|
Power target is pretty much everything, just custom cards generally have it set much higher from the factory. The percentage values you see in overclocking programs are arbitrary and you need to dump the bios to see what actual wattages it's working with. Nvidias GPU Boost is basically a cleverly marketed (And actually very good) power and temperature management system. Core offset helps once you run out of the boost table, but if you're really dedicated to power management you can edit a bios with a new one with whatever values you want and never need to mess with core offset. For benchmarking only there might be some merit to a hosed up boost table and a large offset to get super high speeds in certain areas of a benchmark and slow down in others that cause the GPU to crash, but I've never messed with anything like that.
|
#
¿
May 27, 2013 20:04
|
|
|
|
| # ¿ May 9, 2024 19:49 |
|
|
Sometimes they do end up getting carried over, but it's a good idea to use the reset to defaults command and manually reenter everything to prevent any issues from arising.
|
|
#
¿
Jun 5, 2013 13:04
|
|
|
That vdroop article is outdated and inaccurate relative to any processors you're likely to be using today. The VID can be up to 1.5, but the voltage delivered should never exceed 1.38v to stay within Intels specifications. Vdroop is designed into the power delivery specification and is intentional.
|
|
#
¿
Aug 5, 2013 00:10
|
|
|
That's still from an era where a high FSB is useful. You can probably drop the CPU multiplier to 8 and run a FSB of 360MHz with the memory 1:1 to that and get better performance than with a 320MHz FSB and 400Mhz memory. What chipset does your motherboard have? A lot of them from late in that era could handle a bit over 400MHz no problem. You might be able to drop the CPU multiplier to 7 to do a 411MHz FSB with 1:1 memory, but that's near the edge of where a lot of chipsets topped out at AND overclocking the memory which may not take well to it.
|
|
#
¿
Oct 19, 2013 13:04
|
|
|
Oh, that's a chipset from really early in the era before even 333MHz was an officially supported FSB. Maybe someone else can add in if they remember them clocking very well, but since the specced limit was 266Mhz you're probably already near the limit.
|
|
#
¿
Oct 19, 2013 17:31
|
|
|
VelociBacon posted:Is there this much variation chip-to-chip even within the same model? Yes, there can even be as large as greater than 1GHz differences between good and bad CPUs.
|
|
#
¿
Oct 21, 2013 07:02
|
|
|
SocketSeven posted:PCH voltage is bumped to 1.08 from 1.05 Why? You REALLY shouldn't need to do that. If you're overclocking the bclk, stop and use the multiplier. I've been able to lower mine to .98 just to maybe save a watt or two. SocketSeven posted:VRM Frequency is fixed at 300Khz. Set that to 350, 350 seems to be the magic number for most people. Helped me get an extra 200MHz out of my CPU. SocketSeven posted:VRM Phase Control is set to extreme instead of manual adjustment (which offers a range of options from regular to ultra fast) craig588 fucked around with this message at 22:40 on Nov 16, 2013 |
|
#
¿
Nov 16, 2013 22:37
|
|
|
You might have a faulty motherboard. You're not even overclocking the PCH by running the stock blck. Especially considering you get errors at stock speeds it might be time to RMA the motherboard.
|
|
#
¿
Nov 16, 2013 23:05
|
|
|
K-boost is a bad idea and a huge waste of heat and electricity. If it turns out you have throttling problems modding the bios for a higher power target is a much better idea so you can retain all of the low load and idle power states.
|
|
#
¿
Dec 25, 2013 00:51
|
|
|
You really want a heatsink and some sort of airflow over the VRMs. It's not necessary for the memory. When I first water cooled my 680 I left the VRMs passive with really tiny heatsinks and they got to over 100C according to my cheap IR thermometer. The memory is barely above ambient.
|
|
#
¿
Jan 23, 2014 20:31
|
|
|
The H100 (non I) pump on my videocard is the loudest part in my computer. On my Sandybridge at 4.6GHz I am doing fine with a D14 with the fans at the lowest speed my fan controller goes.
|
|
#
¿
Feb 5, 2014 10:19
|
|
|
MemtestG80 is the best tool for checking for GPU memory errors. I let it run 100 passes with Furmark running in order to heat up the board since memtest itself doesn't generate much heat, if it passes that without errors it's stable enough for games. For GPGPU I think the recommended number of passes is between 50,000 and 300,000. MemtestG80 is nice since you don't need to sit there and wonder if that flash of something you saw for one frame was an error or just a rendering oddity you didn't notice before, you can just see the running tally of all of the errors detected.
|
|
#
¿
Mar 30, 2014 04:08
|
|
|
Wait, it's not normally filled? I thought the problem came from too much TIM resulting in poor conductivity, not that there wasn't enough to fill the space.
|
|
#
¿
Apr 11, 2014 22:44
|
|
|
Thermal paste is practically an insulator compared to copper. I don't remember the exact figures, but copper has a thermal conductivity somewhere around 300 while even the best pastes are somewhere around 10. ShaneB is right. Compared to air gaps it's hundreds (Probably thousands) of times better, but heat will spread through the IHS so much better than it'd pass through a mound of paste that all you'd be doing is making a mess. Edit: It's why you always see custom machined aluminum blocks serving as heatsinks for weird shapes in prototypes, never custom formed thermal epoxy. craig588 fucked around with this message at 23:59 on Apr 11, 2014 |
|
#
¿
Apr 11, 2014 23:53
|
|
|
That's just a heat pipe without the capillary material so it'll only work in one orientation.
|
|
#
¿
Apr 15, 2014 16:46
|
|
|
Heat pipes work by boiling a fluid, the fluid evenly covers the interior surface of the heat pipe through capillary action. When a portion of the fluid is boiled off the other still liquid parts automatically move to take its place because that's what capillary forces do. The hot gas is moved away from the heat source to take the place of the moved liquid where it then condenses and returns to a liquid and the cycle repeats forever. Better heat pipes have their interior surface covered in sintered copper which looks like this, cheaper ones will use a braided wick or even cheaper ones will just have groves formed into them. As far as I can see you're literally describing a less effective heat pipe, using water instead of an engineered fluid and no capillary material.
|
|
#
¿
Apr 15, 2014 18:15
|
|
|
I have a D14 on my CPU and a H100 on my videocard and the H100 pump is by far the loudest thing in my computer. The included H100 fans were also incredibly loud, but I was able to run the H100 passively just through the positive pressure from my case fans.
|
|
#
¿
Apr 16, 2014 19:52
|
|
|
Shaocaholica posted:
I think every consumer facing IVB-E is unlocked. No idea about Xeons, but I'd guess they're all locked.
|
|
#
¿
May 21, 2014 18:59
|
|
|
Pretty sure it's a Q6600, or at least some sort of Kentsfield.
|
|
#
¿
May 23, 2014 20:27
|
|
|
The IHS is the wrong orientation to the centering notches for it to be a Westmere.
|
|
#
¿
May 23, 2014 23:27
|
|
|
What videocard are you cooling? For all of the recent Nvidia cards (starting with the 600 series) you DO NOT need to worry about cooling the ram even a little bit. Just the passive cooling from case airflow passing over the bare ram packages is more than enough. I can't say for sure about AMDs cards, but I'd guess they're probably similar.
|
|
#
¿
Sep 9, 2014 03:07
|
|
|
Yeah, you do want to heatsink the VRMs. I ended up building a heatsink out of a plate and soldering fins to it similar to how the old Thermalright coolers were constructed. I tapped some holes in the plate and reused 3 of the factory mounting holes, springs, and screws. Even that got pretty hot (I think my cheapo IR thermometer was reading in the mid 60s on the top of the heatsink) so I added a fan to it and it's very cool now. I'm still using a H100 on my 680 which is way overkill. I'm able to run the radiator passively and it only gets into the mid 50s even with synthetic loads, games are usually mid to high 40s. That's with extensive power limit modifications in the bios and a 1.21 voltage limit too. The H100 is a waste of money for a videocard, I only used it because I had it laying around.
|
|
#
¿
Sep 9, 2014 04:55
|
|
|
The only gains you're going to see are in terms of noise. A 770 isn't especially limited by heat, especially not one with an open air cooler. You might get .1 to .2 extra Ghz out of the CPU, but it's in the extreme range of diminishing returns. An extra 400$ would be better spent switching to a 780ti or one of the new Haswell-Es, depending on what you mostly use your PC for.
craig588 fucked around with this message at 22:52 on Sep 12, 2014 |
|
#
¿
Sep 12, 2014 22:50
|
|
|
You don't want to get below ambient because then you need to deal with condensation. You'd need to fill the socket with grease and paint the motherboard with nail polish and it'd still inevitably rust and short because of a tiny area somewhere you didn't notice. Phase change is also the cheapest/most sane way to get below ambient, it's why nearly all refrigerators and air conditioners rely on it. The only time they go out of that realm is for very specialized applications.
|
|
#
¿
Sep 24, 2014 02:44
|
|
|
Yes, but just to be sure, you know the daisy chaining molex plugs isn't putting them in series, right? You need to connect the positive from one fan to the ground from the other. You also might have issues getting them to start due to how the fans pulse current up and down as they spin.
|
|
#
¿
Oct 1, 2014 20:54
|
|
|
Use a negative offset.
|
|
#
¿
Mar 10, 2015 00:38
|
|
|
It's overriding what the CPU is requesting. If the CPU thinks it needs 1.163v and you have a -.005v offset the motherboard will give it 1.158v. Those numbers probably will never line up so perfectly because of the intended use of vdroop to control thermal limits depending on different load requirements, if it's requesting 1.163, but you're also doing something that needs 150 watts it'll drop the voltage automatically regardless of what the CPU is requesting, and normally the clock speeds would go along with it, but pretty much every overclocking capable motherboard will ignore power limits for clock speeds. Basically, whatever the CPU requests is going to be offset by -.005 and everything else will move around with that new -.005v adjusted voltage as a target. If you want less voltage increase the offset, I think I'm using -.02v right now.
|
|
#
¿
Mar 10, 2015 20:31
|
|
|
Stop, that's the path to LN2 pots and modding boards to run for only a few hours. Get it stable and ignore benchmarks unless you need to troubleshoot something.
|
|
#
¿
Jan 17, 2016 17:22
|
|
|
28.9 works very well at making the most amount of heat. I haven't been able to find anything that makes more heat. I'm not using anything else but the newest prime 95 for stress testing now because nothing else is as reliable for heat generation. I was using linpack and wasn't even breaking 80C in 24 hours at 4.7Ghz on a Haswell-E, but with prime I was breaking 90C in less than 8 hours. Had to step it down to 4.3Ghz to stay under 80C over 24 hours. Now the usual advice used to be "Who cares if you're going to burn it out in 2-3 years? You'd want a new CPU by then anyways" but that seems to have changed and will probably be even less true as time passes. I kept a Sandybridge for 5 years and if Intel continues to have a hard time pushing up performance I might end up keeping this one for 7-8 years.
|
|
#
¿
May 15, 2016 19:08
|
|
|
MSI Afterburner or EVGA Precision. They hook into the same points in the drivers, choose whichever UI you like more.
|
|
#
¿
Jun 9, 2016 19:52
|
|
|
Nope, check the "apply on startup" checkbox and it'll even be saved between reboots.
|
|
#
¿
Jun 11, 2016 02:28
|
|
|
LLC is cool now. It was just early implementations that either compensated too high or added more jitter to the spec than was supposed to be allowed (and even then there was a lot of FUD from people not understanding how to use an oscilloscope and it was overblown). If you have Haswell or newer I wouldn't say go straight to 10, but move it up as much as you need to to get the voltage you need at load.
craig588 fucked around with this message at 16:34 on Jun 16, 2016 |
|
#
¿
Jun 16, 2016 16:32
|
|
|
Sh4 posted:just flash your cards people. I wish. There's new harder DRM on Pascals NVflash, you can't flash anything that isn't signed yet. It's been cracked before though so I'm still hopeful.
|
|
#
¿
Jul 5, 2016 03:10
|
|
|
Those are extremely cool temperatures. Are those load temps with the latest version of Prime95? You should be able to push the voltage and clock speeds much higher with a chip that runs that efficiently. I wouldn't worry about getting up to 1.3V and 70C loaded lasting for years.
|
|
#
¿
Jul 6, 2016 18:00
|
|
|
cat doter posted:It's just "real world" testing with highly demanding games, I'm not a fan of synthetic tests though I should probably take a look at prime 95 just to see what the worst case scenario is. Yeah, that's why people are happy with getting up to 80C in Prime, it's not particularly indicative of real world temperatures, but lets you know the worst case most extreme and then you can ignore temperatures forever. It seems like you have a really good CPU.
|
|
#
¿
Jul 6, 2016 22:57
|
|
|
The OP was written around the time when Sandybridges were dying kind of more often than you'd like and people were working on theories as to why they were dying. Temperature is an easy quick "Fix this first" thing, but after years of people overclocking and stressing them it's most likely that the reason people were killing early Sandybridges was sending nearly 1.5V through them. The way to keep temperatures below 72C was to reduce overclocks and voltages so everyone ended up significantly less than 1.5V and saved their processors. These days I wouldn't be bothered at all up to the mid 80s under full load (with voltages under 1.4V as well).
|
|
#
¿
Jul 27, 2016 20:27
|
|
|
I lightly played with higher multipliers for lower numbers of cores and it was very difficult to stability test in addition to not being too meaningful of performance increase in benchmarks. By the time stuff needs the performance it's hitting all the cores and dropping you down to the all core multiplier anyways. I gave up on messing with it after a half hour or so, seemed like a waste of time. That was with a 6 core processor too which you'd think would have more room for idle cores, but the way Windows bounces stuff around between cores they all kind of got hit enough to drop to the all cores multiplier with any stress test I tried to run hitting only 5 or fewer cores. I think I had to drop down to 2, maybe 3, threads before it'd use the 5 core multiplier and even then it would bounce around.zergstain posted:By the way, is it likely I'll need to increase the voltage for the XMP profile? zergstain posted:Next up, my GPU. I've heard conflicting information on whether Heaven and Valley are good ways to test the stability of my OC. What videocard do you have? We'll be able to give you a good baseline so you don't waste hours testing speeds that are stable for 99% of people. I like Heaven because it hits most DX11 features, it's able to scale up enough to fully load any videocard out today, and I've been running it so long I know how it's supposed to look so I can tell artifacts right away. The best videocard stability test is something you can stand to live with for a while (as long as it is capable of driving GPU usage to 100% and hits a relatively wide feature set) because you'll probably use it across many GPU generations as you get used to it and know what to look for regarding artifacts. craig588 fucked around with this message at 06:51 on Dec 14, 2016 |
|
#
¿
Dec 14, 2016 06:43
|
|
|
zergstain posted:It's an MSI GTX 1070 with a 1557 MHz base clock. I have MSI Afterburner. Max out the power target and check what clock speed you end up with when 3d rendering first starts up, adjust the offset clock so you start at 2000MHz (You probably could start with 2050 but that might mean you have to move down instead of up and it''s more fun to move up as stability tests pass instead of down as they fail) and move up in 13MHz increments from there. You shouldn't have to stability test long, long enough to get the temperature to level out should be long enough to tell for the initial testing, probably 5 minutes of load or less. The clock speed will drop as temperatures increase, but it's designed to do that, but once you find a speed that crashes in that 5 minute heating up window back off 13MHz and that probably will be stable. If you have good cooling and power delivery you might see artifacts before crashes, Pascal has gotten better about running out of GPU potential around the same time as power delivery capability runs out, if you see artifacts, same deal back off 13MHz and that'll probably be your stable speed. One trap is after almost every in game crash and recovery your computer will still be running, but it'll ignore overclocking settings until you reboot, so I'd just reboot after any driver reset. You'll know it crashed because the screen will momentarily freeze and flicker black and then recover back to running whatever it was doing before. Once you get it stable like that there might be a bit more you can eek out from adjusting the voltage/clock curve. Hit CTRL+F in the latest version of Afterburner and it'll bring up that table, same 13MHz steps there and you can try raising some of the points the GPU drops down to as it heats up. Pascal doesn't seem to get great results out of more voltage, it seems to get a lot more heat for maybe a single 13MHz step increase which doesn't feel worth it to me so I've left the voltage slider alone in all of the Pascals I've overclocked. As for GPU Memory? I'd add 400Mhz and (hopefully) ignore it forever. If 400MHz has artifacts the highest factory memory overclock I was able to find is 250Mhz so that'll almost certainly be a stable setting. It's not that memory performance doesn't matter, but memory overclocking usually doesn't have as much headroom for performance gains as GPU overclocking does. The memory speed isn't stepped like the GPU speed so if you want to get really crazy with it you could get all the way down to the exact MHz that it gets artifacts. Once you get everything dialed in like that try running something for a few hours and see if it's still stable, crashes will almost always be indication to drop the GPU by 13Mhz and artifacts are more likely to be memory, but either could cause either fault.
|
|
#
¿
Dec 14, 2016 14:53
|
|
|
The drivers induce a soft crash to protect the card if they see any sensors starting to read wild values, such as drawing too much current for a given voltage. Flashing Pascal bioses is still uncracked and that's a setting that's only editable in the bios and not software. It's mainly a card protection scheme that has a secondary effect of limiting overclocks.
|
|
#
¿
Dec 15, 2016 02:09
|
|
|
|
| # ¿ May 9, 2024 19:49 |
|
|
The sensors are more reading values post VRM into the GPU and seeing that the 1.05 volts or whatever is drooping to 1.025 across the shunt and that means it's drawing 250 amps which means something is wrong and it's time to go to safety mode to make sure the card doesn't kill itself. The afterburner graph is only a readout of the VID, not the actual voltage, it doesn't mean anything for watching voltage droop. I think the on board sensors might care if they saw the 8 pin 12V line droop to like 10.8 volts or something, but the rest of your system would probably be freaking out before then. I've overclocked 4 or 5 Pascals and I've never seen good results from increasing the voltage, a lot more heat for maybe one clock speed step increase. It's perfectly safe to mess with the slider in Afterburner because it's a factory predefined limit. The catch to watch out for is Pascal will automatically increase clock speeds with more voltage so you'll (likely) have to reduce the offset clock by a step or two as you add voltage to maintain the same clock speed, once you get it running with additional voltage at the same old stable speed then try adding more offset again. Power limit is another perfectly safe slider that virtually everyone should be maxing out even if you're not interested in overclocking, it just slides the card higher up the factory predefined clock table. The only time I wouldn't max it out by default is in HTPC cases that couldn't deal with the extra heat.
|
|
#
¿
Dec 15, 2016 19:27
|
|