D9GKX doesn't need so much voltage to do high clocks, but hits a wall sooner than D9GMH, and even votlage usually doesn't help much. D9GMH is a bit like the old Winbond BH5, if you are willing to pump volts into it you'll be rewarded with high clocks. How long will it last under those voltages still remains to be seen. With fan cooling, 2.4-2.6v has not been seen to cause rapid damage yet.

It seems to be a problem ever since AMD started using SOI technology. The plus point is, idle power consumption has always been low, only recently did Intel manage to beat AMD in that segment, and only using 45nm.

Cache plays quite a big part in superpi (and gaming and certain multimedia apps) because nowadays processors have reached sufficiently high IPC (Instruction per clock) rates that memory latency (not bandwidth) is now a big problem. With enough cache, programs that can fit nicely into the cache will run at incredible speeds compared to those which must run from the main memory.

The single core F3 mentioned here has 1mb cache, which makes the 512mb cache of most of the dual cores look painful by comparision. Yes, I know they have 2 L2 caches, but superpi is a single threaded application which runs on only 1 core, which cannot make use of the other core's cache, which was a disadvantage intel removed by allowing both cores to share cache. AMD's response was a shared L3 cache, but the L3 cache is not yet (keyword here) fast or large enough to make up for the performance difference. Hopefully Deneb will have more cache, with lower latency, and will finally be able to compete with C2D toe to toe, preferably Penryn.

Actually when you sit back and analyse the numbers of Phenom, if you add up the performance differences due to cache, TLB errata, and clockspeed, it actually matches C2D's (conroe's) performance. And Penryn's performance advantage comes almost 100% from the larger cache size and higher clockspeed. How to compare? Use similar spec cases, like the E4xxx series vs Agena, both have almost the same amount of cache (less than 128k difference), set to the same clockspeed, TLB workaround disabled on Agena, and running only single thread apps or apps restricted to 2 cores (using task manager). There has been actual comparision done yet, but from a quick cross comparision of all the current reviews available, my estimate is there should be less than 5% difference, possibly even 1-3% only.

---

Added on December 30, 2007, 12:28 ammemtest (the one that runs at boot) is almost 100% useless for testing memory stabilty for AM2 (it still can be used to detect faulty memory though). I've had this happen to me too often-memtest pass 24hours, but fail prime in less than 30 mins. Mind you, in those cases the processor either wasn't overclocked (memory on 800 divider) or was at a almost 100% stable overclock speed (prime can pass 48 hours if memory not overclocked)

This post has been edited by lohwenli: Dec 30 2007, 12:28 AM

It heavily depends on chipset. Some kingstons come with good chips (Hynix FP-Y5 was not too hard to find at one time, and I've seen pics of some with even D9GMH and D9GKX-for those chipsets DDR1000 is not difficult, and if you're lucky AND skilled DDR1200 is possible). But nowadays they've remarked the chips, so you can't tell any more.

Don't worry, board manufacturers have learnt from their mistake during the K7 & early P4 era (they lost a lot of money cos had tons of RMA claims). Even Abit changed to using 100% Rubycon caps starting with the NF7 boards (those are good & costly caps, very hard to get in malaysia, usually need special order). Cap leaks/failures are now quite rare, and are pretty much impossible on solid state caps (which cost up to 5-10 times more than normal electrolytic caps).

For me, the only thing I don't like about biostar is most of their current boards are fairly limited in vdimm options, and early BIOS often not very good, modded bios is often required to do good OCing (later BIOS fixes this to some extent). But love the Tforce series UV components, and they're decently priced and still have good OC options.

I think you mean the Power MOSFETS..its not possible to heatsink capacitors..
And if you put enough current through them that they need heatsinks, even solid state caps will arc & blow up..

Close to useless putting HS on top of caps..too little contact area to make much difference..
Anyway, PWMs produce less than 40W heat even with 200W processors, a small medium speed fan blowing on them is enough already.

Typical DFI. Not for noobs..they should put a warning label on the box..


I'm guessing its because of the cache.

Why not Phenom Black edition and 3870X2 CrossfireX?

---

Added on January 5, 2008, 8:02 pmI had to increase slightly to hit 360HTT, but not much.

OC log of my NF-M2 nview (warning-lots of pics)
I did not take any time to optimise this, but even then, I've not seen another higher overclock fot this board.


All AMD cores are made in Germany, and the wafers are shipped elswhere to be packaged into the PCB & IHS.

Your case is kind of weird..I had something like that, but bumping up the voltge solved the problem.

This post has been edited by lohwenli: Jan 5 2008, 08:02 PM

Last I heard, it will be available quite soon, within Q1. And it will not cost signifantly more than the 9600, and default clock is same as 9600 but multipliers are unlocked. I hope they don't cancel it, though there is one thing bad, the initial batch will be B2, and will suffer from the TLB issue.

For most applications HT link speed has no effect. I've heard that there is some effect in gaming when the graphic card used is high end, but I cannot verify this yet. Your overclock from 3000Mhz to 3200Mhz also brings little perfomance increase, as its less than 10% increase in clockspeed. Overclocking higher is not practical though, the Windsor F3 you're using can overclock to about 3.2-3.4Ghz at most using air cooling, its possible to hit 3.7+Ghz but only with extreme subzero cooling. Temps below 65C should be ok, though you must check under 100% load (both cores).

From my experience, the CPU reading in speedfan gives a more believeable temp, but the Core reading responds faster, indicating it might be the internal diode (which is supposed to give more accurate temps, but I've seen some somewhat ridiculous readings-suspect that its not calibarated properly).

In my case I did get one weird result when the processor is running cool. In at least one of the screenshots during my underclocking, the CPU temps is below the ambient temp.
http://forum.lowyat.net/index.php?showtopi...post&p=14653750

Read my OC log..should be very helpful for you
http://forum.lowyat.net/index.php?showtopi...post&p=14653840

The first link in my OC log is broken, it supposed to be this
http://forum.lowyat.net/index.php?showtopi...post&p=14603277

Most ram heatsinks are useless la..many of them even trap heat and make overclocks worse.

I have a micrometer, you can borrow it if you can figure how to get it from me in penang and return it when you're done..

Try to beat my HTT record on the nview..hoping you can reach 400.
http://forum.lowyat.net/index.php?showtopi...post&p=14603943

It can clock a lot better than G1(3.0Ghz with stock cooler & stock voltage is quite possible), and the heat & power consumption barely rises even when overclocked.

I'm not sure about this, but the cache latency seems to be unchanged from G1.


Lol..almost like removing the IHS, but with the edge remaining as a shim..
May have some slight contact problems though, you gotta be very even on all sides.

I think AMD has absolutely no chance to take back the crown until their 45nm chips (Shanghai/Deneb) are out, and even then I think it will be a close competition with Nehalem, which will be coming out around the same time (end of this year). It will be a race to see who is first to arrive and take the crown. The good news for AMD is that despite Intel's claims that Nehalam will beat Penryn by 20-30%, I doubt it will be anywhere close to that, since the biggest performance-improving factor known is the integrated DDR3 memory controller and Quickpath bus (Intel's version of Hypertransport). Most also I think there will be a 10-20% improvement, which was what was seen when AMD moved from K7 to K8. Also, Deneb will have a substantially larger cache than Agena (last I heard, ~6mb total L1, L2 & L3 combined), so this removes one of the largest stumbling blocks for AMD (AMD loses big time in anything that uses lots of cache-gaming, multimedia encoding/decoding, benchmarks).

2nd good news for AMD is that its unlikely that Nehalam will clock higher than Penryn, even on extreme cooling Penryn doesn't clock much higher than Conroe (though it produces so little heat that you can hit 4Ghz on air cooling). This shows that there is a major technological block to achieving more than 4Ghz on current silicon technology, and its highly unlikely that it would be solved by the time Nehalem is out. For point of reference, the max overclock for Conroe hasn't changed much since it came out over a year ago, and Penryn hardly improves on that. AMD has a chance to work on their design and produce a chip that can match Intel on clockspeed, but they much get it done within the year or things will get difficult. Clockspeed should improve when AMD goes 45nm, and I'm praying they will start using Si-Ge in Deneb, though I doubt it will happen though research is underway at AMD. Si-Ge integrated circuits can clock spectacularly higher than pure Si circuits, for identical designs a pure Si circuit that maxes out at 4Ghz can reach 6-12GHz when built on Si-Ge.

Another plus for AMD platforms is that they will not require any big system changes, while its unavoidable for Intel users that they'll have to change their motherboard for sure (the circuit is 100% different). Last I heard, even AM3 processors will be able to run on AM2 boards and DDR2, even though they're supposed to run DDR3. Nehalem will almost surely run only DDR3.

The bad news for AMD is that Intel will have plenty of experience with 45nm production by the time Nehalem arrives, and its unlikely Intel will make any mistakes for AMD to take advantage of. Any mistake by AMD will be fatal at this point, and it doesn't help that AMD is having financial troubles due to the Barcelona TLB screw up and the overpriced purchase of ATI.

Thats quite interesting, I need to try it out on my nview. Seem to be getting a slightly worse overclock than most of the F3 users, though its partly because the place where I overclock is pretty warm.

Last I heard, the highest clocked dual core Phenoms (kuma) are clocked similarly to the quad core Agenas. In general though, as I recall most of the kumas are clocked higher than the agenas.

But one thing is for sure, Kumas and Tollimans (X3) will not arrive till Q2. By then 45nm is already on its way (Shanghai/Deneb). In fact, Shanghai is already in production, though the current revisions may still be improved on before producing the revision that will be launched hopefully end of this year.

This post has been edited by lohwenli: Jan 15 2008, 09:03 AM