Gadgetory


All Cool Mind-blowing Gadgets You Love in One Place

Exponential Ryzen Voltage-Frequency Curve (Overclocking)

2018-04-24
one of our biggest findings with rise in two was its reduced voltage requirement at a given clock discussed in our 2700 X review our r7 1700 required at least one point 4 volts to one point four to five to maintain four gigahertz stability in our most torturous workloads but our 2,700 X only required one point one six to to hold the same frequency under today's torture this drew our attention because we already knew that our 2,700 X could barely manage 4.2 gigahertz at one point four to five volts without modifying base clock though it could do so with base clock changes more readily in other words there was a five percent increase in frequency from 4 to 4.2 gigahertz but a 22 point 6 percent increase in reported voltage requirement today we're talking about the Volt frequency curve for the new Rison 2 processors before that this is brought to you by the msi gtx 1080 gaming X and Nvidia's GeForce experience which allows you to retro actively capture key gameplay moments with shadowplay convert captures into gifs with new tools and apply filters to games hashtag no filter Emma sighs gaming X PCBs are high quality with well built power management and coolers that we've previously recommended learn more at the links in the description below frequency with Rison - is beginning to act a lot like GPU boost 3.0 or other modern GPU boosting features where basically your frequency is almost more dictated or predicated on the temperature and the voltage and the power than anything else it's not like it's just a fixed number anymore the boosting frequency Headroom is entirely dependent on other factors so provided you have thermal Headroom provided you have power budget and provide your voltage is low enough that you have both of those things because the voltage will impact the power consumption and the thermals you can boost higher automatically and boost higher or even just manually overclocking you can get it higher with a lower temperature to some degree so we've tested a lot of this and our initial experience with Rison - led us to believe that the volt frequency curve would be almost exponential to the point where we can put one on the screen as purely an example where as you kind of approach the end of the stable frequencies the voltage required to hold those stable frequencies it shoots up like a rocket so this is our hypothesis basically that it's a somewhat exponential curve to maintain stability as clock continues to push upwards and to be very clear here we have been able to overclock higher or more stable at least by using reference clock overclocking what we're doing today though is strictly tuning the multiplier and the core and we're leaving everything else controlled so we only want two variables for this one right now our biggest limiter here will be thermal risin two does actually run a bit warm especially as you begin overclock in it we typically do these tests with an NZXT kraken X 60 to 280 millimeter cooler max out fans and pump as we progressed through testing we realize that we're running into a thermal barrier as the voltage pushed up to 1.4 and Beyond and frequency to 4.2 and Beyond and so we eventually switched over to the flow 360 radiator from thermal take with three maglev fans on it which gave us enough Headroom to push through the rest of our testing so typically again we run out of thermal Headroom before we run into a an unsafe voltage as you might call it so our concern is less about killing the CPU with too much of voltage because we know we're not going to get there without basically exotic cooling or sub ambient cooling or something like that so again the hypothesis here is that at increasing frequencies the required voltage for both chips the 720 700 X rapidly increases out of a maintainable range thermally and the vertical difference between the lines in our theoretical graph at a given frequency should become huge one point four to five volts versus one point one six to for example at four gigahertz however at a given voltage the horizontal difference remains minor four gigahertz versus four point two gigahertz at one point four three ish bolts or 1.40 and based on this knowledge this curve again is what we expected to see but it's just an example let's look at some real data well begin with just the results table for the r7 1700 at the low end we were able to maintain a minimum stable voltage of one point zero six nine volts for 3.5 gigahertz all core producing a TDI of 49 degrees Celsius at 7 amps at the EPS 12-volt rails equating 86 watts of power draw increasing by 100 megahertz to 3.6 gigahertz required an additional 0.037 five volts and increased temperature by a few degrees with power draw increased by seven watts the next 100 megahertz jump required 0.04 four volts with another 100 Hertz on top of that going to 3.8 gigahertz now requiring 0.06 volts it's becoming clear how this is a nonlinear voltage requirement 3.9 gigahertz required an additional 0.075 volts on top of the 1.2 and 2 volts we had previously putting us at one point two eight seven volts we're also now at sixty four point six degrees Celsius and 144 watts of power consumption our final step to 4.0 gigahertz required zero point one one nine volt on top of the previous jump putting us at a total of zero point three three seven over the initial 3.5 gigahertz requirement of one point zero six nine moving on to a chart to help visualize this we now add the 2700 x data points at 3.5 gigahertz the 2700 exa needed just 0.9 volts to hold stable a significant reduction from the r7 1700 the 2700 acts required an increase in voltage to zero point nine five six to sustain 3.6 gigahertz or a jump of 0.05 volts and so on until we eventually hit 4.0 gigahertz at one point 160 volts for this particular test run that then required a jump of 0.08 volts to climb to four point one gigahertz and at this point we switched to a larger cooler as we were heading 70 degrees and up on TDI which was causing instability crashes and limitations in our overclocking for our long term blender burnin test this means that the X 62 and flow 360 data is not perfectly comparable and so is represented by a dotted line for the 360 you can see that the curve gets incredibly steep at 4.2 gigahertz where we had to increase from one point to four to one point three eight volts to hold stability with the larger cooler note that we could achieve this with lower voltage of tuning B CLK instead but that's one of our controls here we can't get into territory of really vertical increases because we're limited by temperature there Bauer included a graph in his 6000 megahertz 2700 x OC video showing voltages and frequencies he achieves without - reformatted to match our own graphs his data looks like this his voltages are lower because among other reasons he's using a higher level of LLC but the shape of the curve is close to our own right down to the sharp increase required at the end of the chart he went as high as 1.5 volts without surpassing the four point three gigahertz he was testing in increments of 0.025 volts and 0.025 gigahertz so we can extrapolate a best-case scenario of achieving four point three to five gigahertz at one point five to five volts on the coolant solution he used with the controls he used you can see our own voltage curve extrapolation is similar to his with both hitting a steep wall at the end that is basically impossible without sub ambient cooling again all core overclocking on the 2700 X is sort of unrewarding same thing for the 2600 X because that X demarcation at the end means that your base and boost out of the box are pushed up just high enough that you don't gain a ton from overclock in all core unless you go through and manually to an individual cores in which case you're doing a lot more work to bin your cores and you will get a bit more performance but is it really worth it it just depends on if you're doing it because you enjoy that process or if you're doing it for performance because if it's the latter it's a lot of work for the performance if it's the former have fun because that's the point anyway so it doesn't really matter what else it gets you but the bigger point here is that as stated previously Rison 2 is able to sustain same clocks with rise and 1 at a much lower voltage but as you push the frequency higher it's sort of exponential or at least a nonlinear increase in that you can go from zero point zero three zero point zero to increase in voltage to eventually zero point three increase in voltage and up which is a big difference so pretty interesting data from that standpoint a couple of notes on test methodology before closing out here I move this to the end just because people get bored of things like being accurate so for methodology our previous test used both prime and blender but in the interest of efficiency in these tests we tested stability by using our most stressful blender benchmark for about 10 minutes that's enough time to perform test it's not enough time to guarantee a stable OC for long up times like 24 hours whatever but strictly for academic purposes it's enough and we're not going to 24 hours stress test things like this because it just it would not be doable so a plenty for the circumstances we're looking at for the testing we're doing there Bauer his chart the numbers we use there are with Cinebench r15 so we have different approaches which means probably our test would be more likely to fail at a given voltage then his just because Cinebench is a bit lighter on the cpu then our blender benchmark which lasts much longer so good enough for the purpose of this in terms of other methodology information we used LLC one step down from the max level which is why in the tables especially in the article below you'll see the input voltage is different from the measured voltage at the CPU but that's fine we present both numbers bios settings and hardware we kept the same everything kept the same other than the CPU V Corps and all core multiplier which were the various variables being tested and the 4.1 gigahertz tests on the 2700 X we're done using again the Thermaltake 360 flow for 4.1 and up and that's as opposed to the two eighty millimeter kraken x 62 and these are marked with an asterisk in the tables and a dotted line of charts so this wasn't a test of how high the 2700 acts can be overclocked we can do higher than 4.2 barely but we were trying to see what could happen without adjusting things like reference clock because it's a test of mote frequency curve not a test of max OC performance and all core overclock in any way is is nearing bordering on pointless these days with how good acts of r2 is on its own so something to keep in mind for the x class CPUs or just by an on x1 and overclock it to an X and get the same performance for cheaper anyway that's the Volt frequency curve if you like this type of content as always go to patreon.com/scishow and exit stubs that directly go to store document access dotnet to pick up one of our mod mats on backorder subscribe for more i'll see you all next time
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.