Gadgetory

this video is a recap of an article that Patrick posted a few days ago this is basically a research piece so mainly we're looking at some quick and dirty testing on Ryze and CPUs we normally use these types of tests for internal reasons like planning and test execution don't really publish them because they're just used to fuel the actual testing that does get published that said Rison and the Zen architecture are still young so we're going to be publishing this data just to kind of get some more information out there there's not really a huge conclusion at the end of this it's just data for the sake of data and learning how Rison works before getting to that this coverage is brought to you by Thermaltake and their contact silent 12 which is a $25 air cooler that supports am for mounting you can find a link in the description below if you are looking for a quiet and affordable option for cooling your cpu so the tests included here if you've already read the article I'll save you the time it's the same information if not stay tuned because we're looking at things like balanced versus performance and the impact of those power settings within Windows on frame rate so that would be the high performance versus the balanced state I'm sure you've heard about that for rise and testing we'll be looking at that impact on frame rate we're also looking at boosting and XF our performance in those power modes and how they change functionality based on which power mode it's in we'll be looking at the impact of EFI versions on fps and the impact of two different motherboards on fps ultimately we found that our decision with the Asus board to use five seven zero four which at the time was the newest bios anyway was the best at the time for benchmarking given our options but this does expand some understanding of how the gigabyte board performs and that board receives a few days later so haven't had nearly as much time with it but we've had enough time to run a few tests of the review week EFI options that were out there on the gigabyte X 370 gaming gaming 5 and the Asus crosshair 6 hero board which is also next 370 board let's start with the efi and motherboard differences mind again that this is entirely a research piece these numbers are not fully inclusive of all performance expectations on all tested platforms that's why we normally keep this internal but the stuff we're looking at today provides a better foundation to help everyone including our own team better understand Ryan's behavior just before we publish the article we received efi updates from both aces & gigabyte the test and was already done so neither were included in this piece because the pace is so fast right now once things slowed down and the motherboard manufacturers stopped releasing basically beta updates will visit the topic again with the newest efi whatever it is at a time for now the purpose is to look at initial performance during review time with the review time efi versions we use the HTS crosshair for our reviews with efi at five seven zero four for all initial testing at the time that was the newest despite AMD at shipping the motherboard with an older efi that was built for unn finalized microcode on Rison we got this version by working with asus directly if anyone tested with the version that AMD shipped it was an older version of efi that could have impacted performance since then asus has released another efi beta version that makes some do CP tweaks we haven't yet tested that version but we'll grab the next stable release anyway here's a look at the frame rates when using the a cs5 704 efi provided during launch and the gigabyte F 3 and F 3 n versions of efi that were available during review week we now also have a version f5 but again we're waiting for the next Rev to test the current plan is to skip a revert to given the high frequency of releases and look at the other stuff with watchdogs - we're seeing the five 7:04 asus board marginally edged out the gigabyte f3 and efi version where we move from 84.3 280 3.7 FPS average and 80 2.3 FPS average for the other efi version 5 7:04 it generally seems superior here when looking at averages and lows but the difference is negligible overall certainly not something you would see but definitely something that's measurable and that is also important to include because that's what we see on the charts this is not a huge jump in performance board to board when looking at the release weak efi versions we need to include msi for a fuller picture but haven't looked at those boards yet we have heard of bigger performance changes though between MSI's initial EFI revs and the stuff that came out closer to launch despite these sort of boring results it doesn't mean that the EFI version is useless we've heard from some folks that they got stuck on twenty six sixty six to twenty eight hundred megahertz on the gigabyte motherboard and know that the later EFI revisions theoretically helped improve this though our particular configuration seems permitting of at least twenty nine thirty three that's not true for everyone this is partly due to differences in Sam's on a rev and Hynix dies on the memory modules the performance differences between the gigabyte and asus boards were up-and-down but overall they reinforced our than only choice of the Asus board for initial testing from a 16 gigabyte ashes of the singularity has about a 1.2 percent drop watchdog's to 0.7 percent and total war Warhammer decreased about 1.5 percent in average FPS on the gigabyte board and the favorite Cinebench dropped 5.7 percent in multi-threaded performance on gigabytes board after averaging 6 passes for the synthetic benchmark normal variants in tests can account for some of the differences but the logical conclusion is that differences in hardware and software can cause repeatable differences in benchmark results this is good news for consumers if there's repeatable variance or unreliability between platforms right now in our tests it would imply that there's some room for optimization and improvement this also means that our test results should only be 100% directly compared against our other test results because benchmarks on other platforms will show different outcomes even using the same CPU memory support has a tremendous impact on this from what we've read and so testers or users incapable of exceeding 20 666 megahertz may be more affected by that quirk than anything else grabbing relevant efi updates for memory support should help the mr guard though and most the vendors have pushed updates for do CP or XMP equivalents now that said keep in mind that gains are limited by the confines of reality while we can anticipate improvement in performance via efi or motherboard changes these initial results that do indicate that they are limited at least for these initial efi versions expect nothing and you'll be best off performance gains it won't be enough to get the 1700 7700 K levels in gaming performance for instance especially not with the 1080p testing that we encourage there will likely be improvement but not on scales that large so keep the expectations checked or just wait for more benchmarks before forming honey let's move on to a stopgap measure to increase performance in Windows and that would be changing from the default balanced power mode to high performance mode this is one that was recommended by AMD though we did test with high performance throughout all of our reviews before they even contacted us so that was covered on our end but we wanted to look back and see what actually changes if he's do switch to the not recommended balanced mode a few things to note here first of all there can be some FPS difference we'll go through that but also performance is not really ideal because it does force the clock to basically run at full tilt the whole time so not great for power it is wasteful in power it leaves less room for voltage and frequency modulation for example in XFR when you're using one core one thread something like that some really light tasks in terms of threading XFR is less capable to hit and sustain those higher frequencies in these short periods of time provided when running already at basically maximum voltage we'll look at that through these results for some quick numbers Metro last light shows no real difference between average frame rates with the configurations tested we see both at around 124 FPS average and that's within range of variants the 1% and 0.1% low values are slightly favored in performance mode but that's more so on 0.1% then on the 1% side moving the watchdog's 2 we see that the power modes have almost no impact on results performance mode is marginally ahead and averages and lows though that's insignificant and the two are effectively tied we saw noteworthy differences emerge in battlefield 1 with DirectX 11 where we move from 117 about 3 FPS average to 130 132 FPS average this is finally a significant we also noticed higher low-end performance with this mode less significant in percentage gains than the averages but still noteworthy overall GTA 5 also opposed today gain with performance mode moving us from one 17.7 average to 120 4.5 fps and that's a change of about five to six percent improvement again our reviews contained all performance mode numbers so this changes nothing with regard to conclusions on the 1700 and 1800 acts as they stand today it does provide some information as to how balanced mode could affect results if you're using Windows 10 and its default power mode and you wanted to know if it's worth disabling or going to high performance instead next we're moving on to XFR and frequency behavior under the two different power states again balanced and performance same thing here that's our a/b for right now we can learn about how the r7 1700 in this instance responds to the different power modes we're really just using Cinebench and pov-ray for these because synthetics provide a really good baseline we just need a foundational understanding of what's going on nothing the crazy and depth here so the point is to gain that through repeatable synthetics that have multi-threaded and single threaded options Cinebench and pov-ray provide that quick note before we start patrick discovered a bug in a hardware monitor when we were working on this so the built in chart generation with hardware monitor means that the peaks and the lows in the charts that are pre generated if you use those that supplies to you will be inaccurate the way it works the way the bug works is hardware monitor captures the frequencies every 0.5 seconds and when it does so it's trying to do some predictions so it might see let's say 37 50 megahertz three times in a row if it does you'll get a plateau so it hits 3750 that flattens and then continues however it may continue that is the max if it captures 3750 once or twice Hardware monitor will spike so it'll jump up higher as if it's trying to predict it because there's no third data point and that number is inaccurate so if you see in our charts I've kind of really hastily fixed that by using just a photoshop Paint Bucket basically that flattens things out but the CSV itself is accurate so if you look through the CSV on the same data point you'll see that the data point that's being extrapolated does not exist in the CSV but the graph generation does something funny so don't trust that Foley it's still fairly accurate overall just that'll give you the wrong impression this also applies to the low-end so anything that dips below the orange line at the very bottom that's not real that's just an error of hardware monitor one more thing the y-axis scale is different between charts not a fan of that but again these are quick tests and we're using a harder monitor to generate the graphs just to give a better understanding that's more Universal overall reisen's frequency adjustment features are explained in detail in a post on the ANA tech forums but the gist of it is this for each rise in CPU there is a base frequency across all cores a boost frequency across all cores a base frequency for a single core and an xfr frequency for a single core for the 1800 X these frequencies are our 3.6 3.7 4.0 and 4.1 respectively empty does not advertise that XF our number on its boxes so you'll see 4.0 instead of 4.1 for example and our power draw testing we found that the 1800 X would keep all cores at 3.7 gigahertz went under 100% load but would boost one core to four point one gigahertz in single threaded synthetic test so that should give you an idea of how it actually works this chart shows the r7 1,700 non x in a Cinebench one thread benchmark with high performance of mode selected in Windows we smoothed out those erroneous outputs by hardware monitor at the top with a quick Photoshop Pass and note that we're seeing the cores that take turns boosting up to X if our frequencies then falling as another core takes over let's get the r7 1700 Cinebench multi-threaded chart onto the screen to provide some perspective when all threads are engaged this is also in high performance mode as seen in the graphs on the screen now the r7 1700 running Cinebench and high performance mode all cores engaged remains at 3.2 gigahertz that's the multi thread limit when everything's under a hundred percent load in single threaded tests individual cores a boost up to three point seven five gigahertz the peaks above this again that we've smoothed out are due to harder monitor graphing software issues so our seven seventeen hundreds at non axe demarkation contributes to its lower XFR range if that was the curiosity to you and of course this also means that overclocking decreases your single-threaded performance provided the OC is lower than the our frequency that's really only likely to be true for the 18-yard ex because it's easy to OCD past XF are on seventeen hundred and seventeen hundred x CPUs this also relates to high performance mode in Windows which sets the minimum clock speed to the current base clock of the processor and in the process of doing so reduces latency as the CPU is forced to ramp in to workloads by minimizing the frequency modulation and power saving functions performance ensures that the rise in CPUs are working at full capacity during workloads or as close to it as permissible given the platform and OS limitations that may exist here's the next chart this one shows the 1700 score frequency during pov-ray single-threaded benchmarks still using high performance mode in Windows again ignore those files at the bottom of the chart those are from the hardware monitor bug we're seeing all cores take turns of boosting to higher frequencies during this test switching to balanced mode for which we have another chart changes things a bit will pop that onto the screen now suddenly we're looking at just two of the core is performing the boosting with the rest being unutilized in balance mode of course four to seven set at 1.5 gigahertz constantly cores a three and two alternate between 1.5 and 1.4 to 2 core 1 ranged anywhere from 1.5 to 3.75 and core zero maintained a relatively constant 3.2 it's a 3.75 gigahertz there's definitely a visible effect from windows power settings and boosting in balanced mode it seems to prefer one or two cores rather than bouncing between them constantly some testing in Cinebench suggests a mild single-threaded performance advantage in balanced mode if anything at all and we see a 145 that to 5 balance first 140 7.5 average performance in CB marks that's after about half a dozen passes or so but it seems that the end result isn't much affected by windows power settings for single threaded as discussed in our 1802 extra view and just a moment ago high performance mode it does seem to have a positive impact generally on gaming or none at all depending on the multi-threading and how the game is behaved high performance mode isn't great for power consumption especially for overclocked CPUs that will be forced to constantly run out there see frequency so that's it for this one no big conclusion or anything just data and research to recap the ACS 5 7:04 efi version we use for our review testing based on these results was the right one to use also the only choice you really had at the time gigabyte doesn't show a whole lot of change between f3 and f3n unless you were memory limited to begin with f5 exists we might look at it we might wait for def 7 or whatever is next and asus has new versions basically efi versions are rolling out almost every day to reviewers and especially internally so we are going to be kind of waiting and following a cadence that's every other every couple efi versions other than that there's that pretty much sums up all of our research so hopefully that helps you understand some of the things better that we've looked at over the past week or so power versus balance versus high-performance States was one of the more interesting things to look at in terms of performance with the games and synthetic tests full articles in the description below if you want to learn more about that and find these charts in a more readable format if you prefer subscribe for more patreon.com slash gamers Nexus if you'd like to help us out directly thank you for watching I'll see you all next time you