where does hyperthreading help except rendering?

[Phoenix70]

Can it help with multi gpu bottlenecking, or is that an architectural/bandwith thing? ive heard it is nearly pointless for gaming. maybe multi tasking? thank you for reading and excuse my dumb question

[qwertywarrior]

http://www.techpowerup.com/reviews/Intel/Core_i5_4670K_and_i7_4770K_Comparison/8.html

[AlwaysFSX]

It's nearly pointless for gaming at this moment in time but DX12 should fix that by putting equal load across all threads, thereby reducing frametimes.

[Enderman]

Hyperthreading helps in multitasking, computing, servers, rendering, virtual machines, and things that use many cores. 99% of games use 4 cores or less so having 8 threads doesn't improve anything. BF4 uses 8 threads, so you get a performance boost with the 4790k over the 4690k

 

Multi GPU support is handled by the chipset controller, not CPU.

[jmaster299]

It's not pointless for gaming, at least not any more. Windows also takes full advantage of it, and multitasking like you said. It will be more important going forward though, making the $100 price premium worth it.

[The_Strict_Nein]

Hyperthreading helps in multitasking, computing, servers, rendering, virtual machines, and things that use many cores. 99% of games use 4 cores or less so having 8 threads doesn't improve anything. BF4 uses 8 threads, so you get a performance boost with the 4790k over the 4690k

 

Multi GPU support is handled by the chipset controller, not CPU.

 

Yes, but seeing as how the CPU has to handle the render calls, if you have more render calls due to increased card power, the CPU needs to be able to keep up, as well as handling the physics and what else.

 

This will be greatly improved in DX12, however. Hopefully.

[i_build_nanosuits]

it helps in anything that use more than 4 main threads...many games such as BF4, crysis 3 and watch dogs make good use of it.

[vm'N]

Hyper-threading (SMT): Is having two threads (line of code) been process simultaneously on a single core. Both threads are dynamically sharing the resources on the core. This means that if both threads are heavily depended on the same resource, it might starve on of the threads (non are giving priority over the other. It is literally first come first served).

To utilize hyper-threading there is only one requirement; You have two process (two threads). However this isn't an improved utilization.

To fully improve the the gains of hyper-threading, you would need to optimize your code for it. You would have to be carefully with your thread- and resource-management. Try not having threads been depended on the same resources. It is all about the TLP (Thread-level parallelism).

In the server environment the software are optimize for SMT (hyper-threading), hence why so many processors (not only Intels) feature SMT (Hyper-threading is intels way of implementing SMT).

IBM have a processor which feature 8-way SMT. This means that a single core can simultaneously process 8 threads at once.

So SMT can help in MANY cases, however it needs a special optimization.

[vm'N]

Can SMT help not bottlenecking the GPU. Yes.

First we need to know what causes a the processor to "bottleneck" the GPU.

In the computer world there is a first-class citizen and second-class citizen. The first-class citizen is the CPU. And second-class citizen is coprocessors like the GPU.

This means that if you want to do any kind of processing on a second-class citizen, it would need to be directed from the first-class citizen. This is essentially done by drivers (You don't need drivers for the CPU, as it is a first class citizen). Drivers are software that allows the CPU to speak to the coprocessor. For gaming and utilizing the GPU, the CPU needs to send some API calls to the GPU (It essentially sends what it needs to process and where the elements is). The dx11 is uses only a single rendering thread (also dx11 is a high-level API (easy to program for, and works all around)). So for high-level APIs needs to be translated further into machine code (The higher the API, the more the translation), so you are wasting resources on nothing.

Mantle will be a lower-level API (needs less translation), and will have a better multi-threaded rendering thread.

Once a processor are already fed up doing tons of calculations and physics and whatnot, it might not have resources enough to do a constant API calls (so it might either skip or delay it).

So, to simplify things:

The GPU is lazy. The CPU is the general officer. The general officer needs to constant yell at the lazy to make him do things. If the general officer don't yell, the lazy don't work.

So, when the CPU is skipping or delaying a API call, the GPU don't do anything meanwhile causing a drop in FPS.