Is HyperThreading worth the extra £80?

[paperwastage]

I'm just wondering if in 2+ years, games will come out that make use of more than 4 cores, and thus HT would be giving large increases.

If that happens, I'll be kind of disappointed that I got the 3570k.

nope... unless a crazy new Crytex engine v3 comes out(or poorly optimized from whatever consoles we have in 2 years), or you want to do something like eyefinity 5.....

you'll still be better off saving that $ for a future GPU upgrade in 2 years

the current Sandy Bridge / Ivy Bridge is more than efficient enough for gaming for the next 4 years


and remember, in 2 years, some new technology will be out... there'll be people upgrading from their 3770k.... buy one used for cheap and upgrade... there

 

[Cythreill]

A lot of ignorance, right here. Just FYI, some of us do more on our rigs than just game.

But hey, HT obviously doesn't do anything at all.

If the main thing you do is gaming and you don't see yourself running much multi-threaded software, then go for an i5.

We're talking about HT in the context of gaming. FYI. Was stated in the first post, and was talked about in the context of gaming in most other posts.

 

[LOL_Wut_Axel]

We're talking about HT in the context of gaming. FYI. Was stated in the first post, and was talked about in the context of gaming in most other posts.

Yes, you are.

He was talking in the overall context, so what he said is patently false. In my comment, which you just quoted, it also said that if you don't run much multi-threaded software and the main thing you do is gaming to go with a Core i5.

 

[2is]

I'm going 3770k from a Q6600 that I've had for over 4 years. It was already out for about a year when I purchased it. I plan on keeping IB around until Intels next "tock" after Haswell. The extra premium over the life of the processor ends up being a small drop in the bucket.

Now if you plan on going Haswell in 2013 then the savings may be more compelling.

 

[ninaholic37]

What I never understood is that technically, a single threaded application would utilize 12.5% of a 4 core proc... whereas with no HT.. that # would be 25%.. NeroAACEnc/MeGUI audio conversion would be an example of how this could be a pain in the ass.. thats just one instance i remember, where it was single threaded.... and a long wait..

I think as long as you have less than 5 different threads running at the same time, there should be no difference in a 4 core proc. Otherwise, hyperthreading would try to optimize the routes each process takes to "fill in the gaps" to make the best use of the 4 cores, whereas without hyperthreading it would just run the 5+ threads back and forth to the 4 cores depending on how the OS wants to assign them, instead of letting the processor itself handle the decisions for all 5-8 threads. At least... I think that's how it would work ideally, so I don't know how hyperthreading could make things worse (I suppose there would be some extra overhead work for HT on some level?)... someone can correct me if I'm wrong though...

 

[bryanW1995]

Hasn't been mentioned in here yet, but if you have any interest in distributed computing, hyperthreading is certainly worth it.

See my post above

 

[bryanW1995]

A lot of ignorance, right here. Just FYI, some of us do more on our rigs than just game:

But hey, HT obviously doesn't do anything at all.

If the main thing you do is gaming and you don't see yourself running much multi-threaded software, then go for an i5.

Did you even read my post, or did you head straight for the personal insult button? Please share with us your personal experience in going from an ht to a non ht CPU. Also, describe your overclocking statistics, power usage, heat output, and noise. And while you're at it, but a 2500k and 2600k, overclock them side by side in the same rig, and then tell us that everything is identical Btwn the 2 CPUs.

Did you miss the part where I mentioned that I run SETI 24/7? I'm right in the power alley for an ht CPU, and I'm still extremely happy that I didn't get one. Take from that whatever you will.

 

[cytg111]

I'm just wondering if in 2+ years, games will come out that make use of more than 4 cores, ...

In ten years? yes problary. The next few ? No.

Haswell and onwards, there is no immidiate plan to release more than 4 cores to the mainstream channel for years.. Therefor there will be no mainstream application to make use of more cores.

 

[LOL_Wut_Axel]

Did you even read my post, or did you head straight for the personal insult button? Please share with us your personal experience in going from an ht to a non ht CPU. Also, describe your overclocking statistics, power usage, heat output, and noise. And while you're at it, but a 2500k and 2600k, overclock them side by side in the same rig, and then tell us that everything is identical Btwn the 2 CPUs.

Did you miss the part where I mentioned that I run SETI 24/7? I'm right in the power alley for an ht CPU, and I'm still extremely happy that I didn't get one. Take from that whatever you will.

All that info is widely available on the internet.

Also, did you go head straight for the "let's make a useless, generic comments about HT not doing anything" button when you were writing your initial comment?

Sorry that you found my comment debunking what you said insulting.

 

[cytg111]

I think as long as you have less than 5 different threads running at the same time, there should be no difference in a 4 core proc. Otherwise, hyperthreading would try to optimize the routes each process takes to "fill in the gaps" to make the best use of the 4 cores, whereas without hyperthreading it would just run the 5+ threads back and forth to the 4 cores depending on how the OS wants to assign them, instead of letting the processor itself handle the decisions for all 5-8 threads. At least... I think that's how it would work ideally, so I don't know how hyperthreading could make things worse (I suppose there would be some extra overhead work for HT on some level?)... someone can correct me if I'm wrong though...

So, like, imagine a case where one core with hyperthreading is better than two cores without. Right? That is the question?

Only case i can think of is two threads working on the same data, modifies said data and that that data fits in level1.
It is problary not many real-life cases at all..

 

[2is]

So, like, imagine a case where one core with hyperthreading is better than two cores without. Right? That is the question?

Only case i can think of is two threads working on the same data, modifies said data and that that data fits in level1.
It is problary not many real-life cases at all..

Two physical cores would certainly be a lot better than 1 physical 2 logical in every scenario except for maybe power consumption.

 

[Blades]

Let's keep it simple and compare a single core running windows against a single core with hot enabled running windows. With SMP comps, back in the day when very few apps were multithreaded, a single core process that was processor intensive would only use 50% CPU in task manager - and that was fine - but it'd have been really nice to have one doubly fast CPU rather than two cpus that are mid grade. Then came along the p4 with HT - fake smp - what the hell... where's the beef when running a single threaded process? This processor is a slacker, thought I! For single threaded apps it really only does half the work...

I suppose the best way to tell would be to run a single threaded benchmark on comparable systems.

 

[2is]

50% utilization on a single core with HT does not mean you're only using half of the processing power at hand. It just means the CPU can handle another thread, it would do so at the cost of the first thread though, just as it would on a single core without HT, except the performance hit on the non HT CPU would be greater. You're still getting all of the processing power available even with single threaded apps. When it comes to HT, task manager utilization gets rather convoluted.

 

[Blades]

Ah, so its not like multi-core/SMP single threaded performace... Where on a dual core comp, a single threaded app is truly using only 50% of the cpu.. I should see an example on this in a discrepancy between the task manager and a CPU load monitor... Where a single threaded app will appear to be utilizing 12.5% cpu in task manager, but the cpu load will be 25%?

 

[ninaholic37]

You're still getting all of the processing power available even with single threaded apps. When it comes to HT, task manager utilization gets rather convoluted.

Yup, my windows 7 task manager says I'm running 609 threads. The Resource Manager says that Firefox is using 33 threads, DosBox is using 16 threads, and Wolf4SDL (a game) is using 10 threads... among other things. I wonder what all these threads in each process are doing, exactly, running in parallel with each other (?)... the deeper I look at this, the more interesting it gets.

 

[2is]

Ah, so its not like multi-core/SMP single threaded performace... Where on a dual core comp, a single threaded app is truly using only 50% of the cpu..

precisely, the issue with task manager is that it doesn't differentiate even though there's a pretty big difference.

 

[Blades]

Well, according to throttle stop - running 7z benchmark on 1 thread indicates that technically, only half a core is being used..

If I had a way to disable HT on my laptop, I would... but I'm guessing that running that benchmark with 1 thread on a quad core, HT disabled.. would yield better performance... yet slightly worse than running it on the HT enabled comp with 2 threads.. But we're talking about single threaded performance here..

 

[Cerb]

Yup, my windows 7 task manager says I'm running 609 threads. The Resource Manager says that Firefox is using 33 threads, DosBox is using 16 threads, and Wolf4SDL (a game) is using 10 threads... among other things. I wonder what all these threads in each process are doing, exactly, running in parallel with each other (?)... the deeper I look at this, the more interesting it gets.

Mostly, they sitting there doing nothing. A CPU thread is, at its most basic, a unique processor state (program counter and registers). While there may be 609 threads, all those threads are not running. Most of them are standing by, ready to be run.

It is not uncommon to set up a thread within a process to perform some specific task or set of tasks, so that it will stay asleep until another thread in that process needs it. Sometimes this can use more cores to enhance performance. Often, though, it is done for the benefit of the developers, just making for nicer code, and possibly easier debugging.

Ah, so its not like multi-core/SMP single threaded performace... Where on a dual core comp, a single threaded app is truly using only 50% of the cpu.. I should see an example on this in a discrepancy between the task manager and a CPU load monitor... Where a single threaded app will appear to be utilizing 12.5% cpu in task manager, but the cpu load will be 25%?

Open task manager, and look at the right-hand graphs in the performance tab. Each one is a logical CPU core. The icon and status bar values are the total of all available cores (thus maxing out at 100/corecount% for any single thread). Task Manager does not differentiate between physical and logical cores, but does give you info for each logical core.

 

[bryanW1995]

All that info is widely available on the internet.

Also, did you go head straight for the "let's make a useless, generic comments about HT not doing anything" button when you were writing your initial comment?

Sorry that you found my comment debunking what you said insulting.

So you have no personal experience then?

 

[pittguy578]

No need to spend that extra money for HT unless.........

You use these apps..

Premiere - video editing
Vegas - Video Editing
Photoshop - Photo Rendering
Sonar X1 - Digital Audio Workstation ..
and other HT 64 bit apps.

If you use these then YES get the 3770k ........

If you don't use any of these then get the 3570k ..... gl

I am not sure if hypethreading really provides that much additional performance in terms of 3570 versus 3770. Just looking at the benchmarks in the reviews thread Photoshop/Premire/After Effects really wasn't all that much faster on the 3770..in fact in Photoshop the 3570 and 3770 were a bout tied..and in applications that will use Intel's Quicksync there was also no difference..I think biggest difference will be 3-D rendering/FPU applications..I think it is more difficult to hyperthread encoding applications such as video/audio, at least at this point in time. Maybe in future those applications will take better advantage of it.

 

[VirtualLarry]

I am not sure if hypethreading really provides that much additional performance in terms of 3570 versus 3770. Just looking at the benchmarks in the reviews thread Photoshop/Premire/After Effects really wasn't all that much faster on the 3770..in fact in Photoshop the 3570 and 3770 were a bout tied..and in applications that will use Intel's Quicksync there was also no difference..I think biggest difference will be 3-D rendering/FPU applications..I think it is more difficult to hyperthread encoding applications such as video/audio, at least at this point in time. Maybe in future those applications will take better advantage of it.

Actually, it's quite the opposite. Hyperthreading works best, when applications are NOT optimized. Because unoptimized code has lots of little "bubbles" in the execution pipeline, and does not fully utilize the hardware resources in the processor available to that thread. HT allows a second thread to share those resources, to more fully utilize them.

Optimized code, will more fully utilize the execution resources on the core, such that there is very little left to gain for HT.

So once code is "optimized", it lessens the benefit that HT can bring. Theoretically, fully-optimized code does not benefit from HT at all, for the most part.

Edit: This is correlated by what you wrote above, with the Adobe software. The HT-enabled CPU was barely any faster than the non-HT CPU at those tasks.

 

[Cerb]

I am not sure if hypethreading really provides that much additional performance in terms of 3570 versus 3770. Just looking at the benchmarks in the reviews thread Photoshop/Premire/After Effects really wasn't all that much faster on the 3770..in fact in Photoshop the 3570 and 3770 were a bout tied..and in applications that will use Intel's Quicksync there was also no difference..I think biggest difference will be 3-D rendering/FPU applications..I think it is more difficult to hyperthread encoding applications such as video/audio, at least at this point in time. Maybe in future those applications will take better advantage of it.

I wouldn't bet on it. They've had a decade to do so, already. Making the most of HT, in a case where HT doesn't help much, will generally involve making performance of the same binary worse when there isn't HT.

If SMT helps, it's worth trying, but if it doesn't, it's not. Optimizations that leads to greater utilization of whole cores will be a better use time for them.

P.S. Ninja'd, between opening a reply tab and getting around to it.

 

[LOL_Wut_Axel]

So you have no personal experience then?

I already gave you numerous benchmarks, and what I do with my personal computer isn't any of your business.

I use HandBrake, folding@home, and Visual Studio 2010 to name a few.

 

[paperwastage]

Actually, it's quite the opposite. Hyperthreading works best, when applications are NOT optimized. Because unoptimized code has lots of little "bubbles" in the execution pipeline, and does not fully utilize the hardware resources in the processor available to that thread. HT allows a second thread to share those resources, to more fully utilize them.

Optimized code, will more fully utilize the execution resources on the core, such that there is very little left to gain for HT.

So once code is "optimized", it lessens the benefit that HT can bring. Theoretically, fully-optimized code does not benefit from HT at all, for the most part.

Edit: This is correlated by what you wrote above, with the Adobe software. The HT-enabled CPU was barely any faster than the non-HT CPU at those tasks.

I would say that there's still a lot to gain from HT in fully-optimized code.. sometimes you are waiting for resources, and it makes sense for HT (swap contexts to another thread)

resources like mem cache misses, data from network / hardware, waiting for multi-threaded program to hit checkpoint to sync..... OS/hardware stuff you can't optimize for in the code, that the OS can optimize via hyperthreading

 

[LOL_Wut_Axel]

I am not sure if hypethreading really provides that much additional performance in terms of 3570 versus 3770. Just looking at the benchmarks in the reviews thread Photoshop/Premire/After Effects really wasn't all that much faster on the 3770..in fact in Photoshop the 3570 and 3770 were a bout tied..and in applications that will use Intel's Quicksync there was also no difference..I think biggest difference will be 3-D rendering/FPU applications..I think it is more difficult to hyperthread encoding applications such as video/audio, at least at this point in time. Maybe in future those applications will take better advantage of it.