BF3 CPU bottleneck? Really?

[boxleitnerb]

HT on often is slower than HT off in games by up to 5-10%. Specifically about BF3, I don't know. But as you cannot reliably bench multiplayer anyway, we will never know.

 

[moonbogg]

HT on often is slower than HT off in games by up to 5-10%. Specifically about BF3, I don't know. But as you cannot reliably bench multiplayer anyway, we will never know.

We did bench multiplayer reliably. HT helps a little. Perhaps further testing of others can confirm this.

 

[boxleitnerb]

We did bench multiplayer reliably. HT helps a little. Perhaps further testing of others can confirm this.

How did you do that? Did you have 63 friends on a private server who all did the exact same thing twice?

 

[Termie]

News flash...HT helps in lots of new games, not just BF3 multiplayer, which is admittedly hard to bench but tends to show that HT works.

Here's my testing of my i7-860 using the built-in benchmark in Batman Arkham City:

Still believe HT doesn't work in new games?

 

[boxleitnerb]

For 2 cores it does. For 4 cores it often does not and can even be counterproductive. Your Batman AC benchmark is
a) not real gameplay
b) capped at 60Hz

Did you run it at least twice, better three times to account for variations? Turbo was off?

 

[BTRY B 529th FA BN]

I may be shooting waay off here but has anyone benched fps in offline mode with 64 bots?

 

[Termie]

For 2 cores it does. For 4 cores it often does not and can even be counterproductive. Your Batman AC benchmark is
a) not real gameplay
b) capped at 60Hz

Did you run it at least twice, better three times to account for variations? Turbo was off?

I agree this isn't real gameplay, but you were criticizing Moonbogg for attempting to benchmark actual gameplay in BF3, which showed an improvement with HT. This is the counterpoint - irrefutable evidence that HT works, without any possible variation in the scene.

And I know it's capped at 60, but the fact that significant differences are still evident makes it that much more amazing. Without the cap the delta would no doubt have been higher. I ran these benchmarks over and over - three or four times each. Sometimes the minimums would change, but the averages (and obviously the max) never would. The core clocks were locked - the reason I used 3.4 with two cores was for a comparison to my old e8400@3.3. I actually run the 860 chip at 3.0.

By the way, as to HT helping with 2 cores but not with 4, that would only be true if the game were only coded for 4 threads. There's plenty of evidence to show that BF3 is coded for 6, as the AMD 6-core chips show some improvement over the quads in BF3. Don't know about Batman, but my performance certainly didn't go down with quad HT.

 

[BTRY B 529th FA BN]

News flash...HT helps in lots of new games, not just BF3 multiplayer, which is admittedly hard to bench but tends to show that HT works.

Here's my testing of my i7-860 using the built-in benchmark in Batman Arkham City:

Still believe HT doesn't work in new games?

someone benched with HT on, verse HT off with the UE3 engine and got similar positive results with HT on. I bookmarked it for later reference. on pc at home though

EDIT: LOL that was you I think, termie

 

[boxleitnerb]

I agree this isn't real gameplay, but you were criticizing Moonbogg for attempting to benchmark actual gameplay in BF3, which showed an improvement with HT. This is the counterpoint - irrefutable evidence that HT works, without any possible variation in the scene.

And I know it's capped at 60, but the fact that significant differences are still evident makes it that much more amazing. Without the cap the delta would no doubt have been higher. I ran these benchmarks over and over - three or four times each. Sometimes the minimums would change, but the averages (and obviously the max) never would. The core clocks were locked - the reason I used 3.4 with two cores was for a comparison to my old e8400@3.3. I actually run the 860 chip at 3.0.

By the way, as to HT helping with 2 cores but not with 4, that would only be true if the game were only coded for 4 threads. There's plenty of evidence to show that BF3 is coded for 6, as the AMD 6-core chips show some improvement over the quads in BF3. Don't know about Batman, but my performance certainly didn't go down with quad HT.

I didn't critcize moonbogg, I just pointed out that it is nigh impossible to get consistent results in multiplayer. I don't know what he/they actually did, but I highly doubt that you can do the same thing twice with that many players and get an error quota of less than 5%.
I wouldn't go so far to call the 2fps a significant difference (44 vs 46). I have Batman AC installed, I can do a test this weekend, but due to repeatability issues, only short segments of course.

Some engines may show no performance hit with HT on with 4 cores/8 threads, but many do. For example I benchmarked Shogun 2 and HT didn't help at all beyond the first core. Only "real" cores did help and HT on was slower than HT off from 2 cores upwards.

 

[Termie]

someone benched with HT on, verse HT off with the UE3 engine and got similar positive results with HT on. I bookmarked it for later reference. on pc at home though

EDIT: LOL that was you I think, termie

Yup, was me. BTW, no bots in BF3, so no option to bench that way.

I didn't ciritize moonbogg, I just pointed out that it is nigh impossible to get consistent results in multiplayer. I don't know what he/they actually did, but I highly doubt that you can do the same thing twice with that many players and get an error quota of less than 5%.
I wouldn't go so far to call the 2fps a significant difference (44 vs 46). I have Batman AC installed, I can do a test this weekend, but due to repeatability issues, only short segments of course.

Some engines may show no performance hit with HT on with 4 cores/8 threads, but many do. For example I benchmarked Shogun 2 and HT didn't help at all beyond the first core. Only "real" cores did help and HT on was slower than HT off from 2 cores upwards.

The more data the better, so if you have some results to add, that would be helpful. I originally ran BF3 with HT off based on the initial impressions on the forum that it hurt in the beta, but shortly after the full game came out, I switched HT back on. I agree benchmarking in that game is very hard (and potentially very frustrating when the other players actually try to kill you while you're attempting to get clean fps data!). I've never actually done an HT analysis in BF3, but I know the multiplayer can be CPU-bottlenecked pretty easily from other testing.

Shogun is a CPU-intensive game, so it would be an interesting example.

 

[BTRY B 529th FA BN]

there are differences felt with HT on. I would say I am pretty sensitive to the movement in Unreal Tournament III after 3-4 yrs of consistent play. While I see no difference in fps because the 670 & clock on my cpu really push the frame limit the movement can be improved with HT on. Simply the character feels like it moves further and the response reaction is faster.

 

[moonbogg]

How did you do that? Did you have 63 friends on a private server who all did the exact same thing twice?

There is a technique to it. Avoid running around and playing. Find a quiet corner of the map, lay down behind a rock and face the open part of the map where the action is. You will notice that FPS only fluctuates 2 or 3 fps with the whole map in view and with you laying still. This lets your CPU simply process everything going on without any major variation caused by you taking a rocket to the face, etc. When you run around and play, you are facing a wall one second, and dodging a crashing jet the next. This makes results inconsistent.
I did this and went from 1 to 12 threads in task manager and the difference between 4 and 8 threads was significant.

 

[Akantus]

Some engines may show no performance hit with HT on with 4 cores/8 threads, but many do. For example I benchmarked Shogun 2 and HT didn't help at all beyond the first core. Only "real" cores did help and HT on was slower than HT off from 2 cores upwards.

Unfortunately Shogun 2 is terribly threaded game, it generally uses core and half of my quad and fps go to teens if you zoom into battle. :\

 

[boxleitnerb]

Unfortunately Shogun 2 is terribly threaded game, it generally uses core and half of my quad and fps go to teens if you zoom into battle. :\

Shogun scales quite well with 3 cores on Intel CPUs (real cores, not HT threads), yet performance is still very low despite that. But you're right, performance tanks once you zoom in. I will open a thread regarding this HT thing, maybe we can gather some benchmarks.

 

[2is]

How did you do that? Did you have 63 friends on a private server who all did the exact same thing twice?

Play on the same maps and do enough runs both ways (ht on and ht off) and the law of averages kicks in, even if you don't want to believe it.

 

[cytg111]

there are differences felt with HT on. I would say I am pretty sensitive to the movement in Unreal Tournament III after 3-4 yrs of consistent play. While I see no difference in fps because the 670 & clock on my cpu really push the frame limit the movement can be improved with HT on. Simply the character feels like it moves further and the response reaction is faster.

I was a clan head back in the quake days .. what you are describing can easily be placebo for something else.

One trick i did to myself when I was getting 'stuck' progress wise was to change the color of my crosshair .. that little change took me a bit further every time. (from a neural construct, pattern recognition perspective, it also makes a certain level of sense).
The sheer fact of 'knowing' that you have changed something will often raise your alertness level, and thus for a period you will play better.

 

[Akantus]

Shogun scales quite well with 3 cores on Intel CPUs (real cores, not HT threads), yet performance is still very low despite that. But you're right, performance tanks once you zoom in. I will open a thread regarding this HT thing, maybe we can gather some benchmarks.

Well, haven't really tried how it scales with cores, all I know is there is usually 1 core pegged at 100% bottlenecking the rest, so even if it did use lots of cores you wouldn't know it because there is already bottleneck in the first one... I was really disappointed when I found that, and now I am impatiently waiting for Haswell (well, Haswell-E if I can wait that long )

And hoping TW:Rome will be better, but from what I have read my hopes aren't high. :\

 

[BTRY B 529th FA BN]

I was a clan head back in the quake days .. what you are describing can easily be placebo for something else.

One trick i did to myself when I was getting 'stuck' progress wise was to change the color of my crosshair .. that little change took me a bit further every time. (from a neural construct, pattern recognition perspective, it also makes a certain level of sense).
The sheer fact of 'knowing' that you have changed something will often raise your alertness level, and thus for a period you will play better.

I've considered that before I even said something. I didn't want to put my self out there for scrutiny like this so I took my time trying to rule out overlooking other possibilities e.g. like the one you just mentioned, and other things like how many people are on the server, what are their pings, are they effecting mine (high ping players will have an adverse effect in UT3 whether in Greed/DM/Duel/etc,) what is going on in my local network, have I just took a little break after being warmed up, am I relaxed, am I on some caffeinated energy drink, is the map small like a HOLP maps that reduce the processor's need to push the fps, while trying to maintain all other things consistent like not changing my xhair style/size/screen resolution/overclocking/etc.

I am very sensitive to the game and I did notice an effect.

 

[soccerballtux]

I found the times when i was sobering up I was the best at UT3 iCTF

 

[BD231]

There's been some complaints regarding the APIs. It seems like a really big weakness they have currently is the draw calls.
http://www.bit-tech.net/hardware/graphics/2011/03/16/farewell-to-directx/2

I think it's unfair to blame the consoles for multithreaded CPUs being held back. Quite a bunch of the early console games, that later got ported to the PC, did indeed show benefits by having 4 cores instead of 2. And so do many still.

It's still going to take time before we get more than 4 cores as a standard.

Not everyone is on a quad core yet which any developer has to consider so PC standards are just as big a culprit. You still see the biggest increase in performance going from one core to two and beyond that the gains aren't nearly as big which suggests developers are still limiting core functions of games to function on lesser hardware. BF3 is just another good example, heavier task loads are being put on some cores than others.

Given the choice between a faster CPU or better utilization of hardware through software I'd take utilization. A need for faster cores due to software inefficiency is not the reason I want to be purchasing a faster CPU.

 

[cytg111]

I've considered that before I even said something. I didn't want to put my self out there for scrutiny like this so I took my time trying to rule out overlooking other possibilities e.g. like the one you just mentioned, and other things like how many people are on the server, what are their pings, are they effecting mine (high ping players will have an adverse effect in UT3 whether in Greed/DM/Duel/etc,) what is going on in my local network, have I just took a little break after being warmed up, am I relaxed, am I on some caffeinated energy drink, is the map small like a HOLP maps that reduce the processor's need to push the fps, while trying to maintain all other things consistent like not changing my xhair style/size/screen resolution/overclocking/etc.

I am very sensitive to the game and I did notice an effect.

I feel you, I really do.
But without a methodology for proof, it is what it is.
If you can think up a way to demonstrate, say on a frame pr. sec basis. one setup to another, that would be groundbreaking.

 

[toyota]

News flash...HT helps in lots of new games, not just BF3 multiplayer, which is admittedly hard to bench but tends to show that HT works.

Here's my testing of my i7-860 using the built-in benchmark in Batman Arkham City:

Still believe HT doesn't work in new games?

how are you that cpu limited in Batman AC? I can get better results running just a single core of my 2500K.

 

[cytg111]

BF3 is just another good example, heavier task loads are being put on some cores than others.
.

.. I'd like to see excatly *what* is being put on those cores.. How does BF3 parallize its execution in reality

 

[moonbogg]

.. I'd like to see excatly *what* is being put on those cores.. How does BF3 parallize its execution in reality

See the numbers again. The standard variation in FPS is about 2-3FPS during this testing. I was laying down in the corner of 64 player Oman and FPS is very steady when doing this. BF3 seems to distrbiute the load pretty evenly when going from 2 to 4 cores. So, your percentage of increased FPS is pretty much proportional to the percentage increase in core count. After 4 cores/threads, the performance falls off. As you can see, BF3 doesn't seem to care much if the core it is using is a HT core or a real core, at least once it moves boyond 4 real cores. Testing with 2 cores/4 threads might have been interesting. See below.

8 Threads is where its at for this game.

1 core = 1fps
2 cores = 20fps
3 cores = 35fps
4 cores = 50fps
5 cores = 53fps
6 cores = 56fps

4 threads = 50fps
6 threads = 57fps
8 threads = 63fps
10 threads = 64fps
12 threads = 64fps

 

[BD231]

.. I'd like to see excatly *what* is being put on those cores.. How does BF3 parallize its execution in reality

Me too. People keep laying around on maps instead of actually playing so the numbers in this thread are useless.