Slashdot Mirror
Overclocked Memory Breaks Core i7 CPUs
arcticstoat writes "Overclockers looking to bolster their new Nehalem CPUs with overclocked memory may be disappointed. Intel is telling motherboard manufacturers not to encourage people to push the voltage of their DIMMs beyond 1.65V, as anything higher could damage the CPU. This will come as a blow to owners of enthusiast memory, such as Corsair's 2.133MHz DDR3 Dominator RAM, which needs 2V to run at its full speed with 9-9-9-24 timings."
This will come as a blow to owners of enthusiast memory, such as Corsair's 2.133MHz DDR3 Dominator RAM, which needs 2V to run at its full speed with 9-9-9-24 timings."
I'll just stick to the mathematics of quantum field theory. Kids these days and their crazy machines!
When we asked Pooh what the opposite of an Introduction was, he said "The what of a what?" which didn't help us as much as we had hoped...
Warning, pushing your components beyond their ratings may damage them!
Wow, never knew that overclocking might be problematic, guess I shouldn't have ignored all those warnings by the manufacturer, the system bios, the warranty pamphlets, the packaging....
I'm out of my mind right now, but feel free to leave a message.....
Since when has a manufacturer said "Yes, over-volt the shit out of our part, it will be fine."
Windows has more viruses because linux has more virus coders.
Agreed. I overclock, but I accept the risk, and do a little research first.
I understand the mindset, obligatory car analogy here, but it is not something I've ever done. Shopping for hardware has become a bit of a mine field lately, as most of the top tier motherboard and RAM manufacturers offer a *ton* of options for boosting the juice to various things all over the motherboard. They advertise this as a feature. I'm glad for those folks who like to go faster. It does make things a bit tricky having to check the RAM voltage, against what the motherboard can handle, and the processor will take, and hoping it all works. I'd like it if Asus and Gigabyte could maybe come up with a 'Get off my lawn!' series for us folks who like stock voltages, and wear onions on our belts.
Have you ever tried booting some of this memory with the default timings on a motherboard to find it will not boot with overvolting? I bought 8gb of OCZ memory this summer and could not get my system to boot till I took out some other memory from a Dell my company gave me and overvolted that memory in BIOS to 1.7 and than swapped in the 8gb OCZ. I should not have to do that, doesn't the memory specify what voltage it needs to run at; and if not, why not?
An Education is the Font of All Liberty
Considering that so many memory modules require running out of spec voltages to operate properly, while the Intel CPU requires voltages within spec, it would appear to me that the memory makers are turning out bad memory.
Maybe instead of requiring users ramp voltages up to CPU damaging levels, they should fix their chips? Now that Intel has brought the memory controller into the CPU, that they have tighter tolerances for the voltages does not surprise me.
I suppose this is a downside(although not a terribly upsetting one) of Intel's move to an on-die memory controller. Typically, CPUs are moved onto smaller and lower voltage processes more aggressively than are the northbridge and southbridge. It looks as though, in this case, that means that the CPU will impose substantially lower voltage limits on RAM than the northbridge used to.
Given the boost that on-die memory controllers gave to AMD, back when they adopted them, I suspect that the tradeoff will still be worth it. On the other hand, I strongly suspect that there are going to be some very unhappy cries of "WTF! How could RAM voltage kill my CPU?" from adventuresome kiddies unfamiliar with the implications of this change. Warning stickers aren't going to deter them.
The're the same kind of people who outfit their base-line stock 4-banger Honda Civics with nitrous and twin turbos and then wonder why their engine blows and their clutches slip when the checkered flag drops.
"This will come as a blow to owners of enthusiast memory, such as Corsair's 2.133MHz DDR3 Dominator RAM, which needs 2V to run at its full speed with 9-9-9-24 timings." I think some one forgot to proof read. Either that or manufacturers are REALLY pushing the data width technology as opposed to clock speed...
You are missing a point here. there are ram chips out there that are designed to run with more voltage then 1.65. So you do not even need to overclock for this to happen.
for example
OCZ Platinum 2GB (2 x 1GB) 240-Pin DDR3 SDRAM DDR3 1333 is a 1.8v standard. that's NOT overclocking
I agree overclocking and you break something your own problem but this product can't even use some decent ram as its stated to be used without blowing the CPU. At that point I would want my CPU replaced thank you,
So you don't buy that memory to use with your new chip--that memory is out of spec.
This guy's the limit!
If you push the limits of a device, you deserve what you get. Maybe good and cool, maybe broken shit.
Considering an entire subset of the industry exists dealing exclusively with parts designed to run 'faster-than-spec' I'm more inclined to lay the blame on Intel. They should know full well by now that the enthusiast market drives a lot of personal buying decisions further down the food chain...
Remember when Tom's Hardware broke this story?
If you can't release components that will run with existing kit, well someone is going to get the short end of that stick... And when it's the high end consumers, well Oops!
On the Oregon Cost born and raised, On the beach is where I spent most of my days
Push what limits?
You're not pushing a cpu, it was designed to run faster! Just bined lower.
You're not overclocking overclocking ram at 2v. Its designed to run with that voltage!
This isn't an overclocking issue, its a design flaw by Intel. Not our fault you can't see the forest for the trees.
Run a CRC on your brain, sparky, you dropped a bit or two.
The Nehalem CPU is designed to run at JDEC Spec of 1.5V, but can handle 1.65 without being binned. Yes, the RAM is designed for 2V, but the CPU wasn't - use the RAM, take a chance on killing the CPU and voiding your warranty.
60nm parts have 25% more area in which to absorb electrons and 25% more dielectric between elements than a 45nm part, so of course they could handle more voltage without damage. It's a design flaw in material physics, not the processor.
"Depression is merely anger without enthusiasm." - Anonymous
I don't really know why there's a sudden flood of people trolling overclockers. Overclocking is a legitimate way to gain more frame rates. For my job I often adjust clock settings on CPUs for benchmark tests, and it's not as if the CPUs that are overclocked weren't designed to do so.
The reason AMD processors used to be popular was that they appealed to enthusiasts and they had the ability for overclocking more so than Intel.
For a company that sits in the lead of the processor market, putting out a high-end CPU that does not allow for enthusiasts to have their way, is kind of lazy in my opinion. Maybe that's a bit much, but they could at least try to cater to the kind of customer that would purchase a high-end part.
and it's not as if the CPUs that are overclocked weren't designed to do so.
I can remember a time when connecting nodes on circuit boards with a graphite pencil was a good way to increase multipliers and voltages. And it wasn't that long ago, these new fangled extreme processors and enthusiast motherboards are a pretty new thing. So I guess this doesn't surprise me too greatly, I'm sure a lot of minds at Intel Corp. remember the good old days when they were the ones OC'ing chips straight off the assembly line to sell as premium stock and motherboard settings were locked down like Guantanamo. To them it's like, if it runs at it's rated speed with recommended board settings its good enough to sell. To us it's like, the last three processors I bought overclocked by .2-.3Ghz ATLEAST... and ran with my enthusiast memory...
On the Oregon Cost born and raised, On the beach is where I spent most of my days
... until somebody solders a crapload of diodes to their motherboard to drop each ram output line voltage by .7v.
Bonus points if they're LEDs.
Still running that Dell eh?
1.8 volts for DDR3 memory is severly out of spec.
The nominal voltage is 1.5. Chips nominally operating at higher voltages are of *LOWER QUALITY* than chips operating at the proper 1.5 voltage.
The ability to increase voltage to offset more aggressive timings than the memory supports is the real issue. At that point you are getting no real performance improvement and the real possibility of random bit flips + additional wear on the memory/northbridge/cpu.
DDR3 and CPU caches are all about bulk data transfers and have zero to do about latency. Whatever silly gains you think you are getting by playing with timings are hidden by the nature of the hardware.
You're not pushing a cpu, it was designed to run faster! Just bined lower.
This is a brand new CPU. I don't think they're worried about the low-end market just yet, and are labeling them as high as they can.
Dewey, what part of this looks like authorities should be involved?
Intel can't do split volts on the cpu and ram like amd boards and older Intel boards can do??
Will any other stuff like this show up in QPI 2+ systems with the QPI bus?
by adjusting the RAM voltage, you are also the voltages on the input pins of the processor. Overvolting an I/O pin can cause latchups, which basically is a short circuit.
If they can get a stock four-banger with nitrous feed and twin turbos bolted on to not slip the clutch until the checkered, they're doing pretty well. I'd expect that setup to fly apart closer to the green.
Looks like there are enough missed points to go around. The JEDEC DDR3 specification (see JEDEC Standard No. 79-3B) explicitly defines VDD as 1.5 V +/- 0.075 V for DDR3-compliant memory modules. Furthermore, the max supported frequency is 1600 MHz. What OCZ and other like-minded manufacturers are doing is intentionally violating the DDR3 spec to enable overclockers. Higher frequencies can only be reached with higher voltages, so they screen the DRAM chips to find the ones that can be pushed the farthest. These are then sold to enthusiasts to enable them to "push the envelope" on their gaming monster. Specifications exist to enable interoperability between different manufacturers. Intel is supporting the spec. OCZ is not. It's hard to blame Intel for not supporting OCZ's non-compliant parts.
They deserve to live with their results, be those increased performance or broken components. Saying they deserve busted components is like saying someone who soups up their car deserves a blown motor. Both endeavors, done correctly, can boost the performance of the tool in question. It's not hurting anyone, so why the sour grapes? Never were quite able to get the CPU overclocked so you want everyone who tries to fail?
Buy DIMMs that work at lower voltages because they use smaller processes or buy motherboards that separate the refresh power circuits from the data circuits on your RAM. It's doubtful that the data lines need 1.8 or 2.1 volts or whatever.
No, you're right. In rare cases an overclocked Celeron performed better than the standard-clocked Pentium 3 of the same nominal speed on most benchmarks. It's been a long time since the Pentium 3 and that generation of Celerons, though, and it usually wasn't worth doing even then.
Back in the day of DDR1 you'd be right, but these days the timings on the RAM are much larger but this isn't necessairly a bad thing. DDR3 runs much faster then it's older brothers and so the actual latency times are quite comparable.
The bigger numbers in timings mean a whole lot less when the clock is ticking that much faster :)
The Refined Geek - Technology, Finance, Space and everything in between
You are missing a point here. there are ram chips out there that are designed to run with more voltage then 1.65. So you do not even need to overclock for this to happen.
You're missing the point here. If there are RAM chips out there that are designed to run with more voltage than 1.65v then those RAM chips are not designed to the JEDEC standard. Legally, they probably shouldn't even be able to sell them as DDR3 since DDR3 is a JEDEC standard and the parts on non-compliant. Of course, most of the memory manufacturers do this anyway, and since they are part of JEDEC nobody complains too loudly...except when things don't work, of course.
OCZ Platinum 2GB (2 x 1GB) 240-Pin DDR3 SDRAM DDR3 1333 is a 1.8v standard. that's NOT overclocking
But it is over-volted.
Says Intel, dipshit.
This is old news, by the way.
That means DDR3-1600 is the max speed as a standard.
Anything faster than DDR3-1600 is already an overclocked memory by the memory manafacture.
However, Nehalem supports up to DDR3-1333 only.
As a hardware enthusiast (but not an overclocker), I would rather be using a DDR3-1600 memory.
Understandably, the overclocking community would want to use DDR3-2000 or faster (if any).
Personally, I would not be buying Nehalem until a newer one comes out
with at least DDR3-1600 or faster support.
You're not pushing a cpu, it was designed to run faster! Just bined lower.
This isn't an overclocking issue, its a design flaw by Intel.
Wow. Did you not notice the contradiction of those two statements?
Skilled in differentiating ravens from a writing desks.
I've never noticed it in gaming, but matching the latency to the CPU timing can noticeably affect video encoding. Changing my ram from stock 3-3-3-5 to it's full supported 2-2-2-3 decreased encoding time significantly with no other changes to the machine.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
No, they're lower binned parts that couldn't make "enthusiast" OR normal speeds at 1.5 volts.
They crank up the voltage and re-rate them at a higher speed, and slap on ridiculous cooling (heat spreaders, heat pipes, built in fans).
They KNOW these pieces will fail at high rates, so they jack up the price and call it gamer/enthusiast RAM. When the return rates start to drop off, they start issuing rebates to move remaining stock.
The data and address lines are connected. No amount of design can change that.
Methinks someone doesn't know what DDR stands for...
So much for intelligent design.
You are welcome on my lawn.
Then how is it proven that ANY processor can be run reliably at its binned speed? You can NEVER prove that it is reliable except by running limitless operations and checking the result of each one.
"There can be little doubt that union activities lead to continuous and progressive inflation." F. A. Hayek
If you're not an idiot you know what the hardware is spec'd to take, and what other overclockers have gotten it to without any problems (reduced life is a given, of course). I've never had to push voltages past what the manufacturer says that specific piece of hardware can handle to keep it stable.
Actually, this is still logically intractable. The best you can do is run it and wait for it to fail. If it doesn't fail, all you've shown is that...it didn't happen to fail. That isn't to say that it WOULDN'T have failed if you had run it for one more cycle, just that in your test space, you didn't get it to fail.
Short form: try to prove that something DOESN'T fail is trying to prove a negative, which doesn't work.
This is what you were getting at, obviously. I just wanted to clear it up for other people. :)
Probably because the IO voltage rating of the Intel technology for the transistor is lower than AMD. Intel CPU is on a 45nm process and AMD a 65nm process, usually bigger process are more tolerant. If Intel IO run at 1.5V we can suppose there are 2 reason for the limit of the ram.
First if the IO go beyond the 1.5V you can either break the protection diode on the cpu pin or inject current on the power line for the IO on the chip. That part is bad because it force the power supply to compensate for that and try to keep the same voltage on the power pin. While you have higher voltage incoming from the digital pins creating a differential on the internal power supply line, who start to carry more current than designed. That cause the line to heat and dissipate a lot of power eventually breaking them.
Second option is that having a higher voltage the transistor aren't made to support, is going to cause more electron leaking trough the gate and eventually breaking the isolation layer. If the isolator become to cracked by the electron a pinhole could form creating a contact between the gate and the substrate. Transistor gate are in reality small capacitor so contact between the 2 side it become a wire. That would cause the transistor to stop working. It also would inject changing voltage on the power line inside of the chip. Worse than the previous problem because now if a pinhole is created you can inject either a positive voltage or ground level on both power line and at different rate everywhere. That would effectively assure the destruction of the IO bank.
Because Intel and other chip fabricators can run lower level tests on the actual electronics of the chip than a nerd on the internet can.
They can physically inspect the chips from a given batch.
The most 99% of overclockers do is run a program to calculate Pi to a hojillion places over night.
Intel and other chip fabricators have set tolerances for the electronics. If a part falls within the tolerances, it is deemed good, if it doesn't, it is deemed bad.
For Intel and other fabricators, if a chip passes physical inspection, and a batch of them meets or beats the MTBF, they are considered good. If they pass physical inspection, but are statistically deviant from the MTBF (in a bad way), the batch is bad.
In a processor, logical failure is often the end result of physical failure, but physical failure usually does NOT end in logical failure.
You CAN prove that any given processor is logically reliable if run all possible valid input sequences on it. This is beyond astronomical (but not infinite, since we're talking about a logical level, and there are a finite number of logical states to any processor, along with a finite number of valid inputs).
You cannot prove that a processor is physically reliable, since the processor physically changes as you use it. This is why we have tolerances. Unfortunately, we want more performance, which means smaller fabrication processes, which means tighter tolerances, which means lower yields.
That extra 20fps won't make your penis any larger.
Sorry..
Gone!
"Among other things"?
Other things like the fucking initial design and engineering process? The processes, parts, and materials we use all have known physical limitations.
We know what the theoretical top speeds are when we design processors. We know that variations in the manufacturing process often alters the capabilities of a design in the real world.
It's not like baking a fucking cake and then being surprised at how delicious it is. We design, manufacture, and test to make sure we get our expected deliciousness. We don't get surprised and say "hey John did you add crack to this cake? It's more delicious than should be possible!".
Speeds can go up as the manufacturing process improves, or as you sell your chips to others who then strap on extra cooling and better power control. Speeds do not go up past the theoretical maximum (unless you've done something really, really wrong). You may get a good group of chips from the center of the wafer (the "golden sample", they call it) that beats your expectations of tolerances in real-world applications, though.
The stress testing chip manufacturers do is much more level than what a nerd with a desktop and the power of the internet can do. Manufacturers can physically inspect a processor, as well as run low-level logical tests on it. Nerds on the internet run 3dMark, SuperPi, Folding @ Home, etc.
60nm parts have 25% more area in which to absorb electrons and 25% more dielectric between elements than a 45nm part, so of course they could handle more voltage without damage. It's a design flaw in material physics, not the processor.
And that looks like a fault in your calculation. 45^2 = 2025, 60^2 = 3600. 3600/2025 = 1.78. So 60 nm parts have 78% more area.
Football Odds
I'll be the one laughing at the pathetic excuses you make up when I still kick your ass.
The Kruger Dunning explains most post on
A properly written software test is what, exactly?
Something that would cause the chip to fail physical inspection may not show up on any software test, especially if it only caused the part to be rebinned to a slower speed.
A CPU can be operating incorrectly in countless ways. Whether it shows up on one specific logical test under certain physical conditions, or whether it continues to show up or not after a certain amount of time is another issue entirely.
Go to school, or go back, or major in something other than retardism.
Although they are related measurements, process names refer to the ram cell pitch, not the size of the transistors.
If it's porn rendered in 0.5 fps on the Dell maybe it will.
This isn't an overclocking issue, its a designed flaw by Intel.
Fixed that for ya
bite my glorious golden ass.
Nope, sounds like they just made a major braindead basic design mistake on their memory controller (in this case it looks like assuming a relationship between memory and CPU voltage that should not have been assumed). Although I do wonder if it's the kind of thing that could be fixed with some OpAmps or Zeiners on the motherboard or something. I'll have to ask my dad, I'm weak on hardware design.
...
The stress of overclocking / overvolting reduces the lifespan of the processor.
The damages and errors may show up sporadically, unter certain environmental / power conditions, or after a period of time.
Even if you had software that could check every fucking part of your processor, and it returned a clean bill of health today, that would mean nothing tomorrow.
Beyond that, no piece of software exists that completely or thoroughly tests a CPU. Stuff overclockers use is basically "throttle it to 100% usage overnight, and hope it doesn't crash/throw errors" type shit. It can only tell you when you're fucked, it can't tell when you're safe.
People assume that if they can run for 24 hours fully stressed and not get any errors, then they're "stable", but the fact of the matter is, their processor is likely being slowly damaged.
I mean, who the fuck should we trust? Engineers and manufacturers? Nah, fuck that, this forum has some guy who says he got his CPU to 8 GHz on liquid nitrogen and it's totally stable!
A high-speed clock and data recovery system like that used to implement the memory controller and RAM won't be fixed with additional mobo components. Put anything in that path and it will very likely break. 2.0V is likely well above the Vmax of the FETs used in Intel's controller. They needed to take care of the voltage conversion at the pads to avoid issues like this. Instead I'm guessing they run the whole controller at the same voltage as the pads. That might allow them to run the controller logic and FFE/DFE faster but it's bad for power and then causes problems like this one. This looks like a bad design on the part of Intel.
Intel transfer the difficult from Hadware to software, for get more power, programmer need more technology. -- chinaitn
bzzt. nope. Process names are half the distance between two adjacent DRAM cells. I know you're thinking CPU's don't even have any DRAM cells, but it is what it is. See: MOSFET and Front-end Process Integration by Zeitoff, Hutchby and Huff.
A few months after the initial release of desktop i7 chips, they'll release a chip that can handle up to 2.0V DDR3 running at up to 2.4 GHz. The CPU will cost $1500, have an unlocked multiplier, and require a $300 motherboard, a $200 power supply, and a $100 cooling device to function with the out-of-spec enthusiast RAM. Gamers with more money than sense will eagerly shell out for it, and blame Nvidia's drivers when they only get an extra 1.3 FPS over JEDEC-compliant mainstream CPU/RAM configurations.
There's no failure quite as dissatisfying as a complete and total solution to the wrong problem.