Tuning your AMD CPU for Gaming

[Frakk]

Overclocking the CPU frequency is not necessarily getting the most out of your AMD CPU, aside from advantages of also getting your memory running fast, AMD Black Edition and FX CPU's also have unlocked CPU-NB.

Before i go on i should say that AMD list there Phenom and Athlon CPU as designed to run 1333Mhz

Bulldozer and Piledriver FX CPU's support 1866Mhz RAM.

Its the CPU-NB that we are concentrating on here.

The reason is Game Physics rely on memory operations or Floating Point performance, by setting a higher CPU-NB frequency you can improve that performance a little, it also makes for a snappier feeling computer.

For instructions on how to overclock your AMD CPU and its CPU-NB refer to Dolk's guide to the Phenom II The CPU-NB guide is at the center of This section.

System used is.

Motherboard: Asus Sabertooth 990FX
CPU: Phenom II x6 1090T
CPU Cooler: CM Hyper 412s
GPU: Radeon 7870, overclocked, 1175 / 1350

Benching software 3DMark 11

I have highlighted the changes to the CPU and the consistency of the GPU clock in yellow

First a bench with the CPU @ 4.2Ghz and the CPU-NB @ 2000Mhz and Memory @ 1333Mhz

As you can see there the Physics score is 5745 and the combined score 5297

Now we are going to turn the memory up to 1470Mhz and CPU-NB to 2200Mhz whilst keeping the CPU at 4.2Ghz.

CPU Physics score has gone up 300 Points to 6029.

Next the memory up just a little to 1520Mhz and the CPU-NB up a lot to 2750Mhz.

As you can see the Physics score has shot up another 370 Points to 6399, much more with a small memory increase and a big CPU-NB increase than a big Memory increase and a small CPU-NB, and the CPU is actually down to 4.1Ghz due to heat.

Overall its close to a 15% increase.

You maybe wondering why i did not just leave the memory at stock and just up the CPU-NB, it would have been beter that way but some CPU's don't like high multipliers on the CPU-NB, mine is one of them.

So to get it much higher i have had to use the FSB, or external clock under another name.

You may find this same problem, or you may not. if you can i would recommend you use the CPU-NB multiplier to increase the speed.

Lastly, setting higher CPU-NB frequency will increase your CPU temps a little, the key is finding the right balance between CPU-NB and CPU performance

Edit, added later. As requested i'm adding some game benches

first a standalone AVP bench.

CPU: 4.1Ghz - CPU-NB: 2000Mhz - MEM: 1333Mhz - GPU: 1175 /1350.

AvP D3D11 Benchmark Report
==========================
**************************************************
* Report Created: 2012-07-25 @ 23:08:39
**************************************************
* Executable Build: V1.03, Apr 19 2010
**************************************************
*DX11 Hardware Detected*
Using Default Video Settings:
Resolution: 1920 x 1080
Texture Quality: 2
Shadow Quality: 3
Anisotropic Filtering: 16
SSAO: ON
Vertical Sync: OFF
DX11 Tessellation: ON
DX11 Advanced Shadows: ON
DX11 MSAA Samples: 1

Benchmark Summary:
Number of frames: 8176
Average Frame Time: 12.8ms
Average FPS: 77.9

77.9 FPS

CPU: 4.1Ghz - CPU-NB: 2650Mhz - MEM: 1450Mhz - GPU: 1175 /1350.

AvP D3D11 Benchmark Report
==========================
**************************************************
* Report Created: 2012-07-26 @ 00:15:14
**************************************************
* Executable Build: V1.03, Apr 19 2010
**************************************************
*DX11 Hardware Detected*
Using Default Video Settings:
Resolution: 1920 x 1080
Texture Quality: 2
Shadow Quality: 3
Anisotropic Filtering: 16
SSAO: ON
Vertical Sync: OFF
DX11 Tessellation: ON
DX11 Advanced Shadows: ON
DX11 MSAA Samples: 1

Benchmark Summary:
Number of frames: 8219
Average Frame Time: 12.8ms
Average FPS: 78.3]

77.9 to 78.3 (+0.6%) I would call that nothing.

With some help from our fine members i have now manged to get the CPU-NB to run higher without the need to ajust the Memory and FSB, so from here on everything will be purly a CPU-NB and HT-Link diffrance.

3DMark11 bench run again.

CPU: 4.2Ghz - CPU-NB: 2000Mhz - HT-Link: 2000Mhz - MEM: 1333Mhz - GPU: 1175 /1350.

Physics: 5809 / Combined: 5368

CPU: 4.2Ghz - CPU-NB: 2800Mhz - HT-Link: 2200Mhz - MEM: 1333Mhz - GPU: 1175 /1350.

Physics: 6478 +11% / Combined: 6027 +11%


Next is NFS Shift 2, Benching software use is Fraps, for consistency its off line on the same map (Alpental Circuit) using the same BMW M6 car driving the same racing line, overtaking the one car i had to at the same spot, i also started the bench at the same time the red lights went out and stopped it driving across the finishing line.
So i don't have to repeat myself i will be doing the same thing with every game benched.

Res 1024 x 768, game settings everything minimal.

CPU: 4.2Ghz - CPU-NB: 2000Mhz - HT-Link: 2000Mhz - MEM: 1333Mhz - GPU: 1175 /1350.

Min 92 / Max 145 / Avg 121.152

CPU: 4.2Ghz - CPU-NB: 2800Mhz - HT-Link: 2200Mhz - MEM: 1333Mhz - GPU: 1175 /1350.

Min 97 / Max 165 / Avg 135.007
(+6%) (+15%) (+12%)


Res 1920 x 1080, game settings everything as high as it will go.

CPU: 4.2Ghz - CPU-NB: 2000Mhz - HT-Link: 2000Mhz - MEM: 1333Mhz - GPU: 1175 /1350.

Min 64 / Max 108 / Avg 90.415

CPU: 4.2Ghz - CPU-NB: 2800Mhz - HT-Link: 2200Mhz - MEM: 1333Mhz - GPU: 1175 /1350.

Min 70 / Max 102 / Avg 93.175
(+9%) ( -5%) (+4%)

There seems to be greater improvement once the work is off loaded on to the CPU, The minimum FPS also seems a consistent ~10% at both high and low resolutions, where as the higher FPS have actualy droped but given that the avredge FPS is still up a little i think i can put that down to a GPU blip


BF3

CPU: 4.2Ghz - CPU-NB: 2000Mhz - HT-Link: 2000Mhz - MEM: 1333Mhz - GPU: 1175 /1350.

Res 1024 x 768, game settings Low preset.

Operation Guillotine
Min 103 / 201 / Avg 147.117

Swardbreaker
Min 88 / 201 / Avg 131.109

Res 1920 x 1080, game settings Ultra preset.

Operation Guillotine
Min 40 / 93 / Avg 55.714

Swardbreaker
Min 31 / 81 / Avg 49.868


CPU: 4.2Ghz - CPU-NB: 2800Mhz - HT-Link: 2200Mhz - MEM: 1333Mhz - GPU: 1175 /1350.

Res 1024 x 768, game settings Low preset.

Operation Guillotine
Min 105 / 201 / Avg 159.953
(+2%) (0) (+10%)

Swardbreaker
Min 95 / 201 / Avg 141.616
(+7%) (0) (+8%)


Res 1920 x 1080, game settings Ultra preset.

Operation Guillotine
Min 40 / 93 / Avg 56.862
(0) (0) (0)

Swardbreaker
Min 34 / 82 / Avg 49.823
(+3%) (0) (0)

No change in BF3 at high res and Ultra settings, yet, again there is a ~10% gain at low res / low settings

More to come....

 

[bassnut]

Interesting ..... Thanks Frakk. Leaves me wondering how some of this will translate over to Bulldozer. Will deff be something to think abbout when things cool off and I can do a bit of testing on my OC.

 

[rescuetoaster]

Cool, thanks for the write up man.

I would like to add a caveat to the Phenom 2 IMC thing. The Thuban cores seem to be better than Deneb, so we generally run our ram at 1600 (some go higher).

Is there a safe limit to CPU-NB overclocking?

 

[Frakk]

Cool, thanks for the write up man.

I would like to add a caveat to the Phenom 2 IMC thing. The Thuban cores seem to be better than Deneb, so we generally run our ram at 1600 (some go higher).

Is there a safe limit to CPU-NB overclocking?

On the Thuban IMC, ok thanks

CPU-NB, i don't know but i have stress tested mine as high as 3Ghz, it seemed fine

@ bassnut, yeah, it would be interesting to see this on Bulldozer, and more importantly Piledriver.

Thanks guys

 

[Mandrake4565]

Very interesting Frakk, I'm going to have to play around with mine and see how it changes things. Thanks for the post.

 

[ssjwizard]

I have not done extensive testing on my setup for CPU-NB but I keep it around 2400mhz with my 8120 as that doesnt really require a voltage bump and is 100% butter smooth and stable at that speed. What I might do here is run some tests at a lowish(~3.6ghz) cpu clock and push my NB only without touching the FSB or memory speed since my 990FXA allows up to 30x multipliers on the CPU-NB. That should minimize the differential between runs and allow us to see only the CPU-NB effects.

I need to get some 3D11 benching done on the my 460s anyways.

 

[Bluefalcon13]

Meh, i hope my PII 955 burns out..... I want a reson to upgrade to a new proc

So far my 955 has been rock solid on 3.8ghz, ddr3-1600, and a cpu-nb of 2400mhz. Are you saying using the ddr3-1600 will kill my imc? I thought the 955BE was specced to run memory @ 1600

 

[EarthDog]

Nice writeup!

Does this translate ti fps though in games? 3d11 is one thing, games another.

 

[manu2b]

IIRC, upping the NB from 2200MHz to 2740MHz (@4/4.1GHz) on my PhII resulted a nice 3/4% FPS increase in most games.

Nice write up Frakk!

@Bluefalcon13: PhII IMC is rated for 1333MHz DDR3.

 

[RGone]

Model number 955BE
CPU part numbers HDZ955FBK4DGI is an OEM/tray microprocessor
HDZ955FBGIBOX is a boxed microprocessor
Stepping codes AACYC AC CACYC AC
Frequency 3200 MHz
Bus speed 667 MHz Memory controller
One 2000 MHz 16-bit HyperTransport link

 

[ATMINSIDE]

667x2=1333
DDR is double.

 

[RGone]

667x2=1333
DDR is double.

Yeppers it surely does translate like that. Which as 'Frakk-man' said, the memory controller is DDR1333 rated.

 

[ATMINSIDE]

Yeppers it surely does translate like that. Which as 'Frakk-man' said, the memory controller is DDR1333 rated.

Yep! I was just posting it for people to see, that question has been asked too many times about "why does my RAM show 800 instead of 1600"

 

[EarthDog]

IIRC, upping the NB from 2200MHz to 2740MHz (@4/4.1GHz) on my PhII resulted a nice 3/4% FPS increase in most games.

Cool.. Yeah lets see that tested on canned game benchmarks, for example, the ones we use in reviews. In the benchmarking team, we are aware of the importance of the NB for synthetic benchmarks. But this is gaming and not benhmarking. And while higher bench scores do translate to higher FPS, its certainly not on the same scale.

So since this is under the guise of gaming... gaming benchmarks should be used to show the results accurately instead of synthetic benchmarks.

 

[manu2b]

^Unfortunately, I sold my PhII rig a couple of months ago...

 

[EarthDog]

Frakk is doing the testing in this thread... Its up to him (or I guess anyone that would want to help).

I left your name out of it intentionally and was really in reference to the OP.

 

[RGone]

I HEAR that man...

Yep! I was just posting it for people to see, that question has been asked too many times about "why does my RAM show 800 instead of 1600"

But no matter what, it seems the majority cannot insert within brain cell transmit/receive buffers.

Gave you a post thank you kindly too for hitting the post on upair.

RGone...ster.

 

[Frakk]

Looks like i have some more work to do

Ok, yes, and a good point

I will set about some game benches.

Yet i don't have any games to hand.

BF3 / Shift 2 Unleashed / Crysis Warhead / Ridge Racer Unbounded / Nexuiz and AVP standalone bench. (if there are any Free standalone game benches that i can download please let me know, please)

What i will do is run each one maxed at 1920 x 1080 and also at the lowest possible setting and res to off load the work to the CPU.

Any advice, ED? or anyone.... would be welcome.

 

[RGone]

Looks like i have some more work to do

Ok, yes, and a good point

I will set about some game benches.

Time to be up and attem' there Mr. "Frakk-man". Hehehe. I have never ever seen anyone post any test of any kind that someone did not want some other or more tests run. Never man. Get after it.

RGone...ster.

 

[EarthDog]

Any advice, ED? or anyone.... would be welcome.

Heh... well, thats the tough part without shelling out any money...

Alien vs Predator bench is free.
HAWX2 (though ancient) is free.
Personally I would prefer one to stay away from 'run throughs' as the results could be off and skew your results unless its a kind of canned area (like sitting in a jeep on the mounted gun and all you do is shoot vs complete manual runs on MP maps)