Will hyperthreading ever be useful for gaming?

[Maximilian]

Maybe when 8 core CPUs are the norm will it allow 4 core HT cpus to still do okayish without being hopelessly obsolete? Or is it completely useless?

Core i3 seems to win vs G6950 in games quite a bit

http://www.anandtech.com/bench/Product/118?vs=123

 

[smakme7757]

Maybe when 8 core CPUs are the norm will it allow 4 core HT cpus to still do okayish without being hopelessly obsolete? Or is it completely useless?

Core i3 seems to win vs G6950 in games quite a bit

http://www.anandtech.com/bench/Product/118?vs=123

I don't think it matters. Most games are GPU limited anyway. Hyperthreading is useful for quite a lot, just not for gaming!

 

[BD231]

If console hardware somehow takes advantage of it in the future it's a possibility, but, highly unlikely considering you don't need a lot of CPU for games anyway.

Terrible bench comparison by the way, HT has zero impact on gaming performance so there are clearly more pertinent factors involved.

 

[mnewsham]

I think when we see the next gen consoles (2015?) we will see hex-core or octo-core CPU's or a quad with HT in both the top systems (sony, M$) then we will get games that start to use more cores, without the console support devs arent going to make games for a small audience (pc gamers) when the consoles make hand over fist more money then the PC market.

 

[tweakboy]

heck ya, but for now 4 logical cores are enough. Give it two years or three and you'll see more DX11 Hyperthreaded games to use all 8 logical cores properly.

 

[VirtualLarry]

HT is only useful for software that isn't optimized. Most games are optimized (somewhat).

 

[Concillian]

There's also 1 MB more cache on the i3.

There's a 4.6% clockspeed advantage for the i3.
On the games that are not well multithreaded (everything except Dragon Age origins), when you remove that 4.6% for clock speed, the i3 is only about ~8% faster due to a combination of 33% more cache + hyperthreading.

Dragon Age: Origins does make usee of multithreading well, and that game sees a rather significant improvement from the i3. This is where you see the true potential of the hyperthreading, with nearly +50% performance vs. the pentium.

Eventually yes, multithreading will be prominent in game programming. But it's pretty obvious that they'll recycle old game engines as long as possible. It takes long enough to do the artwork and gameplay development for a new game, they really can't afford to make new engines every year or two.

So eventually yes, how far in the future before it's really prevalent? Who the hell knows? My feeling is several more years. How long have dual cores been around and we still only have a handful of strongly multithreaded games.

There will always be some penalty associated with multthreading, even if small. 2 --> 4 will be a bigger jump than 4 --> 8. Eventually games will get to the point that 8 cores are much more useful than 4, butI think it's far enough in the future that most games will be there that it really doesn't enter into the equation if you're building a computer today. Especially when the price difference between a 2600k and a 2500k is about +50% on the CPU.

 

[lamedude]

No need to wait for future consoles Xbox already does SMT.

 

[Ross Ridge]

Maybe when 8 core CPUs are the norm will it allow 4 core HT cpus to still do okayish without being hopelessly obsolete? Or is it completely useless?

The biggest problem with Hyperthreading is that it's redundant with out-of-order execution. It's more useful on CPUs that use in-order execution like the Atom or the PowerPC CPU in the Xbox 360. Otherwise, Hyperthreading is never going to be useful for games. Its disadvantages (more contention for shared resources) will outweigh its diminished advantages.

 

[taltamir]

Hyperthreading is useful because many operations the CPU is doing do not make use of the full capacity of each core. HT allows a core to execute two threads in parallel and actually do more work than it would otherwise. Overall improvements vary based on workload, from 0% to 70% improvement.

Many games today are quad core capable and optimized and running those on a 2 core CPU with HT should benefit them (will require specific tests to know for sure).

 

[Cerb]

I don't think it matters. Most games are GPU limited anyway. Hyperthreading is useful for quite a lot, just not for gaming!

This is said quite a bit, but it doesn't hold true, unless all you play are console games ported to the PC.

Overall, HT's benefit, which isn't huge to begin with, reduces as the CPU stalls less, and as memory takes over as the bottleneck. Well, as a game's most important code gets optimized more, the already-minor benefit from HT drops even lower.

HT being good is predicated on the CPU's instruction units twiddling their thumbs a great deal of the time, your having more threads to execute than you have real execution units, one of threads always being ready to execute something, and your having more cache and main memory resources available than either logical thread can use. Now, on the server side of things, all of that happens quite a bit. For games on the desktop? Meh.

 

[Concillian]

The biggest problem with Hyperthreading is that it's redundant with out-of-order execution. It's more useful on CPUs that use in-order execution like the Atom or the PowerPC CPU in the Xbox 360. Otherwise, Hyperthreading is never going to be useful for games. Its disadvantages (more contention for shared resources) will outweigh its diminished advantages.

Why then, does i3 get 50% better performance in Dragon Age: Origins, a game known to perform better with quads than duals, but only sees ~8-10% advantage in the other games?

I wouldn't say that Hyperthreading will NEVER be useful for games with an OoO architecture. It will never be as good as a full dedicated core, but it does have potential to bring improvement... if the game is written in a way that more cores will create a benefit.

 

[taltamir]

This is said quite a bit, but it doesn't hold true, unless all you play are console games ported to the PC.

Even on console ports this isn't entirely true. Mass Effect is CPU hungry as are a bunch of the unreal engine games once ported to a PC.
Besides unreal engine games, a lot of console ports have some demanding feature or another. And a beefy CPU will let you run Apex PhysX on the CPU at high settings (Games like Mafia 2) without an nvidia GPU. So you could have an AMD GPU and intel CPU and still get physX, only if your CPU is fast enough that is.

I wouldn't say that Hyperthreading will NEVER be useful for games with an OoO architecture. It will never be as good as a full dedicated core, but it does have potential to bring improvement... if the game is written in a way that more cores will create a benefit.

I agree with that. I think HT is very situational for games.

 

[Cerb]

Why then, does i3 get 50% better performance in Dragon Age: Origins, a game known to perform better with quads than duals, but only sees ~8-10% advantage in the other games?

DAO must stall all the time. Even so, it appears to greatly benefit from faster cores and those additional cores being physical cores. Just the same, other games that benefit from more threads tend to get far superior benefits from more real cores. Outliers exist for almost every situation you can think of. So that one game performs well w/ HT on an i3--would you rather have an i3 w/ HT, or an i5 w/o HT?

 

[Concillian]

other games that benefit from more threads tend to get far superior benefits from more real cores. Outliers exist for almost every situation you can think of. So that one game performs well w/ HT on an i3--would you rather have an i3 w/ HT, or an i5 w/o HT?

Nobody's arguing that a hyperthread / virtual core is equivalent performance to a real core... are they?


Interesting data point:
http://www.anandtech.com/bench/Product/102?vs=204
phenom II x4 vs phenom II x2 at same clock speed. Phenom x4 is only +55% the Phenom x2, where the i3, once adjusted for clock speed is 49%. There are other factors in the i3 / Pentium than just HT though. Still the full core advantage is definitely there, but in this case it isn't HUGE. Especially if you consider that the x4 is 109 FPS and the i3 is 95 FPS... average framerates at that level are not noticeably different to a real person playing an actual game. Neither should result in any FPS dips, and, for all intents and purposes, these are tied at "perfect" for this particular game.

Of course a real core is better than a virtual core, but that costs money, power usage, and die space.

Also, I have an i3 dual w/HT instead of an i5 quad w/o HT. HT did not enter into the decision, but it's kinda funny that you chose that example.

 

[Cerb]

One thing, as well, it appears Dragon Age: Origins is also an Xbox360 title (the game is not the least bit interesting to me, so I didn't know this until maybe 15 minutes ago ). I wonder if it was made with the Xenon's SMT in mind (it works like HT, but with limited developer control options, IIRC), and that spilled over to the x86 code?

 

[tweakboy]

To me it counts as a logical core. Chip has 4 physical cores, and 8 logical ,,

a i7 4physical but 8 logical. The logicals will count as CPU if the game or app is programmed to use HT.

So in reality you have a hexacore inside a quad core but only if products fully support it. I know most recent stuff likes HT like Premiere and Sonar X1

photoshop would love it too.

 

[Maximilian]

Do apps have to specifically support hyperthreading? Or would an app that can use 6 cores partially make use of the hyperthreading in a 4 core i7?

 

[mnewsham]

An application that is coded for HT would perform better then an app that supports 6 cores and is running on a quad core with HT. In practice it might be different (depends how it is coded)

 

[Accord99]

Going through Anandtech's list of game benchmarks, you'll find that going from 4 cores to 6 doesn't help much most of the time either. And that the 2c/4t i3-2100 is actually a pretty decent gaming processor.

 

[IntelUser2000]

Xbitlabs show a big difference between i3 540 and Pentium G6950 in games: http://www.xbitlabs.com/articles/cpu/display/clarkdale-review_9.html#sect0

So this is how it works in games for Hyperthreading:
-Game takes advantage of less threads than total logical cores: no gain
-Game takes advantage of equal or more threads than total logical cores: nice gain

It's exactly how it would work for substituting it with cores instead, except cores would obviously give greater gains.

 

[Dark Shroud]

Well I'll say this, Dragon Age Origins runs on my system without a hitch. Completely smooth.

 

[Ross Ridge]

Xbitlabs show a big difference between i3 540 and Pentium G6950 in games: http://www.xbitlabs.com/articles/cpu/display/clarkdale-review_9.html#sect0

So this is how it works in games for Hyperthreading:
-Game takes advantage of less threads than total logical cores: no gain
-Game takes advantage of equal or more threads than total logical cores: nice gain

The Xblitlab's tests you linked don't show that at all. The Pentium G6950 loses significantly to the i3-540 on all tests, single-threaded or multi-threaded. There's no evidence that Hyperthreading is helping the i3-540 in games.

 

[tweakboy]

To me it counts as a logical core. Chip has 4 physical cores, and 8 logical ,,

a i7 4physical but 8 logical. The logicals will count as CPU if the game or app is programmed to use HT.

So in reality you have a hexacore inside a quad core but only if products fully support it. I know most recent stuff likes HT like Premiere and Sonar X1

photoshop would love it too.

To add to what I said earlier honestly it may take even more then 2 or 3 years for games to be 64bit , DX11 and HypterThreaded.

By 2015 maybe when the new consoles come out.

Until then a quad core with 4 logical cores or a hexacore with HT = 12 cores if application properly supports it. In due time games will recognize the CPU is HT and take full advantage of it. However I don't think this will happen any time soon. Right now most games are DX9 and 32bit and not really optimized for the i7 8 logical cores or the 990 extreme or Sandy Bridge

 

[IntelUser2000]

The Xblitlab's tests you linked don't show that at all. The Pentium G6950 loses significantly to the i3-540 on all tests, single-threaded or multi-threaded. There's no evidence that Hyperthreading is helping the i3-540 in games.

Nearly every test there is multi-threaded up to 4 threads one way or another. The only one that is 1 thread only is iTunes. You'll notice that from Q9550 vs. E8500 comparison. The Q9550 is clocked even slightly lower than the E8500.

Dirt 2, L4D2, HAWX, Far Cry 2 show better results on the 2.83GHz Q9550 than the 3.16GHz E8500.