Hyperthreading made up for the inefficiencies of the P4 design (the P4 was designed to advertise itself with inflated clock speed ratings that do NOT match up with actual performance). 1.5GHz P4s sounded AMAZING when 1GHz P3s were the highest from intel, but the performance simply wasn't as amazing. HT made up for some of that but not entirely, there's a reason intel is going away from it.
SMT's positive impact is a well-known fact, one minute with google can get you a sample of studies....
http://www.linuxclustersinstitute.org/L...Revolution/Archive/PDF02/11-Leng_T.pdf
http://www-128.ibm.com/developerworks/linux/library/l-htl/
By the way, don't bother posting studies where there is a slight decrease in performance with SMT-enabled machines. There are some workloads known to perform worse on SMT/MP setups, and also one can see a negative speedup if SMT is used incorrectly. The point is to realize an averaged gain.
Please don't call technology you barely understand a "joke", it stinks of fanboism... see above post for example.
http://www.linuxclustersinstitute.org/L...Revolution/Archive/PDF02/11-Leng_T.pdf
http://www-128.ibm.com/developerworks/linux/library/l-htl/
By the way, don't bother posting studies where there is a slight decrease in performance with SMT-enabled machines. There are some workloads known to perform worse on SMT/MP setups, and also one can see a negative speedup if SMT is used incorrectly. The point is to realize an averaged gain.
Please don't call technology you barely understand a "joke", it stinks of fanboism... see above post for example.
I completely understand that technology and it is actuallly a joke, the performance gains aren't impressive at leat in the apps that I use. Also don't show me bechmarsk prepared by intel to show the greatness of some joke technology which is now absolutely useless. Look at you just throwing at us an article by dell, the intel No.1 servant.
If you completely understood the technology, you wouldn't be sitting around spewing bs and not posting any real numbers, and relying on personal anecdotes to justify your points. It took me about a minute to find two studies (and only one of them by dell), and if you actually read them, you'd see they weren't written by yesmen... does that make you a no-man? LOL.
By the way, I don't claim to completely understand every aspect of SMT, and I deal with it at the design level... so please don't say you know this and that.
By the way, I don't claim to completely understand every aspect of SMT, and I deal with it at the design level... so please don't say you know this and that.
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
For games, there will be 0 difference since your graphics card will be a bottleneck at every setting in every current game.
For everything else, dual core is better the more you multitask.
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
It is true that HT is a joke compared to dual-core or multi-core processors because HT is not a dedicated setup like a dual-core system. However, it makes your experience much smoother than any single core setup.
Download THG Video #5 - P4 3.6ghz no HT vs. P4 3.0ghz w/HT - real world apps
Also HardOCP notes:
Subjective Computing Experience
"With my Athlon 64 FX-53 system, I simply could not go about my daily business of running HardOCP, HardForums, and Ratpadz while encoding DivX. Back in July of 04, I built myself an Athlon FX-53 system to play DOOM 3 on. Also at the time, I was in middle of moving my DVD collection to DivX in order to use it on a ?video jukebox? I was building for the living room.
On my previous 3.2GHz Pentium 4 system with HyperThreading, I could easily start a DivX encode using Dr. DivX and then go about my other normal daily tasks that for the most part are not CPU intensive. On my Pentium 4 system with HyperThreading, the DivX encode was all but transparent. Of course, you could not run another heavily CPU dependent application and expect the same smooth user performance. Trying to do my daily work on my FX-53 box while encoding DivX left me with an unresponsive user interface that was slow and laggy. I was simply not able to get my multitasking workload done without a large amount of frustration.
Then:
I used both DivX encoding and Premiere Elements WMV encoding while playing Counter Strike: Source online with other players. Let me discuss my gaming experience while having Premiere Elements encoding a WMV file as it was more intensive than the DivX application. Here are my gaming experience notes in full:
Athlon Single Core ? Extremely Long load time. So long you would miss the first round at map change. Had two instances where the game never loaded till encode was done. Once in the game, the map was very choppy and simply unplayable. A much worse experience than Intel single core play.
Intel Single Core (w/HT) - Almost playable during first and second pass. You ?could? play, but again this would not be something you would likely want to do. You get that laggy and zoomy feel and it negatively impacts your gameplay. Not a gaming experience I would suggest. "
Obviously dedicated dual-cores dominated but HT is not necessary a joke. True it doesnt improve performance significantly, but given it barely cost anything more to switch from P4 2.66, 2.8 to P4 2.6, 2.8 processors with boosted FSB to 800 and gave HT, I dont think it was a waste at the time.
Oh yeah, the topic of the thread is pointless because SMT costs <5% extra die area while dual core doubles it. Different return on different investment. I guess I'm only posting because I was amused by some of the comments.
I won't disscus anymore about an outdated and now useless technology like HT. That's it. The question is HT Vs. Dual core, and actually the only truth is that DC>>>>>HT. And I said HT is a joke from my standpoint, with the apps I USE the experience is as frustrating as with Athlon single cores, good for you if HT is useful to some of you people, but for me and my really heavyweigth tasks HT just sucked as the single core athlon does (in spite of that athlon got the job faster done under some circunstamces), that's why I always use DC and MP configurations. For the normal multitasking Athlon single cores and P4HT are just fine, with a little advantage of P4 in very specific areas.
That's what I mean, If you run multiple Heavyweight apps using HT the system get as unresponsive as the Athlon does, that's useless to me, solution: use a dual core and that won't happen.
In am not agree with that at all, I was using an Athlon 64 3000+ oced to 2.6GHz and encoding a video from a DVD with XVID I could do the daily tasks with no lags, of course I have a dedicated OS Raptor HD. I also used a P4 to do that but using only one disk just produced also some lag as my A64 system with one disk, adding another fast disk solved the problem, also I had to minimize the encoding window and the system gets absolutely responsive even usinfg an old AXP 2400+!. I guess the problem was just a purely I/O bottleneck.
Both situations are ugly gaming experiences , I really hate lags in game, specially suddlenly slowdowns, that is something nobody wants to do while encoding at least that you have a dual core CPU. Once again I don't see any advantage in this situation.
Nobody gives a rat's ass about your "heavywright tasks" when it comes to debating SMT as a technology. And enough with the personal anecdotes already, it is really sad.
You won't discuss it because you don't know jack about it. You only said SMT was a joke from your standpoint in the last post; you've been sniping at the P4 implementation without any justification this whole thread. But you reference IBM's usage without negative commentary. Typical fanboy.
You won't discuss it because you don't know jack about it. You only said SMT was a joke from your standpoint in the last post; you've been sniping at the P4 implementation without any justification this whole thread. But you reference IBM's usage without negative commentary. Typical fanboy.
Whatever you say dmens. I won't disscus about HT because it's not worth it, its useles, it's outdated and a joke compared to dual cores whatever they are intel or AMD, fanboy of what? I am not snipin anything just saying that DC>>>>HT, isn't that true? HT is now useless. Fine by you if you love HT, but you have to accept that DC are a lot superior that HT and that's it. Maybe in the past your beloved HT had some advantage, but now it can't stand against intel or AMD DCs.
Notice how I've been saying that comparisons between DC and SMT are pointless due to vastly different hardware cost. My only goal has been to disprove the notion that SMT is a "joke", and frankly, your assertions are completely off-target:
1) "Not worth it". WRONG. 5% die area increase for >30% averaged wall clock gain on server workloads.
2) "Useless". WRONG. With the introduction of multicore, workloads will move to become even more threaded, and SMT will benefit.
3) "Outdated". WRONG. SMT is a feature now in Intel and IBM cores, both current and upcoming.
4) "Joke compared to dual cores". Not applicable due to investment difference.
Admit it, you just don't know jack. I'm done here, there is no point talking any more.
1) "Not worth it". WRONG. 5% die area increase for >30% averaged wall clock gain on server workloads.
2) "Useless". WRONG. With the introduction of multicore, workloads will move to become even more threaded, and SMT will benefit.
3) "Outdated". WRONG. SMT is a feature now in Intel and IBM cores, both current and upcoming.
4) "Joke compared to dual cores". Not applicable due to investment difference.
Admit it, you just don't know jack. I'm done here, there is no point talking any more.
Outdated?? :disgust: Hyperthreading is not going away. Xeons are and will continue to be hyperthreaded as well as becoming dualcore. Why are you arguing the benefits of dual-core versus hyperthreading when one doesn't preclude the other? Why are you complaining that SMT only shows performance improvements with multiple cores when all future chips will be multicore? This is like complaining that since dual-core in-order processors get more performance than single-core out-of-order chips, OOO execution is outdated technology.
And here're some real numbers for you:
Linux kernel compilation shows 16-25% speedups in compilation throughput with HT vs without.
http://www.2cpu.com/articles/41_3.html
GIMPS (the Great Internet Mersenne Prime Search) shows increases of 10% in Lucas-Lehmer testing and 40% in trial factorization throughput with Hyperthreading vs without:
http://www.mersenneforum.org/showpost.php?p=52598&postcount=6
Lattice sieving (the most time-consuming portion of the number field sieve algorithm for factorizing integers) shows 43% increase in relations found per second with two instances running on an HT processor compared to one instance (38.023 rel/sec versus 26.455 rel/sec with a 177-digit composite number). I've heard reports of up to 60%.
GMP-ECM runs at the 45-digit level on a 164-digit composite number gets 20% throughput increase with 2 curves running at the same time rather than one.
For the applications I care about, hyperthreading shows real concrete advantages. I find it amazing that you would pass them up. Heck, the increase from 8 to 16 GPRs that x86-64 brings only gives you about 10% performance increase on average while bloating code by 15%.
And here're some real numbers for you:
Linux kernel compilation shows 16-25% speedups in compilation throughput with HT vs without.
http://www.2cpu.com/articles/41_3.html
GIMPS (the Great Internet Mersenne Prime Search) shows increases of 10% in Lucas-Lehmer testing and 40% in trial factorization throughput with Hyperthreading vs without:
http://www.mersenneforum.org/showpost.php?p=52598&postcount=6
Lattice sieving (the most time-consuming portion of the number field sieve algorithm for factorizing integers) shows 43% increase in relations found per second with two instances running on an HT processor compared to one instance (38.023 rel/sec versus 26.455 rel/sec with a 177-digit composite number). I've heard reports of up to 60%.
GMP-ECM runs at the 45-digit level on a 164-digit composite number gets 20% throughput increase with 2 curves running at the same time rather than one.
For the applications I care about, hyperthreading shows real concrete advantages. I find it amazing that you would pass them up. Heck, the increase from 8 to 16 GPRs that x86-64 brings only gives you about 10% performance increase on average while bloating code by 15%.
Hyper threading is like having one 3.2 Ghz processor and one 500 Mhz processor. Dual core would give you two 3.2 Ghz processor.
You are just showing your ignorance about what HT is really about. Please, do not try to promote this in forums because you will only confuse people that are trying to absorb real knowledge here. HT gives you an additional logical CPU which is definetly not rated at half the speed of your current CPU. Dribble's post has an analogy that approaches reality.
SMT is not a joke, because it is also used in multicore configs, I am talking specifically of HT.
Netburst uneficcient architecture fault, AMD64 CPUs doesn't need SMT to easily surpass those numbers. With shorter pipelines the level of improvement would be negligible using x86 architectures, if HT is so easy to implement in short pipelines desings with little core increase intel would Have implemente HT in pentium M CPUs and next dothan CPUs, but no they didn't reason: it's not worth it.
Consifering that netburts is such an unefficient desing and considering the huge amount of power XEON consumes (not fault of HT) the improvements of HT of 15-25% are not amazing, besides those get ass kicked by opterons.
Using Netburst unefficcient architechture, AMD opterons CPUs without HT surpass those numbers with no need of SMT single core useless technology.
Multicore >>>> HT, HT alone will not have much more to offer, multicore will increase the perfromance with increasing of cores and threads, look how useless is HT in the PD 840XE it does anything for the performance and sometimes even harm the performance, the situation won't change in the near future.
SMT features of IBM as stated are worth only in multicore CPUs as Power5 in single core desing the advantages are dismal because there would be not enough resources (That's why IBM power5 is a dual core CPU). Intel Netburst CPUs need HT due it's unefficient desing, in spite of HT opterons machines kickass XEONs configs with HT enabled.
A dual core opteron CPUs consumes less power than a single core Xeon that is a great saving, worth the investment, sorry that netburst is so unefficient that can't not even keep up with HT enabled and the Xeon DC are a piece of crap.
Investment difference, lets see.
XEON 3.2 GHz nocona with HT enabled...319USD
Opteron 165 ...366USD
47 USD of difference for a lot of SMT performance, I see the investment is really worth it, specially in the server market where this is more important.
Untrue. While the first wave of Intel's Next-Generation Microarchitecture (Merom/Conroe/Woodcrest) will not have SMT, the later cores will.
http://www.theregister.co.uk/2005/08/23/intel_next_gen_architecture/
"Interestingly, in its "first implementation", NGMA doesn't have HyperThreading, to make software easier to compile, David Perlmutter, VP and General Manager of Intel's Mobility Group suggested. However, expect cores supporting up to eight threads over time, Intel's Digital Enterprise Group VP, Stephen Smith, said. At that time, some CPUs will be single-threaded, others multi-threaded, he added."
One: it is a myth that SMT is not worth it on shorter pipelines or single cores. The Alpha EV8 would have implemented 4-way simultaneous multithreading in a single core with a 9-stage pipeline. They estimated it would have doubled performance with a die-size increase of less than 10%.
http://www.realworldtech.com/page.cfm?ArticleID=RWT122600000000&p=5
"Research to date suggests SMT can approximately double the throughput performance of an 8 instruction-issue wide processor like EV8 for a cost in extra processor complexity equivalent to less than 10% increased die area for the processor core."
Two: Proliferations of the Merom core *will* have SMT. It probably will not double the performance, but I know for a fact it will increase it significantly.
Compared to -20% to 10% performance increase moving to 64-bit on the same applications I listed before, I'd say 15-25% is quite a significant win. And that's on top of any gains multi-core can give you.
Well, if AMD thought as you, I think Intel has nothing to fear for the next 3 years. You remind me of the people that claim register renaming helps register-starved CISC designs more than RISC, and so is a necessary added cost of designing CISC chips. Well, the fact is, any serious high-performance RISC design also implements register renaming, because the performance gain is worth it.
You could say the exact same thing about x86-64. In many cases, it bloats code and data sizes, decreasing cache effectiveness and slowing things down. Many benchmarks run slower in Windows x64 than 32-bit Windows XP. And the Linux applications I run (especially the lattice siever) are sensitive to codesize bloat, and actually run significantly faster in 32-bit mode.
HT was useful on the P4 because it increased the efficiency of the long pipeline. Dual cores is a better technology that wasn't possible in the days of .13 micron cpus. The cost of a dual core .13 micron cpu would be huge and profit margin would be smaller, not to mention yeilds. HT was a good trade off that increased efficiency without a huge increase in die size.
OMG too funny to resist. HT is not SMT? Tell me why, I'm dying to hear this.
Hate to tell you this, but the whole "SMT has no benefit on short pipeline procs" belief was an AMD PR line lapped up like so much honey by the fanboy crowd. I've read plenty of in-house perf studies that say otherwise. You might even be able to figure out why if you think about it instead of regurgitating stale web tales.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter