Hyperthreading vs. L3 Cache in gaming

[tviceman]

An occasional game here or there shows better performance on hyperthreading-enabled Intel CPU's (like the 6700k) vs. non-hyperthreading CPUs (6600k), even when both are running at the same core clocks.

I've seen plenty of people claiming that HT is what is powering the improvement in performance, but I've not seen that definitively proven. Given that the 6700k (and 4790k, etc.) have more cache than the 6600k (and 4670k, etc.), could the extra cache, in addition to or instead of hyperthreading, be helping the i7 pull ahead of equally clocked i5's? I'd love to run some benches myself but I neither own two PC's or any of the modern games that are showing performance advantages on the i7. Anyone want to chime in with anecdotal tidbits or even outright testing?

 

[daxzy]

This guy did a test (link below) with a G3258, E3-1231 v3, i7-5820K, disabling HT and all but 2 cores and clocked the same. The most direct comparison would probably by the E3-1231v3 and the G3258, as they use the same board and memory configuration (i7-5820K uses the X99 with 4 DIMM channels).

Basically helps for Arma-3, not so much for BF4. Probably depends on the engine (since we're talking about games) and how much crap is thrown on the screen (Arma-3 seems to be endless, but that may just be coincidental).

https://www.youtube.com/watch?v=hTQssyGvAIk

 

[BSim500]

An occasional game here or there shows better performance on hyperthreading-enabled Intel CPU's (like the 6700k) vs. non-hyperthreading CPUs (6600k), even when both are running at the same core clocks.

I've seen plenty of people claiming that HT is what is powering the improvement in performance, but I've not seen that definitively proven. Given that the 6700k (and 4790k, etc.) have more cache than the 6600k (and 4670k, etc.), could the extra cache, in addition to or instead of hyperthreading, be helping the i7 pull ahead of equally clocked i5's? I'd love to run some benches myself but I neither own two PC's or any of the modern games that are showing performance advantages on the i7. Anyone want to chime in with anecdotal tidbits or even outright testing?

What you need is a test where all CPU's are clocked same rate and where the i7 is tested vs an i5 both with and without HT enabled. Like this:-
http://techbuyersguru.com/best-gaming-cpus-pentium-vs-core-i3-vs-core-i5-vs-core-i7?page=1

As you can see it all depends on the game. Crysis 3 is way up on an i7 vs i5, whilst Tomb Raider (2013) and Bioshock Infinite are barely any different i7 vs Pentium (not just avg but also min fps). Crysis 3 shows +43% min fps boost (50 vs 35fps) from HT (i7 with HT vs i7 without HT), but no real difference i7 without HT vs i5 (essentially testing only the cache difference).

Also bear in mind that outliers like Crysis 3 aside, the effect of Hyper-Threading is generally lower on i7 vs i5 comparisons than on i3 vs Pentium in games that use 3-4 threads as such games may load up only the real cores on i7's without touching the HT, yet load 2C/4T cores 90-100%. I've seen games where i7's benchmark the same as i5's despite +2MB cache and HT, and then there are other games where an i3-4170 (3MB) gets literally double the min fps (+100% gain) over a Pentium G3258 (with same 3MB cache) purely from the HT.

Overall, I'd say HT makes far more difference than cache size. This is also visible on i3-4170 vs i3-4360 (3MB vs 4MB cache both at 3.7GHz) often makes little difference in gaming. Another example link, i3-6100 (3.7Ghz, 3MB) vs i3-6300 (3.8GHz, 4MB) vs i3-6320 (3.9GHz, 4MB) is already slim enough (up to +8% difference for +5.5% clocks), but if you were to adjust clock speeds to matching 3.7Ghz, that gap would barely be +3% difference for +33% extra cache. Switch off HT though and fps would fall significantly more.