Gadgetory

in some ways AMD has become Nvidia and it's not necessarily a bad thing in this instance the way the new rise in CPU scale is behaviorally similar to the way GPU boost 4.0 scales on GPUs we're simply lowering the silicon operating temperature will directly affect performance and clock speeds under complete full stock settings a CPU running colder will actually boost higher now alternatively if you're a glass half-empty type you could view it as the CPU running hotter causing thermal throttling either way frequency is contingent upon thermals and that's important for users who want to maximize performance or pick the right case and CPU cooling combination if you're new to the space the way it has traditionally worked is just CPUs will run at one spec with one set of frequencies until hitting t.j.maxx or maximum Junction temperature rise in 3000 is significantly different from past CPUs in this regard and we're going to explore this today with some additional work that we did from our livestream where we overclocked the 3900 acts before that this video is brought to you by us and the gamers Nexus toolkit on store cameras axis net our brand new toolkit just launched and contains 10 custom eight drivers for video card disassembly reap hasting and tear downs the eight core tools are made of high-quality chromium vanadium alloy steel that's built for long service life and resistance to wear during use the other two tools are carbon steel hex heads that were custom ground down for capacitor clearance on video cards all the tools are easily mounted to a pegboard or stored in the GN made tool bag for easy transport learn more at the link in the description below an Intel CPU is probably the easiest foil to rise in 3000 series CPUs with Intel you have really only two parameters to consider there's the turbo boost duration which we have a whole separate content piece on including MCE discussion and then there's the power limitations and beyond that thermal is only a factor once you get up to t.j.maxx for the most part so the way this works is if operating within spec outside of the turbo duration limit which is 90 to 120 seconds on average depending on what the motherboard vendor has done on there and the CPU will stick to one all core speed for the entirety of its workload and there's no modulation within that speed so you could be running the Intel CPU at 40 degrees you could be running it at 90 degrees and as long as you're not at t.j.maxx you'll get one frequency out of it whatever they define for the most part in their in their turbo boosting tables it'll be the same frequency so once you hit t.j.maxx say it's maybe 95 or a hundred degrees Celsius you either get thermal throttling on the multiplier to get multiplier throttling or you get a thermal shutdown and the choice between a two will very depend on which motherboard you've chosen for your platform AMD Rison by contrast behaves more like modern GPUs the positive way to look at this is pretty simple out-of-the-box the companies are now maximizing performance to the best of their thermal power and current conditions with granular steps in frequency along the Volt frequency curve this is also using thermal as a gauge for where on the curve the silicon should operate given the current condition this means that there's less overclocking Headroom it means that as a testing environment it's a lot harder to get like-for-like benchmarks because the companies no longer have to tune for a worst-case scenario but that also means that the out of box performance is nearing the maximum reasonable performance achievable by manual overclocking so primarily this is mostly a good thing because just like with GPU boost the horizon CPUs are boosting to their maximum thermal capabilities as opposed to the alternative which would be AMD setting a lower boost frequency to account for the worst-case scenarios and then giving the user more overclocking Headroom so you get one or the other but ultimately the out of box performance does matter quite a bit because most people don't overclock factually that is that is the way it is so that's the way to look at it the best way to demonstrate how frequency scales with thermal which is what we're doing today we just have one chart it's very simple is to simply run a few tests we're using the gigabyte x5 Sony master motherboard for this and we're running a CPU stock so it's the rise in line 3900 X it's a 12 core CPU keep in mind that as you increase frequency across all cores it has significantly more impact than if you increase frequency across one core so if you run a CPU with let's say frequency that's maybe 150 mega Hertz higher all core in blender versus 150 mega Hertz lower option the difference will emerge there more you'll have a more of a percent change than you will if you're running a more single-threaded workload like a game or like Cinebench one threaded something like that where it's it's just an increase on a single-core further what we are doing running stock is allowing precision boost to to do its thing we have a separate video with excruciating detail on what precision boost means and how it's different from precision boost overdrive or PBO you may have seen those letters lately and also how it's different from auto se which is not part of the president boost feature set or the precision boost overdrive feature set it's in the same menu but it's a different thing so to be extremely clear here precision boost overdrive or PBO is explicitly different from precision boost - and the reason we're pointing that out is because we're using just stock full stock settings zero changes all we've done is run the CPU more or less out of the box and we're gonna go from there so there's no PBO enabled here it's just the native precision boost - which is considered stock and if you're uncertain about any of this all this is also in the AMD review guide where they say to disable PBO for stock operation so hopefully that's fairly clear all PBO does is bypass current limits so it bypasses ppt it bypasses TDC and it bypasses EDC and all of those are in the separate video if you don't know what those are so all that said we're controlling temperatures within a range of about 84 degrees Celsius T die or the die temperature of the CPU and down to minus 80 degrees for the 40 pot for the Alan to pilot temperature and this is also there's no mortise ETL so we don't have to worry about that here it's just straight T die until we get sub 0 and then the reading bugs out and we have to go from a thermocouple that we've attached separately so let's get into it the chart coming up will show frequency and Cinebench score versus temperature it's a very straightforward test we're going to be manually tuning the TDI temperature on the cpu for each test pass so the cooling solution here is 100% irrelevant it doesn't matter what we use to cool it as long as we control the temperature the CPU doesn't know what's on top of it it doesn't know if there's a liquid cooler or an air cooler we're now onto pot all it knows is the temperature it's operating that so this allows us to see the range of performance under various cooling conditions it'll allow us to demonstrate the actual real-world impact of good case and cooler combinations on your C you and we'll start with more of a real-world warm scenario then go to not real world - sub-zero and just see if the scaling continues the chart starts at about 84 degrees Celsius which is where you might be sitting with the 3900 X and a stock cooler with the average modern case in a room that is lightly air-conditioned at 84 degrees we measured frequency between 3975 megahertz and 4000 megahertz sticking closer on average to let's call it 39 75 maybe 3980 once he averaged it out on the 12 core CPU as we manually dialed our temperature to reach 78 degrees we averaged 40 50 megahertz all core frequency and 71 degrees Celsius for TDI put us around 40 75 mega Hertz all core for reference with a step down to 61 degrees a 10 degree drop improving all core frequency notably to 41 50 megahertz and remember this is just in a bench here but the the scaling should continue and pretty much any application based on thermal the next step is 255 degrees where we saw an improvement 240 200 megahertz with a bit of averaging in there becomes more of maybe 40 190 or so 40 185 about 4200 megahertz after multiple runs up that sustained temperature this is not delta T over ambient as the only important thing here is the actual operating temperature ambient for what it's worth was 21 degrees but we're showing CP temperature anyways that's not relevant at the moment what is relevant is that an ambient temperature of 21 to 25 degrees might be common for an air-conditioned house depending on where you live and how you like to keep it cooled and adjust as needed if if you do keep it warmer case ambient is often approaching a range of 30 degrees even on a good case and a 21 degree celsius environment with our test systems we used for case reviews so a 55 degree low temperature is achievable primarily with high-end cooling solutions and well ventilated cases it's doable you can actually achieve this if you try but this is the end of our common real-world scenarios where as the 78 to 84 range would represent a stock cooler load condition with a decent case we can go further we can go lower than that and see what it looks like later so down to 36 degrees we see scaling continue to 40 to 25 megahertz again this is across all cores and at 18 degrees we average roughly 40 to 60 with individual cores bouncing around a bit more than we saw previously event we stopped scaling at minus 80 degrees Allen to pot temperatures are a teapot so the CPU is somewhere between 0 and minus 80 we don't know exactly where because we've lost the sensors at this point so the all core frequency at minus AE teapot is 40 300 megahertz all core fixed never change him that's really good so cold bugs were encountered after this we hit them about minus 81 to minus 85 as for CB marks those scale relatively linearly with frequency the range from 84 degrees TDI to 55 degrees TDI which are the max and men of the reasonable temperatures a user might encounter from not spending a lot on cooling to being try hard on cooling that range is about four percent performance increase just from a lower CPU temperature on the CPU zero changes we're not overclocking this is not PBO this is stock the same thing would happen if you did this on a GPU like an NVIDIA GPU so this is why cooler and case selection matter a lot with Rison and why we as a review outlet have to be careful to maintain a fixed cooling condition for testing in a fixed environment we run AC at the same temperature the whole time for all of our CPU tests and and all that impacts the frequency of the scores so the maximum scale not that it's particularly relevant to users is about 6.4 percent from 31 20 points to 33 19 points scaling from minus 80 to 84 degrees positive over zero Celsius we're basically at GPU boost behavior on the CPU at this point in this regard we can use liquid nitrogen as a really neat utility for carefully controlling the temperature rather than dumping L onto into the pot we can run fairly dry and set temperatures to whatever value we want them to be manual application of allen tool allows us to keep fairly consistent temperatures with a range of roughly plus or minus 2 degrees celsius from the test begin to test and the segments a lot of people will probably overlook the fact that while l n2 is indeed cold the cpu is only as cold as we allow it to run with minimal allen to use and just a cold allen to pot we can get the cpu to run at 84 degrees which is really hot for Allen to as a solution but obviously it depends on what you modulate it and we can operate under the same thermal conditions as a stock cooler in a warm case so this allows us to test with greater granularity and accuracy as we can dial the temperature to whatever we want rather than being limited by coolers available and having to swap them out all the time so from here you can extrapolate the thermal data to know how hard you want to brute-force your cooling but colder is definitely better with Ryze and there's obviously a point of diminishing returns but you really can't go too crazy with cooling for this chip we saw scaling all the way to Sub Zero which you're you're not gonna get that I mean most people are not running chillers maybe a pipe and some air from outside in the winter or something but on average most people are obviously running at or above ambient in the best case is maybe at with a low workload or something so yeah point is if if you're you've got some money and the only thing holding you back between buying some crazy tricked-out cooling solution and going to step down is you're not sure if it's worth it it as long as the cooling solution has a difference in temperature in thermal performance it will produce a difference it'll it'll create a different result is it worth it well obviously that starts coming down to the money argument and that's gonna depend on how much money you have to spend on your computer a lot of the time you could take the extra 50 bucks that you might spend on one class better cooler and sink it into a GPU upgrade and going from well going from like a this is low-end but 5757 rxc that's $50 that performance gap will be bigger than the performance increases heat from better cooling but if you if you don't really have much of a budget better cooling solutions certainly aren't going to hurt it's not like Intel CPUs where once you once you kind of have it controlled there's no benefit to going better other than reducing power leakage and for reducing power leakage you get about a 4% reduction in power consumption by the chip for every 10 degrees Celsius or so that you drop the core temperature but is that a value well not really so anyway pretty cool stuff literally in some cases what we need to do next is obviously look at thermal solutions individually and subscribe to make sure you can catch that content this was filmed and I edited this one so our editors didn't get involved they've got another break and I will be flying to California for an EVGA charity livestream where I'm battling with Jay Jay's two cents with kingpin as the referee for an Ellen to overclocking thing so if you want to catch that check on Tuesday we're gonna try and stream that otherwise it'll be on their channel as well so yeah point is we'll look at more thermal solutions once back but really interesting data how useful is it well you see a scaling of about four ish percent under reasonable conditions so it does that matter that'll depend on you but we wanted to present it and show that Rison actually really cares so when we go on rants for 20 minutes about how much a case is terrible at cooling this is the situation where it's going to matter a lot it also means that we can't really use the rise and chips easily in a case test bench because the frequency will scale so you have to start plotting frequency plots in addition to the temperature plots and the temperature plots aren't exactly accurate because your frequency is changing so it's no longer like for like so anyway yeah pick a good case make it cooler if you can afford it and don't feel too bad if you can't you still get reasonable performance it's just that obviously it's a bit better with colder so that's it for this one thank you for watching subscribe for more you can go to store documents nexus dotnet to helps out directly like by picking up one of these toolkits that we have or by picking up one of our mod mats or shirts if you want something a bit cheaper and that all helps fund things like doing the research we've done today and buying more of this stuff so I'll see you all next time