I tried to prove YOU guys wrong... but it was ME who was wrong!
HTML-код
- Опубликовано: 6 сен 2024
- Learn more about Redux PCs at
Get your JayzTwoCents Merch Here! - www.jayztwocen...
○○○○○○ Items featured in this video available at Amazon ○○○○○○
► Amazon US - bit.ly/1meybOF
► Amazon UK - amzn.to/Zx813L
► Amazon Canada - amzn.to/1tl6vc6
••• Follow me on your favorite Social Media! •••
Facebook: / jayztwocents
Twitter: / jayztwocents
Instagram: / jayztwocents
SUBSCRIBE! bit.ly/sub2Jayz...
Videos like these are WHY I watch this channel. "Screw it, lets see if it works" is basically the origin of scientific practice. This is the kind of argument tech nerds have been having for years, and most people don't have the hardware stash to actually test this kind of thing out properly. I love this kind of "lets find out together" approach, and appreciate that the "well that isn't how that was supposed to work" moments don't end up cut out and only in some private bloopers group chat or something.
Agree
Great comment! Totally agree. These episodes are also very entertaining. More videos like these, please!
"the difference between science and screwing around is writing it down"
in this case, documenting it in a video form
The "well that isn't how that was supposed to work" moments
is the only reason to watch most of these types of videos
oh and the memes ofc
Great video and great comment @Samebito. I totally agree. I also remember watching a few months ago a similar video, about using mismatched memory on LTT. However, I think Jay'z approach is more in depth and more explanatory . Very happy he did a follow up video.
"look at it, that's what peak performance looks like!"
I died
Great line lol
Yes that was the best part of the video...
Actually could be a good meme shirt. A t-shirt with a line drawing of four completely different ram sticks on a mobo and "this is what peak performance looks like"
@@BReal-10EC I'd buy it.
So did Nic, by the sound of it🤣
I'd love to see this exact test with the AMD bench, considering AMD seems much pickier with its memory speeds
sadly amd is picky i guess intel doesnt care
Yes. This.
This!
Yes
early on it mostly had to do with XMP and some of the settings in it that don't work with AMD's memory controller(that's changed over the years since first gen Ryzen released). the interesting thing is that if you manually set all the XMP timings they'll work fine, enable XMP and sometimes they don't work.. the other thing is that running 1.38v instead of 1.35v seems to solve most of the memory compatibility issues especially if you're using hynix memory modules.
That video can be summed up with a quote from Christopher Titus: "I don't fail, I succeed at finding what doesn't work." And to be honest, it answered a lot of questions I had about RAM speeds and timings, so it's definitely not a waste. It was educational, it was entertaining, it was JTC.
"When motherboards are like.. fine, I wanna see what happens too" I died bro. Love it.
17:44 I have a friend that used to run a build with 9GBs of RAM, one 4GB stick, two 2GB sticks and one 1GB stick. At the time he told us the whole discord channel couldn't stop laughing
Please tell me you shared this video to your friend 😂
It tracks tho. One GB reserved for the system 8 reserved for raw performance. I’m not saying it’s not weird but I get it.
@@ghadturner5645 Yes but no.
@Ghad Turner something tells me it wasn't done by design 😂 I've been there with random amounts of ram, stealing sticks out of old PCs to fuel my Frankenstein monster. But.... you do have a point with the 1 GB reserve. It's weird but not stupid!
Ah, yes - 2.7 channel RAM.
The peak performance comment Phil made had me rolling!! Great video guys!
Yup I LOLed my ass off when he said that.
Behold the abomination to rule them all. 😁
"this is what peak performance looks like" xD
Please do a vid explaining timings and recommended timings for different applications, As always, fantastic content and keep up the great work. Thx!
Phil: I wanna see somebody with an *odd* number of GB's!
Well Phil, it's your lucky day! In my linux box I have 3GB's of RAM (1x1GB and 1x2GB)
I have an old Asus laptop that came with 3gb of ram.
I have 79 GB of RAM in one of my servers because I grabbed wrong sticks and couldn't be arsed to fetch full set of 8 GBs, so 9 8 GBs, one 4 GB, one 2GB, and one 1 GB. Runs like a champ.
I also routinely fshsrlisdfw amount of RAM into virtual machines. KVM's balooning is on, so it uses just whatever in needs anyway.
@@Vatharian Take off 10 gigs to make it a "nice" amount ;)
If you set iGPU to 1GB, any modern ram amount will be odd
@@kostaskotoulas6542 Agree and who needs more than that number anyway? On that number is so fast that you already go wit your tongue out :)
18:38 I am deceased. Both Phil and Nick dying laughing had me laughing my ass off! 🤣
This was the best. The misfit ram setup proved everyone wrong. Loved seeing the process of discovery followed by the humility.
I think it’s interesting that when you go to buy ram, the timings are probably the least advertised thing
Some times they are actively hidden or misleading. For example all CLxx are not equal
Same with most big box stores advertising their PC's with how much RAM and storage space it has. A lot has been changing over time, but that used to be all it'd say.
Because so few people understand timings the manufacturers can de emphasize it and actually sell ram with looser timings for higher price cus bigger megahertz numbers.
Because the sales staff and advertisers don't understand the tech... and think size and speed are what matters (this vehicle seats x and can do y speed in 10s... they don't care about what the engine and transmission is!)
Least advertised and one of the first things I check out, along with the speed and capacity.
This would be extremely interesting to see on a ryzen system.
Set out before it gets hot😷😠😡😵🤪😳😱
of red velvet, which suited her 🤔😗😆😀
Do I find you here, you old er!🎏🎎🎍🎋🎊🎉🎈
I actually have a Ryzen 1700 system on A320 that I bashed together from spare parts running two different kits - I think a 2133 and a 2400? - that runs quite nicely at memory speeds in the range of 3000 MHz.
@Ladioz intel isn't better, ryzen is just picky with memory
Jay: Gotta prove that this is just a stupid configuration and there's no way it would work out.
Cinebench: Hold my beer
PC mythbusting is some of my favorite content you make
That should be a series.
so she was always😐😚😊😂
what a nice plump mouthful🍍🍌🍋🍊🍉🍈🍇
Honestly I feel like there alot of mythbusting these pc RUclipsrs miss out on it kinda surprises me. They really are the best videos when ya get to see something go from theory to practice ha!
I remember EDO RAM where you would have 4 slots but only 2 banks, it would make the lager stick come down to the smaller one. If you put in a 16MB and a 32MB in the same bank you only had 32MB total (16 x 2). I learned the hard way because the first time I built my own PC it was on a Socket 7 platform.
Yes please do a timing video with an AMD cpu. Timings are one of the things I understand the least about PCs
Then watch Buildzoid, not Jay lol.
@@bronzehd6212 He holds the single GPU record in 3DMark Firestrike Extreme. What record does Jay have again?
@@imadecoy. number of explosions in a sponsor bit
18:35 that was the funniest thing I've heard this week.
Don't forget about ranks if you're going to get into the whole number of dimms conversation
That's where 4 single rank and 2 dual rank sticks at the same speeds technically perform the same, right?
@@jamesiepooh7337 yes but 2 dual rank is generally preferred
@@whoisthatthingwhat @James DeHart I can't find any information on 8GB Corsair Vengeance RGB Pro DDR4-3200mhz. Is it dual rank or single rank? I've read somewhere that 4 and 8 GB are usually single rank, but 16 or higher is dual rank. It doesn't seem right to me.
@@peterletran single rank
@@peterletran 4 and 8gb are almost always single rank, with the exception of old 2 x 8 kits using something like 2015 Samsung E die. Nowadays some 16gb sticks (2x16gb) can also be single ranked, I know some Micron for sure and probably Hynix as well. The Corsair kit you speak of will almost definitely be single rank and could be any number of different IC's including but not limited to Hynix (Corsair Version no. 5.31-5.48) CJR, DJR, JJR and Micron (or Spectek made micron IC's Version no 3.31-3.34) Rev E and J
"just look at it" at 18:35 made me think of that time Kramer showed Jerry his chest regarding how the chesthair would grow back thicker and fuller if he kept shaving it 🤣
Jay is so old that he remembers when his 2 cents were worth a nickel. /s
Love this channel.
As for the oddest amount of RAM I've used recently, I ran 56GB a couple months ago (3x16 1x8) while I waited for a stick to replace one that failed. I had a 8 stick laying around and for my workflow the 56 setup was faster than the 32 2 stick and the 48 3 stick, not a lot faster but every bit helps.
It's fun, watching the logical part of Jay's brain melting down when hardware just decides to say "fuck it" and do whatever the hell it feels like...
This is brilliant content. Uploading a trial and error fail video is just solid entertainment and why I absolutely love this channel.
Jay: *determined to prove a point*
Jay: *proves everyone else's point instead*
Viewers: *facepalm lol* ❤
It shows that he's following scientific process... and not doing what companies want now, where they want result x (usually because sales have already been made for it) and tell you to make the data support it.
No
Proving, once more, that the single most important element of comedy is...timing.
Disagree. It depends on the style of comedy. Comedians like Norm Macdonald show it can be just as much about perverting expectations or repetition. Timing won't help if it's not funny to begin with, but good timing will make funny ten times funnier.
@@smeezer Really? Do you have a tattoo on your ear that reads, "Inflate to 30 PSI"?
@@qwkimball Thank you for proving my point ;)
@@smeezer A double whoosh. Those are rare.
@@qwkimball lol
I always used mismatched ram, no complaints. Just needed to set as slowest...
Only thing to give headache was to mix single sided ram with double sided... either didn't work, or didn't recognize all of it or didn't enabled dual channel.
Respect.. Every Master of a skill knows the learning is never done.. You learned something new today.. The day we stop learning is the day we start dieing.
The last fully mixed setup is a clapped out civic with a laptop. Taking all the other setups to gapplebees.
This was incredibly entertaining, and very informative. This is why I love your channel!
Mythbusters: forces the accelerated expansion of objects by chemical reactions.
Jay: I'll stick to excess air pressure.
Yo Jay, That ‘Office’ Reference Was GOLDEN!!!
🤣🤣 U shld use more analogies of funny crap like that… love to hear one on animals or politics 😂
I mean tbh I used two diff sets of RAM once and the only thing that changed was the lowest I could get my timings down was the lowest in the pair (one pair was CL 14 and one CL 16, so I had to use CL 16) but otherwise it worked
That's what I have now. Has worked well enough for the past few years. Yes, I'm probably losing a bit of speed but it's worth it for the extra 8GB giving me 24 GB of RAM.
Generally that's how it goes!
yep works ram will work at highest latency and lowest speed. so 3200 14 and 3600 16 it would run 3200 cl16.
Is it efficient? nope
Does it work : Yep (except some rare cases)
You can always run slower timings. Even if you decrease the frequency you often can't run faster timings. That was Jay's mistake. He needed to stick to the slowest timing. The slower ram (with the faster frequency) just can't do 15's.
Jay is hands down my favorite reviewer. I like his information ,testing methods, and personality.
Did you guys watch my RAM videos >_
I bought 2 kits of G.Skill Trident Z series 3600 15-15-15-35. Had a pretty good time getting 3600 to boot on all 4 sticks. But easily overclocked them to 4000 15-16-16-36. Had to bump voltage on my system agent to about 1.25v to stop random crashing.
Jay regarding merch... I need a shirt with Phil's face on it surrounded by a subtitle if his laugh... As long as Phil is cool with it of course.
This would be fantastic.
Ill buy two if Phil isnt cool with it lol
Yes please!
And on the back of it a picture of mismatched ram stating "this is what peak performance looks like" 😂😂
The only thing missing was a bootable raid drive striped across a SSD and a 5400rpm HDD.
Like intel Optane lmao. The where the OS is likely to corrupt and boot to bluescreens everytime you update 🤣
@@infernaldaedra Well, you could always use Linux. I don't think the problem is the concept of striping as much as it is Windows' support for it. Windows is still using NTFS, perhaps one of the most antiquated, slowest, featureless file systems still being used.
I'll be installing Linux on a ZFS mirror (RAID-1 equivalent) with 2 NVMe drives soon enough. It'll be tricky to get /boot and /root to install on OpenZFS, but certainly possible and stable (apart from a bug regarding hibernation). But for a gaming PC, you can install /boot and /root on ext4, then have a ZFS pool where you store things like games and files.
Cinebench is not a good benchmark for memory, it mostly uses CPU cache.
Also 4 stick combinations will (most of the time) have the benefit of running dual rank while 2 stick usually run single rank.
isn't single rank better
@@mmmdawe only if you have two single rank sticks on the same channel so you then have dual rank.
@@bronzehd6212 no.
@@mmmdawe Yes/No. A lot of testing has been done on this. There's also contradicting information online as well. But basically from testing (I personally did with my current computer) and research I did on the topic and looking at other channels who tested different scenario and configuration (Like GamersNexus and ActuallyHardcoreOverclocking) I think we can conclude that :
- At same timing and frequency (so let's say 3600Mhz CL16), Dual Rank is better than single rank (note that Memory rank =/= channel, those are 2 different things).
- However, single rank will usually allow you to get better timing and higher frequency than dual rank. So at a certain point it MIGHT, be better. But we're usually looking at really high frequency (like above 4000Mhz).
Since most people don't buy crazy high frequency RAM (most people will shop between 3200Mhz - 3800Mhz. In that range it's possible to get pretty good timing on dual rank (getting single rank RAM that does better is going to be very expensive) so dual rank will usually perform a bit better. The difference is NOT that significant however and I wouldn't worry too much about it. We're talking a few percents in synthetic benchmark.
As a car guy - this explanation ( from Crucial's FAQ) was what made it click for me:
"If we think of memory as a race car, the module speed (MH/z) would be like the raw horse power of the engine, and the timings would be the driver of the car. As the driver of the car gets better at handling turns and responding to obstacles on the race course, the car will perform faster, so much so, that you can have a car with lower horsepower outperform one with more horsepower, if the driver (timings) are faster in the lower horsepower car than the car with more horsepower."
What an awesome video. I love that Jay keep's his hubris in check such that he not only will say "frick" it and try something like this, but also not try to hide when his initial hypothesis/notions are off the mark. It would be quite easy to shoot this video, scrap it, and re-shoot with the narrative that he's an infallible tech guru. I love you for that Jay. We all need to be more okay with being wrong and/or just not knowing something... it allows for so much more learning and even connection to take place.
I'd love to see a video on manually setting ram timings etc. I still use an older system (2600k and 32 GB ram) that runs just fine. Going over ram timings is a chore to find decent info on.
Love the fun you guys have too.
Yes, this
It's really not that hard to find good info. There's plenty of channels just on RUclips with good videos explaining ram timings. BuildZoid has a series explaining some of the basics on ram and such, GN has had part one of their ram explanation out for a couple years at this point I think, can't remember the specifics now.
@@NotThatGuyJD Thanks
Excellent experiment and thanks for taking one for the team (in the name of science)! Would really like to see you try out as many combinations of kits with 2x stix to see if it's actually not a good idea to buy one kit to start, and upgrade to 4x stix total at a later date (missmatched kits). Brands, speeds, latency, ranks etc. Either way, keep up the great work you do =)
I think I speak for the entire community when I ask "WHERE IS THE ENGINE BLOCK BUILD?????" Day 1 of asking Jay for this. I will be dropping this comment on every video until our demands are met.
I second this, though i cannot guarantee i will remember to ask next time xD. WHERE IS IT JAY?!?!?!
Where the block jay?
Super interesting video.
There are two more questions/tests I would be interested to know the answers to - 1) how would the two fastest matched sticks compare in the overall results. 2) Does having the mismatch in sticks introduces some memory instability if pushed to the limit by memtest or similar tool.
You’re going to have to test higher speeds with mismatched timings as well as mismatched capacities for the heck of it. It would also be interesting to see how a matched set of 4 sticks would compare to a mismatched set of 4 where the lowest stick is a stick of the matched set.
HAHAHAHA! This is the greatest channel. An unbelievable source of ridiculously useful, common sense information with a nice humble comedic flare. I’m a mechanic and that whole Honda Fit turbo to supercharger bit was great! I was freaking dying! Keep up the good work
Yay, Jay's 2 cents has lost value due to inflation. It's at least a dime now. Jusskiddin :)
JayzDime. I feel like Jay would make joke about Mrs. Jay here.
Jayztwobits, two meanings computer terms and 2 bits is 25¢
Do have mo
Moo
O
M
L
O
O
Om
L
@@thetalesofdaneandco
O
Lo
L
Lo
Om
Mooop
I remember having a weird number in my pentium 1 days, like 40MB of RAM.
Kinda amazing though, regardless how your timings or capacity is, mixed brand or same brand, you still get decent performance.
because the brand doesnt matter if theyre all some c die ic dogshit
Hah! Love this. Excellent video.
Gotta say though, with modern day mem controllers, you will still start seeing performance losses on larger capacity setups.
Ram virtualization makes a lot of these mismatched setups not only possible, but even perform decently. It might make boot times a bit slower, but I'm not even sure you'd see a human measurable difference.
Capacity will create a difference though.
If you go from 16 gig to 32 gig, no big deal.
From 4x8 sticks at cl16 to 4x16 sticks at cl16, you'll start seeing a lot more heat from the memory controller.
At 4x32 @cl16, if you can get it to boot without lowering timings, your memory controller will absolutely hate you.
Obviously this is with a standard dual channel currently Ryzen or Intel Z490 chipset.
Server applications and higher dim/channel capacities will see the same type of response though.
Hey Jay, you forgot to mention if each DIMM was Single-Rank or Dual-Rank.
8gb sticks these days are all single rank.
I was just going to say the same thing, this explains some of the "weirdness".
14:45 quite easy to explain: the IMCs act smart and use dual channel for the first 8GB and go to single channel, if they have to use above 16GB. So as long as your RAM requirements stay within the limit of the chnalle with less RAM * 2, you get dual channel bandwidth. If you need more RAM, above the dual channel limit, your scores will decrease.
So it's no wonder that you see the same score in CB. Not sure what amount CB requires, but if you fill up the memory, so that CB is forced to access the single channel area, you will see a decrease in performance.
However: Even that decrease is better than swapping to HD or SSD.
The main takeaway: "If It Looks Stupid, But It Works, Then It Ain't Stupid" lmfao
i had mismatched sticks for so long but they were both 8gb each. I've now got a Klevv Cras X kit @ 3200 tightened to 14-16-16-32-427. Pretty happy with that. Thanks for the video Jay!
Yes do a timings video. I'm about to get 64gb for my girlfriends pc and would love to know if 3600 with a little looser timing is worth more money or if it'd be better to get 3000
In my case about the same. Don't recall the specs exactly for the ram, but 4x8 @3000 vs 2x32 @ 3600. Very very close scores on firestrike and cinebench23.
This with a 3900x on a msi b540 tomahawk max.
depends on the system setup whether it's AMD or Intel since there is a slight performance increase running 3600 over 3000 that may offset the performance loss due to timing and what the systems being used for.. it's likely that overall there's no significant benefit to running 64x3600 vs 64x3000 but saving a couple 100 dollars might be worth it.
I just built a top end pc and felt/feel I overspent on the ram. Feel a little bit better now. Paired an AMD 5900x with 4 sticks of Trident z Royale 16gb 3600mhz 14 14 14 34. Going to running some benchmarks now.
@@tchlin it pains me to see you buying overpriced ram and not manually overclocking it.
@@whoisthatthingwhat if I ever understood subtimings properly I'd probably do all manual overclocks. For some reason I enjoy tuning things more than actually using them.
Phil and Nic dying in the background made this video epic.
Jay I was tripping about FPS and about to replace either GPU or CPU and then forgot the ram having a cl18 is probably really why
overclock it yourself
CL18 is definitely hurting. Hopefully it isn't CL18 at like 2400 speed to add insult to injury.
@@Stubbies2003 my ram is cl18 and it makes zero difference to FPS.
@@oxfordsparky probably because you are gpu limited
I have a Ryzen 9 5900X, TUF 6800XT, and 32Gb Corsair Rgb Pro 3600Mhz CL18 (four sticks), tightened the timings to CL17 at 3600Mhz and also tried CL16 at 3200Mhz, was virtually no difference. If you have 3000Mhz CL18 or something like that, you will not get a major uplift even if you get 3600Mhz CL16..keyword being major..small uplift yes, but i would rather have overclocked the ram you have to the best speed and timings it will do and settle for that.
For real, though, my rig is running the same ‘model’ (XPG Spectrux D41) ram all around with the same timings, but with channel A packing 16 gig sticks, and B with 8 gig sticks. Funny enough, each using a different controller in the pair. I might actually try tweaking the timings and slowing the xmp clock after work just to see if it’ll boot and run tighter
Hey, in the end, great video - now I know my next PC upgrade is going to focus on LOW timing RAM sticks rather than purely on baseline OC speed. This is the sort of information us old idiots thrive on.
I've been building my own PC's since 1992, and have also ALWAYS used matched sticks of ram.
PEAK PERFORMANCE! brought tears to my eyes laughing. i cant believe it!
how did you saying "compacity" not get picked up in the editing room lol
that's not a word?
@@aitskimoro9035 Capacity
I actually had issues a year or so back with my DDR4, this was in an AMD ryzen 2700x pc, with mismatched brands of RAM. I had 2 sticks of 8gig 3200 Gskills already in the rig and it ran fine for over a year. A year or so later I attempted to add in an additional 2 sticks of 3200 Ballistics I found a good deal on at the time to double my ram. The system did not like it, and would drop the ram speed to 1800mhz, and any attempts to clock them back lead to the a boot loop until the system set the ram back to 1800 mhz. I eventually got 4 sticks of Ballistic and took the Gskills to another AMD rig that they work fine in. Having had mixed ram stick builds in the past that function perfectly fine, lesson learned is that not all ram likes each other.
Yeah, AMD still lags behind to Intel with it's memory controller. Can't really innovate your mem controller when you are pigeon holing yourself on variations of the same socket chipset for over half a decade (AM4).
The Office reference, the turbo/supercharger reference, computer nerds. This is why Jay is GOATED
My old system that I gave to a friend has 3 sticks of ddr3 totaling 20g. 2x 8g and 1x 4g. The single stick is like 1333 and the other 2 are 1666 but running the lower. Fx 8350 black. Still running like a champ.
Loved his vid btw
In college my professor wouldn't even use RAM that was from different lots
That professor is a dunce.
@@CompproB237 back in the day, you had to be that picky. So.. he's just set in his old ways
Great video - Always good to see ""what if???"
Looking forward to hearing a video with more on timings. Not a G-Nexus style of this is how it works 'technically' of course. More of the classic Jayz this is what happens when you do!
Many high end Motherboards have advanced automatic overclocking options, what do they choose and is it better to just manaully set to get a better day to Day result (not NO2 levels)
Another curve ball to the conversation would be: How would all this testing turn out if you'd take into account that Ryzen benefits hugely from 4 sticks instead of 2.
Really? Last thing I saw was ryzen benefited from single channel
@@Dr.Spatula This is totally down to what you are doing with your ram (and what ram you have used). For overclocking then single channel is generally better, especially if you are using a daisy chain motherboard (which most X570 and B550 boards are). For raw bandwidth then dual channel is better because this will guarantee you are running in dual rank. This can leverage about 10% more performance in certain workloads, but is is generally a lot harder to overclock 4 sticks of ram than it is 2. This can of course vary from board to board and ram kit to ram kit. You can of course also get dual rank sticks of ram (16gb micron Rev E for example) and in this scenario you can have best of both worlds with dual rank and also potentially better overclocking. (Edited - sorry Dr Spatula that was very confusing)
@@malphadour why are you talking about number of sticks per channel?
@@Dr.Spatula Christ I wrote that all arse about tit, need to edit that. I meant single channel or dual channel, not dimms per channel, let me fix it with coffee...
@@malphadour ok. Most people still don't know the difference between channels and sticks so i was concerned you were in that same place
I bought a G Skill 16gig kit 8gig x2 in 2018 and then bought two more in 2020 and they wouldn't run at 3600MT/s together. It actually caused my system to crash even at 3200MT/s. I replaced them with two 16gig Corsair Dominator Platinum Pro 3600MT/s. computer runs a lot better and stable now.
I think buildzoid needs to tech jay about b die, cjr/djr, micron revs. Just just went full mormy. Does jay know about ram ranks?
Yep, but in the general sense that just contributes to a particular set (or stick) of ram being able to achieve tighter timings. He showed that today without losing a lot of the more casual users with technical jargon.
This video is a nightmare. I'm gonna have to stop trying to explain to redditors how RAM works, because they're gonna keep referencing this monstrosity. Jay is way too big to be putting out a video about something he's THIS uninformed about.
This makes me a little happy to see that my purchase of CL14 (14-14-14-34, 67 Clocks) makes a difference. :-D
This has been an episode of cursed PC configurations, see you next week when we try to create a raid with an nvme and msata m.2 set of SSDs
18:40 look at it! And FYI - it's not only Nic dying on the floor there. I have a floor too :D
This sounds like frequency is important to a point and timings are more important after that
Multi channel memory will create islands when each channel is populated asymmetrically. Intel spec from Sandy maybe even Nehalem stated this. At least in server land. So in the triple 8 in dual mode, you have a single island of 2*8 in dual channel and another as a 1*8 single channel.
Phil's laugh makes these videos for me. If he ever leaves, I'm gone too.
My old out of warranty MB B channel stopped working. When I upgraded, I replaced the two sticks with a larger size. I just used those old two sticks later in a media PC I built. I needed more RAM for media art 2D and 3D stuff. My MB also supports RAM cache for the storage.
My takeaway is that from now on I will be paying more attention to timings on my ram kits.
I have 24gb of vengeance Corsair (2x8 3200 c16 b die and 2x4 3000 c15 should be Garbo tier) I found out I can run them at 3000mhz c14 with some tuning and stable 2/300% memtest. But I got lucky. Any xmp profile didn't work by itself. It was fun to tune them ☺️ it's fun to adjust things and not to get the expensive stuff and put it togheter.
Following buildzoid you learn that not every benchmark really benefit from memory speed, other will benefit from memory latency etc... It depends on the workload/benchmark.
Edit: and depending on the typology of motherboards a 4 filled one can be more stable than just one stick, t-typology motherboards even if rare works that way. Daisy chains one yeah works better with fewer sticks
Lost performance isn’t typically the concern with mismatched RAM, rather the errors that can occur to cause crashes. That’s the main reason matched RAM is greatly important in servers, as they need the utmost amount of uptime and data loss is a great concern.
One stick of this, one stick of that, one stick of dynamite... and another of JayZ's projects blow up! Lol
What's really happening with mixed RAM is if you have mixed capacity, when you exceed capacity of the less capable kit you go into single channel.
That makes a lot of sense!
Total agree on this. This did
Happen to me i bought the same ram didn't see the timing but same speed. Had this problem couple of time than just bought another set than have no issue.After looking at the timing had to be the same than it works great.
Synthetic tests are really poor way to measure ram performance. It starts to matter at higher fps. CS Go would be perfect for example.
Jay is not the guy to go to for proper scientific stuff like this. The video, whilst entertaining, is a shambles from a testing perspective. Steve would pull him to pieces on his methodology :) And you are absolutely right - and using Cinebench was utterly pointless.
Timings are a complex topic. Sometimes loosened secondary timings prove an improvement, because they let the rest of the timings work better together.
It is a sign that tinkering with the secondary and ternary timings can be adjusted to perform even better on the tighter primary timings.
With the thumbnail subtitle I thought you were going to discuss how you measure wall thickness. But love this type of material. Keep up the good work
Asynchronous Dual-channel memory has been a thing since Pentium 4, or Core Due/Quad days. And yes, it works :)
Great video, guys!
Update: used to run a system with 12 GB of RAM for 2 years back in the day. AMD Phenom II X2, 2×4GB + 2×2GB DDR-3 1333 MHz CL9 (iirc).
Hello to all.
I am regularly testing RAM sticks for issues with MemTest86 when I got a new opportunity.
And one time I tried to use 80GB follow crazy combination:
- 1x Kingston 2Rx8 16GB 16-18-18-39 in chA
- 1x Kingston 2Rx8 16GB 16-18-18-39 in chB, not from the dual-channel kit with previous one
- 1x Crucial 2Rx8 32GB 16-19-18-35 in chA
- 1x Kingston 2Rx8 16GB 15-17-17-36 in chB
And they works in OC to approx. 3200 20-22-22-45 and still with quite good performance and dual channel :-D
- Z370 Gigabyte Aorus ultra gaming 2.0 + i7-8700K
I think it's more important which chips are on the RAM sticks. I have 2 sets of the same 16GB GSkill sticks, but one with Samsung and one with Hynix Chips and they don't want to work together very well. I needed to juice them up to 1.65 volts and still, they don't really want to work together. Each separately is working fine btw.
Also capacity still matters for high detail, large world games like warzone. Most other "open world" games are only loading a few block's or cells at a time like mine craft, skyrim, fallout etc.
For space engineers and warzone you *need * 32gb ram for best quality settings
If not mistaken Jay, the mainboard will maintain dual channel memory, up to what memory matches up [off the cuff or manually tuning]. Any excess memory gets backpedaled to single channel
Having that 24GB visible, only 16GB would be handled dual channel, where the excess [in this case, the other dimm] be running single
Yes! I researched quite a lot into ram timings when choosing for my current build, and found the best value to performance was the Viper black sticks! Got a 4133mhz kit for £90 and after a lot of trial and error got this kit to run at 3733mhz cl14 at 1.5v (I know right!) Was using cdmark/ cinebench to see score performance as well as SOTTR for real-world performance, and the performance uplift was equivalent to an entire gpu tier! Around 20fps!
Ram overclocking is considerably difficult compared to GPU and CPU, but there certainly is performance to be had for sure!
ps. 1.5v is a safe daily for samsung b-die (researched into this a lot), do not recommend above 1.45v for lesser quality. Also 3733mhz is the best I could run, as my 3800x refuses to boot at 3800 with cl14 (just unlucky there with silicon lottery I guess)
im running 48gb of 2133@1770 9-11-11-27 cr1 in this old 1366 rig, when my crosshair v-z died my uncle gave me his old 1366 rig, and i replaced the 12gb ram with 32gb i had from the ch v-z and was able to get 1 more matched kit for $99 by luck, replaced the i7 920 with an x5675@4.42-4.65ghz for 20bucks... not bad... for a decade old rig... primocache helps with load times alot...
I had a system recently with 48 GB: (x2) 16 GB G.SKILL Trident Z RGB 3000 MHz 16-18-18-38 and (x2) 8 GB G.SKILL Trident Z RGB 3000 MHz 16-18-18-38.
it would randomly throw a BSOD "page_fault_in_nonpaged_area" at random times, all the time. it was insanely annoying. problem was solved when I removed the (x2) 8 GB sticks from the system (and a clear CMOS), and it's been running great.
NOTE: the 48 GB ran great but randomly decided to BSOD whenever it wanted. JUST BECAUSE IT WORKS DOESN'T MEAN IT'S STABLE!!!!!!
I've had mismatched timing but same speed, brand and model RAM. I set the timings to the loosest to ensure compatibility. It worked, but I encountered instability issues after a couple of months. Removing the looser timing modules and just using two sticks of the tightest timing ones and the system was just as fast and the instability vanished. Both Crucial Trident Z kits. Perhaps there's something about Crucial.
I bought my PC off a "streamer" around the corner to me on Facebook, like 4 years ago. I'm only just now learning about PC, how to make them, what everything is, the does and don'ts, and whatnot. I looked into mind and I'm running 2 4GB DDR3 in the A channel, ahahaha. I take it this was a build by the guy I bought it from and didn't know about RAM. Glad I can learn from this and you!
Also going to add, I am building a rig now, so that's why I am learning about it all. I'm currently on an Intel i5 3.5Ghz, 8GB (2x4GB) DDR3, 2GB AMD Radeon R9 200, MSI B85-G43 motherboard. Finally moving up to an Intel i7 11700K, 32GB (2x16GB) DDR4, ASUS ROG Strix GeForce RTX 3070 Ti OC 8GB, ASUS ROG Strix Z590-F Gaming Wi-Fi Motherboard.
The uplift in memory volume compensates the drop in speed rates, but the drop and uplift sizes are different (uplift is bigger), so the 3dmark scores seem to go higher than initial setup, as the memory capacity overhauls the speed drop significantly.
my first pc was 1x1GB ddr2 667 + 1x2GB ddr2 800 and it ran pretty well for a budget build. the aging hard drive that caused lockups unless actively cooled was more of a performance issue than tjw ram.
my current rig has 2x4GB 1866mhz hyperx and 1x8GB 1600mhz xms3 salvaged from my fx3200 rig. i think it's 8GB per channel so should be fully dual channel rather than just the first 8
I do happen to have a rig still running 56Gb of matched RAM because it dropped an 8GB stick.
the bandwidth math doesn't lie higher clocks with loser time are the same silicon it's just binned differently.
I run some custom stuff that silos the channels, as long as all the DIMMs on the channel are the same it's ok.
it does get a little squirrely between 1 and 2 dims because it will load the "bottom" of the DIMMs first and if there's 1 instead of 2 it runs slower even if the capacity of the 1 DIMM is the same as 2 on the other channel.
I used to have, on each channel, matched memory, but unmatched channel vs channel. It worked very well, and this was on an older setup. But since the CPU is the one that actually handles the memory, I figured, it would just do the best it could do, in each memory controller/channel
Jay Patriot make some very good memory. Their Viper Steel series is great for overclocking and is cheap for high frequencies.