Intel confirms hyperthreading is a goner

Moderator: News Team

Post Reply
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Intel confirms hyperthreading is a goner

Post by purrkur »

Most of you know that I have a certain dislike to Intel's Hyperthreading. Not the technology per se but Intel's implementation of it. It was never any good and it was never any more than just a marketing tool/gimmick. It seems like Intel is finally pulling the plug on the thing. They seem to be sayingthat dual/multi core technology will replace it.

Technology similar to hyperthreading is being used by Sun (for example) to run up to 4 threads per core in their latest 8 core CPU for a total of 32 simultaneous processing threads. This shows that one form of technology doesn't rule out the other. This makes it even more interesting that Intel decides to scrap the whole Hyperthreading stuff.
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
Billl
Posts: 211
Joined: Fri Jul 25, 2003 1:52 pm
Location: USA

Re: Intel confirms hyperthreading is a goner

Post by Billl »

purrkur wrote:Most of you know that I have a certain dislike to Intel's Hyperthreading. Not the technology per se but Intel's implementation of it. It was never any good and it was never any more than just a marketing tool/gimmick. It seems like Intel is finally pulling the plug on the thing. They seem to be sayingthat dual/multi core technology will replace it.

Technology similar to hyperthreading is being used by Sun (for example) to run up to 4 threads per core in their latest 8 core CPU for a total of 32 simultaneous processing threads. This shows that one form of technology doesn't rule out the other. This makes it even more interesting that Intel decides to scrap the whole Hyperthreading stuff.
I've never quite got your complaint on this thing? At worst it does nothing. When I was crunching SETI it most certainly made the systems faster. I've never seen any performance hit from using it. So what's the big deal. In the applications that it helps, it really helps. The ones it doesn't help, it doesn't hurt either. So what's your beef here?


Billl
Image
Jordan
Posts: 64
Joined: Sun Dec 11, 2005 1:15 pm
Location: North Vancouver, BC

Post by Jordan »

HT was cool. my friend has an HT pentium 4. It really doesn't impact anything that much in regualr activities. Bill is right saying it hekps what it helps and doesn't hurt what it doesn't help. It is still cool but multi-core-cpu systems are definately the way to go.


Oh BTW- Has anybody ever experienced quad-pentium PCs before? I'm talking original pentiums. I've never seen one but heard they existed. They sound outrageous. I wonder if a quad slot-1 system has ever been made... I've got 5 or six extra slot-1 celeron 333s just lying around...
The collector of the old and unwanted.
Billl
Posts: 211
Joined: Fri Jul 25, 2003 1:52 pm
Location: USA

Post by Billl »

Jordan wrote:HT was cool. my friend has an HT pentium 4. It really doesn't impact anything that much in regualr activities. Bill is right saying it hekps what it helps and doesn't hurt what it doesn't help. It is still cool but multi-core-cpu systems are definately the way to go.
I didn't mean to imply HT was preferable to 2 real cores. It was pretty much a freebee. Which is why I can’t understand why Purrkur is so anti-HT, unless it’s just an anti-Intel thing? Don’t get me wrong here. I like the fact that AMD is giving Intel such fits. Helps make processors lots cheaper for all of us. I’ve just used mostly Intel’s because in general they offer more head room for over clocking. And until the AMD 64 cores came out I think they were still superior. AMD’s newest chips really are a much better design then anything Intel has on the table now, with the exception maybe of the Pentium M’s. Of course everything after the Northwood has just been a disaster. I can’t figure out what they were thinking when they came up with their dual core strategy.


Billl
Image
davd_bob
Confused
Posts: 1043
Joined: Fri Feb 13, 2004 2:30 am
Location: Houston, TX

Post by davd_bob »

Jordan wrote:Oh BTW- Has anybody ever experienced quad-pentium PCs before? I'm talking original pentiums. I've never seen one but heard they existed. They sound outrageous. I wonder if a quad slot-1 system has ever been made... I've got 5 or six extra slot-1 celeron 333s just lying around...
When I first got interested in multi-cpu boards I really looked quite a bit using "multi" in the search. I never founed pentiums with more then 2 nor any dual-486 although they do exist. I never found anything with more then "dual" except p-PRO which can be found with up to 6-cpus.
It would require a version of "MS-Server" or Linux to take advantage of more then 2-cpus if you can find such a board.

FWIW there is a pretty good thread on this board about one of those 6x p-PRO boards. There is an adaptor "not mentioned in that thread" where socket370 PPGA chips could run on the socket8.

Also I once located an "add-on" card that plugged into...maybe an ISA slot...and that card had dual cellys, dual IDE, SDRAM slot, built in video, etc. Basicly it was a BP6 on a card.
There are *almost* no bad BP6s. There are mostly bad caps.

No BP6s remaining
Athlon 2800
Sempron 2000
ViaCPU laptop with Vista.(Works great after bumping ram to 2Gig)
P-III 850@100
Dave Rave
G'Day Mate!
Posts: 894
Joined: Thu Sep 26, 2002 4:28 am
Location: Sydney Australia
Contact:

Post by Dave Rave »

http://www.supermicro.com/newsroom/pres ... 030100.cfm
http://www.supermicro.com/products/moth ... /S2QR6.cfm

much droolage.

current Xeon with HT are usually dual physical cpu
which is quad HT cpu. I got two. ;-)
Jordan
Posts: 64
Joined: Sun Dec 11, 2005 1:15 pm
Location: North Vancouver, BC

Post by Jordan »

Also I once located an "add-on" card that plugged into...maybe an ISA slot...and that card had dual cellys, dual IDE, SDRAM slot, built in video, etc. Basicly it was a BP6 on a card.
:shock: GAH I WANT IT!!!!!!! That is both the weirdest and coolest thing i have ever heard of!

I found one through google: a pentium 4 on an ISA/PCI slot thing
Its so pointless for a regular person but amazing!
http://www.commell.com.tw/Product/SBC/FS-976.htm

And this one is a VP6 that fits into a CD rom space:
http://www.commell.com.tw/Product/SBC/LS-563.HTM

I have no idea how they actually function or anything, but those are the coolest things i have seen in a while. Do they do SMP or are they function as 2 linked PCs or what?
The collector of the old and unwanted.
Dave Rave
G'Day Mate!
Posts: 894
Joined: Thu Sep 26, 2002 4:28 am
Location: Sydney Australia
Contact:

Post by Dave Rave »

it is a total computer system on a card
what you are using is your existing case, cd-rom, hard drives and are doing a mobo/cpu upgrade without taking out the old mobo

you do need a new PSU.
this is aimed at older P1 type systems
==========
http://www-307.ibm.com/pc/support/site. ... MIGR-58434
http://cgi.ebay.com.au/IBM-xSeries-440- ... dZViewItem

read that spec sheet, and this is a live austion, it does exist.
4 xeons, or 8 MP type xeon cpus
I'm looking for a picture of the cpu area
bp6's 3 x dual @ 533
. . . . 1 x dual @ 466
. . . . 1 sngl @ 400
[( 2 x dual xeon 2.4ghz )]
[( 2 x dual xeon 2.66ghz )]
[( 1 x 2.4C ghz )]
[( 1 x 2.4B ghz )]
[( 1 x dual AMD 1800MP )]
[( 1 x P4 1600 )]
[( 1 x 500 ppga )]
3 x piii 866
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Re: Intel confirms hyperthreading is a goner

Post by purrkur »

Billl wrote:I've never quite got your complaint on this thing? At worst it does nothing. When I was crunching SETI it most certainly made the systems faster. I've never seen any performance hit from using it. So what's the big deal. In the applications that it helps, it really helps. The ones it doesn't help, it doesn't hurt either. So what's your beef here?
Thats OK, I can explain. As I have mentioned before elsewhere on this forums, the Intel HT can definitely be beneficial in a small range of software, especially if the computer is only being used for doing a single task where HT happens to work well. SETI is a good example of this.

My first run-in with HT was when a customer started complaining about random crashes after buying new hardware. We had a really hard time figuring out what the hell was going on but in the end I finally managed to narrow it down to their HT processor. We disabled it and the crashes went away. There wasn't anything wrong with our code (it works without a glitch on dual CPU machines) but HT made it crash. We were able to make some changes to our app so it wouldn't go down but then we discovered that with HT enabled our app took a performance hit. Our app is a calc-heavy financial application and it really brought out the worst in Intels HT technology.

After this I started doing some research into HT and comparable technologies. My conclusion is that Intel's HT is a very poor design compared to what others have done (IBM and Sun for example). I have also done a lot of testing where large servers do heavy duty work and I have found that once you bog such a machine down with several hard working apps then HT becomes a load rather than something that will slow down things instead of speeding them up. So that is my beef with Intel :)
Billl wrote:I didn't mean to imply HT was preferable to 2 real cores. It was pretty much a freebee. Which is why I can’t understand why Purrkur is so anti-HT, unless it’s just an anti-Intel thing? Don’t get me wrong here. I like the fact that AMD is giving Intel such fits. Helps make processors lots cheaper for all of us. I’ve just used mostly Intel’s because in general they offer more head room for over clocking. And until the AMD 64 cores came out I think they were still superior. AMD’s newest chips really are a much better design then anything Intel has on the table now, with the exception maybe of the Pentium M’s. Of course everything after the Northwood has just been a disaster. I can’t figure out what they were thinking when they came up with their dual core strategy.
I understand your point of view. What I have found is that HT can work really well in environments where you are running simple tasks but it becomes a problem when the computer/server is running many complicated apps at the same time. Let me explain my viewpoint of Intel and why I am not simply bashing them. They are a market leader but in the past few years they have not acted like it and they have done nothing to mantle the responsibility that it brings. They have been bringing out shitty products, shitty technologies and selling it to customers who buy their marketing crap based on their status as a market leader. I hate seeing that happen.

Don't get me wrong, I don't hate all Intel products, or the company itself. I think that the Pentium III was a great product for its time and the Pentium M (and I know I have said this before) is also a very good product as well. But Intel has been doing a lot of large misses. HT is one of them in my opinion. It almost feels like the engineering group didn't get to finish the product before Marketing pushed it out the door. After all, it is pretty hard for joe-average to really tell if he is gaining or loosing so they hype it up and sell it to the people and the average joe will believe he is king because he has got this 733t HT technology.

Another miss is the development of the P4 processor. Adding pipelines just in order to be able to ramp up the speed in order to give marketing the "GHz is king" selling point ultimately destroyed this processor. You may argue me on this but I won't budge. Take a look at Intel at the time of launching prescott. There was a lot of "Prescott will break the 4GHz barrier" yada-yada even from the top brass at Intel. It is now more than two years after Prescott was delivered to the market and still no 4GHz. In fact, Intel later on admitted that they were too quick with their announcements. Which brings me to question: Did Intel think that they could add pipelines and use smaller transistor technologies to ramp speed forever? It sure seems that way because they were talking about Prescott as a speed demon before the world found out that the only thing they managed to do was to increase the heat output.

Another design miss with the P4 is again related to the piplines construction. Making the pipeline longer means that you need more cache that the CPU can use to store temporary data as it passes through. Longer pipe means that you have to increase cache. This is a terrible design strategy: More cache means more transistors which translates into more heat. This is a fundamental flaw in the P4 which is a power hog. Notice that the Celerons based on P4 have reduced cache. These CPU's are pretty worthless when it comes to performance.

Third design miss at Intel: Their dual core strategy. Again, this seems like engineering was working on the issues when Marketing rushed in and said: "Give us what you have so we can claim that we have what AMD has got". The dual core design of the P4 is absolutely horrible from a technology standpoint. What we are seeing with the "Duo" processors is looking better though, but their P4 dual cores are just embarrasing. Intel has, again, looked bad when they had to announce that their Dual Core stuff was rushed to market.

So that is my view of Intel. They have been pressing out second rate technology to their channels and selling it for top dollar to customers who belive that buying Intel can't be wrong because they are the market leader, or that nobody got fired for ever buying Intel. That is my beef with them.

Oh, and on the HT stuff, just to strengthen my arguments, Microsoft's engineers have said that it is better to turn off HT on servers running Windows Server 2003 and/or SQL Server because performance will suffer. Also the fact that most hardware companies (like HP) deliver their servers with HT turned off is an indication of what they think about the technology. In short, HT doesn't cut it for heavy duty use.

Intel is taking a hit from AMD which seems to have a technology lead over Intel. I would respect Intel more if they could take that fact like a man, go back to their engineering departments and giving them the possibility of creating great products again. Not half-assed, half finished products that they expect their customers to pay top dollar for. It is because of this that I have lost a lot of respect for Intel.
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
Dave Rave
G'Day Mate!
Posts: 894
Joined: Thu Sep 26, 2002 4:28 am
Location: Sydney Australia
Contact:

Post by Dave Rave »

bah, who needs just 4 cpus
http://www-8.ibm.com/servers/eserver/au ... /more.html

not 4, that's 54 !!
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Post by purrkur »

Dave Rave wrote:bah, who needs just 4 cpus
http://www-8.ibm.com/servers/eserver/au ... /more.html

not 4, that's 54 !!
Hehehe! OK, that is not a PC though :)

The nicest computer/server I ever worked on was a Sun Enterprise 15000 machine with 72 processors. The nicest PC I will have the chance to work on, which will happen pretty soon, is a machine equipped with 4 Dual Core Opterons and, hold on, 128GB of internal memory!!!

The nicest PC with the fastest disk IO I worked on was a 4 CPU machine with an external disk cabinett that contained 28 disks running at 15000 RPM. They were striped too! We tested both Linux and Windows on that platform to see what OS could perform better with that disk subsystem and Linux practically killed Windows in pretty much all tests. This was at the IBM R&D center in Germany last year.
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
Billl
Posts: 211
Joined: Fri Jul 25, 2003 1:52 pm
Location: USA

Post by Billl »

Well thanks Purrkur, I was wondering what your point was. I guess it all depends on your perspective here. As an end user I see some benefit to HT. But I can see how it could cause problems in servers. Of course I still stick by the point that it’s free and if you don’t need it or it causes problems just shut it off. I can see how it probably drove you nuts figuring out what was causing your problems. But some of the blame has to go against the code as it was written. Anyway for most users it does give you a speed boost. Intel probably could have better warned server operators to the dangers of using HT with poorly written code. As to Intel’s new CPU’s were in total agreement, the Northwood was the last processor they made that actually preformed well.

Billl
Image
davd_bob
Confused
Posts: 1043
Joined: Fri Feb 13, 2004 2:30 am
Location: Houston, TX

Post by davd_bob »

purrkur and bill,
sorry for the thread hijack.
Dave Rave, man dont tease like that. Dont give a link unless it shows pics.
I mean of motherboards...not cases.

Later,
David
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Post by purrkur »

Billl wrote:But some of the blame has to go against the code as it was written.
I should mention that the software that gave us issues was a client and not a server side component. It is a financial application that does heavy duty calculations among other things. The code itself was faultless and worked on single CPU machines and dual CPU machines as well which is why it took us so long to narrow it down to HT. It simply didn't make sense at the time :) This software, because of its heavy calculation use sees a performance degradation when using HT.

I don't dislike HT technology. If Intel would only have done it right then they would have been able to use it for future products as well. As you mentioned, the changes to the CPU are minimal in terms of additional number of transistors so it is a shame that Intel is now saying that HT is not needed any more because dual core is replacing it. IBM and Sun are doing it the other way around and combining the two but then again their implementation of HT is far superior to that of what Intel came up with.

I might end up getting my hands on one of those 8 core CPU machines from Sun where each core can handle 4 simultaneous threads (32 in all) for evaluation. I would love to put that machine through it's paces.
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
Dave Rave
G'Day Mate!
Posts: 894
Joined: Thu Sep 26, 2002 4:28 am
Location: Sydney Australia
Contact:

Post by Dave Rave »

Last edited by Dave Rave on Sat Mar 11, 2006 9:43 am, edited 1 time in total.
KliK
Posts: 350
Joined: Fri Mar 11, 2005 3:56 pm
Location: Zagreb, Croatia
Contact:

Post by KliK »

the HT is writen only to speed up the processes by cheating system that it has 2 cores instead of one....now they are making dual-core CPUs, so it's normal to abandon the emulated dual-core technology such as HT!!! :D
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Post by purrkur »

KliK wrote:the HT is writen only to speed up the processes by cheating system that it has 2 cores instead of one....now they are making dual-core CPUs, so it's normal to abandon the emulated dual-core technology such as HT!!! :D
HT isn't emulation. It is technology that allows a single core to process 2 or more threads at once. If you had read all entries above you would have noticed me mention that both IBM and Sun are using comparable (although more advanced) technology in their multi core cpu's.

And HT doesn't cheat the system. It actually takes care of two threads at once. The difference is that those threads will have to fight over system resources while dual cores will have resources assigned to them. Intel's HT solution faltered at the point at how those two threads are handled once a conflict arises. However, even Intel's HT solution can give up to 20-30% speed increase if you are running a very simple program where the threads don't have to compete for resources. If you are doing several things at once and making your hardware do work (like a desktop user running many programs at once or a server working with several processes) then resource conflicts are unavoidable, hence Intel's problem with the technology.
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
Billl
Posts: 211
Joined: Fri Jul 25, 2003 1:52 pm
Location: USA

Post by Billl »

purrkur wrote:
Billl wrote:But some of the blame has to go against the code as it was written.
I should mention that the software that gave us issues was a client and not a server side component. It is a financial application that does heavy duty calculations among other things. The code itself was faultless and worked on single CPU machines and dual CPU machines as well which is why it took us so long to narrow it down to HT. It simply didn't make sense at the time :) This software, because of its heavy calculation use sees a performance degradation when using HT.
Now you know I'm going to take you to task on this one! Their is no such thing as "faultless" code. It obviously couldn't be faultless since it failed to run on HT. With all the millions of systems like that out there, they obviously didn't do a very good job in testing. It's hard to believe they wrote this and never once tried it on a HT processor? Sounds more to me like someone had a deadline to meet.


Billl
Image
purrkur
Linux Guru
Posts: 687
Joined: Fri Dec 12, 2003 5:57 pm
Location: Sweden
Contact:

Post by purrkur »

Billl wrote:Now you know I'm going to take you to task on this one! Their is no such thing as "faultless" code. It obviously couldn't be faultless since it failed to run on HT. With all the millions of systems like that out there, they obviously didn't do a very good job in testing. It's hard to believe they wrote this and never once tried it on a HT processor? Sounds more to me like someone had a deadline to meet.
Billl, This was at the time when HT was hot off the press, so hot that it was not very commonplace and still "top dog", i.e. very expensive. Our customers are not really that sensitive to pricing so it is very common that we are among the first to get to play with the latest and greatest. I have on several occasion gotten to work with hardware from manufacturers like IBM, HP and Dell that hasn't even hit their websites yet. In any case, our customer had received samples of HT enabled processors but we had not, hence the problems.

I say the code was faultless in that it executed just fine on single and dual CPU machines. It was HT that caused issues. You might be aware that changes were made to compilers (both Microsofts and Intel's own) in order to facilitate HT, however, this was so early on in the game that we had not received those either.
2x533MHz@544MHz, 2.0V
640MB PC100 memory
Realtek RTL-8139 NIC
Maxtor 6Y080L0 80GB hdd
Debian Linux stable with 2.4.8 kernel
KliK
Posts: 350
Joined: Fri Mar 11, 2005 3:56 pm
Location: Zagreb, Croatia
Contact:

Post by KliK »

purrkur wrote:
KliK wrote:the HT is writen only to speed up the processes by cheating system that it has 2 cores instead of one....now they are making dual-core CPUs, so it's normal to abandon the emulated dual-core technology such as HT!!! :D
HT isn't emulation. It is technology that allows a single core to process 2 or more threads at once. If you had read all entries above you would have noticed me mention that both IBM and Sun are using comparable (although more advanced) technology in their multi core cpu's.

And HT doesn't cheat the system. It actually takes care of two threads at once. The difference is that those threads will have to fight over system resources while dual cores will have resources assigned to them. Intel's HT solution faltered at the point at how those two threads are handled once a conflict arises. However, even Intel's HT solution can give up to 20-30% speed increase if you are running a very simple program where the threads don't have to compete for resources. If you are doing several things at once and making your hardware do work (like a desktop user running many programs at once or a server working with several processes) then resource conflicts are unavoidable, hence Intel's problem with the technology.

you seem to disagree with me, than you say the same thing in different words...
OK, HT is technology that cheats a system that is has 2 core instead of one and that way two processes are calculated in same time...now that they have 2 cores, they use some sort of on-chip-built-SMP developed earlier...and that one doesn't cheat that there are 2 cores, only it devides the tasks!!! hence the more cache is adviseable... :D

maybe there was problem with the technology...but it's the same for real SMP & HT: devide the process, (cheat there are 2 cores on CPU), calculate the process, combine the results...
so now you see why i think it's the cheating technology at place here!! :D
Post Reply