Gadgetory

for today we're talking about volt frequency scalability on our 8086 kay one last time this time coverage includes a manual bending of our cores as we already Illustrated limitations of the IMC in our overclocking livestream the major takeaway with these bintu cpus is that we can sustain a lower voltage and therefore lower power and thermals at a given frequency but there's a bit more to it than that before that this video is brought to you by Dollar Shave club's daily essentials starter set this full grooming kit ships at $5 and includes a trial sized versions of the companies at popular shave butter body wash and wipes and also includes a solid high-quality razor and a full set of cartridges go to dollarshaveclub.com slash gamers Nexus to get the daily essentials starter kit for just $5 and with free shipping or click the link below it's quick overview here we already have a stream recap if you're interested in that go check it out or check it out when it goes live in a day or so after this we also have a video talking about the thermals the power and the noise considerations of the 8086 k what we're doing now is looking specifically at the volt frequency binning so at what voltage is a frequency stable or at what voltage is a frequency incapable of being held stable for whatever the benchmark is so the two applications we're using today are just blender and fire strike and the thing with these applications is blender does AVX work loads so it is somewhat comparable to prime95 a really popular stress testing program except with prime95 you can adjust the FFT size now with smaller F of T sizes like 8 by 8 it's extremely abusive on the core and it's great for finding a points of failure and an overclock but you could also just do the most intensive application you're likely to use in daily use and that might be something more like blender which is still intense is still ABX but it's not nearly as stressful for the CPU as 8k F of T's and probably 95 and ultimately if you're trying to really make sure it's never going to crash yes prime95 is a great way to do that let it run for a while see what happens but what we're doing is keeping it realistic today so blender and fire strike are good fire strike does not require nearly the voltage for stability that's something like blender or prime required because it's more of a gaming application it's a synthetic game functionally so fire strike service is a great stand-in for what kind of voltage we require at a given frequency for the 8086 k when gaming what we're also going to be doing finally is looking at just some rough scalability at stock at a 50 X multiplier 51 52 and 53 and then also we have a 5.3 5 gigahertz clock that we achieved in the stream so that will give us just a guideline for how much performance you can expect for each of those multipliers when overclocking something like a high-end Intel CPU so last note here before we get into this keep in mind that the voltage requirement to keep a given application stable changes based upon the application and what kind of workload it creates on the CPU also keep in mind that this voltage if it goes higher to sustain a frequency in an application is going to require more power it's going to output more heat refer to our previous 8086 K video for that it's a great study on how the processor thermal and heat load behaves and how it can impact your noise levels to a factor of 2x when you are dealing with those higher voltages to sustain higher frequencies and eventually you want to meet a middle ground where you get a decent volt frequency table without being too abusive on the CPU or on your noise levels so we'll start with blender today this is a great example for just bending the Vote frequency in an application that actually is something we can use so it's not like Prime in that way keeps things simple for blender all of these our requirement is stable for 30 minutes for the most part if a overclock is going to fail it will fail on that 30 minute window almost always now if we were selling these CPUs like a company like Silicon Lottery does we'd want to run it for a whole lot longer than that probably on a an order of hours but for realism sake in terms of completing these tests we chose 30 minutes because we know 99% of the time if it doesn't fail in that window it won't fail so keep keep that in mind either way start with blender we found that in blender 5.1 gigahertz at one point three five volts was stable whereas as you'll see in the fire strike table in a minute we were able to hold 5.1 gigahertz with a slightly lower voltage depending on what we were working with four parameters we have a couple of other blender numbers here for volt frequency requirements and you can see where we finally stabilized and stops for this one fire strike is where we spent the most time low so let's move on to this fire strike right away for fire strike this is a lot less intensive than AVX workloads like blender prime but more important for competitive benchmarking purposes and fire strike most impressively we were able to sustain 5.0 gigahertz at 1.30 volts which was sustainable during our live overclocking stream as well we later verified a stable prolonged period of fire strike physics testing with the same configuration so it did survive for 30 minutes and this is functionally a stand-in for gaming it illustrates just how well bend the 8700 ka CPU performs and that's ultimately what an ad 86 is our 8086 K holds 5.0 at 1.30 to 1.35 depending for a fire strike whereas the 8700 K s holds 5.0 gigahertz at one point four zero to one point four five in this case depending on which CPU it is that's the bigger takeaway here at a given clock we've reduced our voltage requirement significantly five point one gigahertz also held at 1.30 but only for long enough to run the 30 second test it did not survive the longer burnin we next found a 53 X multiplier to be stable at about 1.3 9 volts which is impressive when considering at most 8700 KS could never reach 5.3 gigahertz even with high-end cooling you need exotic cooling for that that's even more true now though because most of those higher-end 8700 KS would now be removed to become 8086 KS we were unable to stabilize a 54 multiplier under any circumstances including per core overclocking we were however able to sustain a 101 BC LK shown to the left of most column with a 53 X multiplier and 1.41 volt this gave us a 53 50 megahertz core clock at one point for one volts pretty impressive for purposes of fire strike physics it would appear that the optimal performance the thermal and power configuration would be roughly 5.1 at one point three to two 1.35 volts with our specific CPU and as always this will change based upon the individual CPUs none of them are made equal you'll also see an LLC column here in our notes the ACE use of Maximus export it seems to keep a fairly flat voltage provision for V core when at LLC level six which is what we want going to LLC level seven was sometimes necessary but it tended to over provide voltage when spurious delivery became problematic so you'd end up jumping and voltage in a way that could be damaging if you're using your CPU under a 24-7 overclock with the voltage you think is lower than is reality moving on finally to scores for relative scores and scaling we found the following results across our 8086 K frequency testing firestrike spat out an eighteen thousand and seven score for the complete baseline test that's stock with only XMP enabled clocking trivially the 5.0 gigahertz in one point three zero volts got us to twenty thousand nine hundred points an increase of sixteen percent with about thirty seconds of work from a seven hundred megahertz or so bump in the clock we next observed only a one to two percent increase in score by overclock in an additional one hundred megahertz up to five point one gigahertz our five point two gigahertz overclock at one point three five volts yielded an increase to twenty one thousand seven seventy three points an increase of 21% over the baseline of eighteen thousand from earlier at five point three gigahertz in one point four volts we observed an increase to twenty one thousand ninety two points hitting diminishing returns now another marginal hike over the previous hike to five point two gigahertz each 100 megahertz increase is at this point hitting diminishing returns and we're not individually gaining a lot but it's still measurable and it's repeatable by jumping to a b c lk of 101 which increased the memory frequency and also impacts other things in the system we ended up at twenty three thousand to twenty three points a big jump but a lot of that is from the bc lk increase and therefore the frequency increase and just general responsiveness in the system going to be c lk 101 with thirty six hundred megahertz memory at CL 15s and now we've improved a lot of things at once including an uncor to fifty two we had a final result of twenty three thousand seven forty five points so going back to the discussion then our CPU bend out to basically have definitely stable 52 multiplier for the uncor which is great we haven't tried higher yet we are definitely say about 53 50 megahertz for the core and then for the rest of this scaling once you get into the five gigahertz range beyond that there's not a huge impact for fire strike physics and that will correlate actually better than most games well if you compare fire strike physics to games the correlation is semi linear to a point and the point at which it stops being linear is when you start adding more cores because fire strike is a lot more responsive to more cores and threads than most games actually are but because the core count on the 80 780 86 remains somewhat reasonable for a game to use the scaling is fairly comparable and from that we can draw a rough conclusion and we'll look into this more later if we need to but it's an 8700 K all over again so same idea Jarre rough conclusion that at 5.0 gigahertz if you can keep your voltage and therefore power and therefore thermals way lower it might be worth just running out that in which case you could buy an 8700 K roll the dice and hope that it's a 5 gigahertz chip and a lot of them are but the ones that are 4.9 you're still not losing a ton going up to 5.3 is fun it's good in terms of having something to do and telling all your friends you got the best one but in terms of performance unless you're competitive about it like fire strike overclocking competitively or something like that you're not gonna gain a ton of extra framerate out of another hundred megahertz or 200 Hertz or whatever it is but when you're talking 400 500 700 like we're doing here in some cases that's where it definitely becomes noticeable this of course also reduces render times and applications like blender but not to the same extreme that increasing the thread count would do because it is a more thread intensive application in that instance so that's it for our 8086 k bin that's what we ended up with that's how our CPU performed we actually just got another one in that we paid for for another purpose a build that we'd be talking about in about a week or so hopefully so stay tuned for that subscribe for that content and other content as always been working hard on getting back to a bunch of charts and data because it's been a little while without all the launches and with computation traveling the way so getting back into it had a lot of charts for this week's filming I hope you enjoyed it and you can support us on patreon Thomas lash gamers Nexus go to store doc guarantees access net to pick up this freshly restocked teal tear down logo shirt if you want to pick that up this one's been in high demand since we put it out about a year ago so grab it there and otherwise subscribe for more how see you all next time