witout HT. Your mileage may indeed vary, but I've found that enabling HT indeed gives a slightly snappier experience, and certain programs (esp once that do a lot of floating point operations and that are multithread aware) will clearly benefit from having it on. But given that HT these dayts comes for free (more or less) it's a cheap test to turn it on . Kind regards, -- Gerhard den Hollander Phone :+31-10.280.1515 ICT manager Direct:+31-10.280.1539 Fugro-Jason 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 files transmitted with it are confidential and intended solely for the use of the addressee. This e-mail shall not be deemed binding unless confirmed in writing. If you have received it by mistake, please let us know by e-mail reply and delete it from your system; you may not copy this message or disclose its contents to anyone. Please note that any views or opinions presented in this e-mail are solely those of the author and do not necessarily represent those of the company. E-mail transmission cannot be guaranteed to be secure or error-free. The sender therefore does not accept liability for any errors or omissions in the contents of this message, which arise as a result of e-mail transmission.
On Friday 05 March 2004 08:06 am, Gerhard den Hollander wrote:
* Bruce Marshall
(Fri, Mar 05, 2004 at 07:57:37AM -0500) I would more than likely go with a 64bit chip before I went to two processors. The inter-processor protocols, the physical parameters introduced by the finite propagation speed of electromagnetic impulses, and the need for software to support (external) SMP make it unattractive.
I have a P4 2.8 running quite nicely here with HT. Can we gin up some simple test to see what it does?
I think I have an answer for you and Lee is right.
I wrote a simple looping program that would show the elapsed time for nnnn passes through the loop. Run by itself, a valid time would be 1.55 secs for one pass. Start up another copy of the same program, and the time for each became 2.9. Close enough to twice the time that it's obvious the HT is a sham.
``obvious that HT is a sham'' and that is based on one simple looping program.
It's good enough for me...
Please read up onw hat HT is, what it does and under which circumstances it can be benificial.
I think I know what it is doing...
Or give it a try on your machien, and check the performance of a bunch of common programs with and witout HT.
It *IS* on my machine.
Your mileage may indeed vary, but I've found that enabling HT indeed gives a slightly snappier experience, and certain programs (esp once that do a lot of floating point operations and that are multithread aware) will clearly benefit from having it on.
But given that HT these dayts comes for free (more or less) it's a cheap test to turn it on .
I can agree with all of your points.... see my later message. I've been dealing with multi-engine machines (mostly large IBM mainframes) for the past 43 years so I think I know a little about schedulers (I've written a few) and cpu usage and what works and what doesn't. HT can have some advantages.... just not as much as a dual-cpu setup. And since I am subscribed to the list, please don't copy me on your posts. -- +----------------------------------------------------------------------------+ + Bruce S. Marshall bmarsh@bmarsh.com Bellaire, MI 03/05/04 08:15 + +----------------------------------------------------------------------------+ "Money will say more in one moment than the most eloquent lover can in years."
"Steven T. Hatton"
AMD does hyperthreading as well. The difference is the AMD chip emulates three processors, whereas the Intel emulates two.
It does definitely *not* do any Hyperthreading, believe me. Maybe you're confusing it with Hypertransport, but the latter is a high speed serial bus to connect processor and chips. Philipp
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Saturday 06 March 2004 10:07 pm, Philipp Thomas wrote:
"Steven T. Hatton"
[Fri, 5 Mar 2004 02:02]: AMD does hyperthreading as well. The difference is the AMD chip emulates three processors, whereas the Intel emulates two.
It does definitely *not* do any Hyperthreading, believe me. Maybe you're confusing it with Hypertransport, but the latter is a high speed serial bus to connect processor and chips.
Philipp
I don't claim to be an expert, but the way it was explained to me is the Intel chip emulates two CPUs by maintaining two sets of context registers and switching between them. The AMD chip does the same thing with three sets of context registers. By maintaining multiple thread contexts the chip can switch between threads without context switching delays. STH -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.2 (GNU/Linux) iD8DBQFASrDRwX61+IL0QsMRAi8KAJoDp9hQoh/23jbEZPyv3MKNJCNWdgCgjghn dK03NAjt44XN/PztsE0g+nY= =OHNJ -----END PGP SIGNATURE-----
"Steven T. Hatton"
The AMD chip does the same thing with three sets of context registers.
Which AMD chip are we talking out? Non of the current AMD64 chips does have that feature. If they'd do, you'd see much more marketing blurb about it. Philipp
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On Sunday 07 March 2004 06:59 am, Philipp Thomas wrote:
"Steven T. Hatton"
[7 Mar 2004 00:19:06]: The AMD chip does the same thing with three sets of context registers.
Which AMD chip are we talking out? Non of the current AMD64 chips does have that feature. If they'd do, you'd see much more marketing blurb about it.
Philipp
I'll have to agree with you. And I don't recall the line of chips in question. I tried to find more information on the topic, and didn't see any indication that AMD is selling a simultaneous multi-threading chip. STH -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.2 (GNU/Linux) iD8DBQFASxE5wX61+IL0QsMRAsmQAKDLpqn8H6A7BxHA+49Fw9gpgYOSdACeJ6e1 Uf7D0qqWhad/TNl3oBKpaV0= =975u -----END PGP SIGNATURE-----
participants (13)
-
Anders Johansson
-
BandiPat
-
Bruce Marshall
-
Carlos E. R.
-
Charles Philip Chan
-
Gerhard den Hollander
-
James Knott
-
Philipp Thomas
-
Richard Mixon (qwest)
-
Sid Boyce
-
Steven T. Hatton
-
Trey Gruel
-
Örn Hansen