Slashdot Mirror
More on Athlon Overclocking
The Tech Report is running an article concerning Athlon overclocking. With the 1Ghz+ chips coming down the line apparently soon, will things need to change? Or will it be just a documentation change? The other change is the issue of moving to on-die L2 cache - how will that affect things?
AMD, naturally, will never tell you of the risks you run in using their products, but you have a right to know.
The L2 cache design is really still experimental. The old L1 architecture is proven, but it's just about tapped out performance-wise, so AMD is moving on. Sadly, they're releasing designs that are not yet "production-ready" in a few crucial respects. Under certain circumstances (multiplying > 1k x > 1k matrices of doubles, for example) the L2 design will "underclock", or in engineering idiom "sprag". "Spragging" means, in layman's terms, that a lot goes in and nothing comes out. This happens when fast swapping cuts in during cache defragmentation with a mod zero byte block size in the under-storage. The result can be as small as a die melt ("burnhole"), which will require you to replace the unit (yes, that's "small", relatively speaking) to an extremely fast and sudden heat buildup: The L2 cache warms up only slightly, but it happens to be very near the clock-monitor throwback module. Yes, you heard me. You don't want your clock-monitor throwback to be soaking up any joules from a sprag, believe me. If the "glow-up" is fast, it'll throw a burnhole in the substrate and that's all she wrote. An expensive hassle, but not a major tragedy. Unfortunately, slow glow-ups do happen from time to time (larger byte blocks in the under-storage can cause this) and in that case the substrate won't burn-through before the clock-monitor throwback gets hosed. When that happens, look out, you've got the thing spinning out of control. Have you seen those cryonic supercooling rigs for seriously overclocked chips? Do you have any idea how much heat they're soaking up with all that liquid nitrogen and so forth? A hell of a lot of heat. Your computer under your desk doesn't have any liquid nitrogen. Fry your clock-monitor throwback and you'll wish it did. Here at AMD (that's why I'm an AC, duh) we do keep liquid nitrogen in the lab when we work on this bug. That's how serious it is. A couple of engineers have been very seriously burned, one so badly that he's on permanent disability.
Be safe, be careful, is all I'm saying. It's a rare, freak glitch, but it can happen, it has happened repeatedly in testing. And it can happen to you.
It's been covered up, obviously, with the connivance of the liberal media, but at least three moderate-to-severe head-explodings have been recorded in ordinary household usage of Athlons. Don't let this thing near your kids. It'll pop 'em like popcorn.
As I've explained, the story was never meant for the front page; I left it as an internal joke for the other staff going through the 'stories' list tonight. Unfortunately, I accidentally left it as 'Always Display,' and the rat-bastard thing displayed for like 10 minutes before I realized I had done something waaay wrong. :)
So, before you attack my editorial integrity, get the facts first. Thanks!
--Emmett
Also, check out #slashdot on irc.openprojects.net
Additionally increasing the clock gives the charge carriers less time to move, if it cant make a difference to the far side of the active region of the semiconductor, it isnt working. Increasing the voltage increases the electron drift velocity. Increase it too much and theres a chance of electrolysing your chip ;)
You also dont want to reduce the temperature too far or kT will drop to below the bandgap, turning the semiconductor into an insulator.
The best way of increasing speed is to reduce the size of the active layer, but soon we'll start getting quantum effects, which in current designs, are unwanted.
-Yarn - Rio Karma: Excellent
Boy lemme tell ya, if anyone is gonna spend the $1200 (or whatever) for a 1000 Mhz chip, and doesn't overclock it, they are downright fools. I mean, how can you look at yourself in your monitor's reflection every day with a measly 1000 Mhz machine? Sure you compile kernels in 5 seconds, but look at the guy down the road. He's running his at 1700 Mhz right now, pulling in about 738 fps in . Yeah, that's right, you non-overclockin weirdo, you.
Ladies and gentlemen. Do not let this scenario happen to you. The next time you remortgage your home to buy a CPU, remember, overclock that puppy to heck. Thank you.
This has been a paid advertisement from the Go Ahead and Fry that CPU Foundation.
Anyone know of a motherboard anywhere that supports two athlons, that was sposed to be a big technology benifit when they were introduces over pentium 3.
rated speed to avoid single-bit errors and the like
SBEs, eh? Well, that implies that you invested in ECC memory. Investing in higher quality components is highly unusual for an overclocker. I can assure you that most of them observe SBEs as a BSOD in their GameOS, because they spent all their cash on a higher-clocked processor and then bought Joe's Factory Thirds "PC92-1/2" Memory from the guy on the corner wearing a trench coat.
Fortunately for the engineers, there are FCC requirements for emissions and inteference, so they can tell the markedroids to shove it.
--
"L'IT c'est moi!"
Microwaves are optimized to produce microwave radiation and use a wave guide to concentrate the energy in an enclosed space. Faraday cages are very good at blocking high frequency radiation so not much leaves the box.
In CPU design, radiated power is minimized because it is required to get an FCC license of the device and because radiated power can cause interference within the box itself which is obviously a bad thing.
--
"L'IT c'est moi!"
Hell with that! *REAL* overclocking was desoldering one's 6809 CPU and replacing it with the 6309(?)!
(actually, i may be blowing smoke here... i never did do this, but i *think* i recall seeing plans to...)
--
--
Don't like it? Respond with words, not karma.
One difference between factory marked and overclocked chips is that the factory marked chips have been tested at their rated frequency by the manufacturer. This is an important difference for non-recreational users. A dealer or end user can not properly test an overclocked CPU. The manufacturer uses a very expensive chip test system with proprietary test vectors.
Mea navis aericumbens anguillis abundat
Which is better?
Compile, test, debug
Compile, test, debug
OR
Compile, check email, test, debug
Compile, have lunch, test, debug
?
I'm not so concerned about 1GHz CPUs anymore. I want to see supporting subsystems match the processor's performance, specifically video and memory. It would be very interesting to have a 500MHz memory subsystem and a 1GHz 256bit 3D video subsystem. By way of example, I have recently been able to extend the gaming life of a number of older P/PII systems by giving them Voodoo3 3000 PCI cards. How much more performance could we all get from a bleeding-edge CPU if everything else is running just as fast?
I well remember cranking my Intel Abacusium IV to over 100 BOPS (bead operations per minute), my favourite FPA (first-person arithmetic) games ran lickety-split...
Chris Morgan
According to this Electronics Weekly article, on non-graphics intensive applications like the LINPACK benchmark the Athlon 800 Mhz CPU performs twice as fast as a Pentium III 800Mhz. It goes on to mention that the Quake 3D test which is so widely used is in fact faulty. Underclocking both CPUs to 400Mhz (i.e. halving the speed) and using an nVidia GeForce caused the frame rate to drop by only 2 % at the highest resolution! There is some serious bottlenecks on the video card. Who needs a 1Ghz CPU then?!
BTW.. Dihydrogen Monoxide is a joke, but please don't ruin it, because it's funny as hell reading it and then when you finally realize what DHMO is.
-----------
"You can't shake the Devil's hand and say you're only kidding."
Full-speed cache is the only thing letting aluminummines approach or outperform Athlons, just look at the graphs and you'll see that the P3s get closer and closer as the speed increases, and the Athlon's cache gets slower and slower by comparison. With full-speed cache, AMD will be in a very good position as far as benchmarks go. If you think the Athlon is a good performer now, just wait. :)
Am I the only one thinking we'll be seeing the BiAthlon and TriAthlon just like the P2 and P3 came along?
-----------------------
Nicotine free Amish .sig.
Instead, you'll need a device that sits between the chip and mobo
Has it struck anyone else that this needn't be a separate device, but could be a part of the mobo. Overclocker friendly mobo makers (hello ABIT, are you listening?) could include the circuitry in the board, and configure it from the bios... now that's easy overclocking!
dufke
-
__
Comment submitted. There will be a delay before you understand what you posted.
we have to assume that they are actually designed better.
As far as I know, these are the same K75 core, 0.18 micron die chips.
The difference is probably just in yield. When Kryotech released 1Ghz, there were to few chips capable of running at 1Ghz air-cooled. Now, the proccess has run longer, it is (probably) tuned somewhat, and AMD has had time to save the best chips for a while. Remember, neither Kryotech nor AMD need to sell huge quantities of the top chips... the prices are huge!
dufke
-
__
Comment submitted. There will be a delay before you understand what you posted.
Uh, dont you think that companies would be smart enough to have shielding around the processor itself, rather than relying on the case? I think people think retards work in the R&D section of companies, they ARE aware of severe problems like THIS.
SpamMan
Ok, so like let me overclock my 1GHz to what? 1.1?that's not really neat... the percentage gained in performance by overclocking goes down as the speed increases, because the relationships between die size, speed, and heat are exponential not linear... The faster things get, the harder it will be to overclock, at least without cooling... I see a nice future for kryotech...
la la lalalala
Yeah. There are CPU diagnostic programs for PCs, such as AMIDiag, but few users have them.
Well, being a former overclocker myself, the hey-day(-5 points for spelling) seems to be over to me. No longer is overclocking simply a dip switch or jumper change. With these new cartridge-afied CPUs, you hafta break into the things, and maybe even de-solder and then re-solder those miniscule eyeball straining resistors. With chips running as fast as they are nowdays, I opine that a modern-day overclocker is significantly more bleeding-edge than those of yesteryear, like me. I do admit that I miss it...
mas cerveza, por favor politically incorrect stu
Water can boil at any RF frequency. When you radiate 1kw of power into a steel box its going to cook something. This has nothing to do with your computer. The cpu clock is running at 1Ghz, but at very minimal power, probably milliwatts. What takes so much current is the 40 million transistors on the cpu. If your athlon suddenly started radiating hundreds of watts of RF I would be worried.
Only the State obtains its revenue by coercion. - Murray Rothbard
The article brings up the fact that, with Socket A (as opposed to Slot A) chips coming sometime in the not-too-distant future, the whole goldfinger card issue will be more or less moot. Instead, you'll need a device that sits between the chip and mobo, or a "slotket" type device that plugs into an older, Slot A, board. There are, however, pins on the chip die designated for these sorts of things, so it shouldn't be too hard.
I agree that the on-die cache is going to be the most important advance, performance-wise. The L2 cache was running at 1/2 clock, then got bumped down to 2/5 clock at around 700 MHz (due to problems with high-speed L2 cache chip suppliers, I believe). When this goes on-die, the cache will finally be able to run at the full clock frequency, which will make the difference between 800 MHz and 1 GHz look paltry by comparison.
On the whole, though, we may be getting to the sorts of CPU speeds where overclocking no longer serves any useful purpose. A 1 GHz chip (or even a 800 MHz one) can, for most tasks, outpace any other componant of the system. So is overclocking my chip to 1.3 GHz going to make a major difference in my Q3 frame rates? Probably not. The graphics card (yes, even a modern 3D card) became the major bottleneck about 400 MHz ago. Sure, that might let me compile software faster, but in this case, I'm going to go with the rated speed to avoid single-bit errors and the like. Heck, with a chip that fast, I might even underclock a bit, just for the added stability.
Quantum mechanics: the dreams that stuff is made of.
Most microwaves ovens operate above 2Ghz, but arent there some modes of water down around 1Ghz?
Also, metal cases for microwave shielding will cost money. What happens when people start skimping on the case, or putting the new 10000Ghz motherboards in their old plastic boxes.
Well, power and frequency are different things, I suppose. But one can still wish for a few exploding heads anyway....
I love competition!