Wesley's Tool-Box

Saturday, August 5. 2000

Followup to Unlocking Duron

Followup article, Unlocking Aftermath and Overclock is up for your viewing. See how far I had pushed my now-unlocked Duron.

Posted by Wesley in News:Tool-Box at 06:50 | Comments (0) | Trackbacks (0)

Remember this chip from my previous article? This is a Duron 650MHz. I chose Duron over Thunderbird because of overclockability. The high quality of the AMD chips ensure a fair amount of overclocking already, but Duron's default voltage is 1.5V as opposed to Thunderbird's 1.7V. Since the maximum voltage setting for Socket A is generally 1.85V, Duron has better headroom for overclocking as higher voltage generally helps the chip operating in higher speed provided that a suitable cooling solution is present. Many hardware sites have confirmed that this voltage headroom does indeed help Duron become a good overclocking chip, getting more than 200MHz faster speed in most cases.

Duron's average speed is only about one speed grade (50MHz) lower than Thunderbird and depending on the applications Durons and Thunderbirds perform pretty much the same. Hearing about the overclockability of Duron, I concluded that I had little to lose in performance and a lot to gain from overclocking, saving money in the end. I looked for the local availability of Duron and 650MHz was the only choice. I bought it, only to discover its locked state. That didn't matter for too long, as I was able to unlock it anyway. Now I just needed to see how far the chip would go...

How did it go?

CPU:	AMD Duron 650MHz
M/B:	Asus A7V (KT-133 Chipset)
Cooling:	Alpha PAL6035MUC Fan/Heatsink
RAM:	128MB x 2 Hyundai PC-133 SDRAM
Video:	SUMA GeForce2 GTS 32MB
OS:	Windows 2000 SP1
Else:	Classified ^_^

Speed	Config	Voltage	Status
650MHz	100 x 6.5	1.50V	Perfect
700MHz	100 x 7.0	1.50V	Perfect
735MHz	113 x 6.5	1.50V	Good
750MHz	100 x 7.5	1.50V	Boot Stop
800MHz	100 x 8.0	1.50V	POST Stop
800MHz	100 x 8.0	1.70V	Perfect
850MHz	100 x 8.5	1.80V	Perfect
1000MHz	100 x 10.0	1.85V	No POST
950MHz	100 x 9.5	1.85V	POST Stop
900MHz	100 x 9.0	1.85V	Good
910MHz	107 x 8.5	1.85V	Good
892MHz	105 x 8.5	1.80V	Boot Stop
892MHz	105 x 8.5	1.85V	Perfect

Perfect = No Crashes Whatsoever
Good = Benchmark Crashes and Freezes Occasionally on Normal Use
Boot Stop = Computer Crashes During Boot or Immediately After Boot
POST Stop = Computer Crashes Before POST finishes
No POST = Cannot Boot At All

As you can notice up there, I had 1 GHz mark in my mind. Unfortunately, my chip wasn't able to break that. 900MHz was not as stable as I wanted either. But at 892MHz, with the maximum voltage possible, it was as stable as the initial 650MHz speed due to good cooling provided by Alpha fan/heatsink combo. After further analysis, speed of up to 935MHz would have been possible with 1.90V setting. But 1.85V of 892MHz setting is scary enough anyways, so I decided not to hack the voltage controls. Now for some benchmarks.

Synthetic Benches

Benchmark		650MHz	892MHz	Change
d.net client	RC5 Long, core#6 (MKeys/s)	2.230	3.083	+38.3%
Sandra 2000 7.6.49	Dhrystone (MIPS)	1855	2545	+37.2%
	Whetstone (MFLOPS)	876	1241	+41.7%
	MMX (it/s)	2150	2960	+37.7%
	3DNow! (it/s)	2751	3781	+37.4%
	Memory ALU (MB/s)	423	442	+4.5%
	Memory FPU (MB/s)	531	553	+4.1%

As expected, synthetic benchmarks reflect the exact increases in clock speed as well as the FSB. Mathematical increase of clock speed from 650MHz to 892.5MHz is 37.3% and most of the results mirror this within margin of error. As for the memory categories, speed increase of 5% in memory from increasing FSB was mirrored in the results as well.

If you're into RC5 code cracking with distributed.net client like me(or my team), you can see that you'll get a very efficient cracking machine (over 3 million keys cracked in a second!) without big investment when you buy a Duron system.

3D Benches

Benchmark		650MHz	892MHz	Change
3DMark 2000	Overall (1024x768 16bit)	3975	4665	+17.4%
Quake3 Arena v1.11 (fps)	640x480 HQ	98.5	116.2	+18.0%
	800x600 HQ	96.1	110.7	+15.2%
	1024x768 HQ	83.0	86.3	+4.0%
	1280x1024 HQ	50.0	50.2	+0.4%
	1600x1200 HQ	33.4	33.4	0.0%

Overclocking to 892MHz also brought up the scores and frame rates nicely. The memory bandwidth becomes a big problem for resolutions equal to or higher than 1024x768, so high resolutions don't benefit much from CPU overclocking, but it does help a lot in lower resolutions.

If you've run 3DMark 2000 on your own a lot, you'll notice that the score is quite low for a 900~MHz CPU with GeForce2 GTS. I think this is generally because 3DMark runs slower in Windows 2000 than in Windows 98, but examining details revealed that something else may be holding back the score. Nevertheless, it did effectively show how the overclocking helped increase the score.

Conclusion

What a difference a pencil makes.

By unlocking the CPU that was destined to run only at around 650MHz, I could get a noteworthy performance increase out of it at virtually no cost. A lot of heat was generated while operating at 892MHz, but Alpha PAL6035 was good enough to lower the temperature to acceptable level. You should also note Duron's high overclockability - rock-stable while running at more than 37% the original speed. While it fell a little short of the magical 50% overclock achieved by the legenday Celeron 300A or more recently, Celeron 566, I do not doubt that it would certainly compete well in the overclocked performance arena with its Intel counterpart, 'Celemine(FC-PGA Celeron)'. Now I need to get some sleep...

Posted by Wesley in Tinkerings at 02:00 | Comments (0) | Trackbacks (2)

Unlocking Aftermath (1/2)

Success reports

After posting the initial unlocking article involving pencil, many positive responses came although I had wondered if the results were reproduced in a favourable manner. It seems that this is indeed the case; a good example is this posting from HardOCP on August 3rd in their Fifth Edition posts.

This is what Hard|OCPer Patrick has to say about the Unlocking the Socket A CPUs with a Pencil.

Just wanted to let you know, that you can unlock the TBird w/ a normal #2 pencil by just going over the L1 bridges a few times. I got mine in yesterday, and it worked perfectly. Just thought I would let ya know it worked...

Then just when you thought you have heard the last of it. (which I seriously doubt) Another Hard|OCPer busts out with incredible close-ups of his mechanical pencil microsurgery. Don't you need a license for that??

all i used was a .7mm mechanical pencil. thanks pin0yclocker

New Questions Answered

Some new questions popped up. Most of them seem to be generally questioning the safety of the method further. Here are the Q&As that may help you.

Q: Doesn't the pencil markings gradually fall off and become a trouble?

[A]: This is not likely. Mr. Nam of South Korea noted in the guest book entry #134 that even with his heatsink-fan that has serious vibrating conditions, the markings are holding on well. Also, even in the rare chance of the markings falling off to the point that it becomes ineffective to connect the bridge while the computer is on, the normal CPU operation will not be affected (the system won't freeze) since the multiplier setting is detected only at the boot time and is not used while in normal use.

Q: I still don't like the possibility of this 'falling-off'. What are the ways to prevent this effectively?

[A]: You should try to put on a thin layer of cover on top of the markings. Nail polish seems to be a good choice, since applying it, as well as clearing it off is easy. Lacquer is another possibility, though it would be more permanent than nail polish. Neither of the suggestions have not been reported to have been actually tested, so try it at your own risk. Tapes, like scotch tapes may not be good since the chip gets hot and the adhesive may become messy or lose its stickiness.

Q: How about the markings getting mixed up with the heatsink compound?

[A]: The heatsink compounds are not supposed to be on the markings in the first place. Socket A processors are of FC-PGA type, meaning that the core sticks out of its ceramic casing. Thus the heatsink compound is supposed to be on the core area only and never anywhere else. But if the heatsink compound does ever touch them, nothing should happen unless the compound is conductive(which isn't a good choice).

What to Do with Unlocked Chip?

What's the whole point of unlocking a chip? Overclocking, of course! It's clear that a person reading this article is expecting to overclock his/her Socket A chip (Duron and Thunderbird Athlon all apply) or at least looking at the possibility of doing it. This is what happened in my quest of overclocking.

Posted by Wesley in Tinkerings at 01:00 | Comments (0) | Trackback (1)