Relevance of HT

[UltraTaco]

Howdy everyone! Out of curiosity, doing some testing(again) whether or not to keep HT on or off...

Anyways, during video encoding, there is 15% boost when HT is enabled *BUT* only when you don't have any transitions, effects, and doing same codec. Once you add any sort of effects or anything extra or going to a different codec, all that HT advantage is out the window. In fact, it's even a bit slower compared to HT off.

I have an i7 920 cpu, so has anything changed by now in regards to HT? Or does it still only work well when job is streamlined? Has it become more efficient or "smarter"? I remember from the Pentium4 days that HT wasn't very ideal..

Weekend evening discussion I guess...
Fire away I'm fairly green in all this, so any schooling is welcomed.

 

[trents]

I can't say for sure but I would almost put money on the thought that the implementation of HT has improved significantly since that first generation. If nothing else, you know the instruction sets have progressed.

 

[||Console||]

Isn't HT just a way to better schedule tasks for the CPU ?

 

[bmaxa]

Yeah, with HT you can ran twice tasks without OS context switches, which translates to better performance. Other then that I noted that when I run WCG on all 4 cores and TW: warhammer in parallel game runs smooth, while with HT disabled there is chopping.

 

[mackerel]

I haven't seen much of a difference in functionality in HT since it was put into a P4, and even AMD's SMT is very similar. Compared to without, it gives up to 50% boost in compute throughput. Note however that 0% is also very possible. Some tasks just don't benefit from it at all. In those cases, the extra management overhead from it being there may cause a slight degradation. Also, there are potential complications if software uses a limited number of threads fewer than the maximum and how those are assigned out by the OS. More modern software is more likely to better benefit from it.

 

[UltraTaco]

I remember "lock on"-pc game, kept crashing randomly until I set single thread affinity in task manager. Also FSX would run choppier with HT on. That thing was an elephant though. Never enough ghz. I am a little disappointed in my results because basically anything more than just simple encode means rendering will be slower on HT.

Like mentioned above, New set of instructions perhaps straightens things out better for HT on later versions of cpu for video rendering

 

[EarthDog]

Software taking advantage of it is a thing as well... typical improvements are on the order of 30-50%

 

[Dolk]

HT was originally an attempt to exploit unused parts of a core while multiple tasks were running on a system. Back when a single core or two cores was all you could fit on a chip, you could open up some room by adding a bit of cache and wires inside the core to transition between a couple different tasks that were being pushed through the pipeline. For example, if a floating point multi-threaded instruction was awaiting for another core to finish up its calculation, it could go ahead and do a memory access instruction or something similar. Later it has evolved to much more complex tasks as transistor size has decreased. Now you can issue those two commands at the same time and the core will understand how to prioritize. HT is less firmware/software and more on the micro-code to take advantage of instruction that are placed in the queue. Not to say that firmware/software can't optimize the usage further, its just that you typically don't see a negative gain when software is utilizing HT these days.

In OP's case, the after affects and the like require more utilization in parallel tasks such as floating point and memory calls because after affects are re-samples (think Anti-aliasing or SSAO in GPUs). This means the core pipeline is more filled, and the instructions cannot be exploited by the HT micro-code. Would a newer processor do better? Most likely but I can't say to what degree.

 

[UltraTaco]

Thank you for very thorough explanation, sir. (and everyone else)