Overclock.net banner

Hyperthreading In Games

101K views 43 replies 34 participants last post by  IBooNI 
#1 ·
So you may be wondering, does hyperthreading help or hurt gaming performance? Yes, we've all read various opinions, like "latency penalties", "games are optimized for 2/4 cores only"... and my favorite: "you paid for it, why not use it?" argument. But despite all of these well-informed people, I couldn't find any tests or data to back these claims. So I set out to do some testing of my own. Originally it was just a few benchmarks for myself, but I quickly realized it would do more good if I shared my findings.

I had two goals:
1) 4ghz HT vs 4ghz without
2) 3.8ghz with HT vs 4ghz without
The first is a straight-up comparison of with and without hyperthreading: which pulls higher fps?
The second is to address the majority of i7 920 owners: on air, overclocking limited primarily due to heat. Should they use a lower clock and hyperthreading, or higher clock and disable hyperthreading?

Test system: i7 920, Classified E760, HD5870 stock clocks, 6gb ddr3 @ 1552mhz cas 7, win 7 x64, P64 Corsair ssd. Cpu clocks were 191x20=3810mhz HT, 191x21=4000mhz HT, and 191x21=4000mhz no HT.

Update 2/18: Updated with a Crysis, Fallout, and another Far Cry test.
Update 2/19: Per user request, I did a GTA IV benchie at 2.67ghz, and got the same fps as the 3.8 and 4ghz runs...
So im scrapping all GTA data in favor of a custom walkthrough. Thanks to grossebeaver for the suggestion!
Update 2/20: New GTA and Dragon Age results are up, along with user requested stock runs of GTA and Dragon Age.
Update 4/26: Back by popular demand! Updated with a Bad Company 2 and L4D2 results.
--------------------------------------
***Bad Company 2***
Settings: 3x @ DX 11, 1680x1050, 1xMSAA, 1xAnisotrophic, Detal/Texture/Shadow= medium, Effects= high, HBAO= off

The new smash hit from DICE. The run was the first vehicel chase sceen, with plenty of explosions and AI.
Goal 1) Although averages are the same, non-hyperthreading recieves a 2fps boost. Add a point for non-HT.
Goal 2) Lower clockspeed manages to match average framerates, but falls 4 fps short in the minimum catagory. Higher clocks takes the gold.

***Crysis***
Settings: "cpu test" 3x @ DX 9.0, 1680x1050, AA 0x, medium preset with physics= very high

The pc gaming standard, Crysis. Don't need to say much else.
Goal 1) Although average framerates are within 0.6fps, HT manages to pull ahead with 3.4fps minimum improvement. A win for HT.
Goal 2) The extra 200mhz wins by a landslide, most notably with the 10fps minimum increase. Solid win for higher clock here.

***Crysis Warhead***
Settings: Ambush 3 runs @ dx9, 1680x1050, 0xAA, Gamer preset, physics=enthusiest

Yet more Crysis goodness.
Goal 1) Identical averages, but non-hyperthreading pulls a 3fps minimum lead. Very small difference, but non-hyperthreading wins here.
Goal 2) Again, identical average fps scores, but raw clockspeed wins with a 7fps minimum gain.

***Dragon Age: Origins***
Settings: 1680x1050, Graphics and Texture= High, 0xAA, no Frame-Buffer effects

Goal 1) HT shows a 5fps minimum improvement over non-hyperthreading, as well as 2fps average. HT takes the gold here, although at those framerates you cant tell the difference.
Goal 2) Dragon Age clearly benifits from higher clock speed. Non-hyperthreading ftw.


Identical settings as regular DA:O above, except with stock clocks

***Fallout 3***
Settings: DX9, 1680x1050, 0x AA, 0xAF, Medium preset, object/actor fade= max

A 2 minute walk from vault 101, through Springvale, ending with overlook of city in the distance... measured with fraps.
Goal 1) Scores too close to declare a winner here. Its a tie.
Goal 2) Another huge victory for raw clockspeed. Non-hyperthreading is a clear winner.

***Far Cry 2***
Settings: Ranch small test, 3 loops @ DX9, 1680x1050, 0xAA, fire/physics/trees= very high, everything else= medium preset.

Very close.
Goal 1) Hyperthreading sees a 0.05% increase in average fps. Conclusion: identical performance.
Goal 2) 2fps average and 1.5fps minimum increase with higher clock. Negligible difference considering these are in the +100fps range, but non-hyperthreading technicly wins.

Settings: Ranch demo, 3 loops @ DX9, 1680x1050, 0xAA, fire/physics/trees= very high, everything else= medium preset.

The ranch demo test features more fire, explosions, shooting, and AI npcs.
Goal 1) Very small, but HT takes 1st place with the 2.3fps lead over non-hyperthreading. Chalk up another HT win.
Goal 2) Although HT sees a 1.3fps minimum gain, it also looses by 0.8fps average. Too close to call a victor here (tie).

***GTA IV***
Settings: view=50, detail=70, 1680x1050, texture and render=medium.

New and improved walkthrough: the first mission (drive cousin to his place), measured with fraps.
Goal 1) Suprisingly, non-hyperthreading pulls some impressive numbers, up by 4fps in both average and minimum catagories. Cores ftw.
Goal 2) Another hands-down win here, the extra mhz is what GTA likes.


Identical settings to regular GTA IV test above, but with stock clocks

***Left 4 Dead 2***
Settings: 3x @ 1920x1200, 0xAA, Shader/Texture= Medium, Effects= High, Paged Pool Memory= High

This test was pretty fun. I loaded "The Parish" level, turned on God mode, infinite ammo, and grenade launcher. I then called 3 horde/panics, recording each with FRAPS.
Goal 1) HT gets a 2.5fps average lead over non-HT, but looses by 5fps minimum. Tough choice, but I feel non-HT recieves another point.
Goal 2) Its pretty clear the extra 200mhz over HT is the way to go.

***Oblivion***
Settings: 1920x1200, 8xAA, Bloom lighting, Textures=Ultra High, View/render distences=max, Quarl's HiRes texture pack mod

There was no timedemo for oblivion, so i devised a 45-second run through trees, water, and a view of the Imperial City.
Goal 1) Hyperthreading gets a 3fps average and 11fps minimum boost. Nothing interesting, but HT wins this battle.
Goal 2) Again, hyperthreading manages a 2fps average and 5fps minimum increase, despite the lower clock speed. Another win for HT.

***S.T.A.L.K.E.R: Clear Sky***
Settings: Default preset, full dyn. lighting, 0xmsaa, 1680x1050

Although it was a bit buggy, the stand-alone stalker benchmark program is a great idea. I also noticed that this game only utilizes one core...
Goal 1) Hyperthreading looses by 1.7fps average. Again, very small difference, but non-hyperthreading wins.
Goal 2) Hyperthreading sees a 4.4fps average and 11fps minimum loss to its higher-clocked opponent. Solid win for non-hyperthreading.

***3DMark Vantage***
Settings: Performance preset

Goal 1) Identical gpu test scores, but omg look at the AI and Physics scores! Wow! Hyperthreading hands-down winner.
Goal 2) Exact same result: gpu-wise they're the same, but the AI and Physics tests blow non-hyperthreading out of the water.
---------------------------------------

Goal 1 Conclusion: It seems there isn't any real pattern between hyperthreading and non-hyperthreading. They trade blows and all results are very close, but if one had to choose a winner, turning HT off seems to yield slightly better framerates. But for simplicity's sake, the verdict is: Hyperthreading neither helps nor hurts when gaming.

Goal 2 Conclusion: Excluding the Vantage results, an extra 200mhz over hyperthreading is always the better choice. Not surprising, as HT in goal 1 showed no real advantage or disadvantage. If you couldn't guess it, the verdict is: Always choose a higher clockspeed over hyperthreading for gaming.

Thank you for watching! I may come back with a Crysis 2 benchmark when its released, but based on what we've seen so far, don't expect anything ground-breaking
 
See less See more
16
#2 ·
Wow! absolutely steller guide sir, One big rep for you!
Now post a link in your sig.
I was pretty shocked by the results myself, I would have thought HT would have made more of a difference than it showed in your benchmarks, I can now feel confident in leaving HT on w/ my core i7. Thanks!!!
 
#3 ·
Great benchmark!

It seems like for me the i5 would have been the ideal choice given the HT issue you brought up. Only reason I even went i7 was because of the PCI-E lane limitation on P55.
 
#5 ·
So it looks like the "you paid for it, why not use it" guys were right


...well unless turning HT off allows you a higher overclock.
 
#6 ·
Quote:

Originally Posted by BinaryDemon View Post
So it looks like the "you paid for it, why not use it" guys were right


...well unless turning HT off allows you a higher overclock.
From what I've heard, this is the case. Again, I haven't had the chance to find out through personal experience yet, but I heard that HT leads to higher heat output. By disabling HT, apparently your CPU will run cooler allowing for higher VCore and higher OC.
 
#7 ·
Quote:

Originally Posted by r31ncarnat3d View Post
From what I've heard, this is the case. Again, I haven't had the chance to find out through personal experience yet, but I heard that HT leads to higher heat output. By disabling HT, apparently your CPU will run cooler allowing for higher VCore and higher OC.
I can confirm this, I have seen a 10C drop on my core i7 without HT, I was also able to get a higher maximum OC.
 
#8 ·
Try Fallout 3?

Also, you might want to run each test 3 times to help remove some of the noise since the results are so close.
 
#10 ·
Quote:

Originally Posted by Nautilus View Post
Helps me understand why Phenom II X4 965 and i7 920 is the same at gaming. HT has no effect in gaming.
Well true, if amd cpus had HT they probably wouldn't show much difference either but the 1366 and 1156 cpus generally score better because of higher IPC.

Quote:

Originally Posted by BinaryDemon View Post
So it looks like the "you paid for it, why not use it" guys were right
...well unless turning HT off allows you a higher overclock.
Yup. Although nothing shocking or concrete, it does seem that higher raw clock speed does slightly better than hyperthreading.

Quote:

Originally Posted by r31ncarnat3d View Post
Again, I haven't had the chance to find out through personal experience yet, but I heard that HT leads to higher heat output. By disabling HT, apparently your CPU will run cooler allowing for higher VCore and higher OC.
Very true. As i pointed out in Goal 2, the i7s on air are mainly limited by heat, which HT is responsable for 5-15c extra during load. With it off, this lowers the cpu temp and vcore requirment, thus giveing the user more oc'ing headroom.

Quote:

Originally Posted by DuckieHo View Post
Try Fallout 3?

Also, you might want to run each test 3 times to help remove some of the noise since the results are so close.
Yeah, L4D2, Dragon Age: Origins, Fallout 3, and Bioshock 2 tests are planned for the near future. But there are no benchmark programs or timedemos so it'll take a little longer to devise a consistant environment.
Also, the crysis and farcry tests were averaged from 3 runs, and the stalker program runs 4 tests (different weather). But i will make sure the rest are multiple-run averaged!


To all the rest, thank you for your feedback and kind words! The next 4-6 tests are on the to-do list, so stay tuned for more results!
 
#13 ·
I would actually recommend using a much lower clock speed (3.0-3.6) to really demonstrate any difference between HT enabled/disabled, since the increased core speed reduces potential bottlenecks with the CPU and instead makes them GPU bound. In particular I would like to see the difference in GTA IV, and DAO since they do a nice job of loading up 4 cores.
 
#14 ·
True, a cpu at 3ghz instead of 3.8 or 4 would definitely ensure it was the bottleneck, thus exaggerating HT vs non-HT results. However, i dont see too many people with their i7 oc below 3.8ghz, meaning the 3ghz results would be purely academic and almost irrelevent (because it doesn't test a common real-world senario).

However, you still have a valid point, so i will retest GTA IV at 2.67ghz (stock clocks), and will do a normal and 2.67ghz run of Dragon Age. Thank you for the idea, and you can expect the results later tonight (10pm-ish, 2/19).

EDIT: ran gta benchie on stock clocks and pulled better fps than 4ghz... I'm going retest all settings (including stock clocks) with a custom walkthrough. Thanks grossebeaver for helping me discover that! Sadly all gta and dragon age results will be posted tomorrow
 
  • Rep+
Reactions: Artev
#16 ·
I'm not buying that HT would rightfully give a 5x increase in the AI score, or that it means anything real-world. Then again, I've never considered 3dmark to be real-world, just a good tool for refining/comparing.

Honestly L4D2/L4D would be the benchmarks I'm interested in. By my own tests (timedemos) it scales well to more cores but not linearly. Most games these days don't do any better on a quad from a dual core except for L4D/L4D2.
 
#22 ·
Wow, I made this thread almost 3 months ago and im still getting rep
I was just going to let it go but, but if people are still interested, I could do some more benchies? Like L4D2, Metro2033, Bad Company 2, and Borderlands?
 
#23 ·
Quote:

Originally Posted by MrLinky View Post
Wow, I made this thread almost 3 months ago and im still getting rep
I was just going to let it go but, but if people are still interested, I could do some more benchies? Like L4D2, Metro2033, Bad Company 2, and Borderlands?
PLZ DO L4D2 and BC2 PLZZ!!!!
i have this paged bookmarked lols
 
#24 ·
Nice benchies
 
#26 ·
Those vantage scores are interesting! I would leave HT on IMO, but I also use virtual pc running linux, windows server, and xp while in windows 7. I dont know that it actually makes a difference but im assuming it does. I love being able to play Starcraft 2 while I have vpc or photoshop on a second monitor!
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top