When hyperthreading comes around the discussion

LeHibou · 2012-05-01 10:00:08

Hello there,

I was wondering one thing, as I am about to create a new system with Arch.

The next configuration will be about Blender (Cycles), Scribus, Gimp, Inkscape, encoding and some games, Komodo and some libreoffice stuffs.
A work-around configuration.

All working on a SSD + normal HDD to store datas and a solid graphic card 560Ti/570.

As far as I am concerned, it seems that those applications won't change a lot of thing if used with hyperthreading as they don't manage it very well/don't use it (Cycles uses the GPU instead, which is great).

All the benchmarks tells that 2600K is a beast. I am pretty sure I can do the same on a linux architecture with a 2500K/3570K.
Even if both are OC, think the 2600K is pointless on linux, for now.

Gaming : who really uses the CPU ? The GPU are so powerful that the difference will be elsewhere. But where ?

My opinion is just : nowhere.

What is yours ? It is not a troll but a franck debate concerning the place of hyperthreading nowadays in Linux distrib, especially Arch, even in non-gaming apps.

Regards,

LeHibou

Gusar · 2012-05-01 10:27:50

Every time you have something highly multi-threaded, hyperthreading helps. Most common scenarios are video encoding and compiling stuff. Now whether it'll make a difference in you select few scenarios, who knows, likely not. But calling an i7 "pointless" because of it goes way too far.

graysky · 2012-05-01 10:31:21

Average boost in povray with HT on my machine is 22 %

LeHibou · 2012-05-01 11:30:54

Way too far, well, not my thought. A bit farther than usual is slightly more correct, isn't it ?

It is way too far if many applications used multi-T. Not the case, or not on a productivity basis. It is still a struggle for many apps.

Compiling : As far as a former gentoo user, i think I should feel concerned. Arch let me breath with that and AUR is just not a pain (not compiling KDE, libreoffice, firefox and so on).

Moreover, who compiles everyday ? (just asking, no attacks) Developpers ? Their case is appart, and I agree a HT is a positive thing (many compilations, and let's don't forget that time is money) But on a general purpose, not seeing the point.

Graysky, what is you cpu ? Is it OC ?

anonymous_user · 2012-05-01 16:53:45

Not all software can take advantage of numerous cores (real, modules (AMD's Bulldozer) or HyperThreading). Some people can benefit, and some can't. If your system and the software you use does not see much benefit from having so many cores, then don't buy it. Simple as that. Don't think that technology has to benefit EVERYONE to be useful.

LeHibou · 2012-05-01 17:14:42

I will have to, since I have to do stuffs under windows via bootcamp.

Real question is not if i7 worthes it over i5, but what is the linux strategy concerning HT for creative non-IT peoples.

How far is the day in which HT will be supported by majors apps ?

Pierre · 2012-05-01 17:17:15

HT has to be supported by the kernel not the application. And Linux supports HT for a long time.

Gusar · 2012-05-01 17:24:25

What kind of "strategy" are you talking about, and why would there need to be one?

HT is not something you directly code for, like is the case with instructions sets like AVX. HT is just there. It allows my netbook to play 720p h264 video. And the ffmpeg developers didn't need to do anything HT-specific for that, they just made video decoders multi-threaded. The only thing aware of HT is the kernel scheduler, there's SCHED_MC and SCHED_SMT.

LeHibou · 2012-05-01 18:16:00

Please, let me make this straight.

HT has to be supported by the kernel not the application. And Linux supports HT for a long time.

Not the subject.
Out of hibernating for years, everybody knows that.

HT is just there.

No. HT is not just here. Did you even tried Scribus, Gimp and many other apps to say that ?

You would know that HT is not enabled or mismanaged on Linux.

Out of Dev task about compiling which work just fine and ffmpeg-like apps, any other apps under linux will struggle with HT.

Didn't you noticed ?

Now, my question about the strategy : is there a moment in which devs will make apps in linux really productive with HT, like it is on other OS.

That is the question.

anonymous_user · 2012-05-01 18:30:47

Simply put, most apps are not very multithreaded. If an application doesn't benefit from HT then it probably won't benefit from additional cores either.

It could be the work done by a program cannot be multithreaded. Sometimes one thread needs to be finished before the next one can start.

Gusar · 2012-05-01 19:03:21

LeHibou wrote:

No. HT is not just here. Did you even tried Scribus, Gimp and many other apps to say that ?

This has nothing to do with HT, but with plain multi-threading. If a particular Gimp filter is single-threaded, well, then it's single-threaded. If, on the other hand, a Gimp filter is multi-threaded, the kernel will take care to schedule the threads properly. HT is irrelevant to the app, scheduling is the kernel's job.

LeHibou wrote:

You would know that HT is not enabled or mismanaged on Linux.

Err, what? As I said, the kernel has SCHED_SMT. And that's all there is to it, all that's needed. A scheduler that takes into account HT when scheduling threads. If you mean here that not many apps are multi-threaded, that's an entire different discussion that has nothing to do with HT.

LeHibou wrote:

is there a moment in which devs will make apps in linux really productive with HT, like it is on other OS.

Err, you said you understood that HT is a kernel thing, not an app thing. So why are you now talking about apps again?

LeHibou · 2012-05-02 06:30:30

Simply put : If HT is activated by kernel, we should provide from it up to the apps.

Question is not the system, but apps.

Benefiting from HT + multithread in Linux.

So, as a buying concern on Linux, HT processors are not necessarily the best match for the purposes my computer will be asked to do. If apps are not taking advantage of it, why would I spend so much more buck for it.

So, i5 3570K is better money-wise than a 2600K on a linux machine, since HT won't really benefit the whole system and my apps.

Both are multi-threaded, and will be limited by the poor multi-thread support of the apps, so will they do the job much equally.

In which cases will HT be a real advantage out out compilation and MASSIVE video encoding (an i5 can do the job for your all the encoding since you don't have to gain money from it, as a professional purpose, in which every second gained is ringing dollars).

Am I wrong ?

Last edited by LeHibou (2012-05-02 06:31:55)

Gusar · 2012-05-02 07:49:40

I think you're still a bit confused what HT is and how it works.

And when you talk about "poor multi-thread" in apps... it's a) not specific to Linux, and b) most apps wouldn't benefit much from multi-threading anyway, as they're not CPU bound, but I/O bound or spend most of their time waiting for user input. But there are apps that are multi-threaded. So saying the whole system won't benefit from HT is just plain wrong. Especially when there's also another angle you haven't taken into account - a machine will not be running just one app, but pretty much always several apps at the same time. So the more cores, the merrier.

Now if you think HT and the larger L3 cache of an i7 compared to an i5 isn't worth the extra money, that is a valid individual choice. Keyword individual, because there are other individuals who may think otherwise. In any case, your crusade against HT is just silly.

Lone_Wolf · 2012-05-02 10:38:46

LeHibou wrote:

Simply put : ~~If HT is activated by kernel, we should provide from it up to the apps.~~
Question is not the system, but apps.
Benefiting from HT + multithread in ~~Linux~~ any OS.
So, as a buying concern on ~~Linux~~ any OS, HT processors are not necessarily the best match for the purposes my computer will be asked to do. If my apps are not taking advantage of it, why would I spend so much more buck for it.
So, i5 3570K is better money-wise than a 2600K on a ~~linux~~ machine, since HT won't really benefit the whole system and my apps.
Both are multi-threaded, and will be limited by the poor multi-thread support of the apps, so will they do the job much equally.
In which cases will HT be a real advantage out out compilation and MASSIVE video encoding (an i5 can do the job for your all the encoding since you don't have to gain money from it, as a professional purpose, in which every second gained is ringing dollars).

I corrected it for you.

LeHibou · 2012-05-02 11:52:57

That is what I call fair !

I have got my answer : I will go for a 3570K

Many thanks everybody, seems that the case solved !

Regards,

LeHibou

Mr.Elendig · 2012-05-02 11:52:57

Total sidenote, there isn't any good reason to by a 2600k now that IB is out, unless you find it on a really good sale.

Anyway, HT can benefit non-SMP apps too, if you are running enough of them
But that said, the price premium on i5 vs i7 is quite large, In many cases you are better off getting the i5, and spend the money saved on more ram/better gfx card/ssd or something instead, when it comes to getting the most performance for your money.

LeHibou · 2012-05-02 11:54:07

Same time response.

I got it Mr Elendig, many thanks, all of you.

I can sleep like a baby now

Arch Linux

#1 2012-05-01 10:00:08

When hyperthreading comes around the discussion

#2 2012-05-01 10:27:50

Re: When hyperthreading comes around the discussion

#3 2012-05-01 10:31:21

Re: When hyperthreading comes around the discussion

#4 2012-05-01 11:30:54

Re: When hyperthreading comes around the discussion

#5 2012-05-01 16:53:45

Re: When hyperthreading comes around the discussion

#6 2012-05-01 17:14:42

Re: When hyperthreading comes around the discussion

#7 2012-05-01 17:17:15

Re: When hyperthreading comes around the discussion

#8 2012-05-01 17:24:25

Re: When hyperthreading comes around the discussion

#9 2012-05-01 18:16:00

Re: When hyperthreading comes around the discussion

#10 2012-05-01 18:30:47

Re: When hyperthreading comes around the discussion

#11 2012-05-01 19:03:21

Re: When hyperthreading comes around the discussion

#12 2012-05-02 06:30:30

Re: When hyperthreading comes around the discussion

#13 2012-05-02 07:49:40

Re: When hyperthreading comes around the discussion

#14 2012-05-02 10:38:46

Re: When hyperthreading comes around the discussion

#15 2012-05-02 11:52:57

Re: When hyperthreading comes around the discussion

#16 2012-05-02 11:52:57

Re: When hyperthreading comes around the discussion

#17 2012-05-02 11:54:07

Re: When hyperthreading comes around the discussion

Board footer