Search the Community
Showing results for tags 'vdroop'.
-
So, learning a bit about overclocking, and so far I think I'm only scratching the surface. Also doesn't help that I'm trying it with a mini-ITX that already tends to hit 85°C or more in IntelBurnTest when the room temperature is high. (Looking at getting a different cooler, though, might be the tight space.) Anyway, a seemingly important but often overlooked part of overclocking appears to be LLC or Load-Line Calibration. I read it commonly has steps between 0% and 100% and that 100% let's the system have free reign to adjust power to prevent overheating but can introduce instability with higher overclocks, correct? Now... On my motherboard (specs in signature)... It's in levels... Nine of them. So, this confuses me as I wonder what each level represents, might it be like 9 for 90% or is the percentage divided by nine levels so 9 is 100% and 8 is 88.88%? Hope someone can clear this up and darn motherboard manufacturers, stop confusing us with your fancy changes and stuff...
- 2 replies
-
- overlocking
- vdroop
-
(and 1 more)
Tagged with:
-
So I currently have a i7 6700k and a MSI M5. I hate the M5 so what is the best OCing and gaming mobo for z170. Dont recomend MSI Mobos I have had so much trouble with them its not even worth it.
- 14 replies
-
Hello! I thought I'll post my testing results here also. This is aimed for people who are interested in overclocking, to give some sense how the VRM (Voltage Regulator Module) deals the voltages and how accurate the software readings are compared to measured and so on. Background: While doing some overclocking and BIOS modding tests on my R9 290 GPU(s), I started to wonder how accurate the software voltage reading actually is, and where exactly does it take the measurement. I googled for "Vcore measurements with DMM", "VRM voltage multimeter", "GPU voltage measurement" etc, and came up with nothing. I found only a couple of tests that had been done on a CPU, some threads with speculation and how one would measure the Vcore. I couldn't find any factual results on GPUs, or VRM voltage output vs Vcore both compared and so on. I figured it was time to take out my own multimeter and see the stuff for myself! With the caveats of me not being any kind of expert and the limited accuracy of the results, I hope these results can give an indicative picture of the matter. All tests were done on one of my Gigabyte Windforce R9 290s (OC model), water cooled with an EK full cover block and a beefy custom water loop. The second GPU had pcie power cables unplugged = off. Here's a picture of my card's PCB (I originally took this picture when I upgraded the VRM thermal pads, to see how the TIM had spread..): This GPU board design is very similar to the AMD R9 290 reference design, the main difference being a different choice of chokes and capacitors used in the VRM. It uses the same 5 phase design with IR6811 & IR6894 "DirectFET" mosfets. The IR3567B controller on this card is used on numerous AMD GPU generations, and has also been used in some motherboards (for example Gigabyte X79). For the physical voltage measurements I used a digital multimeter with test probes. PCIe current was logged with Corsair Link software from my digital AX860i PSU, that can report individual connectors currents etc. Here's a picture of the card in my system and the voltage probing spots marked on the back of the card: What I mean with the different voltages: Vcore = GPU core voltage, measured with DMM over a ceramic capacitor on the back of the GPU core (I used the one shown in the picture). VRM Vout = VRM output voltage for the GPU core, measured with DMM over a capacitor right after the VRM (I used the one shown in the picture). VID = Voltage Identification, it basically means the programmed voltages in the BIOS. I also refer to it here as the result of core VID + core voltage offset - it's what you would call your set voltage. Vdroop = A processor power delivery design feature, that is intended to increase stability and lifespan of the products under normal operation. I understand it as VID - Vcore. Vdrop = Voltage loss, results from current passing through resistive material. Here referred to as VRM Vout - Vcore. LLS = Load Line Slope, is used to define the load current - Vcore drop relationship in VRM operation design. LLC = Load Line Calibration, a feature found on modern motherboards to adjust Vdroop out of specification, or as Asus has called it in the past "CPU voltage Damper". Early models only had on or off setting, implementations vary. There are also others, but we will not concentrate on those now. Test settings, unless specified otherwise: GPU BIOS: Original Gigabyte R929OC BIOS, modified where needed for this test, OCP, OVP, TDP etc limits were modified, GPU core @1000 MHz, memory @1250 MHz, VID 1250 mV, AUX 1000 mV, Vmem 1500 mV. VID: Set to 1250 mV in BIOS and voltage offsets were applied to it with software tools ranging from -100 mV to +400 mV. Thus the effective VID testing steps were 1150, 1250, 1350, 1450, 1550 and 1650 mV. Stress test program: Furmark (v.1.11.0), resolution 1024x768. Gives about the hardest load one can encounter with GPU. It's fast, repeatable and doesn't load the CPU much. LLS: Different levels of "Load Line Slope Trim" were applied to the VRM controller via BIOS modification. Settings: +40%, default=+0%, -40%, off. Okay, now that's out of the way, let's looks at our first graph, the measured Vcore over VID range with different Load Line Slope Trim settings: The VID setting is on the x-axis and measured Vcore on the y-axis. As you can see, the LLS trim settings affects how close to the set voltage we get under load. Here is the same data, but turned into more easily visualized Vdroop (shown as negative voltage difference): Okay, I guess it's time to mention something about how the Load Line Slope works. Of course you can google the real technical data, but basically LLS is a (small) value of resistance. The controller uses this to calculate how much of a Vdroop there should be at any current (load) level. Let's say we have defined in our VRM design, that the LLS value should be 1.000 mOhm. That would mean that at say 100 A core current, we want 1.000mOhm * 100A = 100 mV Vdroop. When the load increases, so does the Vdroop. The lines are not linear (except LLS off), because the core current doesn't scale linearly with voltage. The grey line is what you would normally get with an unmodified R9 290 BIOS. Setting the LLS trim to +40% means the LLS value becomes larger, thus the Vdroop is also increased, and vice versa. Setting the LLS off makes the controller ignore the Vdroop calculations altogether, and the Vcore is kept steadily at the set level, regardless of the load level. However, doing this increases the transient voltage spikes, which is one of the reasons Vdroop is used. Now, let's take a look at how the VRM Vout relates to the set voltages. Measured VRM Vout difference to the set VID: Due to resistance, the voltage will drop by the time we reach the GPU core from the VRM, and the current also needs to make the trip back to the VRM. The controller ultimately adjusts the VRM Vout voltage, and needs to compensate accordingly to the set VID, LLS, and load. The VRM Vout is intentionally dropped as the load increases to meet the specified Vdroop. With LLS off, we can see just how much does the voltage need to be boosted, to be at the VID at the core. From all the VRM Vout and Vcore measurement data points, we can make out this Vdrop chart: At the highest settings the actual physical Vdrop due to resistance is 150 mV. Taking in the AX860i PCIe current measurements we can make a combined chart that shows Vcore, VRM Vout and Power: The power is calculated from the logged GPU PCIe power current, and multiplied with 11.8 V (see additional raw data dump I also did a test where the VID remained the same (with default Vdroop) and GPU frequency was changed instead. VID 1250 mV + 250 mV = 1500 mV, default LLS, scaling with GPU core frequency: As you probably guessed, the power, current and thus Vdroop scale pretty linearly with frequency. At the highest level there is some leveling off, maybe we start to hit architecture bottlenecks (like ROPs, memory bandwidth etc), or just a bit unstable. I was doing somewhat throughout testing, so I figured I need to test the VRM switching frequency too. Here's the VRM switching frequency impact at two voltage levels. Tested range 200 - 2000 kHz: Well what I can tell you, I have done a successful 1380/1700 MHz Superposition run at -40% LLS and the default 500 kHz switching frequency. From my experience with R9 290, going over 500 kHz doesn't have other impact than making your mosfets run hotter. Anyway in our overclocking science test we can see that the switching losses increase by ~60 Watts going from 500 kHz to 2000 kHz. I haven't yet talked about the software voltage readings, even though they were one of the main reasons for doing this. For me, they were actually pretty spot on throughout the test. You can compare the raw data below. Anyway, there are tweaks that can throw the readings off (for example VRM Tool ;)), but I would guess they are accurate enough for average user. When in doubt check with multimeter. Additional raw data dump I will just drop here: Total system draw at idle ~120 W. Max PSU reported Power draw @ VRM Vout 1620 mV: 1089 W from wall, 988 W out, 948 W 12 V rail out, 29 A (6-pin) + 41 (8-pin) = 70 A to GPU PCIe. Go over this, and the PSU fast OCP will kick in. Measured idle PCIe Voltage at connector pins, both: 11.99 V. Measured under load (1250 + 300 = 1550 mV setting) PCIe Voltage at connector pins, 6-pin: 11.94 V, 8-pin: 11.80 V. VRM Phase in voltages, 2@6-pin: 11.88, 11.88, 3@8-pin: 11.74, 11.74, 11.74 V, GPU-z report 11.50 V VRM Vout phase voltages, 2@6-pin: 1438, 1437, 3@8-pin: 1435, 1433, 1432 mV Power lost to heat just due to current traveling to the core and back at highest settings ~ 70 Watts. VRM Vmem_out, idle: 1500, load: 1550 mV Vmem differences, closest chips: 1520, furthest: 1495 mV Raw excel: Some earlier tests (not directly comparable, different BIOS with software Vcore messed up!), but also show other loads than the Furmark, raw excel: Well there you have it! There are definitely some inaccuracies, for example the PCIe current measurement was done with pretty coarse accuracy. Also the voltages increase as the GPU and VRM get warmer (especially at high V levels), so I tried to take the readings quickly after the start of the Furmark. I hope this was interesting to someone. Also, if you think I erred somewhere feel to leave comment! I sure enjoyed doing this and learning about the VRMs workings. Also had some exciting moments with sparks flying from the back of the GPU, I guess I'm not cut out to be a doctor judging by my probe handling skills
-
Hey guys, I have some overclocking questions. In the past I have overclocked my CPU a bit irresponsible, just slapping a stable voltage at a moderate frequency without giving it much thought. Now I want a good overclock, since I am a video editor so an extra 20% boost, means about 20% less render time and on a 10 hour render thats 2 hours less which is very important to me. I am running a Intel 3930K on a Gigabyte X79-UD3 1. First question... Leaving LLC at auto gives me MASSIVE Vdroop ... If I set the votage at 1.385V CPUz reports that my CPU is running at 1.300V at full load. At LCC set to Turbo I am getting 1.368V out of 1.380V ... the settings available to me are Normal, Medium, High, Turbo, Extreme. Should I push it to Extreme or should I settle with this. 2. Running Prime95 at 1.368V out of 1.380V I am getting a bit high temps... at around 85-90C across all cores. The setting Im using is for maximum heat output. Should I care about this, when Im stable, since when I am rendering temperatures are significantly less (about 10-15C less). 3. Such voltages and overclocks are pretty useless to me 95% of the time during the week and I noticed that when I set my voltage and frequency overclock manually, it just stays there no matter what. When I am running at defaults my CPU drops to 0.9V at 1.2GHz which is more than enough during my daily tasks. Can I get the best out of both worlds? Thank you.
-
Now I'm a complete amateur when it comes to overclocking, so I may be misinformed or missing something entirely here, which is why I'm asking. The way I understand it, when a processor (particularly one that has been overclocked) abruptly comes off of high load, there is a spike in voltage. The way to combat this is Vdroop, which lowers voltage under load so that in such an event the voltage spike does not exceed the processor's specifications, or the voltage the user has defined. Vdroop, however, is a double edged sword in this manner, because reducing voltage under load means reducing stability, sometimes to the point of causing errors or crashes. So here's my question - Would it not be possible to design software, or perhaps even better an integrated circuit, that would be able to detect the processor coming off load, and if utilization was dropping too rapidly feed more or less arbitrary instructions to the cpu to reduce load more gradually, thereby potentially eliminating voltage spikes and in turn vdroop to allow more stable voltages and more stable overclocks? If it is plausible, has it been done? If not, why? If it isn't, could you explain to me why so I might have a better understanding of the issue?
- 3 replies
-
- overclocking
- voltage
-
(and 1 more)
Tagged with:
-
WAN show featured! LLC no longer affects vcore stability in Haswell or Devil's Canyon processors, as all the voltage regulation happens on the cpu. LLC only affects Vin (the voltage supplied to the cpu as a whole), and as far as I can tell, Vin stability has no effect on Vcore stability. More information can be found in this thread. First a disclaimer: this is my experience with Load-Line Calibration (LLC) during my overclocking. Some of the information in the post may be incorrect, although I will try to only post information which I feel is validated given my experience. I don't have any certification in making motherboards or programming, so this is simply my understanding of how this stuff works. Please remember that every motherboard manufacturer may use a different bios setting to implement LLC, so make sure that you look at what YOUR motherboard/manufacturer says about their LLC implementation. Why I'm writing this: This subject has been briefly addressed by several people on several different forums and several different overclocking guides. But a quick google search for load line calibration gave an article from 2010 (http://www.overclockers.com/load-line-calibration/) concluding that LLC was good and overclockers should use it. Other threads that come up with a google search are usually two-liners asking whether they should use LLC for their overclock, with the general consensus being yes they should. I couldn't find anything about LLC on the LinusTechTips forum, and I felt that this should be addressed. There seem to be no recent go-to beginner-level threads about LLC, and there seems to be no general understanding of what it is actually doing. This thread is intended to be an introduction to what LLC is actually doing, and why you should use it with care. Background on LLC: For those of you who don't really know what LLC is: LLC was a featured added to motherboards several generations ago to combat vdroop. Vdroop is a drop in voltage supplied to the CPU as load increases; basically when you go from idle to load, the voltage would decrease. Given the small voltage tolerance that overclockers are working with (increased voltage is proportional to the CPU frequency/multiplier that an overclock can achieve), a droop in voltage applied to CPU can make a theoretically stable overclock unstable (dropping the voltage below that required to achieve the set frequency). LLC applies additional voltage to the CPU to combat vdroop so that when switching to load, there is sufficient voltage to keep that frequency stable. So LLC is great and you want to turn it on? Yes, but... For most modern motherboards, there are different levels of LLC that you can set in your bios. At certain levels of LLC (these may be different for each motherboard), the LLC can overcompensate for this vdroop, and actually apply vboost. Vboost is when the voltage actually supplied to the CPU is above the value that you set in your bios. This can be a nice way of ensuring that your overclock will be stable, but you have to be careful, because each CPU has a death voltage (the voltage where, if applied to your CPU, it will likely die). If you are toeing the line near your CPU's death voltage to try to squeeze every last MHz out of your overclock, LLC can bring your actual voltage above this level, which is a great way of killing your CPU (or making it degrade much faster). So although LLC is great for overclockers, it should be used with care, because you may just end up killing your CPU. Now each motherboard is different, and may label their LLC settings differently, and this thread will be based on my settings on my ASUS Rampage IV Extreme. Make sure that you check your motherboard manual (and do a bit of googling for other people's experience) for how LLC is implemented in your case. (For instance, Asrock motherboards from the H77 generation had their LLC values reversed from the values that ASUS uses.) Now I had heard of all this LLC mumbo-jumbo (actually it was thecrazyrussian who told me to be careful with my overclock), and I wanted to see just what actually happened when you try different levels of LLC. So I went through my motherboard and tested each LLC setting and ramped up the set voltage, seeing what the actual read voltage was. I used a digital multimeter (just a basic one that I bought at the Source), but even that is more reliable than the values that your motherboard bios (or software) can read. Testing Setup: ASUS Rampage IV Extreme, i7 3930k, 16Gb Corsair Vengeance CL8 ram in quad-channel, XFX PRO 1000W PSU, Corsair H100i cooling. The amount of Vdroop changes with the CPU frequency, so I set my multiplier at 40 (stock is 32, and I have a stable overclock for this CPU at 45). I started at an arbitrarily chosen Vcore of 1.325V and ramped up until 1.4V (the voltage past which my CPU degrades much faster, and is considered by some to be the near death voltage of the CPU), or until the temperatures hit about 78 C. To avoid having to reboot with every voltage change, I applied all of my voltage tweaks using ASUS' AI Suite 2. Idle voltages were taken at the Windows 7 desktop with no programs open, load voltages were taken after Prime95 (small FFTs) completed its first pass. My motherboard has five settings for LLC: Regular (0%), Medium (25%), High (50%), Ultra High (75%), and Extreme (100%). My Vcore (CPU voltage) can be changed by steps of 0.005V, which may seem very small, but keep in mind that a change of 0.005V in Vcore can destabilize an overclock. Results At a LLC setting of Regular (0%), the voltage at idle was an average of -0.018V from set (idle: blue line, set: black line), the voltage at load dropped an average of -0.054V from set (red line), with a droop from idle to load of -0.036V. Immediately we can see that not enabling LLC can seriously destabilize an overclock. Moving up to an LLC setting of Medium (25%), the average voltage changes from set were -0.007V at idle, -0.023V at load, with an idle to load droop of -0.016V. The idle voltage isn't that bad, being pretty close to the set voltage, but the load droop is still more than enough to destabilize an overclock. Now up to an LLC of High (50%), the average voltage changes from set were +0.005V at idle, +0.011V at load, with an idle to load boost of +0.006V. This LLC appears to be pretty good, with the motherboard actually putting out a similar voltage to the one we set in the bios. There is a small amount of Vboost, but the magnitude is unconcerning, putting us nowhere near the death voltage of the CPU. For this LLC setting I only went to a setting of 1.380V, because CPU temperatures were becoming concerning. This is where things start to get interesting. Setting a LLC of Ultra High (75%) gave average voltage changes from set of +0.018V at idle, +0.045V at load, with an idle to load boost of +0.028V. This is hugely different from what we set in the bios, idling at ~3x and loading at ~9x the increments we can increase and decrease by in the bios. Here I stopped increasing values for two reasons: the first being that the CPU was at 77 C, and the second being that the actual read voltage was just barely below the fast-death voltage for my CPU. I was ready to stop at Ultra High, but to do my due diligence, I tried Extreme (100%) LLC. The idle voltage was +0.031V above set, and the load voltage was an insane +0.086V above set. Just switching it to load brought the voltage well above my 1.4V ceiling. I didn't even let my prime95 get to the first pass, I just took the reading and brought the computer down as fast as I possibly could. Conclusions Quite frankly I was shocked to see the effect that LLC setting has on actual voltages, especially at Ultra High and Extreme. I do understand that that every motherboard may implement LLC differently, and the Vdroop/Vboost changes may not be as incredible as I saw on my board. I can easily visualize someone trying to get the highest overclock possible, but ignoring the LLC setting (or worse setting it to extreme) and frying their CPU. I hope this thread illustrates my experience with LLC and persuades the reader that LLC should be used when overclocking, but must be used with care. Personally I chose an LLC setting of High (50%) for my overclocking, because it resulted in no Vdroop, but didn't result in enormous Vboosts. I also took into account the small observed Vboost, and made sure to never bring my voltage to a level where the Vboost would touch the fast-death voltage of my CPU. I have what I consider to be a stable overclock with this motherboard and CPU at 4.5 GHz at a Vcore of 1.325V (stable for 24h of prime95 small FFTs). Note to the reader after additional testing: Vdroop and Vboost will not behave in a fixed manner! Idle and load voltages follow a linear trend, but the slopes of those lines are not equal. Read my follow up post for more details. TL;DR: LLC should be used while overclocking, but used with care. If you don't and you're not careful, you could kill your CPU or degrade it very quickly under load voltages. It can also be chosen logically, see part 2 for more details. Read Part 2 here!
- 50 replies