Reacting to way too many RAM timings
ฝัง
- เผยแพร่เมื่อ 3 ก.ย. 2022
- My Patreon: / buildzoid
Teespring: teespring.com/stores/actually...
Bandcamp: machineforscreams.bandcamp.co...
Twitch: / buildzoid
Twitter: / buildzoid1
Facebook: / actuallyhardcoreovercl...
#RAM #overclocking #ddr4 #ddr5 - วิทยาศาสตร์และเทคโนโลยี
Ah, the best low effort content.
This AFTER the RAM timings series is perfect. The reviews/occasionally savage roasts are educational AF.
@@erisium6988 does anyone know of this guy has a discord. He is awesome and he is 100% knows what he is talking about.
I laughed my ass off when the guy showed all of his timings on auto and begged for forgiveness
Hey Buildzoid, I know you said knowing what memory timings mean has no effect on your ability to overclock. I was curious if you could make some videos on where to start on memory overclocking. For example, on DDR4 or DDR5, how to figure out what timings to start with at a frequency and how to walk them in to tighter timings.
Basically what I'm saying is I need a memory overclocking for dummies video.
amen brother
As an old school overclocker, (P4, s754, s939, etc):
Back in the day on EOCF and xtremesystems... competitive overclocking and truly 100% optimizing a system overclock was a matter of bruce forcing ever-tighter timings, higher frequencies, and stability testing plus benchmarking the results. You’d set a low hanging fruit ‘OK’ overclock based on what ICs* your ram is, what chipset your mobo was (a very different affair compared to today), and what stepping your cpu was, you’d just start about 85% of the way there, and then incrementally change bus speeds, tighten timings, etc.
It feels less surprising to me than it does to buildzoid in this video, that weird out of place non-critical (in a performance sense) timings look oddly UN-optimized.
This kind of thing happened to me when I was younger all the time. You’d hit a golden “milestone” number on bus sped, and in a benchmark, but just not be P95 stable... until what do you know, but some bullshit completely inconsequential timing getting loosened made the system 24hr P95 stable. And then boom, you’d be done with it.
@@smillman437 I guess I just haven't figured out how to know what to loosen, what to tighten, or by how much. I know there's trial and error involved, but is there some sort of method for start with one group of timings and then walk it in? How do you know good loose timings to start with at high frequencies? For example I have DDR5 Hynix and if I just increase the frequency to 6600 MHz the computer doesn't boot. If I load XMP which is for 6200 with a couple of tighter primary timings and then just raise the frequency to 6600 MHz, the computer is nearly stable. But from there I don't know what to loosen to bring stability back. I can brute force lowering the other timings afterwards, but it would be nice to have an idea where to start. I don't like the idea of brute forcing loose timings to try and regain stability as it seems like you're moving in the wrong direction and I don't know how much a timing needs to loosen for stability.
@@SlimeyGuitarStrings I must emphasize... as an old school overclocker. I don’t presume everything to be the same now as it was then. I started watching Buildzoid’s videos over the last year and a half to try and get myself somewhat back up to speed!
In my haste, in my earlier comment reply, I totally skipped over modulating voltage and cooling, as part of the tedious tweak>test>benchmark, sort of overclocking that I was so familiar with. Things like VTT, and the different SOC/Chipset voltages that we use these days, didn’t even exist mostly, and whatever analogs did exist, they were called something else. And back then, overclocker were 24/7 OCs... not boost OCs.
Anyway... I think Buildzoid’s videos on timings are as good as it gets. Used to be (outdated), one would get the best benchmark results you can with the best (highest) frequency one can on the memory, and with the lowest CAS latency possible, then you’d bring the 4 “main” timings down as low as possible. And be 1T command rate obviously (with the exception of machines running 4 sticks of RAM... as they generally wouldn’t do 1T back then). Also, preferably without running extreme dividers... which usually didn’t make much sense anyway, unless you were ‘OCmaxing’ memory and didn’t bother to invest in a CPU that had a suitable bus multiplier... it generally wasn’t worth overclocking memory to a high degree, when you were workiing with a 1.4 to 1.6ghz dual core that’d maybe do 1.8ish tops. For real world performance, it wasn’t worth the effort of min/maxing of that memory, unless you were tweaking and just finding the memory’s OC ceiling for fun... I hope that makes sense.
I will stay tuned and continue to try and get myself back up to speed, eventually via practice with my own new system, eventually. I just had my second kid, and I think my next system will be second hand intel 12th gen, unless there is a undisputed price:performance cpu in my price range, from the upcoming Ryzen 7000 lineup.
I'm upset I missed this again!
me too =(
man, i missed it again
edit: did not know the trfc register went past 1023
same 😬 and I've been quite lazy with my timings, just bumped freq to 3600 from 3000
will wait for the next react one
this video is so cool, you reviewed 2 kits of 2x32GB SK Hynix, and you commented on the timings with info that i couldn't find anywhere else, thanks buildzoid! I got 2x32GB hynix CJR 16Gbit, at 3733 (fclk won't go higher) with decent secondary and tertiary timings, but for the primaries, tCL is stuck at 18 ( i know you said this timing doesn't matter so much), and the damned tRCD(RD) won't go lower than 19 without errors, i'm at 1.4V and i am just unwilling to risk the sticks by going higher in voltage. Also running tRFC at 500t, stable, not so bad! Almost finished tweaking the kit, but as you said, testing 64GB is kind of a pain. Still, i'm having tons of fun! Thank you for the info in your videos, it helped me a lot, you even helped me with choosing the sticks, i just went with the cheapest 3600 18-22-22 kit i could find and was not dissapointed! I'm going to try do disable geardown mode and go to 1T CR , perhaps i can get tCL at 17 then, but I don't have high hopes.
As much as I love your "rambling" videos, I really wish we could see NON-B-Die overclocking; like CJR for those of us that got too excited about a sale price and didn't look closely enough at the part number on their Trident Z Royal kit...
I guess that's almost like asking to oc cpus with locked multiplier because that's what you got, but in essence IMO overclocking is all about getting the most of what you got regardless of what it is. Luckily with memory the process is the same, you're limited by what the chip can do and not something artificial and knowing what to expect will just save you some time.
If you're buying TZR because it's cheap and you like the look then i have no forgiveness for you if you get DJR instead of B-Die.
I owned JJR for a while and it was pain in the ass. Sold them and got vipers instead. 4400 C19. Now tuned to 3733 CL14.
Frankly im not sure why people even bother with Hynix anymore. Cheap B-Die is better not does not cost a kidney.
People who post stock timings are expecting bz to give out the values for them. Which works
you are a genius thks for your time and information on ram always greatly appreciated
8:24 That 16Gb Micron Rev B kit is impressive though. XMP 4400 19-19-19 timings used to be Samsung B-die only. Somehow Crucial manage to bin this kind of thing out.
I am really enjoying this video it’s so awesome. It’s Incredible a person on TH-cam is looking and peoples timings and letting them know if it’s ok. All I can say is you are awesome!! I hope to get mine in one day were you can tell me how bad it is lol.
I think people want it now because they've realized that maybe their RAM timings aren't good and they want your approval.
Anyway, my X570S Aorus Master hates when I try to make my tRCDWR under 14 on *dramatic drum roll* dual rank B-Die :D
13:00's timings look so clean. So nice to look at.
Great video BZ! At least you didn't get too upset at the low effort content! Overall your viewers sent in some not too shabby content. Thanks for entertaining us BZ!🙃
20:00 it is two sticks of micron e die and two sticks of micron b die, both are xmp 3600@18 (same corsair vengeance kit, but one year between the purchase) but at 64gb even xmp is not really stable so i never really tried 3600 and just went with 3533@16. Setting the timings down to 4-4 6-6 seams to work.
Configuration of putting Micron Rev E sticks (dual rank) into one channel and Micron Rev B (single rank) into anouther one may be worth doing. At least that configuration avoids mismatched memory sticks within a same memory channel.
@@volodumurkalunyak4651 I already thought about trying that, i probably should do it tomorrow
Got micron e die DR corsair LPX at 3600 16 19 19 19 39 84 at 1,4V suggested by my motherboard maybe you are going to have more luck starting from here.
Great content! 🙏👏
Can completely understand your frustration about the views. People just seek OC approval from BZ, that's it. (me too, but couldn't make it in time xd)
It's maybe disincentive to you, but the other type of content (ddr4 timings explained for example) requires more focus and attention from a viewer standpoint.
However, I'm enjoying it all, thanks ✌️
the few cursed Configs in this were very amusing to see.
Is DDR3 architechturally similar to DDR4, especially in terms of memory timing importance? I've been on DDR3 for quite some time and was always curious how much DDR3 performance can be improved upon, as DDR3-era didn't have any videos in regards to this topic.
ig it would also depend on the system.... such as intel 4 gen vs AM3+?
DDR4 has t_RDRD_sg, t_RDRD_dg instead of t_RDRD_sd of DDR3. Reads (same for writes) to same DIMM are now split beetween same group and different group. DDR4 has 4x as many banks arranged in 4 bank groups per memory rank compared to DDR3. t_RDRD_dg and t_WRWR_dg should always be 4 for DDR4.
Intel naming convention here.
The rest of timings should be very similar.
Soon after AMD users and Buildzoid popularised memory timings tweaking, but before I upgraded, I tweaked my Sandy Bridge system. The results in artificial memory benchmarks were outstanding, but I never bothered with any real-world tests. Did it mostly for fun.
Of course, the general concept of how RAM works hasn't changed but there are key differences and it always depends on the use case. Personally I still find DDR3 OC'ing to be a lot of fun and I think it's great for beginners since it can be a bit simpler to wrap your head around. In general, overclocking back in the day was awesome because there was typically a lot more headroom to be found and it was more straight forward. Getting another 1 GHz out of a CPU for daily use just isn't going to happen today. Feeding that extra frequency with tuned memory could certainly make an impact.
for the most part DDR3 functions quite similarly to DDR4, yes. DDR4 Chips are broken up into smaller chunks, but the procedure that the Memory operates in is quite similar.
CPU's have been DRAMbound for multiple Decades, so making the DRAM not perform like crap like it always does by default, yes, can quite significantly increase real world performance.
however, good DDR3 Overclocks are tricky. because the Chips that Overclock well are also all the smallest Capacity ones. so for really good Memory Overclocks, the maximum possible Capacity you can have is 8GByte, and that's just really not enough Memory if you're using a DDR3 Platform right this moment. you really gotta have 16+ realistically.
versus DDR4 where there's many good performing 8Gbit and a couple good performing 16GBit Chips. making having enough Memory and still having some Memory performance so, so much easier. like even though 64GByte isn't recommended due to limited Overclocking performance, you can still make it pretty good overall, and 32GByte has wonderful performance.
but anyways, as long as the Chips you have aren't awful, getting 10-20% real world from Memory Overclocking should be very doable. a lot more if they were fast Chips.
being on one of the later DDR3 CPU's rather than the earliest ones would also help a lot.
I love that content, just picked up a Patriot Viper b-die kit to see how far I can go on a good X370 motherboard with a Ryzen 5500.
Whyyyyyyyy im misssing this aggggaaainn. Noooooo . Next time please announce it firsssst sir
Man, this series of videos about dram timings was great and useful for anyone into memory oc, I don't think such videos can be made short, or very entertaining.
The problem is that it's a niche topic, and another thing is that there have been irl problems of different kinds for many people recently, so the viewcount may be disappointing.
Gigabyte absolutely messes with tRFC, and I have no idea why they do it.
I've only ever encountered it on MSI.
Same happened to me, when I left the setting on auto the bios told me a different tRFC value than it actually was. The motherboard is a Gigabyte X370 Gaming 5.
@@snowhawk4049 oh that's normal. The gigabyte BIOS doesn't readout tRFC correctly. However it shouldn't exceed 350ns on auto.
@@snowhawk4049 autos don't show full truth until you do first cold boot with XMP (they show JEDEC 2133 default)
So there may be some XMP profile wankiness, imo.
my Asus B550 board also defaults tRFC to 550 ns. Currently using 16Gb Micron Rev. B single rank
Your “low effort” content is top notch BZ!
So I am on 5800x3d and I was going through your DDR4 timings explained series. As I am watching the first example here - I am a bit confused. Both in this video and I also found that information somewhere else - you suggest going to 2 with the tRDRDSCL timings. I think you said in the DDR4 Timings Explained series that read to read should not go below 4 as the read burst takes four clock cycles? Or am I confusing those right now?
That's a BOMB hilarious opener! Yes, BZ, we fucking enjoy it
Hi guys, i need your help: i have always kept my samsung b die 3800c14 at 1.53v (watercooled), 100% stable no whea (tested with tm5-anta extreme). Then about a couple of months ago I had to have my motherboard replaced (x570 dark hero) since it became defective and with the change of the motherboard I noticed that the overclock of the ram and the CPU was not 100% stable anymore so I slightly increased the voltages bringing the rams to 1.54v. Now after about a month, without any apparing issue, I have found other malfunctions that have led me through the change of many different bios and many cmos and at the present time my pc is not able to boot if the ram is set above 1.50v (the pc does not turn on at all and I'm forced to the cmos). It seems fine at 1.50v and the cpu overclock seems stable. Is there anything i can do about it? I'm not able to find any solution by my self or looking on the internet.
Is it better to run 1T in general ? then need geardown off right, cause if you activate it, it won't run run odd number on the primary.
Also I had my kit almost stable at 3866 1933 fclk but had audio issues.. reducing it solved it, I wonder if there is some voltage somewhere I could crank to try to stabilize, otherwise I guess I'll be OCing at 3600 (currently trying 3666 c14-16-16 need to tweak the other timings...) got a 4266 c19.. sux to not be able to get good clocks on these, needs alot of tweaking with subtimings, so time consuming, I'm thinking instead of 5600x b550 I should have go for 12600k..
Edit : I just tried for half an hour to run 1T and god this cpu memory controller and or mobo (b550 tuf gaming plus) shit itself, giving up on 1T sadly.
thanks, was able to lower a few of my CRJ timing lower
Got 3600x that can't go above 4.35Ghz but can do 1900IF :(
sounds like my 3700X.
@@ActuallyHardcoreOverclocking 51:05 +1 for a video ✌️
17:45 Yeah, that looks like stock XMP. I have another 64GB 3600Mhz CL18 kit (mine are 2x dual rank), and the XMP tRFC1/2/4 IS 991/631/469. Thankfully asrock bios tightened it up to 312/192/132
The kit is Patriot, chips are hynix.
edit: oh, yeah, and your timings work, checked them, tho only on 3600Mhz, not the 3733 he set.
Can You explain how GDM compares to running in 2T mode?
I noticed on auto, compared to your gigabyte am3 w/ the Teamgroup 3600 C14, my asus x570 with that same kit, sets some really wild (loose) sub timings.
Wonder if i made the cut… probably not
Regardless time to watch and see if there are any other timings i can touch cus the ones over touched ive tuned to within 1 even 1 less on any of what ive touched and it wont make it to even 100% mestest let alone 400% or more
Already at 3 to look at and less than 5 mins in… 7 to look at in under 20 minutes… 9 to look at under 50 minutes… finished watching its tuning time
Well tuning the final thing tRDWR and then i run out of things to tune for now
I know it's a dumb comparison but after recently playing with my decade old X79 rig I have to chuckle at how my quad chan DDR3 bandwidth and latency is still slightly better than most of these in AIDA.
I wish I'd saw your post, that would have been a funny curve ball to throw in here.
I wish i knew how to oc my ram a bit. i can't even find what chips this uses.
Should RD and WR SCL be the same? I can't boot with both at 3, but I can with RD at 4 and WR at 3. Hynix H16C ICs.
Love the Cezanne imc xD
Yeah would be good to do this again, hopefully some older stuff gets in so I can understand how to apply it to my system (since I'll probably miss the tweet again). I'm only 25m in so maybe some older stuff comes up later.
TeamGroup Ram. It’s the chips they sweep off the floor at night.
I still don't understand trdrdsc, trdrdsd and trdrddd.. should i set 1-1-1 for SR 2 dimms 8gb each??
Ah, low-effort background RAMblings while I sort out a TrueNAS server.
I can listen to Buildzoid just flame people on their memory timings all day! 😂😂💀
Hey Buildzoid, my timings at 1:09:04 on tREFI and tRFC have a reason. They just won’t go any better with 1.425V. On that kit default tRFC is stupidly high, per SPD 880 at 3200. Maybe that has something to do with it being a 16Gbit die, but i wouldn’t go lower than 672, at lower i had random errors in mem test so i stopped there. Same with the tREFI, i just kept increasing by 1024 until it threw errors . That random number is just 48 x 1024.
Have a nice day.
what does 1.45v get mean
@@phr3ui559 i put 1.425V in the bios settings. But what gets actually applied is 1.45V as messured later
@@CHA0SHACKER is that because of ur LLC or is it just the board misreporting?
@@phr3ui559 could be both. Don’t know
"Everything about this is AWFUL ... Disaster .... " 🤣🤣🤣🤣
"Oh ...Oh ... No ... Theres NINES ...." 😂😂😂😂
You so funny man 👍
i had some djr and got 4000cl15 but the rest of the timings were horrendous. tFAW 50, tRCD and tRP 22, 300ns rfc with 20000 REFI
tFAW 50 could crater performance to worse than JEDEC level. tFAW 28 at 2667 JEDEC isnt as bad as tFAW 50 at 4000.
th-cam.com/video/lDnTuYFv2KY/w-d-xo.html
FILLER
tRC = 73 ? NO FORGIVENESS 🤣🤣🤣🤣🤣🤣 41:22
Honestly my memory timings werent all that, either. making some changes now after watching this.
What DDR5 should WE be looking for AM5, which IC this time?
Hynix ICs do pretty well
We can borrow Buildzoid's knowledge in these videos, as there is a wide variety of setups submitted.
For example, I am looking for some knowledge on 4 DIMM Quad Rank JJR with Zen 2.
I had 6 there too. Ryzen DRAM Calculator proposed this to me
I was too late, again
arg 1 minute in and i already feel like there's something I did wrong with my ram, I need to start using twitter or something >.> see if I can get on here :P
I tried getting scl's to 2 on my B-die I played with the driver strength's a little, pumped in some extra voltage it will post and is stable with 3's but 2 is just no bueno :( (3900X, X570 Strix-E, 4X8GB SR B-die)
but then I also spent a bunch of time trying to reach above 3600mhz and...i don't think my 3900x wants to do more then 3600mhz...i mean I got 3733 pretty far , passed a couple of stability tests and then my PC started to bootloop for ???? reason and I could not get it to post with 3733, then spent some time on 3666 which eventually reached the same yea I don't wanna do this anymore point. 3600 still bulletproof so guess I'm sticking with that (tinkered with the voltages SOC/CCD/IOD as well CPU seemed...fine but then I doubt the ram is what's actually causing the issue..though interesting note that pumping more voltage in might at some point do nothing >.> should keep an eye on that).
I really need some help with my ram, I cannot do any overclocking with it at all! Not sure if I’m terrible with ram overclocking or if my ram’s just terrible (Corsair 3600 c18).
Typical Samsung C-die bin if yours is 3600 c18-22-22-42 at XMP settings. Those things lose stability if you push VDIMM above 1.4V . 1 of the worst chips for memory overclocking.
@@fleurdewin7958 they’re the timings I have, strange thing is taiphoon reports them as b-die. I can’t touch the timings or speed at all without it going into a boot loop.
@@thesilverwang False reading by Thaiphoon Burner. I had a similar kit like yours but it's a G-Skill Trident-Z . 1 way to verify it is to bring your TRFC down to 350 . If yours is a real B-die , it will boot, C-die will never boot at such low TRFC no matter what you do.
@@fleurdewin7958 it won’t even boot with trfc set at 400 🤣
Any chance of a vid for improved timings for Samsung ddr5 like the ddr5 Hynix vid?
Samsung DDR5 is really annoying to stabilize.
@@ActuallyHardcoreOverclocking ha ok
Given that I expect to completely skip DDR4, I'm sure the number of hours this channel's consumed of my life on DDR4 tuning, confirms I'm weird! \o/
Just skip ddr5. Ddr4 was best
@@dvr1337 Either you're new/young or just trollin'.
That's been said early in every generation ... at least since SIMMs were the standard, when I started with PCs.
@@ChrispyNut im 34...
I think some if those timings were deliberately to get you to bag on them 🤔
Response for intro: We actually do enjoy watching these videos. Keep it going 👍😄 @actually hardcore overclocking
1:26 tRTP could go way lower too.. my crucial ballistix max at 4000 xmp cl18 does trtp 10 rock stable, 8won't post no chance, 9did not try, won't try 9 probab 6:41 omg just realized i have the very same ram (8gigs one stick tho, 16 total, and it's B, not BL) gonna try these settinx roight now yay. Btw it's micron e (or h?) die.
Is there a point of DDR5 speeds on Zen 4 when its FCLK is still limited to 2000MHz on good samples and you'll never run it 1:1 with the UCLK and MCLK? Isn't the IF on Zen 4 a bottleneck for DDR5 since the IMC is connected through it to the CCDs even if the IMC runs at 3000MHz? DDR5 seems such a waste on Zen 4 and AMD might as well released it on AM4 where you could use DDR4 4000MHz and benefit from the lower latencies.
I know nothing of this. Is a multiple of ram clock speed beneficial or possible?
They may have done something to IMC to optimise latency
AMD says DDR5-6000 should be optimal for Ryzen 7000-series. That kind of implies FCLK could be 3000Mhz instead of 1900Mhz or 2000Mhz.
Amd could also use 384bit IF (not 256bit) at 2000Mhz so now 1:3 IF to memory speed is optimal, but I dought that.
15:50 I think its a troll or maybe just someone who has given up on overclocking stability. MUH HIII SPEEEDDZZZZZ!!!!
Too late by around 10 minutes, gfdi
👍👍👍
I have some micron D9XPF (MT40A2G8VA-062E:B); 4x16 SR kit that looks similar to th-cam.com/video/o3HpXNkrQqM/w-d-xo.html
I was able to pull in the timings to 16-18-18-18-32 with a bit of voltage at that same frequency of 3733, but trying to bring in trrd/twtr/tfaw/trfc always produced errors
A lost cause for benching, but really nice for capacity
Yo everyone should I buy a kit of 32gb(2x16) 4400Mbps cl19 rev.B single rank crucial ballistix max for $250cad? I will try to negotiate for $200 instead if this is worth it to try for the next timings vid.
@@shraf2kay somewhat good news those kits the dude is selling them for $200 instead, might push it down to 150 since these rev.B kit ain't Samsung B-die.
nooooooo i was to slow ;'(
to post my ones XD
Lol
On the topic of b-die voltage scaling, my kits of Team XTREME 4133CL18@1.4V refuse to stabilize at 1.45V, and refuse to post at 1.55V regardless of settings on a B550 Vision D. That said, they'll do 3733CL14 at like 1.38V.
if you're saying you're trying to do 4000+ on B550, it's probably the Memory Controller/Fabric to blame here, rather than the DRAM itself.
you're trying to push the CPU extremely hard for the Platform.
@@taiiat0 no, I'm just referring to voltage. My 3733cl14 profile works great at 1.38-1.42V, is unstable at 1.45V, and fails to post at 1.55V. If I run the IF desynced, I can run 4133 no problem and I've toyed around with 4800. I just can't scale the timings low enough to have it make any sense (not that it ever would but...)
@@williambrown1094
so yes exactly what i thought. that DRAM Voltage isn't your problem, it's the Memory Controller and/or Fabric. you need to focus on those and tweaking their Voltages to get them to be willing to accept what you're trying to do.
@@taiiat0 🤦♂️
I'm gonna play this video from time to time just to cope that people tried to brag about worse timings than mine.
15:00
Low effort is what we want!
Samsung “E” die on the 2666 ram
why does tWRRD go as low as 1 on Ryzen but is >20 on Intel?
intel and AMD implement the timings differently.
@@ActuallyHardcoreOverclocking is there any accurate reference for those? Funny how even same name does not guarantee aything.
@@ActuallyHardcoreOverclocking I am struggling to imagine how retarded the AMD naming is then
equalize it
All your videos are awesome but this one is even more awesome. I’m going to have to pin this one and come back to it. I sure would like you to take a look at mine and tell me what is is wrong. Lol
can I send you my Timings via E-Mail? I don't have Twitter or Facebook or Istagram or Snapchat
no
@@ActuallyHardcoreOverclocking tough, but understandable... Email is too personal way for something like that
Sadly all these ways to communicate you are kind of blocked in my country, for example.
But i still want to thank you for these videos as i managed to tighten up some timings i didn't bother to do before.
Understanding that y-cruncher is extremely sensitive to memory really simplified stability checking.
++ low effort content
I just run some stock 3600 cl18 clocks at 4000
34:16 Stock SOC voltage for AMD 5000 APU's is 1.25V. SOC also powers iGPU and that needs 1.25V for high iGPU clocks AMD ships those processors with.
I got an 5600G, that thing does do DDR4-4400, IF 2200 Mhz, iGPU 2200Mhz fully stable. DDR4-4533, IF 2266Mhz, iGPU 2200Mhz - does boot Windows, doesn't run Furmark (SOC is stock at 1.25V)
15:00-19:30 getting mad at 16Gbit ICs for having high tRFC
I was not aware that 16Gb ICs are allowed by spec to have a 550ns refresh.
Karhu RAM Test pro tip: use half of the available threads and your testing speed goes up by a few MB/s. You read it here first ;)
the reason that happens is because you're putting less load on the memory controller. You should every memory stress test with as many threads as your CPU has.
That's not a pro tip, that's just dumb.
Don't understand half the stuff your talking about, stay for the rambling.
Eh, don't have a twitter. I'd like to post, my set up had somehow degraded, and from doing this i.imgur.com/GTmCU4D.png or i.imgur.com/WTmZgPy.png, now can only do this i.imgur.com/p0L6U1v.png. 4133 straight up refuses booting, 4000 MHz is unstable and tRDRD of 7 unbootable. System Agent 1.32-1.45 V, VDDQ 1.4-1.5 V, VDIMM 1.56 V (wrote a range of tried voltages). Wonder if a contact frame would fix that.
rip