Tuning your AMD CPU for Gaming

[Aldakoopa]

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!

It seems I need to learn more about HT link and CPU-NB... research time.

 

[ATMINSIDE]

It seems I need to learn more about HT link and CPU-NB... research time.

I've always been told to keep them the same.

 

[Frakk]

Have now managed to overclock the CPU-NB only. (without the need for memory adjustments)

Edited: NFS Shift 2.

Added: 3DMark11 with 'Only' CPU-NB set higher.

Added: BF3 Bench.

http://www.overclockers.com/forums/showthread.php?t=712972

 

[Jagged_Steel]

My Signature with my rig specs are not showing yet, but I am running :

Fx-4170 @4.86 Ghz
ASUS M5A97 EVO
16 Gb SAMSUNG 30 nm WonderRAM @ 1890 8-9-9
2 x PowerColor AX-6790 Crossfired

Here is what I score on the Shogun 2 1080p scripted benchmark :



That is with my 6790's OC'd to 900/1100 over the stock 840/1050

Here is what I score on MaxxMEM.



The MaxxMEM "score" does not reflect actual throughput performance on AMD chips, although it seems consistent in the amount it lowers the "score" among AMD CPUs, so it is a fairly good way to measure relative throughput potential. It is a fairly popular benchmark so it is worth running just because of the mass of other benchmarks to compare to.

More thorough and properly coded memory benchmark programs like SiSoft Sandra rate my setup at 19 GB/s

 

[EarthDog]

What do you mean it lowers it b/c its an amd cpu?

Isn't thus thread about nb sped testing? Did u test that or......what's going on?

 

[Jagged_Steel]

I've always been told to keep them the same.

That isn't quite right. What you want is your NB to be as high as possible, and your HT should be around ~2000 Mhz. There seems to be a wall of performance return around 2000 Mhz for the HT link. This may vary a bit from chip to chip but in my experiments I found performance dropping when I set it higher on both Phenoms and FX. The NB is of course the opposite, every tick higher you can get on the speed there is worthwhile.

 

[Bubba-Hotepp]

Nice to see you made it over from OCN Jagged. I think you'll find this is a much les "hostile" forum for AMD owners.

E.D. You must have been tired. Your typos are showing

This is just what I noticed on my 6100 so take it for what it's worth. I leave the HT link on auto as it doesn't really affect results all that much. As for CPU/NB I keep it at 2200 @ 1.2v. When I overclocked it more I couldn't justify further bumps as the performance differences I was seeing didn't justify the added voltage (and thus added degrees C). I might have to play with it some more when I get my rig back together (using the PSU and GPU for something else) and see if I can up it anymore and stay at 1.2V.

 

[Jagged_Steel]

What do you mean it lowers it b/c its an amd cpu?

Isn't thus thread about nb sped testing? Did u test that or......what's going on?

I am sorry that I was not clear about why I was posting these and their relation to tuning your NB for Gaming, my post was not quite thorough enough, so I will explain.
Those benches I posted show the results of a couple weeks of tuning this FX-4170 exploring to find the absolute best throughput possible, experimenting over and over with inching the bus up and then checking each bus frequency with it's best NB speed at that setting, repeating, etc etc. Going back and then finding the top memory speed and experimenting with tightening the timings until failure, etc etc. The point I was trying to make was that I agree with the OP about the importance of maximizing your NB speed, and therefore throughput, as it shows noticeable gains in Games. Which is why I tossed out the S2 1080p bench. I thought I had a before fine tuning one, but I don't see it in my pics. I was scoring around 56 before I fine tuned the NB and memory with this setup, so we are talking about ~5% fps gains, which is definitely worth taking the time to fine tune your memory and NorthBridge for IMO.

About MaxxMEM and AMD CPUs: I like to use MaxxMEM to dial in memory/NB settings because it is a really quick test, and it finds major instability instantly. It does not give true throughput values for AMD for whatever reason. From what I can gather Intel systems score pretty much equal on both MaxxMEM and other memory benches like SiSoft Sandra, but with AMD rigs the MxxMEM total throughput score is lower by ~30-50% compared to results from other major memory benchmark programs I have used. You can see what I am talking about when you look at the Sandra bench I added. Sandra is great but from boot it takes a few option choices and then about a 3-4 minute test to get your results, whereas with MaxxMEM you can pull it up after boot in seconds and with 1 click know the results of your tuning effort 30 seconds later. The score does not reflect the real throughput of the system, but it is consistent so you can use it to check settings yields when tuning. That's what I meant by MaxxMEM not being completely accurate yet still useful.

@ Bubba: Good to see you. I should have my CHV Monday (already got my new case and TT W2 Pro W/C), so I will be burning up a bunch of benchmarks next week trying to get the new incarnation of the Cannonball Fxpress tuned up. I will probably be around this weekend too, just to compile a decent record of how this rig functions now so I have something to compare it to after the big overhaul.

 

[Kalistoval]

My Signature with my rig specs are not showing yet, but I am running :

Fx-4170 @4.86 Ghz
ASUS M5A97 EVO
16 Gb SAMSUNG 30 nm WonderRAM @ 1890 8-9-9
2 x PowerColor AX-6790 Crossfired

Here is what I score on the Shogun 2 1080p scripted benchmark :

View attachment 113948

That is with my 6790's OC'd to 900/1100 over the stock 840/1050

Here is what I score on MaxxMEM.

View attachment 113949

The MaxxMEM "score" does not reflect actual throughput performance on AMD chips, although it seems consistent in the amount it lowers the "score" among AMD CPUs, so it is a fairly good way to measure relative throughput potential. It is a fairly popular benchmark so it is worth running just because of the mass of other benchmarks to compare to.

More thorough and properly coded memory benchmark programs like SiSoft Sandra rate my setup at 19 GB/s

View attachment 113950

Hey jagged and bubba I've gone ahead and switch on over here too heres

 

[Aldakoopa]

I have a possibly noobish question but it's had me confused; Isn't the NB on the chip ever since Athlon, therefor including Phenom and FX, and, if so, why is there still an north bridge on the motherboard?

 

[Kalistoval]

I have a possibly noobish question but it's had me confused; Isn't the NB on the chip ever since Athlon, therefor including Phenom and FX, and, if so, why is there still an north bridge on the motherboard?

you have 2 north bridges the cpu its self has a northbridge usally known as a cpu/nb and then you have your mother boards Northbridge the chipset on motherboard say for example my crosshair V has a 990fx chipset its also known as my northbridge

they both operate at different voltages too

 

[Bobnova]

The memory controller moved onto the CPU on socket 754.
The motherboard northbridge still has the PCIe controller, IGP, a few other controllers and the link to the SB.

 

[Lordred]

It's a dirty job, but some one has to put those Samsung 30nm users in place!

G.Skills steping up to the plate (ambients are still a bit high for running this fast, winter is going to be fun )

 

[Jagged_Steel]

It's a dirty job, but some one has to put those Samsung 30nm users in place!

G.Skills steping up to the plate (ambients are still a bit high for running this fast, winter is going to be fun )

I did not realize this was a memory competition thread. That is good score, but you are running only 8 Gb. I am pretty sure I can get above 13 Gb/s on MaxxMEM with 8 Gb too. I lost about .5 GB/s when I added the second set back when I was running a 555 so I think it will rise again if I pull out a couple sticks. When my CHV arrives in the next couple of days I will probably do some speed runs to find out. I am not aware of anybody claiming that Samsung 30 nm RAM is the absolute fastest. What they are is an incredible value for the price you pay. Your post supports that. I paid $39 for my first 8 Gb and $47 for the second 8 Gb set I have. You can barely get a higher score with crazy settings pushing your rig to the brink of thermal meltdown on 8 GB of more expensive RAM compared to what I can get with 16 GB of Samsung 30 nm with my daily driver settings. I am ok with that.

P.S> If you really want to get your throughput score higher you should work on getting your NB above 2600 Mhz. Tuning your Gaming system by getting the the highest NB you can is in fact the subject of this thread, BTW.

 

[EarthDog]

This thread has to do with the NB speed and how it affects your FPS.

 

[tikithorsen]

This is a very interesting and exciting thread, props to Frakk!

I'll do some tests with my CPU-NB at stock and then 2600MHz (or @ 2800MHz if I mange to get it stable) and post later tomorrow the results.
¿The "max" voltage for the CPU-NB would be 1.250v?

Also, I was left a bit confused regarding the HT Link;
What I understand about the HT Link is that, If we ain't running Crossfire there is no need to increased past the default value of 2000MHz. ¿Is that right or I'm way of here?

 

[Frakk]

This is a very interesting and exciting thread, props to Frakk!

I'll do some tests with my CPU-NB at stock and then 2600MHz (or @ 2800MHz if I mange to get it stable) and post later tomorrow the results.
¿The "max" voltage for the CPU-NB would be 1.250v?

Also, I was left a bit confused regarding the HT Link;
What I understand about the HT Link is that, If we ain't running Crossfire there is no need to increased past the default value of 2000MHz. ¿Is that right or I'm way of here?

I found turning up the HT Link to 2200Mhz (and no more) is because it can help with stability when running a high CPU-NB. Other than that it has no effect.

Oh and thanks

Anyway, i might add 1 or 2 more game benches later on when i have time, i will also add that HT Link comment to the initial post, and see if i can clean the whole thing up a bit, i think it might need it.

On the memory, i honestly don't mind at all if my threads go off topic, i actually think its a good thing as long as its hardware related. And the memory discussion is interesting. It might deserve its own thread.

 

[BeepBeep2]

The only reason you need to turn up the HT Link to 2200Mhz (and no more) is because it can help with stability when running a high CPU-NB. Other than that it has no effect.

Oh and thanks

Anyway, i might add 1 or 2 more game benches later on when i have time, i will also add that HT Link comment to the initial post, and see if i can clean the whole thing up a bit, i think it might need it.

On the memory, i honestly don't mind at all if my threads go off topic, i actually think its a good thing as long as its hardware related. And the memory discussion is interesting. It might deserve its own thread.

No, and I don't know where you heard that. Also, the statement about Athlon and Phenom IMC being damaged by OC is absolutely ludicrous...you need to be able to understand the topology of these CPUs and the way they work...the IMC is the same design on all Athlon II and Phenom II parts, even the Semprons.

I don't know where you heard a lot of things, but changing HT Link between 200 (yes, two hundred) and 3000 MHz (provided you have the CPU-NB to get there) makes no difference in stability.

 

[Frakk]

No, and I don't know where you heard that. Also, the statement about Athlon and Phenom IMC being damaged by OC is absolutely ludicrous...you need to be able to understand the topology of these CPUs and the way they work...the IMC is the same design on all Athlon II and Phenom II parts, even the Semprons.

I don't know where you heard a lot of things, but changing HT Link between 200 (yes, two hundred) and 3000 MHz (provided you have the CPU-NB to get there) makes no difference in stability.

Just seems to run smoother. every Chip is different, you know that, that's why i say things like 'can' and not 'will' if people are having problems or it does not feel quite right they then know to try that.

Also, will you lay off the IMC already, repeating it 3 or 4 time in a thread does not make it any louder, relax.

 

[BeepBeep2]

Just seems to run smoother. every Chip is different, you know that, that's why i say things like 'can' and not 'will' if people are having problems or it does not feel quite right they then know to try that.

Also, will you lay off the IMC already, repeating it 3 or 4 time in a thread does not make it any louder, relax.

Frakk, please show proof. "Seems to run smoother" has nothing to do with "can help stability", ...because placebo can go a long way.

I can show proof.

Thanks.