Is HyperThreading placebo?
I notice no difference with it on or off.
>>51821146
for some tasks it improves the speed, for some, it gets worse
>>51821146
Hyperthreading for modern games has no effect, yet. Future games will start to make use of the extra threads, which of course makes your cpu more future proof. Also hyperthreading offers less stuttering during high demanding games like witcher 3
Cannonlake will come without HT.
So yes, it's a meme-tech.
>>51821402
It actually isn't. Please back up your point.
>>51821162
I was fooling around with python hyperthreading and processes to chop up and run a large batch of tests. Creating 8 threads and assigning a workload to them didn't seem to utilize all 8 virtual cores.
I have no idea how to create threads and have hyperthreading utilize them, didn't dive much into it. Perhaps I was doing something wrong at the time and all threads created were sequential instead of concurent.
If I created 8 processes however, then hyperthreading seems to use all 8 virtual cores and finish a lot sooner.
But setting up queues for interprocess comminication was a bitch.
>>51821592
It's been a long time since I studied operating system theory, but I believe the threads have to do different things that the processor figures out it can run in parallel.
Sure it is.
HT can be useful only when code is antioptimized for normal cores
In most cases, yes. Hyperthreading is just a CPU core pretending to be 2 cores. In the end you still have 1 real core that does work.
>>51821146
Depends on if you can actually utilize those extra cores. Asking that is pretty much the same as asking if there's an advantage to a turbo charger. Going around in inner city traffic it it doesn't help very much, but if you're doing a lot of fast accelerating and specially if you're dragging heavy loads on highways it does help quite a lot.
I've done quite a lot of multi-threaded scientific computing and in my experience the addition of computing on more threads doesn't provide as big of a boost after you've run out of physical cores than before you've run out of physical cores, but it definitely speeds things up.
>>51821592
>python
https://wiki.python.org/moin/GlobalInterpreterLock
>>51821146
I have a i7-3517U and i do a lot of stuff that uses cpu, and i kinda feel like it was worth the extra money
It allows unused execution units on a cpu to function as a virtual independent cpu. It's like 1/5 the performance of additional real cores with none of the power and cost downsides of additional cores.
So, pretty nice imo.
Gaming proves it works. i3's are very close to i5's in benchmarks in most games.
HT is like out-of-order execution, except that it can schedule a completely different thread instead of just reordering instructions. Sometimes the CPU sits idle because it's already reordered all of the instructions that it can, and it's still waiting on data or something else. HT allows another thread to be ready to run while the first is waiting.
>>51822547
>Gaming proves it works. i3's are very close to i5's in benchmarks in most games.
Idiot doesn't understand what he's talking about.
>>51822515
good goy
It's useful if you're using a program that makes use of it.
I've done some photo rendering of 3d models with a program that makes use of hyperthreading, and it helps speed things up a good deal.
Hyperthreading is alright for running lots of different processes at once. However, if you try to split up a single, homogeneous task into different threads, it won't help. So it's pretty useless for actually speeding up a single application, like a game. CMT like in the FX CPUs is a bit better since each core has multiple of certain resources, but it's still not the same as having an entire other core.
>>51822547
That's because most games don't even use 4 threads to begin with.
>>51821146
Depends on the task being run. Single threaded programs see no boost, and programs that can thoroughly saturate the pipeline would see at best no performance boost and at worst a per thread performance regression.
Applications that dont saturate the pipeline (data heavy programs, programs with a fuckton of branching and/or bubbles that come about from execution) however can see a significant boost, as while one thread is waiting for data to arrive from main memory, the other thread, who's data is likely already on the CPU, can run.
>>51823251
>However, if you try to split up a single, homogeneous task into different threads, it won't help. So it's pretty useless for actually speeding up a single application, like a game.
It actually does work well in (most) games, specifically because they are *not* homogeneous. A bunch of unrelated threads are spun up for different tasks and it does help a lot that more of them can efficiently share cores without context switching.
>>51822486
python is fucking trash for this reason
>>51821146
> Is HyperThreading placebo?
Not entirely, but you could consider it a workaround for just doing Windows business workloads better.
On Linux the difference between HT and simply having Linux schedule tasks is usually... minimal. Very minimal. Unless you're explicitly producing benchmark cases, you can't get to 5% difference or so.
its a server thing
>>51823660
It's just as much an interactive desktop thing.
>>51821146
>Is HyperThreading placebo?
You can easily answer your answer by searching on your search engine of choice what it is and reading the relevant results
>>51823695
>answer your answer
meant answer your question
>>51823251
I'm referring to games that use more than 2 cores. There are hundreds of benchmarks to back up my claims. Celerons and pentiums do poorly and the i3s can handle ir.
>>51821354
>hypterthreading
>future proof
>gimped compared to actual cores
if you want to future proof you buy a broadwell-e
>>51821146
Your cpu figuratively has two pipes in, some instructions are so fast the bottleneck happens from the input pipe, so hyperthreading was invented by intel.
HT is useful in improving the throughput of heavy fetch/load oriented code since it presents an extra register store that the memory controller can fill while the active thread is busy with executing its time slice. When that time slice is over, the core doesn't need to stall while waiting for a context switch to finish, it can just swap registers and restart the execution pipeline.
But this is only useful when a context switch is extremely expensive. In systems with a simple cheap thread concept the performance gains are smaller.
At the end of the day it's not actually an increase in performance as much as a decrease in wasted silicon time.
>buy an i7
>turn off hyperthreading
>don't overclock it
>>51824015
>>turn off hyperthreading
That's actually a good idea in some cases. An i7 has more cache to better facilitate HTT, meaning that an i7 without HTT will outperform an i5 clock for clock.
There are also a few edge cases where programmes will refuse to run on a virtual thread, making it crash. This is extremely rare though.
It's a thing mostly because intel is jewing big time, refusing to put in more cores simply because AMD is so far behind.
But no, hyperthreading works, but it's nowhere near another core.
>>51824119
Activision's game Prototype does not like hyperthreads and crashes on startup.
While the cache differences between Intel chips plays a factor you can see that hyperthreading does not give linear scaling - far less in fact.
Compared to AMD's CMT approach where you typically do get linear scaling as cores increase, even with the shared resources.
>>51825092
I don't see the difference
>>51825166
An 8350 has double the score of a 4300 as it has twice the core count. A 4770k does not have twice the score of a 4670k despite presenting twice as many threads to the OS. Thus hyperthreading - while clearly giving gains - is not as effective as having my physical cores.
>>51825166
AMD's module design actually has more base resources to execute from (2 sets of integer pipes and the FPU can run 2 separate float operations at once even though its slow about it), whereas intel simply adds some things to keep track of architectural states. The other available resources remain the same.
With AMD, going from 4m 4t to 4m 8t produces a noticeable improvement in performance for the tasks that can use it, as there is actually an entire set of integer cores now available for compute that has minimal interference on the other half of the module.
With Intel, going from 4c 4t to 4c 8t produces a significantly lower performance boost, as the second set of threads is executing on the same hardware, and if the first set of threads is tying up cores, the second set has to sit there fiddling their thumbs while they wait for resources. And if all 8 threads are hammering the cores, total throughput can actually drop compared to running just 4t.
>>51825219
But it's not linear.
FX-6300 from FX-4300 is 2 additional cores of 150% increase in cores yet the increase in performance is only 134%
>>51825294
That is due to the differences in clockspeed - the 6300 runs 300mhz slower, thus the difference. At identical speeds it would be linear.
Intel's implementation of SMT yields around a 30% scaling advantage.
30% more multithreaded performance from that additional logical thread.
>>51825312
So a 7.5% difference in clock speed decreased performance by 16%?
>>51825454
The pictures came from this review, so make of it what you will.
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/62166-amd-fx-9590-review-piledriver-5ghz.html
>>51823161
Who cores?
>>51821354
>yet
What makes you think games can me much more threaded? I'm not saying they wont.
>>51821146
it helps on some specific benchmarks, made specifically for hyperthreading. unnoticeable jewing otherwise
>>51821354
>Hyperthreading for modern games has no effect
Fucking retards who have no idea what they're talking about are always spouting this epic meme when it is demonstrably not true. Unless you have another explanation for a stock 2600K outperforming a 4670K? Or the 12fps difference between the latter and the 4770K? Didn't think so, fuckstain.
>hyperthreading offers less stuttering during high demanding games like witcher 3
The Witcher 3 isn't CPU-intensive at all you stupid, worthless fuck. Kill yourself.
>>51828728
Not him but not all games are that well threaded. Be honest, that's a cherrypicked example, isn't it?
Also, you're a very angry person, aren't you?
>>51828632
DX11 really cannot into more than 4 threads as its essentially DX9: turbo edition. DX12 which is more or less built off mantle (along with vulkan) can into threads which is great for modern cpus.
>>51828728
Witcher 3 is a graphical powerhouse for sure, but on the cpu side its not as fantastic as you would think. Ironically the best game for cpu usage in recent times is Tomb Raider simply because it gives zero fucks about cpu in general - an athlon is as good as a 4770k in that game (and the game scales really well with gpu horsepower).
>>51828819
True that might help a lot. Even though I was talking about threading game logic, which runs on the CPU. But yeah, if DX12/Vulkan manages to saturate the threads then it's 100% effective on achieving many-core multithreading.
>>51828920
Even if DX12/vulkan removes driver overhead (which is their primary goal) then thats is a huge amount of extra cpu resources available for vidya. You don't need to be thread aware to see benefits if you aren't cockblocked at every turn.
you know that you can just turn off HT ?
Just run a few benchmarks and see for yourself.
The last time people did some in depth shit on this was like a decade ago,
from what I remember HT had a positive impact over all, although some applications suffered slightly for it.
Not sure what todays landscape looks like as we have come a long way since then.
>>51828943
Logically, but we were specifically talking about increasing the number of threads a game can use.
>>51829006
>The last time people did some in depth shit on this was like a decade ago
There are benchmarking sites that do this for all their benchmarks.
The result is a mixed bag. Some programmes are threaded well enough to perform drastically better than others when they're allowed the additional threads. Others don't care about the number of threads but enjoy the extra cache given when presented with an i7 without HTT, though the performance difference between the i7 and an i5 at the same clocks is negligible.
Overall, the sanest thing is the same as it's always been; Get an i5 and save your money for something else unless you have something you know runs really well on extra threads.
It's not a placebo at all. It does exactly what it's supposed to do which is to increase performance in multithreaded applications.
Is the performance difference between for example an i7 4770k and an i5 4670k huge? No. Those 4 extra threads will at best give you 25% better performance only in multithreaded applications. You will get little to no extra performance in anything that uses only up to 4 cores and 4 threads. It's as simple as that.
In my opinion, if you use multithreaded applications for work, you're better just saving up and getting a 6core, 12 thread or even an 8 core, 16 thread CPU.
>>51821354
people have been saying this for years now.
>>51823905
>buying the last CPU to support the outdated X99 platform
>future proof
>>51829390
Should I bother to get a 5930k from IREP ?
I could sell it down the road with no loss, just not sure if its worth typing up money until the the summer deal.
>>51829488
bumping for opinions.
Not sure what the summer deal is gonna be , maybe finally some skylake cpu..
>5930k for 194.00 or wait until next year
>>51830053
6 cores cpu > 4 cores cpu. Especially now that Win10 is being shoved up our asses by MS.
>Future proof
In 5 years time you'll be buying the next best thing anyway, especially if you're a gaymer. Don't try to kid yourself into thinking your hardware will last even close to a decade unless you practically do nothing but browse the internet in which case future proofing is a joke.
>>51831399
Also and unless AMD somehow overtakes the markets your motherboard will be useless when you go to upgrade your jewtel since they change sockets all the time
>>51830579
looked at the cost of mobos for the 5930, wow they are costly.
Even with the discount price, the DDR4 requirement is really gonna push the price up.
Not sure if its worthwhile getting a system I might be ditching in the summer time.
>>51821146
slit your throat imbecile
>>51831416
That's not entirely true, an overclocked core 2 quad will still perform adequately. Still,, you'd be expecting to turn down a few settings.
>>51831399
AMD really aren't that much better. If you look at the pattern the past few years, Intel have supported two generations of CPUs per motherboard on the consumer versions. AMD haven't been that much better with FM1 only supporting Llano, FM2 only supporting Trinity and Richland and FM2+ supporting Trinity and Richland but being required for Kaveri. The Phenom/FX side of things is far more messy with the chipsets working but the sockets being slightly different and manufacturers not exactly jumping on board. This has given us wonderful motherboards with HD 4000 integrated graphics and the 760 chipset, which was so outdated at time of launch that no one in their right mind should have bought them but they made up the entirety of the FX matx mobos.
