just how bad is mixing memory

 (dramatic music) - Wait a second. Everyone knows it's bad tomix and match your memory. But how bad is it? I mean, if you found a really sweet deal to double up the RAM and your gaming rig, would you be such an evil personif you pulled the trigger? What if the RAM that's in your system isn't available anymore? Are you stuck with it forever? Let's find out. After we find out justhow smoothly Linus here is gonna segue to his sponsor. With GlassWire you caninstantly see your current and past network activity, detect malware, and block badly behaving appson your PC or Android device. Use offer code LINUS toget 25% off GlassWire at the link down below. (upbeat music) We never wanna feed youmisinformation or myths, particularly if it causes you to miss out on a free or cheap upgrade. So we took five differentmemory sets from our inventory, rated for various speeds and latencies, throw them on an AMD test bench with a Ryzen 5 3600X and2060 Super, and went to town, checking configuration after configuration after configuration after... Well, you get the point. And after many grueling hoursof mind-numbing benchmarks, we're here to give youthe exciting results. Guess what? It barely matters. Or at least, it sometimes barely matters. We created a baseline for reference using two eight gig modulesclocked at 3,200 megahertz with primary timings of 6-18-18-36. We ran our benchmarks, took notes, and then started swappingout modules one by one. And you might be as surprisedas we were to find out that, honestly, for the most part, with two sticks you canget away with quite a lot as long as you're willingto tweak some values here and there in the system BIOS. If you don't overclockyour memory subsystem. And I'm talking eitherwith XMP or manually, RAM compatibility withDDR4 has come a long way to the point where basicallyit just freakin' works. Faster modules will generallyhave no issues clocking down and running steady with the slower ones, especially if they'reunder four gigahertz. And our performancemeasurements demonstrate this with pretty much all of our configurations within the run-to-runvariants of our tests. But the enthusiasts among you will know that if you wanna get themost out of a modern system, increasing your memoryspeed is really important. What if we take thenhalf of our reference kit and half of a slower kitand run them in dual-channel at the reference kit'sfrequency of 3,200 megahertz? Our slow kit here is timedpretty similarly, 16-18-18-35. But the traditional rule of thumb is that for every speed bin bump, you would expect to alsoraise your CAS latency by one. I am really not sureif this is gonna work. Crazily enough, though, theresults look pretty good. Almost as good as our RAM styleTech Scarf on

LTTStore.com. And the benchmarks look identical. So what we've managed to showhere is that if you get lucky and you're mismatched RAMwill run at your target speed, there's unlikely to be anyperformance difference. But remember how I promisedwe'd find some differences? What happens if we up the ante, and mix our 2,666 with ourhigh-speed 4,000 megahertz kit? Well, the reason DDR4 4,000costs more than this stuff is that it's been sorted orbinned by the manufacturer according to their internalperformance targets. Now, sometimes you get lucky and this puppy'll overoverclock to the moon, but other times, you get ourlittle friend, Mr. 2,666 here. He managed to be compatiblewith the big, bad, extreme gaming modules, butnot at their rated speed, where the system failedto even power on properly. Things broke down even further when we tried to populatemore than two DIMM slots. The thing is the more modules you install the greater the load on your CPU's integrated memory controller. This didn't cause anyperformance issues per se, as long as we could getthe RAM all running, but it did cause some flakyand undesirable behavior. For example, when we populatedthree slots with a mixture of our reference and slow modules, the system would post and benchat 3,200 speeds just fine, at the start, but here's the thing. AMD's Ryzen processors havea feature called flex mode that allows two modulesto run in dual-channel, with a third running in single-channel. That's great for gettingmaximum performance out of a non-ideal configuration, but you've gotta be very careful how you populate your slots. As soon as we put our 2666module in the B1 slot, it got real unhappy, spittingout memory error codes and getting stuck in a restart loop until we reset the CMOS anddialed it all back to 2,666. As for populating all four DIMM slots and getting the system to work? That was no easy feat, with only a single configuration posting, and only when we toned it all down to meet at 3,200 megahertz. That was below our highest rated modules 3,400 megahertz speed. Funnily enough, though, we tried back to this configuration later, and despite using the same BIOS settings with the same sticks in the same slots, it actually failed to postproperly during round two. Which goes to show that while your friend might have no troublerunning a poop mix of memory with no performance penalty, you can't count on the same result. And here's the big problem with that. Two mixed modules was pretty much fine, but most people aren't mixing memory at the time that they buytheir brand new system. It's when they're upgrading down the line and trying to max out an older system, that these kinds of questions come up. So the chance that mixing and matching will result in unpredictability when you're maxing out yoursystem and filling all the slots is just too high to ignore. Speaking of unpredictable, get subscribed so you don'tmiss the next installment in our crazy Intel home upgrade series. Andy is doing a man-cave.(clicking tongue) Back to mixing memory though. Is there anything that youcan do to improve your odds? Well, yeah, actually. For starters, buying an upgrade that is rated for the samespeed and primary timings will help your chances. And so will finding amodule that was built using dies from the same manufacturer. And when I say manufacturer,I'm not talking G.Skill or Corsair or Kingston, I'mtalking about the actual chips that are soldered to the module. So if you've got a modulehere with Samsung B-dies, good for you, by the way, those are nice. Try to get more of the same, even if they have a differentlogo on the heat spreader. "But, Linus," you might ask, "what about different capacities? "Can I pair up similarly specced "eight gig and 16 gig modules "and finally join team dual-channel?" Yes, you can. It's called asynchronous dual-channel. So we also ran benchmarks withdifferent capacity modules of similar speeds and not onlydid it all register properly but as long as your workloaddoesn't exceed the capacity of the smallest module you shouldn't noticeany performance impact With that said, we still cannot in goodconscience recommend going out, buying whatever set of modules is on sale, rubbing them together for good luck, and throwing them at your rig. Both the best and theworst thing about PC gaming is that every system is wildlydifferent from the next. To the point where whatworks for your friend, or for us today, might not work for you. You might hit the silicon lottery and get something thatoverclocks just fine and has perfect compatibility, or you might get something that runs at exactly its rated speedand not a hertz more! Worse yet, you might landyourself in a restart loop or even run intounpredictable blue screens on a rig that otherwiseseems to be working fine. So we would strongly recommend if you're upgrading your memory flip your current moduleson Facebook Marketplace or eBay or something andbuy matching replacements. If however, you've got thespare parts lying around and you wanna see if you cangive yourself an easy upgrade with minimal or no performance loss, and you're not afraid of a little bit of fine-tuning in theevent of some crashing, the good news is that it can work. So please do some of your own testing and let us know in the comments below just how far you were able topush your mismatched hardware. Who knows, maybe you've had access to a free upgrade this whole time. Just like I upgraded my segueto this one from our sponsor, Micro Center. Get the best prices and bestselection on PC hardware, and anything else techat one of Micro Center's 25 locations across the United States. Check out Micro Center'sDIY arcade cabinet kits. Each kit includes the arcade cabinet, a 27-inch monitor, two joysticks, 20 arcade buttons withswitches, a Raspberry PI, Atari-themed graphics, and all the components needed to connect and power the system. The included Atari games SD card comes with over 140 licensed Atari games from their systems, theAtari 2600, Atari 5600, and Atari 7800. I mean, even back then they had console generationalnames figured out. They add 2,600 to every one.Why can't Microsoft do this? Anyway, the point is, ifyou wanna get more creative you can choose your own components with the Micro CenterRetro Arcade Builder. So click the link below and build your perfectarcade cabinet today. If you guys enjoy these kindsof deep dives into memory, check out our previous look at the performance impact of ECC memory. It is slower, but it might help youkeep memory errors at bay. You can check out findingson that sweet stuff at the link down below Or I think we've probably gota thing here. (clicks tongue)