Find the Bottleneck

[n0x1ous]

the details - BF4 1080p all ultra 4x MSAA @ 120hz Vysync OFF - AMD rig in sig

In mantle, gpu usage hovers around 55-60% on each GPU but the CPU usage is roughly 50%......I do get the cpu lag spikes on a regular basis in mantle. FPS is anywhere from 70-120 but usually below 100

In DX11, the cpu lag spikes are gone and CPU usage is around 70% and gpu usage hovers between 60-70% with FPS range of 70-95.

Questions is, if neither CPU nor GPU are maxed out why am I not hitting a solid 120hz consistently?

 

[3DVagabond]

the details - BF4 1080p all ultra 4x MSAA @ 120hz Vysync OFF - AMD rig in sig

In mantle, gpu usage hovers around 55-60% on each GPU but the CPU usage is roughly 50%......I do get the cpu lag spikes on a regular basis in mantle. FPS is anywhere from 70-120 but usually below 100

In DX11, the cpu lag spikes are gone and CPU usage is around 70% and gpu usage hovers between 60-70% with FPS range of 70-95.

Questions is, if neither CPU nor GPU are maxed out why am I not hitting a solid 120hz consistently?

How is the scaling across cores? For example, are you at ~50% CPU usage but it translates to ~100% on 4 cores (again, just as an example to demonstrate my question)?

 

[n0x1ous]

Fairly even across cores. No cores over 85% at any point. No cores parked and all cores in use.

 

[Bubbleawsome]

RAM speed? Might see if kicking up to 2400 helps.

 

[n0x1ous]

I thought of that but on my intel rig in sig I bounce off 120 consistently and that's with slower ram. Would an amd platform be more sensitive to ram speed?

 

[Bubbleawsome]

I think they tend to be as their L3 cache is pretty much as slow as the RAM IIRC. (I'm not that great with AMD cpus though.)
It could also be that the 8350 just isn't strong enough to push 120. :| It's a bit slower than the 4670 at stock and doesn't scale quite as well with overclocking.

 

[n0x1ous]

Well some interesting findings after put-zing around with it tonight. Turning on frame pacing in catalyst has fixed the cpu spikes under mantle. I had it off since 13.12 drivers as it always was causing stutter for me, and I had played around with the different frame pacing methods built into bf4/mantle with no effect either way. Still get the VRAM leak which necessitates restarting bf4 after a few rounds, but thats not too big of a deal to me.

I ran some more dx11 tonight under 14.4 whql's and its fine, but Mantle is so much smoother and the input lag seems less as well. Still not hitting 120 consistently on either API but its high enough to feel really good to me......

 

[3DVagabond]

Well some interesting findings after put-zing around with it tonight. Turning on frame pacing in catalyst has fixed the cpu spikes under mantle. I had it off since 13.12 drivers as it always was causing stutter for me, and I had played around with the different frame pacing methods built into bf4/mantle with no effect either way. Still get the VRAM leak which necessitates restarting bf4 after a few rounds, but thats not too big of a deal to me.

I ran some more dx11 tonight under 14.4 whql's and its fine, but Mantle is so much smoother and the input lag seems less as well. Still not hitting 120 consistently on either API but its high enough to feel really good to me......

Has your GPU usage percentage improved?

 

[Silverforce11]

CPU/Thread bottleneck (MP is very CPU heavy). Put the 7970CF into your 3770K and watch it fly with 100% GPU load.

Frostbite from my past reading only scales up to 6 cores, 4 core scaling is excellent.

 

[n0x1ous]

CPU/Thread bottleneck (MP is very CPU heavy). Put the 7970CF into your 3770K and watch it fly with 100% GPU load.

Frostbite from my past reading only scales up to 6 cores, 4 core scaling is excellent.

yeah the gtx 670's are pegged on the intel system. but if theres thread bottleneck shouldnt i have 100% use on some cores on the FX?

 

[Silverforce11]

Probably Windows 8.1 auto balancing the load beneath it all.

 

[n0x1ous]

Has your GPU usage percentage improved?

not really no. I think I jumped the gun anyway. It was perfect on Firestorm but the next map I played was Altai Range and the cpu spikes were back. I guess I just go back to my intel rig until mantle hd 7000 optimizations are implemented......:hmm:

 

[n0x1ous]

Probably Windows 8.1 auto balancing the load beneath it all.

So cpu bottleneck then ay? And I thought FX was supposed to be so good for Bf4.....

 

[Silverforce11]

So cpu bottleneck then ay? And I thought FX was supposed to be so good for Bf4.....

LOL no. FX is average at best in games when its 4-6 thread supported. Otherwise we all know how awful it is.. AMD CPU just blows for too many generations already its quite sad.

Your best BF4 experience would be Intel CPU + AMD GPU combo.

I reckon your CPU stutter is due to 8.1 moving the thread around the 8 cores when its being hammered. Throw your CF into the intel and I think you will not observe the Mantle stutter (it was apparently fixed via a patch awhile ago already!).

 

[n0x1ous]

LOL no. FX is average at best in games when its 4-6 thread supported. Otherwise we all know how awful it is.. AMD CPU just blows for too many generations already its quite sad.

Your best BF4 experience would be Intel CPU + AMD GPU combo.

I reckon your CPU stutter is due to 8.1 moving the thread around the 8 cores when its being hammered. Throw your CF into the intel and I think you will not observe the Mantle stutter (it was apparently fixed via a patch awhile ago already!).

the GTX's are solid though IM sure mantle would tighten things up a bit, but I dont really want to mess with the intel system as its used for more productive things in addition to gaming. the AMD rig is just to putz around with cuz I never had AMD stuff before and wanted to give it a try. Maybe I just grab a Z97 and haswell refresh and take out the FX......

 

[Pottuvoi]

CPU bottleneck is very easy to test, especially on Battlefield4.
Drop resolution and scaling to minimum and if your FPS doesn't really rise you have found it. (Remember to turn off v-sync during testing.)

 

[escrow4]

Its that tried old AMD chip. It just can't keep up. If you do go Haswell refresh may as well go for a 4790.

 

[tulx]

So cpu bottleneck then ay? And I thought FX was supposed to be so good for Bf4.....

I play BF4 on my 8350 setup below on 1440p and do not have any CPU bottleneck issues what so ever.
All 8 cores are at about 50-65% load during 64 player matches. That said, my target is 60fps since I have a 60Hz monitor. I'll take a close look at frame-rates during the weekend and will report back. Maybe the fact that I have one card instead of two makes some difference as well.
Anyway, will let you know how my exact fps look like.

 

[Silverforce11]

I play BF4 on my 8350 setup below on 1440p and do not have any CPU bottleneck issues what so ever.
All 8 cores are at about 50-65% load during 64 player matches. That said, my target is 60fps since I have a 60Hz monitor. I'll take a close look at frame-rates during the weekend and will report back. Maybe the fact that I have one card instead of two makes some difference as well.
Anyway, will let you know how my exact fps look like.

That's the difference, his setup aims for 120 fps and its not going to happen with a weak CPU.

 

[cbrunny]

I would wager that the bottlenieck is the CPU. This is reasoned speculation, but is probably also laced with pure ignorance:

I see the same thing with my FX-8350 and CF 7970 while playing AC4. It is the CPU, but it doesn't look like the CPU - that is the CPU doesn't hit near 100% - because AMD cores are not true cores - that is these games are only able to utilize the components that make up a single Intel core once per AMD core, thus essentially halving the number of AMD cores these games actually use since some component of an AMD core is shared between two AMD cores. So 50% utilization is the highest that games not tuned for the FX chips will go, since half AMD cores are going full throttle, while the other half sit more or less idle. I suspect this doesn't show in task manager intentionally. Plus, games that are heavily single-thread dependent get choked performancewise by the FX series' relative poor single threaded performance.

 

[Gloomy]

Have you tried changing the frame pacing method for Mantle?

 

[moonbogg]

Do the resolution test by reducing res but keeping details maxed. See if FPS changes.

 

[n0x1ous]

Will do the Resolution test this weekend. And yes I have tried all the frame pacing methods without any difference in the cpu spikes.