|
Alereon posted:One thing to keep in mind is that we're talking about a voltage spike so brief you're not actually going to be able to measure it without extremely sensitive equipment, so take the numbers you're seeing in hardware monitoring programs with a grain of salt. There's really no good reason to have LLC enabled, as it reduces overclocking ability and exposes your CPU to more voltage than necessary. ... even on an Asus board? For me setting it to high was the key to 4.7GHz rather than 4.5GHz. Totally stable, very safe temperatures, Offset mode with all power savings in the Intel specification enabled because I am a fan of the planet when I'm not pretending to kill aliens; without LLC on, it's not stable unless I go with a much higher voltage offset that causes it to idle and run at unnecessarily high voltages. What I look at is how much wattage it's using and that seems fine, so what's wrong with LLC if you aren't using Gigabyte? The chips seem to be able to handle the juice just fine.
|
#
¿
Jul 24, 2011 03:22
|
|
|
|
| # ¿ Apr 23, 2024 22:06 |
|
|
Alereon posted:You'll always get better overclocking performance with LLC off and just raising the voltage than you will with LLC on. This will increase power consumption when not at max load, but the alternative is pumping high voltage transients into your chip, which is a Bad Thing(tm). Here's a good article from Anandtech about power delivery and Loadline Calibration, using a 45nm Q9000-series quad-core and an Asus motherboard for testing. The principles are the same on current CPUs, but they're even more sensitive to higher voltages due to the smaller manufacturing process. Thanks a lot, you jerk, now I have to redo my whole overclock oh who am I kidding this is gonna be awesome, I love overclocking and this is the fun part Edit: Hm, reading that article, it seems like over-currenting (I gerunded that word, sorry) is a big part of the problem with overvolting (which makes sense from an electrical perspective, I'm guessing there are bumps doing stuff that really can't comfortably handle current that high for very long?). But the Asus P/Z boards allow between 100% CPU current and 140% CPU current. Just a :ricer: crazy rear end thing, or what? Also, okay, what are the real-world consequences of using LLC on a moderate setting (goes auto which is the same thing as normal - you can't turn it off - then normal, then high, then very high, then extreme or ultra or some crap but it's higher than the others)? My eyes are glazing a bit trying to make sense of the maths in the linked article and I am not entirely sure whether I am going to kill my processor in a real-world time span. Which I define as replacement/obsolescence within 4 years, anyway, that's my usual cycle, though I like it when my computer can make it a few years into the next one for backup purposes, can't hurt to be able to resume work immediately - and I overclocked the Q9550 in my old PC without ever touching LLC. So... Help me out, it's A Bad Thing but overclocking in general is bolting big gently caress-off heat sinks and taking extraordinary measures to rip stability and performance out of exceeding the factory specifications, so isn't overclocking in general, if it exceeds stock voltage, A Bad Thing? Also, do you happen to know how in -the- gently caress Asus calculates the normal voltage? Offset is ridiculous, the stock voltage is clearly 1.24V for the 2600K (yes I do use the hyperthreading, kicks rear end for audio if your DAW is efficient and you use a lot of high-resource software inside it, ain't all about 2500Ks  ). Yet if I set everything else to stock, and offset to +0.025, that'll park it at about 1.35V. Turning up LLC a tick past normal to "High" puts it at 1.36-1.37V, with occasional excursions to 1.382V that don't bother me or the gigantic heat removal unit bolted onto my motherboard as far as I can tell. Now, as you've noted, I don't have a true picture of the voltage conditions because their polling frequency is deeply insufficient to capture the much smaller scale, quicker overvolting that occurs with LLC... But I'm not going nuts with it. Anyway, the question is, where the hell does the voltage come from? It's never explained in the manual but it changes with LLC, it changes with Phase Control, and the thing that affects it least of all seems to be the damned Offset value. But I am concerned about running my processor at 1.365V all day long including when it's parking at 1600mhz just so it has the juice at 4.7GHz stably...But if you can explain to me some more about how I'm being very, very dumb because *drops knowledge* I will be happy to listen and appreciate the insight, maybe you can solve the mystery of Asus overclocking for me and some of the other guys in the overclocking thread. It's a silly maze right now and guides will tell you to do completely absurd poo poo like FEED IT MORE CURRENT, etc. Also, PLL Overvolting: Bad thing? Was another key to getting my OC at 4.7GHz to not screw the pooch in Prime95 on one hyperthread after a few hours (and only one, which was frustrating - the rest would keep on truckin' but that one part of that one core really wasn't up for it). I can't sleep the PC (thanks Asus) but with an SSD and a lean installation and all the screens after the BIOS turned off the limiting factor is still "how fast after hitting power on can I get the monitor to warm up." It's very fast, it finally exhibits perfect stability in synthetic and real world processing, and it runs cool and quiet thanks to the Noctua NH-14D big bastard of a heat sink. But tell me how I'm loving up and I will  , you clearly know what you're talking about, formal electrical education or no. And thank you.
Agreed fucked around with this message at 18:17 on Jul 24, 2011 |
|
#
¿
Jul 24, 2011 17:53
|
|
|
El Bandit posted:I had this problem and disabling PLL overvolting fixed it. It may limit overclocking (I've no idea), but it's worth it for the convenience. It is a known issue with Asus and Sandy Bridge. Since my computer never sleeps and it was part of taking it from 4.5GHz to 4.7GHz I am reticent to remove it without good reason. I also have my VCCSA and VCCIO very, very slightly overvolted which allowed a stable change from 2T to 1T on my RAM. An extra 200MHz per HT-enabled core and real-world speed improvement with my RAM (unlike synthetic-only improvements if I go past 9-9-9-24) makes a lot of difference in the DAW where I am working in real time with very resource intensive plugins at extremely low latencies. If I need to change my approach from what has worked out to be an extremely fast and stable overclock, I will do it. But I would like to understand why I am abandoning a working setup that seems to have some practical differences in implementation from the bad Gigabyte LLC issue. And despite the smaller process I am having difficulty finding much evidence that LLC on Asus is the big bad wolf. PLL is just a lovely known issue but my boot time is 13 seconds or so, I am okay turning it off when it isn't in use rather than hibernating. Still, if he has info on Asus like he does on gigabyte then I would like to hear him out.
|
|
#
¿
Jul 25, 2011 05:40
|
|
|
Alereon posted:Here's the Anandtech article I linked earlier, it's their original investigation of LLC using an Asus motherboard. They haven't done another article for newer CPUs/motherboards, but they do occasionally remind readers that disabling LLC is their standing recommendation. Asus boards are much less likely than Gigabyte to result in hardware damage or instability, but the reality is that you'll get better overclocking performance with LLC disabled and manually managing voltages, and if you're at your max safe voltage and are enabling LLC to achieve stability, you're just tricking yourself into using a higher voltage. Well, the default setting still has LLC, you cannot disable it altogether, and there is a separate adjustment to control CPU current (allowing for between 100% and 140%). Further, in offset voltage control mode, it seems you don't have especially precise control over the voltage anyway; I and many other Asus and Asrock users have noted that it has brief excursions to wherever it feels like it ought to be when overclocking. The older Core 2 processors were overclocked via changes to the bus and voltage primarily. I am not sure how much of that translates to the weird offset method... But two things I do know: one, I am not uncomfortable with higher voltages than I am using, and two, I am not willing to set the voltage manually and have it running at a constant 1.382V even when parked at 1600MHz just because it can under extraordinary stress briefly reach that above it's usual full 47x multiplier 1.36V. I wish that we had more fact-based info on offset overclocking, because it almost seems like it's just going to do what it's going to do, but for real power saving it is vastly superior to running at a manually set voltage. Interesting to read that 1600 is the sweet spot for RAM at higher clocks, I had heard 1333 before now. Guess I made an alright choice going with 1600 9-9-9-24 and kicking it up to 1T then? Agreed fucked around with this message at 19:17 on Jul 25, 2011 |
|
#
¿
Jul 25, 2011 19:11
|
|
|
The problem is that I can use LLC set to high and get stability with voltage and thermal behavior I'm comfortable with, with only a +0.025 offset; so, my CPU when at its lowest 16x multiplier for power saving when using firefox or thunderbird for the day to day of my job it runs at a comfy 1.025V. If I turned LLC from High to Normal, I'd have to up the offset voltage to compensate and it'd be idling at 1.150V or something like that. So it's really more about managing my lowest voltage while keeping the highest voltage within my comfort zone than anything else. If it allowed more precision it'd be great. Maybe I should tell LLC to stop and just manually set the phase control steps and up the offset, but I really don't want to. I'm using an Asus Sabertooth rev3, it "only" has an 8-phase VRM but they're robust as hell and it is in general an exceptionally nice and stable board that's proved to be a great overclocking platform. The reason I don't want to turn off LLC is because it seems like the lesser of three evils here in terms of power consumption and stability. The alternatives to using LLC are either adjusting other stuff manually to try to deliver voltage and current to reach stability, basically, and result in a much higher net power draw over time since it only gets into the 1.382V maximum value once every six to ten IBT runs - even games that were CPU limited on my Q9550 don't get it above Intel's comfortable 95W TDP at the 4.7GHz overclock, and my audio software doesn't peg it like that either. I mean, unsurprising, I guess, that stuff made specifically to try to crash your damned machine pushes it harder than real-world tasks, but still, if I did a manual voltage control and it idled at 1.382V (since it does very occasionally need that, under extreme load), it'd use a lot more power overall just to get to the maximum possible voltage needed for stability. If I use the manual offset, same problem, voltage idles way higher than I want it. I've never hosed with LLC on previous chipsets, but... it seems useful with Sandy Bridge. Not because I'm "fooling myself" with the voltage, I'm fine with the voltage, but because it allows me to set the voltage without having it there all the damned time. Just on-demand.
|
|
#
¿
Jul 26, 2011 07:23
|
|
|
Zhentar posted:For starters, all modern processors (pretty much everything made since the Pentium 4 or Athlon 64) is TDP limited. In other words, the speed of the processor is limited not by the capabilities of the chip, but by an agreement not to dissipate more than a certain amount of energy. An overclocker with a big heatsink can essentially tell the processor it's okay to violate that agreement, and achieve a substantial speed increase without exceeding the electrical specifications of any of the components. Thanks very much, this is the information I was looking for. Well done pulling the right question out of the jumble, I didn't quite know how to ask what I wanted to know but you answered it nonetheless.
|
|
#
¿
Jul 26, 2011 22:02
|
|
|
movax posted:Ever since my 650i board (DDR2) I've always bumped up RAM voltages by .1 or .2V and I think it does help stability, especially when you have multiple sticks / all the slots populated. Don't expect any ill effects from this either, as I think JEDEC requires them to function up to 1.575V and resist damage up to 1.9V or so (don't have to function, they just need to survive getting that voltage). I'm at 1.36-1.38v to get my 2600K stable with HyperThreading enabled. Can kick that off and achieve lower temps and stability at lower voltages, but it's damned useful for what I do and I've got a high-airflow case and an efficient cooler so I'll let it do its thing, no problems. I wonder if I should up the RAM voltages. I've got 16GB, populated all four slots of my Asus Sabertooth Rev3 with 4GB DDR31600 9-9-9-24 2T modules. At 1.5V I'm able to lower that to 9-9-9-24 1T and pass memtest86+ no problem, just slightly bumped (like, one incremental step as permitted by the UEFI) the VCCSA and VCCIO voltages while pegging the actual memory voltage at 1.5V. Wonder if I could lower my vcore if I raised my RAM voltage a bit. They can survive, boot and run at 1.65V, which I know because updating the BIOS resets everything and the newer BIOS doesn't read them as 1.5V modules even though the March-era BIOS did. Noticed it pretty quick but not before running them into Windows and a successful stability test at 1.65V, d'oh. If this is representative of G-Skill quality, I'll pick 'em in the future, I just went with what would fit under the damned heat sink and could be had quickly from Amazon. Anyone have thoughts on PLL overvolting? Safe/unsafe, apart from the known bug with sleep that I don't particularly care about? Is there a nice, fat white paper I can read that goes in depth on the P67 platform so I can stop asking dummy-level questions in a smart thread?
|
|
#
¿
Jul 27, 2011 06:00
|
|
|
Not exactly... On an Asus Sandy Bridge motherboard, you have to use offset overclocking if you want speedstep and the C1E to function. Which they will, just fine. You ALSO have the option of trying your luck with their EPU power saving which disables Intel's power saving (and, hell, might actually work with a manually set voltage, who knows) but that has always been a poo poo sandwich for me at least going back to the P45Q-E so I leave it off in favor of Intel's fully compatible power saving.
|
|
#
¿
Jul 27, 2011 17:27
|
|
|
movax posted:I always leave all the power-saving options on in BIOS, though now that I think about it, I only ever noticed CPU-Z bouncing between 1.6GHz and 4.7GHz...so I guess some form of saving is still in effect. Yeah, that's just how Asus rolls when overclocking, basically. It used to allow base clock overclocking but at some point in updating the BIOS that stopped and now it's all turbo, so in order to ensure performance as per your requests, if you've got turbo-by-all-cores selected it automatically tells turbo to ramp that poo poo up as soon as your processor gets remotely involved. I haven't experimented with turbo-per-core.
|
|
#
¿
Jul 28, 2011 17:17
|
|
|
MeruFM posted:What's the disadvantage of running at high core voltage if you can keep the CPU cool? Nothing at that voltage, just when you start getting higher it can get a bit dangerous and lead to early failure.
|
|
#
¿
Jul 29, 2011 07:17
|
|
|
I've thought about that, too. My system is letting it hit 120-130W under stress test loads because my cooler can handle it and it's got no reason not to, but most applications will only benefit from one or two cores turboing up. However, I have done measurements and when running Starcraft 2 for example, which is the lovely combination of CPU limited and poorly multi-threaded, it doesn't get anywhere near the stock TDP even at 4.7GHz, so it's not like the Asus board is forcing all those cores to be active just because their multiplier jumps up at the same time. My concern is similar, potential BSODs and stability issues instead of my powerful and stable overclock in exchange for ___________ where I can't really fill in the blank.
|
|
#
¿
Aug 3, 2011 04:07
|
|
|
Intel said 1.5V then chips died. Oops. Now Intel has said 1.38V, but who knows, its TDP is 95W and if you're overclocking and giving it voltage that high it'll hit 120-130W under the right conditions. I haven't had any issues with mine running 1.36V-1.38V, offset method, and it almost never reaches 1.38V even when running stupid long iterations of stress tests (and has never exceeded it); I'll let you know with a massive  if my system suffers an early death related to my processor deciding enough's enough or something.I do feel obligated to note that while a dedicated liquid cooling setup will remove heat faster for extreme (dangeorus) overclocking, the "turbine" thing is a relic. Today's most powerful and effective air coolers are generally still very quiet, using 120mm and 140mm fans and impressive heat pipe arrays. My case is a Corsair 650D, its fans are 200mm front intake, 200mm top exhaust, 120mm rear exhaust. I've installed an optional third 120mm fan on my Noctua NH-14D, which is one of the current contenders for the top spot. I can't hear my processor cooling over my case fans, which I can only hear if my AC is off and the house is silent. High airflow and great cooling is not necessarily coupled with loud noise as used to be the case thanks to efficient heat wicking and push-pull fan setups. Agreed fucked around with this message at 17:24 on Aug 4, 2011 |
|
#
¿
Aug 4, 2011 17:14
|
|
|
Combat Pretzel posted:They were yapping something about "overly increased" latencies due to larger packet sizes and what not, interfering with general audio playback, introducing clicks and such. It turns out 1600 is the magic number, not 1333. But that's okay, the difference is pretty small. It's not like everyone who has been happily using 1333 all this time all of a sudden have second-rate computers, there's just some more real performance gain (small) at 1600 (and above that, gains are basically theoretical only) if I understand the anandtech article correctly.
|
|
#
¿
Aug 5, 2011 20:48
|
|
|
Combat Pretzel posted:Naw, I thought the Sandy Bridges all run at 1333 and the upcoming Ivy Bridges would bump it to 1600. I run four 4GB DIMMs, the CPU would probably not dig 1600, anyway. I'm running four 1600mhz 4GB DIMMs with 9-9-9-24 1T, the CPU appears to be digging them fine with a slight bump to VCCIO, VCCSA and 1.53V over the 1.5V stock. Memtest+, Prime95, and occasionally even using the computer  I initially had it at 1333mhz same timings 1T, no overvolt. Bumped it up to 1600mhz 9-9-9-24 1T, slight overvolt. It's doing fine, idle and load well within safe limits. So... I guess what I'm trying to say, is... I want that 2%, drat it.
|
|
#
¿
Aug 5, 2011 22:23
|
|
|
Nothing tremendous, but it does seem to help with loading stuff (from an SSD) and getting lower stable latency in my DAW when working with real-time recording and monitoring of the processed signal. But it's part of an overall "do everything short of bolting a 500W peltier to the motherboard and manually wick moisture away with a complex system of tubes and a huge amount of cotton" effort, I haven't taken the OC apart piece by piece to see where I'm getting X performance, you know what I mean?
|
|
#
¿
Aug 5, 2011 22:38
|
|
|
DJ Commie posted:With a few billion cycles a second, it might! THANK you. 
|
|
#
¿
Aug 6, 2011 15:36
|
|
|
Motherboards are many layers of relatively fragile connections sandwiched together... I imagine a lot of it is for the same reason you don't overtighten motherboard screws. Crunch the semi-flexible silicon and you ruin important connections, except these are around the processor and NB and stuff so if you crunch there you're extra special hosed? I guess it's sort of a "have faith in the manufacturer" thing with older processors and motherboards. I just stuck a 212+ on my LGA775 motherboard and it has no back plate, but that's about 70 pounds of clamping force. (And because of that big, heat-wicking fucker, without even getting close to the voltage ceiling for the 45nm process, I went from a 600mhz overclock to a 1000mhz overclock, score, the Q9550 is going to be doing its thing for some time yet!) Nothing broke, but it sure would have sucked if something did. I was very careful to tighten it all in small steps to distribute the force evenly. It wasn't a relaxed process in the least. Installing the Noctua NH-D14 on my recent Sandy Bridge build felt, in every regard, much safer. The backplate engineered onto the motherboard itself was a nice confidence boost to bolt that 3 pound bastard onto there and tighten the screws until they stopped without so much worry about potentially making a crunching noise. Still, while it was much more psychologically anxiety-inducing to work with the 212+ in the LGA775 and its naked silicon contact surface, I have to say that Coolermaster did their job and gave it sufficient contact area to take the tightening force without any issues, naked silicon and all.
|
|
#
¿
Aug 7, 2011 08:58
|
|
|
I suspect but don't know for sure that it's a combination of slight tolerance differences with motherboard makers and heat sink makers; the only issue I had with the LGA775 cooler (after market, AC Freezer Pro 7) was that it did bend the motherboard slightly on installing it (before mounting the motherboard to the standoffs). I was much better prepared for that when I swapped it over last week to the 212+ (which didn't flex it nearly as much, either, thanks to the mounting bracket distributing the force much better), but installing the AC back in 2008 and seeing the motherboard flex while knowing that it's got tons of thin metal traces running all through it was pretty scary. And the AC Freezer Pro 7 has good mounting hardware, it was not difficult to install, much easier than Intel's finicky stock cooler from that time period. Nice, firm lock into place, no screwdriver required for any of it, big plastic knobs to turn to open and close the retention clips... But I'm never going back to a cooler that doesn't use a mounting bracket ever again. gently caress that. Mounting brackets are a massive improvement over the old way of doing things. Putting the NH-14D onto the LGA1155 board was just dead simple. You could not possibly gently caress anything up. It didn't cause the motherboard to flex, either. Well done by Noctua, much better mounting hardware than the 212+, but then for an extra $60 I guess you'd have every right to expect that. The 212+' mounting hardware was still much better than the push-pins on the LGA775, but the single mounting point on it allows for about 5º of rotation and if you don't notice it before you tighten it down, it'll be a bit rotated. No big deal, though, it still wicks heat like crazy and makes great contact with the processor. Agreed fucked around with this message at 00:44 on Aug 10, 2011 |
|
#
¿
Aug 10, 2011 00:41
|
|
|
I described the AC Freezer Pro 7. I have the damned thing in a box in the next room, I'm not talking out my rear end here. While they are similar in concept, in execution I found it much easier to work with the AC Freezer Pro 7 than the Intel one, which didn't seem to be as well made, the pin retention turning mechanism seemed iffy by comparison.
|
|
#
¿
Aug 10, 2011 01:07
|
|
|
WhyteRyce posted:Locks firm in place, no screw driver required, big plastic knobs to open and close...that is exactly what the stock heat sink has. I have used the same heatsink on my Yorkfield system, I liked my AC Pro Freezer 7 but I don't understand how it's so much easier to setup or more fool proof when it has a bunch of loose pieces and a "fits on any socket if you use the right pins and line the mounting bracket up correctly" design. Maybe it's because you've done it hundreds of times, while most of us have done it once or twice and that's plenty to decide it loving blows compared to a mounting bracket. I don't have to think about it in the big picture sense that it seems you do - do you build systems professionally, it seems like you must to have done it hundreds of times. To draw an analogy, I'm really good at setting up a guitar. I do it all the time, if you told me to swap out pickups and electronics and give a full setup including a bit of extra work to shield the cavities, no sweat. I've done that a lot, I'm good at it now. But I would understand if someone else who has maybe one main guitar digs into that process and finds that it is complicated and difficult, and the first time they use a drop-in replacement pickguard that's pre-wired and only requires a single component soldered (the output jack) they think "holy gently caress, this is how I'm doing it from now on if I ever need this done." Your experience with the platform is affording you a broader perspective and I'm not trying to say that for you it's hard, I'm just letting you know that for me, a couple times with that clip poo poo (and while the AC Freezer Pro 7's version was better than the Intel stock one as far as robustness/build quality goes, it was still poo poo) is for the birds and I will use a mounting bracket every time from here on out because it's so much easier for the new computer I'll build every two or three years. I hope that clarifies my stance on it, I'm not suggesting that you suck at it, just letting you know that I do and as a result I say gently caress that. Cool?
|
|
#
¿
Aug 10, 2011 09:04
|
|
|
People freak out about temperature way too early. So you're kissing 70ºC with literally one of the most demanding loads that it is possible to run? You could run the computer headless as a dedicated Prime95 machine for years at that voltage and temperature and be fine, that's well within 24/7 safe temperatures. Under many iterations of IBT my overclock (1.36-1.38vcore, 2600K, hyper-threading enabled) kisses 80ºC, and will run in the mid to high 70s for Prime95 on the most CPU intensive test (one core's a bit hotter than the others or it'd just be "mid 70s," nothing to do but put up with its propensity to exceed its siblings by about 2-3ºC). I'm not worried about it, that's very far away from the temperature at which anything would even start to throttle, let alone any protections kick in to shut it down. Processors don't have to be supercooled to run, they just have to be within safe limits. Under 70ºC is super duper safe. Overclocking consensus is often misinformed, with ideas like "turn off speedstep and C1E for a more stable overclock" - no, turn off speedstep and C1E for a higher electric bill. If your processor is running within safe temperatures, you're doing fine. Don't worry so much Under the most demanding day-to-day conditions you'll probably never see anything nearly that high; when doing large project renders across all 8 logical cores I don't get past the mid 50ºC range. Torture tests/stress tests are well named. They show you the extremes of what your processor can be subjected to under its power delivery, clock speed and thermal dissipation. They aren't normal temperatures... And even if they were, again, you're well within safe limits for running the processor full-time at extreme load. Something else would break before it would.
|
|
#
¿
Aug 17, 2011 00:15
|
|
|
I disagree on a few points. First, and perhaps controversially, I don't disagree on LLC. For one thing, the Auto and Regular are both Intel's stock vdroop implementation. Second, while there are good technical reasons why at higher voltages it can be potentially harmful to processor 24/7 lifespan, it's an effective tool to set the rough parameters of your voltage when using offset mode. Within reason I haven't seen evidence that it's going to cause a CPU failure before the system is entirely deprecated anyway. I've used a more aggressive prior implementation of it on a Q9550 system for some years now and recently moved it from 3400mhz/core to 3800mhz/core, for a 1000mhz total overclock, on an Asus board - it's been fine, tests fine on stability tests, etc., etc., and it doesn't have fancy options, it's just off or on. Stability tests and "does it work in real situations" are both fine, though the system was built in 2008 and has been used extensively. That out of the way... PLL Overvolting - Off is definitely the proper choice there for any clock under 4.5GHz, and some folks get away with up to 4.7 or even 4.8 before it shows any stability improvements. And with it on or set to Auto it won't let you sleep, which sucks for some people I understand. VRM frequency - Under most circumstances fine to leave at Auto, but if you get instability that you can't root out, try setting it to 350 manually - that's a pretty safe, stable setting that can help you determine what might be causing instability. Phase control - I recommend setting it to "Extreme." The ones below it, including "Optimized," will not use all of your VRM phases, meaning you're feeding more power through less of your potential power delivery system, which unnecessarily stresses the components compared to utilizing the full VRM phase array. I don't like their naming of this option as I think it encourages people to under-utilize a very safe and beneficial feature of the motherboard. Duty control - While this is, I believe, the default option, ensure that it's set to T-Probe. Safety first, the other option will maintain power on a current load basis even under dangerous conditions for your parts. It's important to keep perspective. A 2500/2600 non-K is already REALLY REALLY FAST so it's better to not overdo it here than to risk your hardware for that extra 100mhz/core. By combining your full VRM phases and thermal, it will more safely deliver solid and steady power to your processor, which is what you're after. CPU current capability - I really don't recommend exceeding 100%. Even a going a little bit out of spec here can be dangerous, there are important bumps that may not be rated for higher current, especially since by overclocking it and raising the LLC and voltage offset you are already increasing current as you raise the voltage. Just opens it up to risks that are completely unnecessary at most overclocks, even high ones. And scrolling down, set CPU Spread Spectrum to disabled. Hurts stability, and it's really unlikely you're going to get some kind of a resonance with it off anyway. Finally, just in case someone told you to do it, don't turn off the damned power saving features. Enhanced Speedstep and C1E don't hurt stability on this processor (or, in my experience, on the Core 2 Quad that I have overclocked either). They do, however, save you a hell of a lot of electricity. Using LLC and a voltage offset to achieve higher voltage under loads for stability are perfectly compatible with the power-saving features, and why would you want your processor running at maximum speed all the time anyway? You only need it when you need it, and provided your VRM is up to the task and you've got a stable overclock, when you need it, there will be no trouble transitioning from the 1600mhz resting state with C1E halts to the full multiplier value you've set under load. Please note that some chips are sensitive to using negative offsets; while negative offsets are very useful for controlling the voltage under load, especially when using LLC to set the rough voltage range (which is a function it serves on Asus motherboards), nonetheless you can end up in the unhappy position of getting BSODs, hanging or crashes if you use too high of a negative offset, because the voltage offset is always applied. I use a +0.025V offset, so I run at 1.024V rather than the stock ~1.0V at the resting state of 1600mhz. If you use a negative offset, you'll be going down from the stock ~1V, and if you go down too far, the processor might become unstable. Keep that in mind... it can seem like a weird gremlin issue because, what the hell, it tests fine at full load under stability tests but then crashes randomly when you're just browsing the internet or whatever - but it's because it isn't being fed the voltage it needs to run when power saving is working. Don't sacrifice power saving to keep a negative offset, though. You're better off not using LLC and instead just biting the bullet and raising the voltage via a positive offset than you are sacrificing power saving. The processor can use as little as 10-15w when running in its power saving mode, a fraction of its load wattage. Agreed fucked around with this message at 19:35 on Aug 19, 2011 |
|
#
¿
Aug 19, 2011 19:26
|
|
|
They have a tremendous amount of leverage at the moment what with AMD not really having swung a bat in their direction for so long and apparently plagued by issues getting it out the door and getting markedly better performance than the current generation. Why wouldn't they wait and see rather than rush them given the current market conditions? Maybe they can fine-tune the manufacturing process in the meantime for better yields while they're at it. I just wish they didn't fragment their motherboard market so badly. I bought in at 1155, what upgrade paths with Ivy Bridge does that decision lock me out of? No QPI processors, or...? (Edit: Or anything, apparently, backwards compatible, not forwards  )Speaking of the roadmap and 2011, they're supposed to be bringing the extreme Sandy Bridge models in pretty soon, aren't they? I wonder if the ease of overclocking the 2500K and 2600K play into the delays at all. Getting to 4.5GHz was trivial, getting to 4.7GHz took only slightly more effort, I could probably go higher if I wanted - and it's really, really fast, makes my Q9550 at 3.8GHz look slow. Did they anticipate the easy overclocking, and if so, would that factor into a decision that affects the timing of releases? Agreed fucked around with this message at 01:56 on Sep 2, 2011 |
|
#
¿
Sep 2, 2011 01:53
|
|
|
Ivy Bridge is tempting in the "if I had unlimited resources, that would be cool" kind of way. Hell yeah hook me up with a couple hex cores and let me stuff so much RAM in that fucker that I could compete with SSDs for space, sure. But then the bubble goes "pop" and I remember that I just spent quite a bit putting together a really salty Sandy Bridge 2600K setup that is absolutely loving blazing fast and that 16GB of RAM is plenty for what I do - which is sometimes pretty processor and memory intensive audio work! - and Ivy Bridge, while great in theory, is going to be ungodly expensive to even start into, let alone get the enthusiast parts to overclock well with. So gently caress that, but if I had the disposable income it sure would be cool.
|
|
#
¿
Sep 12, 2011 15:29
|
|
|
Yeeeeep. To be fair, I rely on a GPS unit which hasn't been updated since 2009 to go from place to place, so this isn't the only roadmap I'm having some issues with  So, what's the deal, then, Ivy Bridge will be LGA1155 socket compatible, faster clock for clock, and around the same price? That's an upgrade I can get behind, if so, but I thought I remembered reading some new chipset being required to support it. Higher numbers. Wait, that's for integrated USB 3.0 support and stuff, right? Well, huh. If they can pull clock for clock as promised I might upgrade, but I'm not really sure; would it have made sense for someone with a pre-Penryn C2 that they had massively overclocked to chance it on a post-Penryn C2 that they might be able to massively overclock, or not? 22nm, what kind of voltage range is going to be safe for the processor I wonder? We're getting small enough now that shorter term damage to the microarchitecture becomes possible with voltages in excess of specification, aren't we? Certainly not going to gently caress around beyond Intel's stock LLC algorithm with Ivy Bridge if I do go for it.
|
|
#
¿
Sep 12, 2011 16:26
|
|
|
Welp I'm fine with my Sandy Bridge now thanks! At least this is some good news for AMD, poo poo was bad enough when they had to do their expectations management to shave their 50% clock for clock down to 35%, if Intel did manage a 20% clock for clock improvement with Ivy Bridge that would have had a sound kinda like a nail going into a coffin for this generation of product - now, not so much, maybe AMD can get some more efficiency out of Bulldozer. Edit: That MOV instruction trick is really clever. Intel is kinda cool. Agreed fucked around with this message at 17:59 on Sep 17, 2011 |
|
#
¿
Sep 17, 2011 17:43
|
|
|
I know they're advertising an increased maximum multiplier and better power efficiency, but I am really curious what kind of overclocking performance enthusiasts will be able to pull out of this thing. The physical space occupied by individual components in the microarchitecture is getting really tiny, and doesn't the much smaller die mean that the bumps themselves are getting smaller as well? More vulnerable architecture and power delivery, offset by more efficient power and heat characteristics. Wonder if we'll be seeing these things at the kinds of clocks people are getting with Sandy Bridge. Only time will tell, I guess, but it's pretty amazing how many people are running 4.5GHz on their 2500K/2600K, and some of us are lucky enough to get 'em even higher without getting into unsafe power and heat territory. I don't expect anyone apart from the "stable enough to boot and take a screenshot before the chip cooks" [H] guys to be coming anywhere near topping out its multiplier, but it will be interesting to see if Ivy Bridge's architectural balance in favor of efficiency means that they'll keep up with or even surpass Sandy Bridge's nearly trivial overclocking and put that 6% clock for clock to work for enthusiast setups. Everything's so small, but I guess a big part of their jobs designing the processor and chipsets is to consider all that and deliver a final product that can keep up with and surpass its predecessor. Still, going to be fun to watch for awhile as people do the trial and error involved with establishing apparently safe parameters! Agreed fucked around with this message at 01:14 on Sep 18, 2011 |
|
#
¿
Sep 18, 2011 01:06
|
|
|
Alereon posted:Desktop doesn't really have much to push a CPU right now. That's true, I don't know of much on the desktop that requires the kind of horsepower we already have and have had for years. Still, the next year should be interesting for the enthusiast desktop user. Even though Intel is making smart moves, AMD is finally positioned to do something that might actually attract competitive business for desktop users (as well as hopefully get their prices right for server use, god knows they need it). You probably caught Bulldozer's world-record 8+ghz with some absurd custom cooling system, as usual for that kind of showy silliness, but the message is clear. AMD is pushing for big numbers and attached to that successful record's press release was the calculated off-hand remark about how easy it was for them to achieve 5GHz+ with air cooling. Given that Sandy Bridge K-sku users mostly run at around 4.0GHz to 4.5GHz, with some exceeding but few getting within a multiplier or two of 5GHz. If AMD's not completely dishonest here (and I make no judgment there, their high clock was verified and it'd be a pretty lovely time to lie), they may be able to leverage more realistic big-number overclocks and really compete for performance (not to mention that we still don't really know how Bulldozer's semi-cores are going to affect real world performance). If they can make it affordable, anyway. Intel seems to have the advantage at every turn but it's something. Their expectation management regarding clock for clock improvement doesn't exactly impress, but if Bulldozer really does turn out to be an extraordinary overclocker as they're trying to demonstrate and they can keep their prices down, it could make the time between now and Haswell much more interesting. Still, Intel is absolutely flush and their resources mean they can afford to keep their eye on the future, prepping their technological base for future improvements. It's hard to beat that much money and the architectural efficiency of Sandy Bridge and Ivy Bridge are impressive. I've read that Intel's GPU is architecturally more integrated than AMD's. Anyone make sense of that claim for me? AMD is focusing so heavily on integrated graphics, is it true that Intel has superior integration (and if so, does that mean we can expect them to be able to scale their on-die graphics at a faster pace as well)? Agreed fucked around with this message at 09:03 on Sep 19, 2011 |
|
#
¿
Sep 19, 2011 08:59
|
|
|
Thanks, folks. Interesting times.
|
|
#
¿
Sep 19, 2011 15:16
|
|
|
Wonder how many of those guys survived the merger on the ATI side and how many are still around on the Intel side. A big small world if some of the guys who used to work together on graphics are now competing directly on graphics. Although it wouldn't surprise me under any circumstances, really, specialized industries are pretty incestuous and no noncompete is going to last 13 years.
|
|
#
¿
Sep 19, 2011 15:41
|
|
|
They've demonstrated that it's a good overclocker. I imagine that's going to be their push in the desktop market. Intel seems better positioned (due to having dump-trucks full of hundred dollar bills) to bring their prices down if it means staying competitive, though. At the proposed price points, AMD is going to have trouble, especially since Intel's going to be offering the 2700K at the 2600K's price point (which is probably just going to be a binning process with the current 2600Ks, but still, it'll push prices on the 2600K and 2500K down by design to make it that much harder for AMD). AMD has talked about "easily getting 5ghz on air" as part of their press release regarding their record-holding 8GHz+ clock on a Bulldozer chip; the only problem is that their claimed clock for clock improvement, unless it's intentional misdirection (why would it be, let their stock take a dive while Intel pushes forward with high-powered architecture and even greater power efficiency for the markets where the real money is?) means that if every single Bulldozer buyer gets 5GHz no problem, it's just a big number. Assuming 35% improvement over current AMD CPUs, that's still lagging behind Intel's performance, and the clock advantage means that even if most Intel CPUs aren't likely to hit 5.0GHz right now, most of them will do 4.4-4.5GHz, so even among enthusiasts they've got an advantage in performance at the same price point as AMD's Bulldozer chips. I guess it remains to be seen what the more-than-just-hyperthreading, less-than-additional-cores extra parts on the Bulldozer chips will do for performance, but they better all overclock like crazy if AMD's got a shot at regaining some performance competitiveness this generation. And bring the server parts down, Intel runs the market there, how will AMD compete apart from cost savings?
|
|
#
¿
Sep 23, 2011 18:17
|
|
|
Yeah, this is an extraordinarily impressive power efficiency improvement. I'm not knocked out by the clock for clock - I remember the 20% figure, obviously they meant power:performance:clock rather than just 20% more, eat it Sandy Bridge - but I am impressed by it nonetheless. Seems like a very well put together processor and the move to 22nm and a superior transistor construction had to happen sooner or later, what better time to get it up and running than when your nearest competitor is dead in the water, I guess. Their current laptop lineup is already extremely power efficient, I imagine Ivy Bridge is going to kick some real rear end in that market. Most importantly it remains very clear that Intel knows what the hell they're doing and continue tick, tock, tick, tock while everything that could go wrong is going wrong for AMD. I don't view that as a good thing, really, but they sure do know how to keep to a schedule. drat.
|
|
#
¿
Oct 10, 2011 05:49
|
|
|
They were publicly saying 50% clock for clock. Then publicly lowered expectations to 35% clock for clock. Which is obviously a crock of poo poo for the majority of applications, and a patch isn't going to fix that. Edit: Oh, sorry, I get what you mean. Clock overall, not clock for clock.
Agreed fucked around with this message at 16:33 on Oct 12, 2011 |
|
#
¿
Oct 12, 2011 16:29
|
|
|
Intel Roadmap Discussion: Bulldozer =  Those TDP numbers for the performance chips, hot drat. I may get one just because there's every chance that the power draw will pay for itself over the lifetime of the system, and I can write off the expense.
|
|
#
¿
Oct 18, 2011 06:47
|
|
|
The Great 2011 Turkey Shoot
|
|
#
¿
Oct 20, 2011 22:34
|
|
|
The fact that they're bringing them out at higher clocks, and that there is some indication that Sandy Bridge isn't going to so easily scale to greater core density with the Sandy Bridge E fab (shipping initial parts with "only" 6 cores instead of the proposed 8, but also meaningful), tells me that power consumption isn't the only consideration. Sandy Bridge is a killer processor, it looks like Ivy Bridge is going to exceed it. The only thing I'm not sure about is overclocking capability but if they're scaling the clocks up and expecting a modest but meaningful 6%ish improvement clock for clock I have to figure they'll be able to overclock really well too. Given that the wait is only a couple months 'til they get chips out to consumers, given that Ivy Bridge is going to be replacing Sandy Bridge except in Sandy Bridge E, there are several good reasons to wait, even if they're just speculative-but-probably right now. So it's not really about power consumption alone, there are a number of things. Things that are making me consider going with an Ivy Bridge processor even though I've got a 2600K that's running 4.7GHz stable. If Ivy Bridge turns out to be a good overclocker, I'll make the swap. But I can write it off and have all the other underlying supporting hardware in place - I don't care about the integrated GPU, PCIe 3.0, or the native USB 3.0 support, since my motherboard packs the elements I -do- care about on its own. I'd just enjoy a better processor. Every bit counts. If they turn out to be mediocre overclockers I'll stick with my 2600K. I have to imagine they will be great, though, the double efficiency improvement (22nm, and 3D transistors to replace the last generation lithography) is really attractive. But I'll make sure to watch a bunch of people at overclockers.net and [H] cook their chips first to get a handle on what the deal is first. If it weren't an expense I'd just skip it, and most people with a Sandy Bridge processor probably should. But you guys who are running Core 2 processors should really hold onto your wallets for the time being because it's shaping up to be a hell of a chip.
|
|
#
¿
Oct 22, 2011 06:43
|
|
|
Magic Underwear posted:Nope. Read the article again. It's going to paper launch in Q4 2011, but "according to Otellini, first Ivy Bridge systems should become available in Spring 2012". I read that as Q1, if it really is Q2 then yeah there's no point. I wonder if they're getting them to OEMs first as a moneymaker or as an opportunity to fine-tune the process with chips that can be fabbed as higher end hardware and blocked off to leash them to a lower performance tier since getting enthusiast hardware in a Dell/etc. is prohibitively expensive. That would make the launch for brand partners partially a paper launch, or at least a lucrative but also useful trial run to get things good and ready for the high-end chip push for enthusiasts and higher performance but smaller scale server applications, would it not? Seems like a mixture of seizing the moment to just make AMD look loving foolish with Bulldozer, adhere to the Tick/Tock despite odds, and also get a chance to work out any kinks before opening up the market completely.
|
|
#
¿
Oct 23, 2011 09:13
|
|
|
Hell yeah, 2500K/2600K sticking it to the man with best IPC The power draw on full load is pretty nuts. Not Bulldozer nuts, but, you know, loving high. Then again, that's 6 real cores and it's still not drawing that much. No interest in this part, will check back when they get those two disabled cores hummin' but only to see "wow that's fast at stuff that can use 16 threads efficiently" and then go back to not being willing to spend $1000+ on a CPU.
|
|
#
¿
Nov 14, 2011 11:37
|
|
|
It's a hex-core Sandy Bridge with higher clocks, what's it supposed to do? Anomalous nosedive in performance? I mean, yeah, it's powerful, but what did we expect? (Well, I guess we kind of did expect 8 cores until they made that announcement but still you get what I'm saying.)
|
|
#
¿
Nov 14, 2011 17:05
|
|
|
|
| # ¿ Apr 23, 2024 22:06 |
|
|
This is pretty much exactly what we've been expecting all along, it's weird to see people in that thread getting all bummed out about "I was hoping for 10%-15% clock for clock" - like, really? Okay, I remember when Intel was throwing out funny numbers half a year ago too, but that was pretty clearly marketing and they've since clarified for the record that the clock-for-clock improvement is modest, it's intended to be a power efficiency and integral process change. As the second gent points out, the power requirements for the same clock rate are crazy good. Unless there's something unforeseen about the tri-gate transistor lithography that makes it prone to electromigration or other degrading nano-scale electrical phenomena, it looks like in addition to higher stock clocks, dramatically better power efficiency, they should also be really good overclockers, with a higher trivial and higher "sweet spot" overclock, not to mention much lower power draw as they approach very high clocks (2600K/2700K can get up to around 300W at 5GHz). The low TDP of the efficiency-first SKUs in particular means they'll be able to bring a serious, genuinely 4-core CPU to laptops where battery life is still a necessity, and the lower thermal waste means they'll be easier to cool in that application as well. All in all, it looks like just what we expected, doesn't it? Very impressive release, especially in the climate where their competitor is tripping over their own feet ... ...  AMD, why  Anyone complaining about Ivy Bridge is misguided or had unrealistic expectations, I think.
|
|
#
¿
Nov 30, 2011 18:20
|
|