By: Exophase (exophase.delete@this.gmail.com), May 14, 2013 8:10 pm
Room: Moderated Discussions
RichardC (tich.delete@this.pobox.com) on May 14, 2013 5:54 pm wrote:
> Here's one: http://www.overclock.net/t/671977/hyperthreading-in-games
>
> I wouldn't expect 5% core clock increase to give a 5% overall speedup.
> Too much stuff depends on DRAM. And of course for gaming the GPU is very
> much involved as well.
>
> However, by "significant increase" I didn't mean to imply as much as 5%.
> Just that HT vs non-HT was a nothingburger, and 3.8GHz vs 4.0GHz was a
> detectable and consistent speedup ("significant" more in the statistical
> sense of being outside the range of random variation).
>
That test predates even Sandy Bridge. I don't know exactly what situation is with CPU load over thread count is in current games but I know for sure that it has been trending higher over time, and the results from HT have improved from Nehalem to SB.
Check out this recent review:
http://anandtech.com/show/6934/choosing-a-gaming-cpu-single-multigpu-at-1440p/5
In each game the X4-960T shows instances of beating the X2-555 by a higher margin than a 3.4GHz turbo clock vs 3.2GHz fixed clock would allow (and that turbo couldn't be reached while all cores are active). Often a much higher margin. And where more cores benefit more threads usually benefit as well. So at the very least there's evidence that > 2 physical cores is an asset to at least some selection of current games. Since many sold Intel CPUs are of the 2C/4T variety this gives a pretty good justification for HT with games.
> Here's one: http://www.overclock.net/t/671977/hyperthreading-in-games
>
> I wouldn't expect 5% core clock increase to give a 5% overall speedup.
> Too much stuff depends on DRAM. And of course for gaming the GPU is very
> much involved as well.
>
> However, by "significant increase" I didn't mean to imply as much as 5%.
> Just that HT vs non-HT was a nothingburger, and 3.8GHz vs 4.0GHz was a
> detectable and consistent speedup ("significant" more in the statistical
> sense of being outside the range of random variation).
>
That test predates even Sandy Bridge. I don't know exactly what situation is with CPU load over thread count is in current games but I know for sure that it has been trending higher over time, and the results from HT have improved from Nehalem to SB.
Check out this recent review:
http://anandtech.com/show/6934/choosing-a-gaming-cpu-single-multigpu-at-1440p/5
In each game the X4-960T shows instances of beating the X2-555 by a higher margin than a 3.4GHz turbo clock vs 3.2GHz fixed clock would allow (and that turbo couldn't be reached while all cores are active). Often a much higher margin. And where more cores benefit more threads usually benefit as well. So at the very least there's evidence that > 2 physical cores is an asset to at least some selection of current games. Since many sold Intel CPUs are of the 2C/4T variety this gives a pretty good justification for HT with games.
