To my knowledge, the only G2s available currently available in Malaysia are the 5000 Black Editions. Xtreme systems managed to overclock to 3.5GHz on good aircooling/normal watercooling, and 3.7-4.0GHz on extreme cooling (DI). After looking over again, G2 overclocks just as well as F3, but unfortunately still lags in actual performance by 0.1Ghz due to L2 cache latency.

Not we all just left..I just got back a short while ago.

Not gonna overclock mine yet, but will do some serious underclocking..wanna see how much performance I can squeeze out of it at ultra low voltage..

Visit silentpcreview often and you'll kena poison eventually. The silence of passive cooling..
Gotta clean my worktable tonight or I'll never get to play with the F3, nowhere to work on that system with the desk cluttered with 4 dinosaur PCs..
Btw, underclocking is really easy with any K8 processor..much easier than overclocking..its with intel systems that its difficult, often need harware mods.

For thermal tape, 3M is the best, but even then it cannot compare with even generic thermal paste. Best of all is thermal epoxy, like Artic Alumina Adhesive, but the bond strength is ridiculously strong (thermal performance can easily rival AS5). Both share the same problem, once you put it on, its hard to take off; for epoxy its 100% impossible without special techiniques (for ram, even the 3M tape may pull off BGA chips if you're not very careful). If you don't want to cause a permanent bond, use normal thermal paste, but put a very small dab of glue at the corners of the chips. UHU glue is commonly used. Another way is suggested by artic silver, mix artic silver with the thermal adhesive/epoxy, it will not be so sticky (test on something else first though).

memtest is hopeless for testing memory stability on AM2. When I overclock the memory, can easily pass memtest 24 hrs, but fail prime95 like crazy (not even 15 mins). Its not the CPU because when I overclock only the CPU at the same speed (ram left at stock or underclocked), its 100% stable (everything pass 24 hrs, prime, memtest, 3dmark, superpi)

Correct, but its not an effective check for AM2 because it does not use the full bandwidth of the AM2 memory controller. Even if memory is stable in memtest, its not guaranteed to be stable in other stress tests (heck, you might not even make it to windows).

Thats the one I used. Good for earlier systems, but not for AM2. Use tests under windows like prime95/orthos. Open the task manager and check the amount of ram free listed under the performance tab, and set the stress test to use up at least 90% of the remaining free ram (not 100%, or the system will start paging). Try to quit as many background programs as possible before checking the free ram amount to make the stress test more effective. Btw, if using orthos, you must halve the number first, as orthos will tell each cores to use the amount of ram you set, which results in double that number of ram tested for a dual core.

This post has been edited by lohwenli: Dec 1 2007, 07:15 PM

PWM temp has got very little to do with the motherboard chipset. It depends on what cooling is used on the PWM, and how much power the processor is using. Usually overheating PSM is due to poor thermal contact between the heatsink and the MOSFETS (dunno why board makers like to use themal pads aka 'bubblegum' which has pathetic themal conductivity). For boards not using heatsinks on the PWM, its almost always because the airflow from the processor HSF is blocked from reaching the PWM (usually due to a whole row of capacitors), also applies to heatsinked PWM, but not so serious.

All busy kua..I also din get chance to crack open my AM2 rig to play with mine yet..got a lot of stuff to test to put up for sale..pokai like hell after buying the F3 and Annys..

HaHA..try OCing Pentium D 800 series...that one TNB meter jalan like car on dyno...

You can use any program, lots of them free. Btw, don't sweat the accuracy of the readings, most of them are just 'lebih kurang' only. Motherboard monitoring chips are not known for their accuracy. Even Brisbane's built in thermal diode is messed up.

DFI has more overclocking options than most mobos (especially voltage), but are highly manual. Have to be very experienced and set everything yourself.

Cool! You should be able to reach 3GHz on just the stock voltage of 1.2v. Most of the Agena B2 achieve that. Tried the AMD Overdrive utility already? Can overclock each core seperately..

Not bad, at least 12-15% improvement on most things even running at the same clock as Athlon. But Phenom badly needs faster cache, and more of it. And also the capability to clock well above 3GHz (3.6 would be nice) on normal cooling. If Shanghai/Deneb does not solve this then things will be very painful for AMD when Nehalem comes out.

More cache is not enough, the cache has to be faster..a hell lot faster..
Even my Northwood can pwn my Windsor in cache benchmarks like I can't imagine how bad it'll be if compared with Brisbane which has pretty slow cache.

RAID driver only works in Windows. Before windows fully boots, your RAID will be running in what I think is called legacy mode (can't recall the actual word). If you do benchmarks in old versions of windows or dos, you will see that transfer rates for most things (HDD, RAM) are badly bottlenecked.

Wohoo, just started overclocking my F3, never has this much fun since my first major overclock back in P4 days (previously main main only)..gonna put up a OC log soon..its gonna be good, though don't expect to see another result like dblooi's..no time to prepare cooling like his record overclock..my tweaking is of a different flavour Sad thing is, only have today to do my overclocking, as that PC is shared by the rest of my family and they are complaining that they don't get to use it..barely even have enough time to stress test the overclock..

Should I post my OC log here or at the POC thread? Its gonna be pretty long..

Both..finished underclocking in just a few hours..
But overclocking took me whole of last night (just to get settings right, stress test short while only)..too long already..mom & sis complaining about comp..


XP-120. When you see the pics in my OC log then you'll know why I picked it. Its definitely no longer close to being the best CPU cooler available.

Wow, you beat my overclock by quite a bit..but then again I didn't have much time to stabilise mine.

Haha, my idea of stable is probably too severe for most overclockers..must pass every stress test minimum 24 hours (prime, 3dmark, ntune and a few others), prime preferably pass for 1 week non stop..I cannot tahan if crash while I halfway doing heavy proc work..render video 3 hours then at 90% crash-that one really tekan man..

It can, but sometimes errors may still make it past the error checking system of the grid processing network. In that case, your PC will actually be a liability, as it will be giving inaccurate results to the network. But the stress on the processor often reveals unstability early (crashes), so this is rare. Anyway, even processors running at stock speed sometimes produce errors due to EMI and power fluctuations, so most processors will still fail stress testing once or twice every year or so.

That might be hard, even with a non-overclocked processor..just one power fluctuation or windows crash and everything have to start over..

Even EMI can cause a very occational memory error and mess things up for you..

Well, EMI & other things are basic problems for electronic devices, especially memory (including the cache of the CPU). Eventually there will be bits that lose data, even if they're not faulty, because of external influence. Its because of this servers built to handle critical applications use ECC ram, which detects and corrects single bit errors(sometimes double bit, depedning on algorithm used). For the average user, windows crashed by itself often enough that you'll probably never realise one of them is probably due to a memory error

Yup, small FFTs which are smaller than CPU cache size will stress the CPU without much stress on the ram. From my experience, testing small FFTs usually reveals CPU unstability in less than an hour. Testing for memory errors takes longer, must use large FFT sizesand usually more than a day to be sure of decent stability. Large FFT sizes are poor for checking CPU stability, as memory is nowhere as fast as cache, and the CPU will end up waiting quite a bit and not being as stressed as it should be.

I don't think you would notice much improvement because of HT3 because its just a faster version of HT2, and even HT2 underclocked doesn't show much difference, meaning its heavily underused. HT3 is most useful for Barcelona & Agena FX multi-CPU systems, as the 2nd HT link joining the 2 CPUs will also be linking the 2 memory controllers together (effectively quad channel ram you really gotta see the memory benchmarks..touching 10k GB/s )

Yeah, its actually slower on the quad core because each core is clocked slower..which is why penryn has a built in OC function to speed up one core above stock speed when it detects that the application is only loading that core and leaving the other idle.

I've heard about it, but no conclusive proof. Conventional electronics theory states that the more stressed a component is, the shorter its lifespan, which is usually characterised with lower and lower performance (eg-transistors switching slower, and requiring more voltage). However, we don't know if there are other effects on semiconductors under stress which may change the end result of that theory.

Btw guys, I won't be around for roughly a week starting tomorrow. About my OC log, still uploading pics to photobucket. Should finish it later today.

This post has been edited by lohwenli: Dec 7 2007, 10:05 AM

Its on idle what..already finish 32m superpi, not calculating any more..

Its not the LEDs but the voltage regulator for the LEDs.


Those are defintely Ballistix Tracers. No way they're 10th anny's.

Uploaded the first part of my OC adventure with the F3. Will put up more this weekend, dad's laptop batt dying now.

Decided to revive an old thread..
http://forum.lowyat.net/index.php?showtopi...post&p=14603277

Yup, cleaned it up before I started. If you look at the later pics there is no dust at all.

There is some effect, but hard to say exactly how much because most motherboards don't have chipset temp sensors.

Using wax is not a good idea as it can form a thin layer which may reduce performance. Even polish is not recommended unless you thoroughly wash the surface later.

It depends, most important is that both processor IHS and the heatsink surface must be perfectly flat and matched to each other. Mirror finish is not so important, I will do a test on this to prove it when I'm free enough. Intel 775 processors suffer a lot from this, the process of attaching the IHS on 775 procs often warps the IHS, causing it to be concave. Many 775 users have reported an average of 3-7C drop after lapping, which shows how bad is the flatness is. However, all the AM2 processors I've examined have pretty flat IHS, even better than most Intel chips, even the 478 ones which did not suffer as bad as the 775 procs.

Metal polish is something like 3000 grit, but keep in mind that while high grit allows for a smoother finish, its also much slower to work with. Need to use low grit (rough) sandpapers first, trying to flatten an uneven heatsink/IHS with 600+ grit is painfully slow, often taking half a day or more. Ever since then, I've always used 300 grit or lower when flattening, only moving on to the next grade once I'm satisfied I cannot improve any more with the existing grade. That way I managed to cut down lapping time to around 1 hour or so, though its not as perfect a mirror finish as some might like. Can still see your own reflection though.