CPU Hyperthreading and Gaming: Real-World Benchmarks

[wand3r3r]

I posted a link in GPUs but the correct place is probably in CPU. Being this thread is being resurrected I'll post it here.

http://www.techbuyersguru.com/CPUgaming.php

This was an interesting analysis on dual vs quad with and without hyper threading.

Maybe we need a thread to combine all the good info.

 

[BrightCandle]

Part of the problem here is the wide difference in GPUs. Gamegpu.ru is using a 690. They have considerably more performance on offer than a lot of the other tests and thus I think they end up in CPU limited scenarios more often than some of the other tests. Not reading Russian I also can't ascertain the quality of the testing they do, whether they run many runs or really what areas of a game they are using. I accept the limitations in the data they present, but there is no one else with such a wide set of games with multiple CPUs. What it does give us is a picture into which games are worth investigating.

But bare in mind that CPU usage is a very very poor marker for how much CPU a game uses. We don't yet have a tool that shows thread level information for CPU usage that we can use for determining the amount of parallel execution.

There are a lot of variables and sites aren't comparable to each other when they use such widely different hardware.

 

[Shmee]

Very interesting thread, and for those who have done the testing great job!

One interesting conjecture that comes to mind when reading this is that as games and software progress to take advantage of the multiple threads, even existing and older hardware will actually see improvements, and even if it becomes obsolete, it may actually do better than it did earlier as long as the program's increase in resource demands don't exceed the benefit of more optimization for more threads.

 

[bjd223]

UPDATE: For more info on this topic, see a follow-up article I wrote here: http://www.techbuyersguru.com/CPUgaming.php

I just want to say that you should clarify the difference between HT and physical cores. The OS can not tell the difference between physical and virtual cores. Games can not request to run on physical cores as appoased to virual cores. People on the net are incorrectly asking if games support HT when they should be asking if games are multi-threaded.

Most poeple agree that a very fast dual core CPU is better than a slow quad core CPU for gaming. This falls in line with that assessment. Does HT a dual core have less game performance than a non HT quadcore?

Anyway HT doesn't really have anything to do with it, it is more about whether games are multi-threaded past 2 threads.

 

[zir_blazer]

I just want to say that you should clarify the difference between HT and physical cores. The OS can not tell the difference between physical and virtual cores.

It used to be like that with WXP, supposedly was improved with Vista and 7 Task Schedulers. After all, the OS should be able to check if the CPU supports Hyper Threading (Via a bit flag or something), then count that for each pair of logical CPUs, first is physical, second logical, and it should always fill first all physical CPUs before logicals. If it doesn't work like that already, Windows is still living in the stone age.
There was also a more complex Task Scheduler for Bulldozer that was supposed to put two Threads from the same application in the same module so they can share the Cache and use a higher Turbo that if they are on two separate Modules (Which I think was a Windows 8 feature, not 7).

 

[Sweepr]

Interesting results. I've done some testing with my old i7 920 a while back. GTA4 performs better with 2 active cores (HT on) than 2C & HT off (10-15% higher average FPS). 4C/4T vs 4C/8T didnt show any meaningful difference though.

 

[SPBHM]

Interesting results. I've done some testing with my old i7 920 a while back. GTA4 performs better with 2 active cores (HT on) than 2C & HT off (10-15% higher average FPS). 4C/4T vs 4C/8T didnt show any meaningful difference though.

even on my i3 GTA 4 was not one of the games that used HT the most...
the best games to see huge gains with HT with dual cores that I've seen so far are from Codemasters, like F1 2012, Dirt 3... Battlefield 3 also seems to love HT (with only 2 cores)

 

[Maximilian]

I just want to say that you should clarify the difference between HT and physical cores. The OS can not tell the difference between physical and virtual cores. Games can not request to run on physical cores as appoased to virual cores. People on the net are incorrectly asking if games support HT when they should be asking if games are multi-threaded.

Most poeple agree that a very fast dual core CPU is better than a slow quad core CPU for gaming. This falls in line with that assessment. Does HT a dual core have less game performance than a non HT quadcore?

Anyway HT doesn't really have anything to do with it, it is more about whether games are multi-threaded past 2 threads.

Are you posting from 2007?

Im pretty sure the OS can tell the difference, vista couldn't, win 7 and up can. Updated vista might be able to, i don't know.

Slow quad > fast dual, this has been proven by all the people still with Q6600s after all this time, people have had to ditch their E6xxx or E8xxx long ago.

 

[AtenRa]

Slow quad > fast dual, this has been proven by all the people still with Q6600s after all this time, people have had to ditch their E6xxx or E8xxx long ago.

That is missleading, if you wouldnt OC then at that time you would have lower Gaming performance, you would spend more and you would ditch the Q6600 sooner than the cheaper and higher performance dual cores.

Everything changes if you would OC that Q6600 to 3GHz+ but then again, you would spend more and you would have to keep it longer in order to justify the extra investement you made.

Those that chose to go the dual core way at the time, they spend less and had to upgrade soon. Thats how things work even today

 

[BTRY B 529th FA BN]

The UE3 engine will benifit from HT in some cases.