BF4 CPU usage

[soccerballtux]

I don't venture in VC&G much anymore and today found this which is applicable to our interests. Looks like we're finally getting on to next gen CPU resource management.

Seeing as it only uses 8 cores/threads seems clear they are targeting next gen consoles.

notice the HT on cores 2+ have no workload.

I'm rather impressed it can load a 6 core FX-6300, seems like a first for gaming.

source: http://gamegpu.ru/action-/-fps-/-tps/battlefield-4-beta-test-gpu.html

 

[BallaTheFeared]

Looks like it's using 6, not 8, and it looks like the 6300 is maxed out whereas the 8350 gets a pretty lofty clock speed advantage over it and two extra cores for background/os.

 

[Arkaign]

Looks like it's using 6, not 8, and it looks like the 6300 is maxed out whereas the 8350 gets a pretty lofty clock speed advantage over it and two extra cores for background/os.

Could be, but another possibility is that with 6 cores, it is bottlenecked and is forced to fully load them to try to maintain performance, and with 8, it's able to balance better.

After all, it does show BF4 using all 8 cores, and loading them about 2/3rd to 3/4th of the way.

 

[BallaTheFeared]

Yes that is possible, looking at the 6300 usage it is at or near the point where it will be limiting that GPU.

Doing stupid math, if you add up the total usage of the 8350 vs 6300 the 8350 has similar total cpu usage compared to the 6300. Which could mean the 6300 is bottlenecking slightly, and the Titan doesn't have enough power to push the 8350 further, or simply the game uses 6 threads and will use 8 with an AMD GPU /w Mantle.


Which is unlikely because the 8350 is still bottlenecking compared to the Intel x6 chips.

 

[DownTheSky]

Then you see per core Intel is 2x as fast as AMD. Which is :awe:

 

[frozentundra123456]

Kind of surprised the 6300 doesnt do better. Just 2 more cores and a slightly faster clockspeed gives 50% performance increase. Be interesting to see if mantle improves the 6300 relative to 8350.

The cpu test is kind of strange though. The raw framerate graphs are at 1680 x 1050, while the core load data is at 1080p. Wonder if that is a misprint, and if not why they showed raw FPS and core load at different resolutions.

Sure would be nice if Intel came out with a mainstream hex core, wouldnt it? Six real cores is clearly ahead of 4 cores plus hyperthreading (even though apparently well utilized) even when at a clockspeed disadvantage.

 

[BallaTheFeared]

8350 is 25% faster and has about as much clock speed over the 6300.

 

[frozentundra123456]

8350 is 25% faster and has about as much clock speed over the 6300.

We're back to how you do the math I guess. If you use the 6300 as baseline, the 8350 is almost 50% faster, i.e. 60/41 = about 50% faster. However if you use the 8350 as baseline the 6300 is 41/60 or 2/3 as fast.

I honestly dont know which is the most valid way to look at it. Either way, not sure where you are getting the 25% number though.

I guess what I am surprised at is that I consider the 6300 as a good starting point for a low/midrange build, but it is kind of borderline in this game, although it is just the beta. Personally, 40FPS is good enough for me, but apparently a lot of competitive online players want 60. For that you need a hyperthreaded intel quad or FX8350.

 

[Carfax83]

Now my question is, if the Frostbite 3 engine has such excellent CPU utilization, why do we even need Mantle?

Native DX11 engines like Cryengine 3 and Frostbite 3 have no problems with multithreading, and so presumably, they aren't draw call limited.

 

[inf64]

I guess for the guy having FX6300 valid point of view would be that he would get 50% more performance with 835 upgrade.

But this is just Beta and it runs awful at the moment. Check battlelog and you will see that vast majority has performance and lag issues. Hopefully it will be fixed.

 

[tweakboy]

It will happen eventually my friend. Be patient. gl

 

[Saylick]

Now my question is, if the Frostbite 3 engine has such excellent CPU utilization, why do we even need Mantle?

Native DX11 engines like Cryengine 3 and Frostbite 3 have no problems with multithreading, and so presumably, they aren't draw call limited.

What your suggesting is more or less valid if everyone had a 6+ core CPU. Mantle helps alleviate CPU bottlenecks, as does having more cores. With that said, Mantle should be most useful in computers without a lot of CPU power, i.e. allowing dual cores run as if they were 4 cores, for example. Allowing current gen consoles run as if they had more CPU power would be another example. This is not to say that Mantle won't be of help at the high-end, because it will, but the performance benefit probably won't be as noticeable.

 

[0___________0]

Now my question is, if the Frostbite 3 engine has such excellent CPU utilization, why do we even need Mantle?

Native DX11 engines like Cryengine 3 and Frostbite 3 have no problems with multithreading, and so presumably, they aren't draw call limited.

Draw call limitations are only a result of weak CPU power in cases where someone coded their engine crappily. The CPU itself isn't the bottleneck, it's the overhead. You also can't just multi thread stuff wherever you want, some operations are train wrecks to sync. The real problem is there's a lot of work necessary for the CPU to do in order to get commands to the GPU. It requires the CPU to take time to switch from user to kernel mode, then revert, you have to access and modify memory, make state changes, there's some safety mechanisms to prevent crashing, you have commands that have to go through drivers, the HAL, etc; you can't directly write to the GPU buffer.

BF's utilization of many cores is mostly due to all the processing required for 64 player mayhem; you get the exact same graphics at better FPS in single player with an i3 than you do with an i7 in multiplayer.

 

[SPBHM]

interesting results, the game is really coded to use many cores/threads, it's rare to see HT being used as much, even on the i3.

also a strong performance for the 8 core FX, but OC for OC the graphic makes me think the 2500K would be holding well against it,

also it's a shame this website is stuck with Sandy Bridge when it comes to Intel.

I think the Xbox One versions is exactly the same, and their target is 60fps

it just shows the level of optimization going on on the consoles, because I think a 8 core jaguar a t 1.75GHz would be getting around 30fps on this list, running windows.
but... fixed hardware, they can adapt the level of details, oh well,

edit: another performance test here
http://pclab.pl/art55028-3.html

 

[frozentundra123456]

interesting results, the game is really coded to use many cores/threads, it's rare to see HT being used as much, even on the i3.

also a strong performance for the 8 core FX, but OC for OC the graphic makes me think the 2500K would be holding well against it,

also it's a shame this website is stuck with Sandy Bridge when it comes to Intel.

I think the Xbox One versions is exactly the same, and their target is 60fps

it just shows the level of optimization going on on the consoles, because I think a 8 core jaguar a t 1.75GHz would be getting around 30fps on this list, running windows.
but... fixed hardware, they can adapt the level of details, oh well,

edit: another performance test here
http://pclab.pl/art55028-3.html

Wow, those pclab results are brutal for AMD in a game they have supposedly spent years optimizing for their hardware.

 

[AtenRa]

I get the same CPU utilization with the FX8350 as it did in BF3.
I dont believe the BF4 BETA is using more than 6 cores/threads.

 

[Maximilian]

Thats weird, IIRC BF3 was 8 threads. New consoles are 8 threads too...

 

[AtenRa]

BF3 uses up to 6 threads not 8. This is the reason why FX6300 has that high CPU Utilization in BF3 and BF4. 8-core FX utilization goes up to 60%, if the game would be able to use 8 threads, the FX8350 would have more than 60% utilization.

 

[SPBHM]

Thats weird, IIRC BF3 was 8 threads. New consoles are 8 threads too...

new consoles have 8 core CPUs, but I think 1 or 2 are not accessible by games for now.

 

[ShintaiDK]

Thats weird, IIRC BF3 was 8 threads. New consoles are 8 threads too...

New consoles are 6 threads for games. 2 cores are reserved for other tasks.

 

[ShintaiDK]

The Pclab is rather interesting. A 4770K scores ~20% more than a 2600K. And is on pair or better than a 4.5Ghz 2600K.

 

[Borealis7]

also, very small difference between i5-2500K and i7-2600K, game uses multiple cores but doesn't like HT?

 

[SPBHM]

also, very small difference between i5-2500K and i7-2600K, game uses multiple cores but doesn't like HT?

well, if you compare the stock 2500k vs 2600K from both tests, there is a bigger difference on the Game GPU test, correcting the clock and cache difference you would be gaining more than 10% because of HT for the GameGPU test, and nothing on the PClab test

so the potential is there, but it's not a huge gain

 

[seitur]

PurePC.pl results:

http://www.purepc.pl/karty_graficzn...battlefield_4_test_geforce_vs_radeon?page=0,3

3930X @ 4500 MHZ and HT OFF

Ultra, HBAO and 4x MSAA

High, HBAO, MSAA OFF

PS. They are gonna add more GPUs results later on.

 

[Dresdenboy]

Very interesting topic!

Could be, but another possibility is that with 6 cores, it is bottlenecked and is forced to fully load them to try to maintain performance, and with 8, it's able to balance better.

After all, it does show BF4 using all 8 cores, and loading them about 2/3rd to 3/4th of the way.

The core load of BF4 on the CPUs in the first thread posting leads me to a different hypothesis:

• The use of 7 threads on the i7 3970X vs. 8 on the FX-8350 implies, that there is at least one heavy additional process not started by the game. This could be some drivers, DX11 and background tasks.
• Core load at 83% on the 2600K and 95% on the 8350 (average!, which could mean, it's often at 100%) while most of the other cores are in the 60-70% range, indicates, that the involved core is holding back the others (some serial bottleneck).
• With HT (Intel) or CMT (AMD) this also means a slow down of both the first and that additional thread. If 2T HT performance is 130% of 1T, then both threads run at about 65% of non-HT performance, meaning 50% higher processing latencies.

It would be interesting, how this multithreading performance could be improved by applying core affinities.