DFI NF4 Ultra Infinity Overclocking

[HoTsHoT]

Hey guys, I hope someone can help me shed some light into my overclocking problem.

My specs:

Athlon X2 2.0GHz, S939 Toledo core at 1.35v
2x1GB Apacer D43s
DFI NF4 Ultra Infinity

So far, this is my first hand in K8 overclocking. I'm trying to overclock the motherboard while keeping other memory and processor settings stock but I have difficulty passing 230MHz unlike other users which can settle on 250MHz with ease. I have lowered down my HTT to stay within below 1000MHz, the processor multiplier to stay within safe range, dropped the memory values and boosted the chipset voltage.

Funny thing is, I've already hit a block where as much people are able to proceed further. Also, putting the FSB higher than 230MHz (incorrect term here, but I'll just use it) causes the NVIDIA RAID BIOS to appear during boot even though I've disabled it.

Maybe flashing a beta BIOS might help?

 

[Tweaked!]

What bios revision are you running? And what other specs do you have setup, like ram timings, locked pci, etc?
Also, at 230fsb, what voltage is the cpu getting, and whats the temps? (at full load) And is that stable under prime for any real length of time?

 

[HoTsHoT]

What bios revision are you running? And what other specs do you have setup, like ram timings, locked pci, etc?
Also, at 230fsb, what voltage is the cpu getting, and whats the temps? (at full load) And is that stable under prime for any real length of time?

Currently I'm using the latest official BIOS by DFI. MY PCI's locked at 100MHz, RAM timings set to 3-4-4-8 1T, the rest left to auto. CPU's voltage is still at default 1.35v, full load at 60C at stock (I'm planning to get something soon to alleviate this, I won't be pursuing any processor overclocking at the moment) as I manipulated the multiplier to stick to stock speeds as I'm testing out how far my DFI can go. The temps report about 36C - 40C for the chipset which I don't think it's a limiting factor just yet.

On stock settings, every test is stable for more than 6 hours on Prime95 Multithread.

I'm currently running Prime95 Blend test with the FSB at 220MHz, 1:1, HTT lowered to 4x, vDDR upped to 2.7v and chipset to 1.6v. Memory settings are similar to above, 3-4-4-8 1T. Two hours have passed, and still running. One funny thing is overvolting things, particularly the vDDR and chipset throws me BSODs and hang ups. I had to test every combination to find the sweet spot.

At first I thought it was a power issue, as I only had an Enermax Coolergiant 480W coupled with a 7800GT. All rails stayed in range, including the 3.3v and 12v.

Another question though, I might be replacing this 7800GT with a G80 8800 GTS, hopefully there won't any power mishaps? I'm just afraid my PSU can't take it anymore with overclocking + new graphic card.

 

[QuietIce]

I take it you're testing your RAM speeds? If not you should turn the ratio down as you increase clock speed. Everything on the K8 is controlled by the clock one way or another. Chances are with RAM at 1:1 you're maxing out the RAM. Most RAM hits a wall for stock settings near the 220-225 MHz mark.

Is 3-4-4 the stock RAM setting?



What PSU are you using ...???

 

[HoTsHoT]

I take it you're testing your RAM speeds? If not you should turn the ratio down as you increase clock speed. Everything on the K8 is controlled by the clock one way or another. Chances are with RAM at 1:1 you're maxing out the RAM. Most RAM hits a wall for stock settings near the 220-225 MHz mark.

Is 3-4-4 the stock RAM setting?



What PSU are you using ...???

Yeah, stock RAM settings at 200MHz is 3-4-4-8. I'm using an Enermax Coolergiant 480W on my rig.

I'm trying to determine how high I can go for my board. I roughly know how D43s scale in overclocking from my NF2 days, and I'm about to get a new budget cooler for my processor. I have two heatsinks in mind: the Thermalright XP-90 or the stock AMD cooler with four heatpipes that comes with the S939 Opterons. I read that the said AMD cooler works about on par with the XP-90, and probably cheaper so I'm eyeing that.

 

[QuietIce]

That 4-pipe cooler works very well - especially if you strap a 92x38mm fan to it! But I hit almost 3.0 GHz with an Opty 165 on a stock cooler - just a hair short at 331x9.

If your stock RAM is 3-4-4 then the only option you really have is to turn down the RAM ratio in BIOS, though you may get a few more MHz raising the vDIMM (but not beyond 2.8v IMO). With tighter stock timings you could loosen them for higher speeds but with CAS 3-4-4 sticks ...

You might find a more favorable ratio using an x9 CPU with a higher clock. 230x10 should translate to 256x9. Running 5:6 (166 RAM) would put you right at 209-210 MHz.



But are you still having issues with the clock speeds? DFI's shouldn't do that, and you shouldn't need to boost the chipset voltage either, not yet anyway ...

 

[HoTsHoT]

That 4-pipe cooler works very well - especially if you strap a 92x38mm fan to it! But I hit almost 3.0 GHz with an Opty 165 on a stock cooler - just a hair short at 331x9.

If your stock RAM is 3-4-4 then the only option you really have is to turn down the RAM ratio in BIOS, though you may get a few more MHz raising the vDIMM (but not beyond 2.8v IMO). With tighter stock timings you could loosen them for higher speeds but with CAS 3-4-4 sticks ...

You might find a more favorable ratio using an x9 CPU with a higher clock. 230x10 should translate to 256x9. Running 5:6 (166 RAM) would put you right at 209-210 MHz.



But are you still having issues with the clock speeds? DFI's shouldn't do that, and you shouldn't need to boost the chipset voltage either, not yet anyway ...

That's the issue that I'm currently facing now, it just refuses to get into Windows over 230 MHz even with the processor set to the lowest multiplier and the memory running at its lowest ratio.

Actually, I got my board fresh out of RMA, I think a little 'training' needs to be done

By the way, do AMD64 PCs work well with ratios? The last time I remember AMD during the NF2 era, performance will absolutely suffer with them. I know this trait was back then, but I would like to know if it's still the same for the NF4.

As for the heatsink, will probably be getting one quite soon. I have two 92mm Vantec Tornados standing by for testing purposes, but not for everyday use as I sleep less than a few feet from my PC.

 

[QuietIce]

I'd keep the processor, RAM, and HT Link set low then try playing with the chipset volts and push to BSOD. If you've already tried that did you put everything over to manual, (RAM timings, voltages etc.) before you did it? IMO, BIOS has gotten too smart in the newer systems. At low multipliers it'll [Auto] change timings and voltage right into a BSOD ...

 

[HoTsHoT]

I'd keep the processor, RAM, and HT Link set low then try playing with the chipset volts and push to BSOD. If you've already tried that did you put everything over to manual, (RAM timings, voltages etc.) before you did it? IMO, BIOS has gotten too smart in the newer systems. At low multipliers it'll [Auto] change timings and voltage right into a BSOD ...

Yeah, was aware of that because the RAM would change its timings accordingly with the SPD table, effectively causing lockups.

So far, memtest86+ stable for 12 hours on 220MHz 1:1. Let's see if I can take this further, will reply here if I get any improvements.

 

[QuietIce]

I forgot about the ratio efficiency you asked about.

1:1 is still slightly more efficient than other ratios but the effect isn't nearly what it used to be. You only see the difference in benchmarks and some argue there is no effect, that it's inside the limits of testing error - that's how small the difference is. The IMC changes everything. Unlike previous systems RAM speed is based off CPU settings, not the clock ...

 

[HoTsHoT]

I forgot about the ratio efficiency you asked about.

1:1 is still slightly more efficient than other ratios but the effect isn't nearly what it used to be. You only see the difference in benchmarks and some argue there is no effect, that it's inside the limits of testing error - that's how small the difference is. The IMC changes everything. Unlike previous systems RAM speed is based off CPU settings, not the clock ...

In that case, running on ratios shouldn't be much of a hindrance to performance.

Just a little update, so far I'm able to do 230 MHz on my board now after a little tweaking on the miscellaneous memory settings from a guide. So far it's only 1 full pass stable on Memtest86, no time to test further.

Funny thing is, I assumed that these settings would only affect memory, but somehow affects the chipset as well.

 

[QuietIce]

Which settings?

With the K8 you've got the following basic settings:
clock speed (stock 200 MHz)

CPU multiplier (per CPU)
HT Link multiplier (stock 5X)

RAM ratio (stock 1:1)
RAM timings (per manufacturer)


I'm not sure what you mean by "chipset" ...???

 

[HoTsHoT]

Which settings?

With the K8 you've got the following basic settings:
clock speed (stock 200 MHz)

CPU multiplier (per CPU)
HT Link multiplier (stock 5X)

RAM ratio (stock 1:1)
RAM timings (per manufacturer)


I'm not sure what you mean by "chipset" ...???

Sorry for the confusion, what I meant in my previous post that the clock speed is now running at 230MHz, where 'chipset' is regarding to my NF4.

To put it more clearly, those miscellaneous memory settings will not only cause extra strain or less of it onto the memory sticks depending on what your settings, but will affect the motherboard stability as well, eg. tighter timings on RAMs force the NF4 to shorten its internal timings too.

 

[HoTsHoT]

240 MHz eight hours Memtest stable, with 5:6 ratio for the memory. Finally I'm getting somewhere.

Also, a very happy Chinese New Year to you guys

 

[QuietIce]

2400 MHz on the CPU? Nice! But you're still pushing, right?


I'd like to see what you've been reading about nF4 v RAM, not in disbelief but for seeking more knowledge. My understanding doesn't go into that much detail on interactions with the chipset. AFAIK the chipset isn't effected by RAM timings at all. As I understand all system RAM requests go through the HT Link then on to the crossbar and IMC, both of which are on the CPU ...

 

[HoTsHoT]

2400 MHz on the CPU? Nice! But you're still pushing, right?


I'd like to see what you've been reading about nF4 v RAM, not in disbelief but for seeking more knowledge. My understanding doesn't go into that much detail on interactions with the chipset. AFAIK the chipset isn't effected by RAM timings at all. As I understand all system RAM requests go through the HT Link then on to the crossbar and IMC, both of which are on the CPU ...

Haha, I wish I could have my X2 at 2.4GHz. Sadly, I haven't started clocking my processor yet, remember?

As for the guide I'm currently referring to, it's right here. I don't have solid proof that the settings DO affect chipset overclocking, i only came across such a situation which is likely so. Reminds me of the 865 PAT days, where the FSB of your installed processor actually determines the type of PAT you have enabled on the chipset.

Also, I've been exhibiting funny things during overclocking. With 260 MHz clock speed on 7x the boot process would get stuck with the NVIDIA RAID BIOS as described on my first post, however, upping the multiplier to 8x solves the problem. There's definitely something fishy going on.

 

[QuietIce]

Old age can be a bane - avoid it at all costs!


It's not fishy if you thing about it - K8's/A64's and above are not like the older systems. The Integrated Memory Controller (IMC) changes everything. Like I said (and your reference source agrees), the chipset really isn't effected by RAM. Since the memory controller is on the CPU I would expect CPU speed to effect RAM timings. And because of that, I wouldn't go into too much detail with RAM timings until you've got your CPU running at it's top speed. (Though I admit, getting to know your RAM intimately is a good thing! ) As you've already seen, chances are the timings will need to be changed when the CPU speed is radically altered - in fact it's one of the things you'll have to correct to get a maximum OC on the CPU.

Testing the top end of the system clock is important but it's just one small step in the process and may change when the CPU multiplier is cranked up. The system clock isn't the chipset, exactly, though it is located in the "NB" chipset. And, on these systems, the chipset speed is almost non-existent in importance.



Let me put it to you this way (and I've actually tested this myself).
Compare:

system clock 250 MHz
CPU speed X12 (= 3000 MHz)
RAM ratio 1:1 (= 250 MHz)
HT Link 4X (1000 MHz)

system clock 300 MHz
CPU speed X10 (= 3000 MHz)
RAM ratio 5:6 (= 250 MHz)
HT Link 3X (900 MHz)


For 99% of all benchmarks the above settings are exactly alike - no differences in performance for the CPU, RAM, RAM latency, nothing. The only place there is any difference at all (that 1%) is extreme benchmark tests of the HT Link but in normal usage you'd never see it. The CPU would top out trying to do enough things to over-tax a 900 MHz HT Link. But, again, all CPU and RAM tests yield exactly the same results. There is no difference in these settings as far as CPU and RAM are concerned and only a small difference at the extreme end of HT Link usage (even benchmarks have trouble taxing an HT Link over 900 MHz) ...

 

[HoTsHoT]

Old age can be a bane - avoid it at all costs!


It's not fishy if you thing about it - K8's/A64's and above are not like the older systems. The Integrated Memory Controller (IMC) changes everything. Like I said (and your reference source agrees), the chipset really isn't effected by RAM. Since the memory controller is on the CPU I would expect CPU speed to effect RAM timings. And because of that, I wouldn't go into too much detail with RAM timings until you've got your CPU running at it's top speed. (Though I admit, getting to know your RAM intimately is a good thing! ) As you've already seen, chances are the timings will need to be changed when the CPU speed is radically altered - in fact it's one of the things you'll have to correct to get a maximum OC on the CPU.

Testing the top end of the system clock is important but it's just one small step in the process and may change when the CPU multiplier is cranked up. The system clock isn't the chipset, exactly, though it is located in the "NB" chipset. And, on these systems, the chipset speed is almost non-existent in importance.



Let me put it to you this way (and I've actually tested this myself).
Compare:

system clock 250 MHz
CPU speed X12 (= 3000 MHz)
RAM ratio 1:1 (= 250 MHz)
HT Link 4X (1000 MHz)

system clock 300 MHz
CPU speed X10 (= 3000 MHz)
RAM ratio 5:6 (= 250 MHz)
HT Link 3X (900 MHz)


For 99% of all benchmarks the above settings are exactly alike - no differences in performance for the CPU, RAM, RAM latency, nothing. The only place there is any difference at all (that 1%) is extreme benchmark tests of the HT Link but in normal usage you'd never see it. The CPU would top out trying to do enough things to over-tax a 900 MHz HT Link. But, again, all CPU and RAM tests yield exactly the same results. There is no difference in these settings as far as CPU and RAM are concerned and only a small difference at the extreme end of HT Link usage (even benchmarks have trouble taxing an HT Link over 900 MHz) ...

Thanks for the very detailed explanation QuietIce, it made me think twice about my pursuit for a high board clock speed

Actually, my intentions to acquire such a high clock speed is due to my CPU's low multiplier - 10x. Initially, if I had to settle on 220 MHz, the resulting CPU speed would be only 2.2 GHz. Let's just say that I want to be more ambitious than that

Truth be told, I have no idea on whats to come when I push my CPU speed further. My plans in a nutshell - find the max speed on the motherboard, then jack up the multipliers, and add CPU voltage as needed within tolerable heat levels.

I absolutely forgot that the memory controller is now on the CPU till you reminded me, which means that I should have placed more importance onto that component in the first place. What I have been doing for the past few days could probably be for nothing!

I'll finish off what I'm currently doing now (my board won't hold out the increasing pressure I'm giving it anytime soon), and get to overclocking the CPU as soon as I find a cooler

 

[QuietIce]

The DFI should clock into the upper 200's. They change between models but most good boards will go 280-90 MHz. With the RAM set to 133 (or 2:3), the CPU at 7x, HT Link at 3x and stock voltages and timings (set manually) you should be able to boot 280-85 with that board. 280x7 = 1960 MHz CPU speed and 1960/11 = 178 MHz RAM speed. That may need a little chipset voltage boost but probably not ...

 

[HoTsHoT]

The DFI should clock into the upper 200's. They change between models but most good boards will go 280-90 MHz. With the RAM set to 133 (or 2:3), the CPU at 7x, HT Link at 3x and stock voltages and timings (set manually) you should be able to boot 280-85 with that board. 280x7 = 1960 MHz CPU speed and 1960/11 = 178 MHz RAM speed. That may need a little chipset voltage boost but probably not ...

I've finally hit a barrier for my board - about 260 MHz. Increasing the chipset voltage to 1.7v doesn't seem to increase overclocking headroom.

I've got another problem here. I did 225 MHz 1:1 yesterday on 9x multiplier, 5 hours straight on Memtest86 flawlessly. Under P95 Blend test it fails less than a minute. Any insight on this one? I've tried going to 8x and 10x, increasing voltages and decreasing them, but no go.