Slashdot Mirror
Asus A7V Overclocking Confirmed
NoWhere Man writes: "It seems that a few Asus A7V Socket A motherboards have reached the market despite AMD destroying the hopes and dreams of overclockers everywhere. A&I Computer has a few boards and has been able to reach 857Mhz. Buyoverclocked.com also got a few, and overclocked to 900Mhz, a picture of the switches is here.
For those interested in overclocking the Thunderbird, Tweak Town seems to have found a way to remove the cpu mulitiplier lock"
Say the average life of your processor is 10 years.
Say then that overclocking cuts your processor's life in half.
You still have a chip that will last 5 years. Who has a 5 year old computer? I mean they exist, but we're talking a 486/33 as a top-of-the-line kick-ass computer from 5 years ago. Besides, most people that overclock buy cheap chips so that if they do fry them, they can afford to replace them.
Example: The 333 Celeron, the most overclockable chip in the world - I have a friend who is running a dual celery board with 333's at 500 - a gigahertz clock rating at 1/3 the price! These chips are only about $40 bucks. And 500 mhz isn't bad, not even today.
insert clever line here
sig?
From Tweaktown:
"AMD have announced to us that they will be introducing a special Fab of Athlon with performance enhancing cache that will be multiplier unlocked. Award and AMI have also told us that they will be helping with a new BIOS code that will prevent remarking of the unlocked Athlon with performance enhancing cache. Also we have heard that Award and Ami will be introducing a scheme to prevent resellers relabelling the CPU. According to the information received, Award will introduce a feature that will read the AMD Athlon CPUID code and will display the actual CPU's designated Speed and the overclocked speed. Eg. If you have a Athlon 2 700Mhz and have it clocked at 900Mhz on boot up the motherboard will bios will show up as: AMD Athlon 700 at 900MHz. This will prevent resellers from relabelling the 700Mhz CPU as a 900MHz for example."
I didn't think it was possible to love AMD any more than I already do. Then AMD comes out with a way to solve their problems (reseller fraud) AND keep the fringe users happy. All it took was a little extra effort by AMD, and it WILL pay off for them when it comes to PR. Kudos AMD.
-GS
------
Let me give you the lowdown
Electromigration is the only thing that can cause physical damage to the chips besides frying it due to overvoltage. But they both end up doing about the same thing to your chips.
You don't know what you're talking about. I have a running 6 and 7 year experiment with electromigration, and guess what? This hasn't been a problem, period. I have a 486/66 and a Pentium 100, bother overclocked, that have been running overclocked the respective times, 24/7, barring power outages. The motherboards are about the only thing in them that hasn't failed. The fans went, the drive (in the P100) failed, the CDROMs long since died, the floppies died, but the CPU's keep on mackin'. The power consumption is to the point where it almost equaled the cost of the machine, but, it's an experiment to see if the CPU (will ever) fail. That and they make nice servers.
Electromigration is something that while well documented in theory, I have yet to see cause a problem. One of my electrical engineering professors has a old transistor collection, and you can see a decay in Beta (B) of these devices, so the gain characteristics have changed a fair bit. This is _much_ more of a problem in a analog transistor, where the preformance is supposed to be linear. In a CPU, the transistors are acting as switches - and as long as they switch, everything is fine.
Heat failure is completely different. Heat goes up, resistance goes up, circuit goes to hell, and the magic smoke comes out. This is what happens when the chips fail due to overvoltage - you physically burned out the circuit. THIS IS NOT ELECTROMIGRATION!
Anyway, that is an explanation of what happens when overclocking goes bad. There's a few white lies in there, but without spending weeks explaining electrical theory and physics to you, it's the best way to explain it. :)
You don't need weeks of physics and theory. If you overclock, you up the clock rate of a CMOS process. Power consumption goes up correspondingly. More clocks, more power, more heat. Too much heat, you burn it out. Better chips can handle more current, thusly running cooler, and therefore be able to run faster. Your watch is CMOS too; It can run for years because it's clock rate is 1Hz.
The problem is that you now have less energy available to generate those higher frequencies.. the clock signals are weaker. Of course, this results in instability and other problems (retransmits, blah blah blah). The solution is that if you want it to run at that higher frequency, you gotta increase the voltage to boost the signal... and take the chance of frying your chip.
This doesn't make any sense. See above comments. Right now, I mildly overclock my athlon. There's nothing that even comes close to making it run slow, so I'll overclock it heavily when games catch up. I suspect this is what most people do.
..don't panic
They overclock for the same reason people tweak cars. It's fun, it's enjoyable, and it's something to boast about. It's the geek equivalent of hot-rodding.
When people modify their cars to extract that extra 10hp, same reason people put on huge watercoolers to go that extra 50MHz.
Sure, it's mostly boasting - I mean, some of those high MHz computers don't get great uptimes, others do, etc.
I don't know what we call people who perform 'safe overclocking' - like bringing a Celeron 533A/566 to 800/850MHz (almost always guarenteed), though. Casual tweakers?
None of those problems should be happening. Better hard disks have MTBFs well above 1 million hours, you get error statuses when they fail, and RAID is an option. Fan failure should be detected; most servers now have internal temperature monitoring, as well as multiple fans. "Spontaneous memory failures caused by atmospheric raditation" are almost unheard of, and if you have ECC memory, they're corrected. Cabling problems remain a big issue; the CEO of Inktomi reports that mounting the power cords on their servers so they can't come out fixed their biggest cause of downtime.
Now OS uptimes need to improve to match.
CPU: Geniune AMD Athlon 900 MHz
CPU Clock: 1000 MHz
the idea that AMD and teh chipzilla multiplier-lock their cpus to "protect consumers" is a load of bs that they dump on us. the reason that they mulitplier-lock is so that if enthusiasts want 1000 MHz, they have to purchase the 1000 MHz part ok bye.
loev,
Axel
mhm23x3, alt.fan.karl-malden.nose
I build a lot of systems for people and make them deals on my old CPU's some I frequently go up. "I'll make you a deal on the 500Mhz Athlon so I can buy a Thunderbird" and so on. Overclocking doesn't pull me very much when I can keep my CPU speed up there pretty good. My wife had some surgery awhile back so I haven't upgraded since the Athlon 500Mhz but it is getting to be time again.
I used to work at a computer store testing returned equipment. You be surprised how much stuff is obviously O/Ced and messed up by ignorance.
He still blamed the Taiwan earthquake on Q3 99 stock loses
I know Michael Dell is a little egocentric, and a lot of people seem to think the world revolves around money and profits, but..
How the hell does Dell's stock price dictate the stability of tectonic plates?!?!?
.sig: Now legally binding!
> This won't work, due to how chips are
> manufactured.
Sure it'll work. Use laser trimming to cut traces on the die before it is packages like they do for precision analog components. Or embed a couple bits of eprom on the die and program it after packaging and testing. It's all really quite simple. 1970's technology.
Ryan
Ahem, you don't know how cmos works. Allow me to explain...
...ohm's law...
A cmos circuit is composed of complementary pairs of mos transistors (hence the name cmos). A mos transistor works more or less like a switch. A voltage on the gate pin allows a current to flow on between the other two pins (transistors have 3 pins). The gate of a mos transistor looks like a little capacitor. You need to charge up the capacitor to turn on the transistor. The output of each cmos gate has two mos transistors, one goes to the supply (about 2 volts), and the other goes to ground (zero volts). The gate turns on one (and only one) of the transistors to select a binary output level. If both transistors were on a short circuit would occur.
There are three sources of power consumption in a cmos circuit. (1) leakage currents (very small), (2) charging and discharging the gate capacitors, and (3) so called "class A current."
The first item, leakage, is very very small. Several orders of magnitude smaller than the other two. It is independent of operating frequency.
The second item on the list, charging/discharging current, is quite important. It takes a fixed amount of energy to charge the gate capacitor and force the transistor to change state. At higher clock rates you change state more often. Since power is simply energy per unit time, the power consumed from these charging currents is linear with respect to operating frequency. Double the frequency, double the power consumption. BTW, you can force the capacitor to charge more quickly if you increase the supply voltage. Sometimes this is necessary for overclocking. Higher voltages will lead to higher power dissipation (though not necessarily increased current (for a given frequency)).
The last item, "class A current" is caused when both transistors in a cmos output are on at the same time. Since the transistorgates have small capacitors, they can't turn on and off instantaneously. Therefore, when the gate changes state, there will be a short period where both transistors are partially on. This overlapping conduction causes a sizeable current to flow; a short circuit of sorts. This is the so called "class A current." The exact magnitude of this current is dependent on the particulars of the transistor design and the clocking waveforms. It's hard to say what happens to class A current as clock rates go up except that power consumption increases.
BTW, in the process of dumbing down something that isn't all that complicated to begin with you said some pretty silly things:
Electromigration is the only thing that can cause physical damage to the chips besides frying it due to overvoltage.
Heat, static, current, etc. Electromigration pales in comparison to soldering irons, drills, and other tools of the trade.
as those gates expand, they start passing more current
MOS has a negative tempco (unlike BJTs). They conduct LESS current when they are hot. That's why you don't need to worry about thermal runaway when playing with power mosfets.
More current = more heat
No. P = IV, you know that.
Now the more current that is passing, the more likely it is to "jump" the gate
WTF?
without spending weeks explaining
Doesn't take weeks. Cmos is pretty simple.
Current is what causes heat, NOT voltage.
No. See above.
Your processor is not a resistor. It's not that simple.
a (slight) increase in heat production due to the higher frequencies
Large increase. Greater than linear.
When I wrote the above, I was responding to the original poster who was complaining about the situation where an overclock-damaged CPU would reduce the reliability of a server.
I'm assuming the "server" he was talking about was an off-the-shelf or homebuilt PC, because if you're buying serious hardware for a high availity machine you're not going to be picking out the CPU in a place where you have to worry about buying an overclock-damaged CPU marked as new.
Most "servers" in this class don't have RAID, fan sensors with built in audible alarm, hot-pluggable drives, redundant power supplies, quality motherboard, or ECC memory. You will indeed run into the problems I've mentioned, and a few that I left out- most notably power supply failures. Even big power supplies that are underused and fed clean AC from a high end UPS fail, sometimes frequently. I have a dell poweredge server that has lost 2 450Watt power supplies in 4 years. Not a stellar record.
The stock Linux/BSD distributions don't appear to support hardware monitoring either. I bet a daemon to do this exists somewhere but I haven't run across it in either the latest redhat, mandrake, and openBSD distros.
Also, about the stray radiation: Studies have shown that a computer that is on 24x7 will be adversely affected in such a way approximately once every 3 months. Most of the time it doesn't do any harm, but occasionally it does. I can't prove that it's been the cause of any failures I've seen, but I've seen a couple sealed boxes that are normally 100% reliable run for years and one day reboot due to a memory parity error.
I have to keep a bunch of machines up 24x7. Most are quality boxes, but a few are normal PCs. The normal PCs do have the problems I've described, and while most of them fail less than 3 times a year occasionally one does fail that often, so I still think that 3 strange failures a year is within the range of "normal" behavior for such beasts..
> With regard to dumming down try looking
> up 'thevenin load' which is the consolidated
> impedance at any point in a circuit model.
Try looking it up yourself. What's the thevenin equivalent of an oscillator?
Ryan
For me, it's all about getting the most out of my money-- as a college student, resources are limited, and if I can pull my 466 Celerons up to 700's (or something similarly insane) I've saved money that I really can't afford to spend.
I think the second reason has to do with "stickin' it to the man", as with Linux. It's some people's way of telling Intel (or AMD or Cyrix or...) we don't have to take it anymore.
Or if nothing else, it's something for nothing.
Kit
But is the performance gain really worth it? If you need those few extra clock cycles every second, why are you cheaping out and then risking not only your warranty (or blowing it altogether) but the life of your CPU? If I really wanted some extra speed in my brand-new machine, I'd spend the extra $200 or so, not blow time, energy, and then my entire investment by overclocking.
Please - teach me. I am but your sponge.
Mr. Ska
Back in the days of Pentium I's and II's many hardware vendors would overclock CPUs and sell them as the speed they overclocked them to, like a Pentium 150 becoming a Pentium 200 or 233. How was the consumer to know it was overclocked? I'm not even sure my Pentium 233 I got from Compaq a few years ago really is a 233, because there's no sticker or multiplier jumper pin location on the motherboard. How quickly we forget.
AMD put the clock multiplier lock on their CPUs to prevent this; if you bought a Thunderbird that was supposed to be 866MHz overclocked from 800, for example, then used it in a benchmark between it and a Pentium III 866, the Pentium would win and AMD would look bad, even thoguh a genuine 866 may have beat the Intel chip (MAYBE)
The same thing happens with modems now. A 56K modem is actually a "Mo", in that it modulates but receives data digitally over the phone line. THat's why the send speeds are less than the receive speeds. and most cheap computers come with a 56K/14.4 modem, i.e. 14.4 modem without analog converter for downstream. It's a ripoff and people don't understand that. If the CPU had the same problem, there would be chaos.
By the way, I'm not entirely sure how a 56K modem works in terms of downstream/upstream and every detail of how it's different, but that's the basic idea.
So maybe you could get a little more juice out of your processor, it should be your choice. But many hardware vendors are not trustworthy enough (since it yielded soooo much profit) so the clock multiplier locks should be obeyed, IMHO.
# debian/rules
I have a celeron 300A in a Tyan Tiger S1832 MB. I can overclock the celeron to 450, but my system is incredibly unstable. yes, it gives me a performance increase of 50% on benchmarking programs, but everything else I run works fine with just the 300.
I still think that more RAM speeds up a system more than a faster processor would. Aside from that, is it really necessary to have a 900 MHz chip? Or is this just some insane pissing contest? No, really, I can buy a 550 PIII for cheaper than a 450 PII (check pricewatch). So I can have dual 550's running for under $300. I simply can't fathom what would programs would require more processing power than that! Isn't the bottleneck for most games the graphics cards? Isn't the bottleneck for most desktop apps RAM? (when you have 8 windows open and you start spilling into virtual memory).
I just really don't see the point to overclocking, it seems to be an inordinate amount of work for a small gain.
Moller
I was initially excited about the Athlons, but honestly, who cares if they can be overclocked or not!!! Why hasn't AMD developed support for a dual or even quad board??? Until they do so, I'll keep buying Intel, and I'll pit my Dual p3 600 machine against any overclocked Athlon any day.
"These are not people who use Linux because it is better; these are people who use Linux because they like the elitism t
1. Reputable retail outlets will not resell returned, potentially defective merchandise as new. They will return the product to the original manufacturer. The bulk of the major retailers fall into this catagory (compusa, cdw, necx, pcwarehouse, etc) as do many smaller places. If you are buying from a place that is not reputable, you have a lot more problems to worry about than overclockers.
2. Overclocking is safe and easy these days. AMD and especially Intel underclock some chips for purely nontechnical reasons. You might as well see what your CPU can really do.
3. Your reasoning here:
is outright wrong. I'm not going to justify this statement here because it would go offtopic, but it should be obvious.4. As someone who runs unattended servers year round, 3 lockups a year is not unusual. You're going to have problems with hard disk failures, CPU and case fan failures, spontaneous memory failures caused by atmospheric raditation, as well as misc. cabling and environmental-related failures. I'm also assuming that you have a quality UPS. I would not consider a machine that locked up for mysterious reasons 3 times a year suspiciously unstable.
Please help keep overclocking an option for intelligent enthusiasts.
-OT
There is even a rumor going around about an Athlon 2 in the future, which will be an unlocked Tbird (nevermind the fact that they still seem to be unlocked so far), and that AMD will be working with AMI and Award to make BIOSes that will help prevent the remarking of AMD CPUs by displaying the true original CPU speed alongside the overclocked speed (e.g., "AMD Duron 600 @ 800MHz").
>well first of all K6's were never a highly recommended potential OC chip,..
??? I don't know about later ones, but at the beginning, AMD all
but told you how fast you could take them.
Essentially all of the 166's could go to 200, and most to 233.
AMD also kind of coughed and pointed at the bus speed with a chipset that
was designed for 83mhz . . .
hawk, who got the 166 one day for a mere $99 when a retailer probably screwed up the day's price
(2) Once it fails, these people may return their CPU to the store as "defective".
(3) The store, rightfully never trusting the word of consumers, will test the CPU.
(4) The store's test is short and in a nice cool optimal climate controlled environment. And subtle problems that show up when the CPU is warm and been running for a long time will not appear on the test.
(5) The CPU goes back into the glass case to be resold.
(6) You (after saving up a long time): "Hi, I want to buy that CPU right there."
(7) You are now on a trip through hell. Your CPU may fail later that day (if you're lucky), or next week, or "lock up" only 3 times per year. You cannot now run your unattended server on your DSL/Cablemodem line. Your system is forever unstable.
(8) By the time you realize the true cause of the problem, the warranty/exchange period is long gone.
(9) You are screwed. And it is the fault of overclockers.
I don't want to hear about how "most" overclockers are responsible, blah blah. One person getting screwed on a machine that takes years for him to finally save up for is one screwing too many.
Overclocking needs to be blocked by the CPU itself, where it can't be circumvented. Just as laws are set up to occasionally let many a guilty man walk free rather than wrongfully convict so much as one innocent person.
Overclocking must end for the good of the consumer.
Thanks! I did just that. Much appreciated.
With the ignorant masses, I don't know if there really is another way to prevent people from swindling maxed out processors and pretending they're running like out of the box.
A CPUID utility would be sufficient. Intel & AMD could strike a deal with microsoft to include their CPUID programs in the next rev of windoze. Before someone jumps down my throuat about how not everyone uses windows, it was YOU who limited the scope of this to "the ignorant masses". Those ARE the people who will be using only Windows.
When one of these people hits a stability problem, what is the first thing that they do? Re-Install Windows. A fresh install of Windows will insure that the CPUID util hasn't been tampered with.
If you're running linux,*BSD, BeOS or whatever else you should have the ability to compile the source yourself provided that they make it freely available.
LK
"Hi. This is my friend, Jack Shit, and you don't know him." - Lord Kano