A3400+ = equivalent speed in P4?

[Albuquerque]

HT is more like SSE, you need to be codded to work with HT to take advantage of it, just like SSE. WHen rendering certina plugins, only like 1 processor, so they are getting 1.4 ghz outa my 2.8 ghz.

Very incorrect... You don't get half the clockspeed (or anything remotely close) when you run a single threaded app on a hyperthreading-enabled CPU. If (and it's a big IF) you get any slowdown at all, the difference is less than 10%. So at the absolute worst, you're getting performance somewhere around a 2.53ghz processor... The only way you get performance that bad is to use HT on a non-supported operating system (Windows 2000 is a prime example, so is Windows NT 4 and anything earlier), while running a single application with multiple simultaneous threads that are meant to run on a single processor. When a misunderstanding OS seperates those multiple threads to be scheduled across (what it thinks are) seperate processors, things go to hell.

Those cases are few and far between.

The reality is, a large number of applications don't even notice the HT portion, and in fact perform BETTER on hyperthreading (without any sort of special programming needed) because a proper OS can schedule multiple threads appropriately.

Microsoft Office 2000 wasn't programmed to work with hyperthreading, but running Excel 2000 in combination with Access 2000 while doing goofy database work runs really well on a hyperthread system -- much better than on a NON-hyperthreaded system. Same goes for a ton of apps that are already out there...

 

[TC]

im going to argue a few points with you TC

While i agree HT is a step in the right direciton, it is not the final solution.

HT is more like SSE, you need to be codded to work with HT to take advantage of it, just like SSE. WHen rendering certina plugins, only like 1 processor, so they are getting 1.4 ghz outa my 2.8 ghz. Yet when they are programed to use more than 1 proc, specailly more than 2, it is really really cool to see how much xeons with HT rock.

So until more appz are programed to take advantage of HT there is no use letting those cycles go to waste.

I would have to agree with you. I don't think HT is the end all. I think it's another useful addition to cpu technology, but like anything else the software has to be designed to take full advantage of it, and that takes time. HT is just another tool, one with limitations, but clearly some obvious advantages when used correctly. It cracks me up that some people think it's little more than a smoke and mirror deal to hide a problem.

 

[DayUSeX]

Very incorrect... You don't get half the clockspeed (or anything remotely close) when you run a single threaded app on a hyperthreading-enabled CPU. If (and it's a big IF) you get any slowdown at all, the difference is less than 10%. So at the absolute worst, you're getting performance somewhere around a 2.53ghz processor... The only way you get performance that bad is to use HT on a non-supported operating system (Windows 2000 is a prime example, so is Windows NT 4 and anything earlier), while running a single application with multiple simultaneous threads that are meant to run on a single processor. When a misunderstanding OS seperates those multiple threads to be scheduled across (what it thinks are) seperate processors, things go to hell.

Those cases are few and far between.

The reality is, a large number of applications don't even notice the HT portion, and in fact perform BETTER on hyperthreading (without any sort of special programming needed) because a proper OS can schedule multiple threads appropriately.

Microsoft Office 2000 wasn't programmed to work with hyperthreading, but running Excel 2000 in combination with Access 2000 while doing goofy database work runs really well on a hyperthread system -- much better than on a NON-hyperthreaded system. Same goes for a ton of apps that are already out there...

I argue greatly, How come when I render and the plugin at hand is the only thing running (when i pause F@H (team 32)) it will only use 25% of the 4 *cpus* avaialble??? (this is the dual xeons 2.8 oced to 3.15). So if im only getting 1/4, where when i turn HT off i get 50%. Explain that to me?? Maybe im its just that i use software all of 1% of the whole world uses I dunno. I just know that what i have seen of HT it really dosent impress me. To many friends have bought the hype of HT when they got a PC for their home videos and have been sorley dissapointed when I do the same thing 2x faster, while doing content creation on my SMP.

TC

Yup we agree, while i would love to see apps support HT, i would rather the designers make their software to support SMP than HT really. (yes there is a differnce, but again maybe only for the 1% out there tha use max i dunno).

 

[Albuquerque]

I argue greatly, How come when I render and the plugin at hand is the only thing running (when i pause F@H (team 32)) it will only use 25% of the 4 *cpus* avaialble??? (this is the dual xeons 2.8 oced to 3.15). So if im only getting 1/4, where when i turn HT off i get 50%. Explain that to me??

With hyperthreading enabled, you have four logical processors. On a thread that isn't SMP/SMT aware, it's only going to fully utilize ONE processor. Full utilization on ONE processor of four = 25% Yay, I'm a genius!

It's the EXACT same thing as running prime with Affinity turned on (so it uses only one processor if multipe are available). On a HT processor, it will peg the processor utilization to 50% and go no further.

What you are completely ignoring is that Windows' utilization graph DOES NOT equate to a mhz performance scale on multiprocessor (logical OR physical) systems. Hell, even on a single processor machine, that graph does not equate to a true Mhz performance scale. Saying that 50% utilization is exactly half of your clockspeed is incorrect, just like saying 25% utilization is exactly one quarter of your clockspeed is also incorrect.

Once again, an application does NOT have to be programmed with SMT/Hyperthreading in order to benefit from it. I've already given you one example and you simply wiped it away in order to blurt out your ignorance about the goofy Windows CPU utilization bar.

 

[OC Noob]

I argue greatly, How come when I render and the plugin at hand is the only thing running (when i pause F@H (team 32)) it will only use 25% of the 4 *cpus* avaialble??? (this is the dual xeons 2.8 oced to 3.15). So if im only getting 1/4, where when i turn HT off i get 50%. Explain that to me?? Maybe im its just that i use software all of 1% of the whole world uses I dunno. I just know that what i have seen of HT it really dosent impress me. To many friends have bought the hype of HT when they got a PC for their home videos and have been sorley dissapointed when I do the same thing 2x faster, while doing content creation on my SMP.

TC

Yup we agree, while i would love to see apps support HT, i would rather the designers make their software to support SMP than HT really. (yes there is a differnce, but again maybe only for the 1% out there tha use max i dunno).

Why would your friends be sorley disapointed that your 2 HT CPUs are twice as fast as their 1 HT chip when doing stuff that supports SMP?

That makes no sense and you are still comparing SMT to SMP while no one in this thread is claiming they are comparable.

 

[DayUSeX]

because you could have built a smp for just as cheap........ OCnoob, again though for the average consumer this is not an option probably (l5 mod xps, i doubt many know how to do that). But still SMP's are the unholy price they were a few years ago.

 

[DayUSeX]

I argue greatly, How come when I render and the plugin at hand is the only thing running (when i pause F@H (team 32)) it will only use 25% of the 4 *cpus* avaialble??? (this is the dual xeons 2.8 oced to 3.15). So if im only getting 1/4, where when i turn HT off i get 50%. Explain that to me?? Maybe im its just that i use software all of 1% of the whole world uses I dunno. I just know that what i have seen of HT it really dosent impress me. To many friends have bought the hype of HT when they got a PC for their home videos and have been sorley dissapointed when I do the same thing 2x faster, while doing content creation on my SMP.

TC

Yup we agree, while i would love to see apps support HT, i would rather the designers make their software to support SMP than HT really. (yes there is a differnce, but again maybe only for the 1% out there tha use max i dunno).

or teh fact that MAX has a detailed report telling you what took X long to render. Therefore by using times its pretty easy to equate just how much processor powre you are using.

Mb i miss spoke, but im just saying comparing the log files that are reported, rendering the same frame, with the same setup, when HT is disabled it renders almost 2x faster, then when i have HT enabled.

I think that explains itself right there.

Again I think I and you are failing to realise that i am not the average consumer. So ill admit in some instances HT is very good, but realise in applications where speed actually matters (lets face it excelle, and axx have nothing on a max) HT can be alot of a hinderance rather than a real benefit.

Sorry i do more on my pc then excell and axx.

 

[Albuquerque]

So howabout this: in the rare two instances you mention, HT gives a slowdown. In 98.3% of the remaining apps on the planet, they don't care.

See the difference?

 

[DayUSeX]

albuquerque thats just what i said, no need to get alle xcited about it. All im saying is it seems to often that intel fan boys jump all over HT, im pretty sure 99% of the people here would rather have the future head towards SMP than SMT.

OH yeah 98.3% of apps dont care, dont make up figures, its pretty obvious you are wrong.

 

[Albuquerque]

It's pretty obvious to anyone who uses an HT system that you're wrong about a full 50% drop in "speed" when enabling HT and using your app, but you seem fine with stating it.

Thus, I'm fine with stating that you're blurting ignorance, and that the other 98.287% of apps don't give a flying rat's posterior.

 

[DayUSeX]

It's pretty obvious to anyone who uses an HT system that you're wrong about a full 50% drop in "speed" when enabling HT and using your app, but you seem fine with stating it.

Thus, I'm fine with stating that you're blurting ignorance, and that the other 98.287% of apps don't give a flying rat's posterior.

yes i was wrong in some instances, maybe im behind on CPU's becuase i dont have an unholy amount of time to read up on them, but when HT first came out, people were turning off HT to get better bechmarks, was it a 50% drop, no, it was more like 7-13% but it is still a drop.

But like i said, people who do more with their computers than excell, axx, or whatever, and actually do video/cad/max/maya for a career, generally dont like HT (im not speaking for the indusrty as a whole, i know alot off ppl who love HT), just those around me who constantly have cpu usage at 100% with 2+gb always full wiht a 8gb swap always full dont like HT.

 

[zangler]

so which is best for checking email?

 

[DayUSeX]

LOL p133 of course =P

yeah really i kinda got of course and sorry bout the thread jacking. Still i stand by that i would rather see the future all SMP thn SMT

 

[KOXC2003]

First:

" In the majority of cases, if you were to enable Hyper-Threading on a desktop PC you would not see a performance increase, rather a 0 - 10% decrease in performance."

umm.. no, although Intel markets HT to make it look much more helpful then it really is, it does help.

What would be interesting is a overclockers benchmark. But this would be basically impossible to do this accurately.

 

[DayUSeX]

First:

" In the majority of cases, if you were to enable Hyper-Threading on a desktop PC you would not see a performance increase, rather a 0 - 10% decrease in performance."

umm.. no, although Intel markets HT to make it look much more helpful then it really is, it does help.

What would be interesting is a overclockers benchmark. But this would be basically impossible to do this accurately.

2 of my friends who are stricktly gamers run 2.4c's and after ocing them (nothing really fancy just past the 3.0 mark) and obsessive benchmarking i found that with HT disabled they benched better (ut2k3, 3dmark, super pi), it wasnt much better but they were getting better results with HT off. One is running a ti4200, the other a fx5900. And these are gamers bench marks basically agreeing with my workstation benchmarks.

 

[OBLIVIONLORD]

show us proof

 

[Albuquerque]

AcesHardware, among several others, found performance gains in UT2K3 when enabling HT -- not a whole lot, but typically between 3 and 8%.

Originally posted by Aceshardware
With HyperThreading on we achieved 82.7 frames in the Asbestos deadmatch benchmark and 81.3 with HyperThreading disabled. [snip] It is quite remarkable how almost every single threaded benchmark still got a small performance boost from HyperThreading, between 1 and 5%. This shows that HyperThreading has matured as it almost never decreased performance, as it did in the first hyperthreaded Xeons.

Most multi-tasking scenarios were measurably faster with HyperThreading on, and HyperThreading is a very smart way to improve CPU performance.

As for hyperthreading in SuperPi, three different websites (Envy, Aces and Tweaknews) all reported identical scores when comparing the same system with HT enabled or disabled on an XP platform.

The ONLY app of the three you listed that can be confirmed as going slower on a HT system is 3DMark01. It is multithreaded, but was built long before hyperthreading (or any form of SMT) was available in the desktop mainstream market. In 3DMark01, the total measurable slowdown was less than 3% when comparing with a 2.4C HT vs non-HT system.

So, I too would like to see some proof of these slowdowns you mention that your friends are seeing.

 

[KOXC2003]

Even in 3dmark03 which is hugely video card based, results are usually higher with ht on.

 

[racpuc]

FROM ACEHARDWARE. You can check out the article to get a full picture. I only saw one game that was bad because of HT. Every other game either stayed the same or was better which HT on. MOST apps improved with HT on

 

[Malichite]

I think one thing that lots of people also forget during these dicussions about HT perf drops is that most examples people give from experience/reviews were done using Windows 2000.

I was recently trying to decide between a 3.0C or a A64 3400+ and I found it almost impossible to find agreement on the benefits/failings of HT. Most seem to be too entrenched with either a pro Intel/AMD attitude that it skews feedback since there ARE reviews out there that show both ways, so it is up to the poster as to which one they believe and refer to. After searching around for weeks, the conclusion I came to was that in several specific apps (typically media encoding related or stuff like 7Zip) there was a noticable increase using HT, in most games there was little to no improvement (current games, since HT may be optimized in future games), and using Windows 2000 or NT caused a performance drop with HT enabled (Windows XP looks at the HT a bit more efficiently so no drop with HT enabled).

In the end I decided to go for the A64 since they run cooler, have better gaming performance (not sure how much since both UT2k3 and Serious Sam have rumored AMD optimizations), close to P4 3.2-3.4C media encoding performance (only 10-15% slower), still using Win2k, and better performance in Scientific Apps/WinRAR due to low latency embedded memory controller. I am still waiting on the parts and I hope I am not disappointed once I slap it in my system. I loved the theory behind HT, but in the end I decided that it wasn't likely that I would be able to do too many intensive things (big WinRAR encryption or DiVX encoding) in the background while playing Far Cry or UT2k4. I can always just try to force the application to use a lower priority level or cap its utilization if I feel the need to try this on my A64.

The one HT question I never found the answer to was since the system sees 2 separate processors does this always insure that a rouge process or instance of IE can never run amok and hang the system (not BSOD just stealing all resource and swapping to disk like a mofo)? It doesn't happen too often, but I use streaming webcast for school and some times the imbedded instance of Media Player 9.0 hangs and steals all processor resources until it resets (i.e. task manager can't kill it, so either wait several mins or reset). I have asked before, but never got much response.

-Mali