Tuning your AMD CPU for Gaming

[Frakk]

Per OP {Frakk}"Before i go on i should say that Phenom and Athlon CPU are designed to run 1333Mhz, you should never run Phenom or Athlon CPU's with a memory clock much higher than that, if anything, as that could damage the CPU IMC which is inherently fragile on those CPU's.

Some may disagree but i have looked into this extensively and there are enough stories out there of people burning out there Phenom / Athlon CPU IMC while trying to run 1600Mhz RAM speed for extended periods, i run mine at around 1400Mhz and no more, my advice is don't, However just for the interests of this post i will make an exception."

Hi Frakk, thanks for your efforts, nice work...

Your advice made me a little worried about my current set up. I have 2 sticks Kingston DDR3-1600 PC12800 RAM installed and working w/a Phenom II X4 965 BE and M5A97 EVO MoBo. I set timings to 9-9-9-27-36-1T and CPU/NB to 2600MHz (x13 NB FID) w/ Voltage @ 1.2v (+0.1v over stock). The RAM is the XMP type, so CPU-Z shows max bandwidth=PC3-10700 (667 MHz)...My impression AFTER buying this RAM (long story about why...) is that it is really made for an Intel system...but Kingston "guarantees" this RAM to run at this higher bandwidth/frequency, so I set BIOS to timings/voltages, etc. to use it at this rated (non-JEDEC) spd.

I benched/stability tested a lot of different set ups - and my current one is "working". IntelBurnTests + Unigine Heaven BMs run fine w/ no errors...have not yet run Memtest86 for 3+ hours so not 100% sure it is stable set up.

The problem w/ BF3 crashes/instability was vexxing me to no end...I narrowed down the problem to the Overclocking of the CPU. I mean I tried lots of different things to approximate a BF3 Full Screen situation. At first, it seemed like the number of threads was an issue lol. (ok I am not Computer Engineer) Like there was a "cap" on the number of threads that the CPU could handle (approx. 840 +/-), so I ran 12 IntelBurnTests simultaneously w/ custom settings of like 8 MB RAM/128 threads/2000 runs...No problems..Ended up stopping this line of testing at around 2200 +/- threads.

Testing was irritating because I did not know how to monitor things while running tests...luckily AIDA64 has that SensorPanel feature...So I set one up with ALL sensors displayed and finally noticed NB vid was not updating as I expected...i.e. Running I.B.T. + Unigine's Heaven simultaneously on Extreme Tesselation got me close to BF3's Load Demands (CPU 80%/GPU 99%/4GB Ram useage +/-/GPU dedicated 600MB +/-/GPU Dynamic 200+/-) , but under these tests NB vid did indeed change to 1.2v (as set in BIOS). BUT, when running BF3..NB vid would drop/remain @ 1.1v??!??. AS you are aware I am sure NB must be above stock to run @ 2600MHz. Changing P-States via K10Stat forced NB vid to 1.2v and everything seems nice and stable atm...

Happily "tested" BF3 yesterday for maybe 3-4 hours w/ no crashes or freezes.

Any thoughts or advice on improvements I could make - either monitoring/testing/stabilzing?

Thanks.
Lothyr

Hi Lothyr

Can you copy that into a separate thread in this room? i'm not pushing you off, that just looks like something that may need something more substantial than a quick fire answer, and there are lots of people around here who can help you out with that.

Also list all your hardware in detail, that helps with diagnosis.

 

[Lothyr]

NP...Sorry if I went off on a tangent

 

[Frakk]

Yeah, your right. I had a brain fart. I forgot for a moment he was talking about a phenom II.

To Frakk, since I was just reminded that you're using a Phen II , you should be keeping the CPU/NB at 2000 for that CPU. Upping it above the HT can create some "wonkiness". On top of that I'm 99% positive the increases you're seeing from the first post have more to do with the mem speed increases than from the CPU/NB speed increases. If you were running a BD with OC'd RAM (2133+) then it would be a little bit of a different story but even then the gains are very minimal. When I'm done testing the Via Nano board I'll put my BD rig together and post screenshots showing the differences.

Ok, i will try and adress this now, i don't know what you mean by "Wonkiness" but i have set the memory to a lower deniminator which has now given me 1220 Mhz on the RAM, thats now 100Mhz Lower than the RAM i ran with the CPU-NB at stock.

And yet the Physics score is still way up.

The Memory speed is now entirely out of the equation, that is purely the CPU-NB now and still scoring 10% higher.

I ran the NFS game again.

Res 1024 x 768, game settings everything minimal.

Before

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

Min 100 / Max 167 / Avg 133.728
(+8%) (+15%) (+10%)

After

CPU: 4.1Ghz - CPU-NB: 2750Mhz - MEM: 1220Mhz - GPU: 1175 /1350.

Min 97 / Max 161 / Avg 133.218
(+7%) (+14%) (+10%)


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

Before

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

Min 70 / Max 109 / Avg 94.331
(+9%) (+1%) (+5%)

After

CPU: 4.1Ghz - CPU-NB: 2750Mhz - MEM: 1220Mhz - GPU: 1175 /1350.

Min 69 / Max 107 / Avg 94.345
(+9%) (+1%) (+5%)

As with 3DMark 11, a very slight dip but still ~10% up with the CPU-NB

 

[Mandrake4565]

Frakk, I ran the tests I said I would do with Aliens vs Predator

1st one Multi 20, FSB 200.68, Mem 668.9, NB 2000, 2nd everything exactly the same with NB at 2600

 

[Frakk]

Frakk, I ran the tests I said I would do with Aliens vs Predator

1st one Multi 20, FSB 200.68, Mem 668.9, NB 2000, 2nd everything exactly the same with NB at 2600

That's the same difference, or lack of, i got. See page one

 

[EarthDog]

Hmmm.. I wonder if turning up the details show any difference due to more data and bandwidth used? There is a 'front end'(GUI) to the benchmark BTW... I cant link to it from the office though...

 

[Frakk]

Hmmm.. I wonder if turning up the details show any difference due to more data and bandwidth used? There is a 'front end'(GUI) to the benchmark BTW... I cant link to it from the office though...

in plain english?

 

[Bobnova]

I'm interested to see continued testing on this. Ideally with the CPU-NB speed isolated from the ram speed. That should be doable by running at stock with the ram at 1600 (or 1333) and then upping the HTRef till the next multi lower gives the same speed of ram.

In the benching world the usefulness of the CPU-NB speed is known as well

On the IMC thing: Memory frequency will not damage the IMC. The IMC may not be capable of running faster, but that is very different than being damaged.
I know you don't want discussion of that in this thread, and to that end I strongly recommend that you take the erroneous statement out of the first post. Being as that it is in the first post right now it is, IMO, on topic to discuss it.
Alternatively feel free to post/PM some proof on that statement.

 

[Bubba-Hotepp]

^---- I agree.

 

[Frakk]

I'm interested to see continued testing on this. Ideally with the CPU-NB speed isolated from the ram speed. That should be doable by running at stock with the ram at 1600 (or 1333) and then upping the HTRef till the next multi lower gives the same speed of ram.

In the benching world the usefulness of the CPU-NB speed is known as well

Is that any different to setting the RAM to 1066 and then using the FSB in conjunction with the CPU-NB multiplier to gain a higher CPU-NB, (IE 2750Mhz) and a RAM speed the same or less than the bench with the CPU-NB at stock (IE 1333Mhz) or less?

In other words the CPU-NB is listed above you @ 2750Mhz and RAM at 1220Mhz provides a ~10% performance improvement in 3DMark 11 and the NFS game vs CPU-NB @ 2000Mhz and RAM @ 1333Mhz.

 

[Bobnova]

Fewer variables, mostly.

 

[Bubba-Hotepp]

The problem is you're changing too many settings at once. Look at the first two. You go from cpu/nb-2650 mem-1450 to 2750/1220.

 

[Aldakoopa]

Is overclocking the CPU-NB on an FX processor the same story as with Phenom II?

 

[Frakk]

The problem is you're changing too many settings at once. Look at the first two. You go from cpu/nb-2650 mem-1450 to 2750/1220.

Ok, i'm not going to go a 3'rd time into why, i already feel like i'm going round and round in circles with that

Would you agree that setting a lower RAM speed along with a higher CPU-NB resulting in a ~10% improvement eliminates the RAM speed as a cause for that improvement?

[Edit] What i can do is give you 2000 / 2200 and possibly 2400Mhz at stock, i have to keep it within 200Mhz of the HT Link unless i use the FSB for reasons already given, twice

@ Aldakoopa, don't know...

 

[Bobnova]

Is overclocking the CPU-NB on an FX processor the same story as with Phenom II?

More or less, yeah. You're still OCing the memory controller and L3 cache.

 

[Frakk]

Have not forgoten about this, more benches to come and existing ones redone with Bubba-Hotepp and Bobnova's suggestion implemented, as much as i can

 

[RGone]

As much as you can with this problem >> 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. << If you have that problem, you have a rough cpu to do the benches for sticklers. I don't really see how you can do the testing with that cpu issue. Well not without going around the world.

You might try the benches with a lesser NB multiplier and test within the multipliers your cpu will actually lock down. We have all seen enough oddities result when upping the FSB/CPU Frequency to greatly wish you could test by only changing the NB frequency.

 

[Frakk]

As much as you can with this problem >> 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. << If you have that problem, you have a rough cpu to do the benches for sticklers. I don't really see how you can do the testing with that cpu issue. Well not without going around the world.

You might try the benches with a lesser NB multiplier and test within the multipliers your cpu will actually lock down. We have all seen enough oddities result when upping the FSB/CPU Frequency to greatly wish you could test by only changing the NB frequency.

The problem with small CPU-NB bumps is the performance increase is also small, its like the diffrance between 3.6GHZ on the CPU and 3.8Ghz... what we will have then is, oh its only 3%, its nothing, its not happening, get real.
So to counter that i need to push it up a bit more to show a real diffrance.

Its fine if i churn up the HT Link with it, i just don't know how high i can get that to go with the NB, anything over 2200Mhz on the MA597 and it would drop all imput devices.

On the Sabertooth i have had it to what i think was not far off 2300Mhz and it was fine.

time to find out how high it will go, i guess

 

[BeepBeep2]

anything over 2200Mhz on the MA597 and it would drop all imput devices.

Then your board is defective, please RMA.
This is not a front-side bus like old systems, and not like AMD Llano where everything is tied to BCLK.
Boards are rated 2600 HT. Almost all will go 3200+. Max multi is 13x (2600) which is the rating, I personally have done 3400.

"Before i go on i should say that Phenom and Athlon CPU are designed to run 1333Mhz, you should never run Phenom or Athlon CPU's with a memory clock much higher than that, if anything, as that could damage the CPU IMC which is inherently fragile on those CPU's."

Bobnova touched on this too, it is incorrect, and if it is correct, it is better to be specific, because many Athlon II CPUs are Phenom II die (however both use the same IMC) and some even go higher than the normal Phenom IIs.

 

[ScorpiDragon]

My PC is running a lot faster and this is why Windows 7 is now performing better with my AMDs Bulldozer based FX 4100 CPU.

Four reasons why the FX 4100 was slow:
1/ Default HT link is 2400 MHz. It’s now 2520MHz.
2/ CPU-NB link was 2100 MHz. It’s now 2520MHz (same as HT).
3/Bumped HT FSB from 200MHz to 210.
4/ Had to tweak voltages (not by much either).

The board I'm using is Asrock 970 Extreme4 and it's rock stable now running at 4.3GHz and Windows 7 is so responsive, I've already tested ConvertXtoDVD4 and it encoded three 1 hour TV shows at 190 FPS. Now that was Fast!