Which ram

[MiszS]

Im making a build with ryzen 5 7500F, and i'm not sure which ram to choose, i've got 3 options, for nearly the same price, cl30 6000MHz, cl32 6400MHz, or cl34 6800MHz, i know that ryzen cpus usually like lower frequency, but considering that the ram is the same price, would it be better to buy the faster ram, and run it at lower speed?

[podkall]

@Schnoz sounds like same RAMs to me roughly, at least speed wise, each 400Mhz increase increases latency by 2 also, so perhaps 6000Mhz would be the best pick

[podkall]

Just now, Schnoz said:

That's not necessarily how latency works. CAS latencies are shown in clock cycles; faster memory typically has higher CAS latency values but the same "real" latency, computed by CAS latency value / clockspeed. CL30 6000 MHz memory will have the same latency as CL34 6800MHz memory.

exactly, which is why 6000Mhz is better due to "sweet spot" reasons, and avoiding potential stability hindrences

[MiszS]

14 minutes ago, Schnoz said:

Ryzen 7000 CPUs have a "sweet-spot" memory frequency of 6000  MHz, since this is an integer multiple of the CPU's bus clock. However, it is possible to get more performance out of a non-6000 MHz kit by basically brute-forcing frequency and therefore memory bandwidth.

 

I'd say if they're at the same price, go for the 6800 MHz kit. If the 6000 MHz kit is noticeably cheaper, get that instead.

its 5$ cheaper lol. Will i be able to run the 6800MHz ram in 6000MHz cl 30 mode? or is the cl the same no matter frequency?

[RONOTHAN##]

1 minute ago, MiszS said:

its 5$ cheaper lol. Will i be able to run the 6800MHz ram in 6000MHz cl 30 mode? or is the cl the same no matter frequency?

Yes, though by the same token you'd also be able to run the 6000 CL30 kit at 6800 CL34. They're all going to be Hynix based, and Hynix DDR5 is very consistent with what settings it does. 

 

The 6800 kit technically has the advantage that it is guaranteed to do higher frequencies. There's two different die revisions by Hynix for 16GB DIMMs, A die and M die, where A die tends to clock significantly higher. 6800 is basically guaranteed A die, while 6000 still has a chance of being M die. Unless you're trying to get DDR5 7600+ though, that doesn't really matter, and I'd still go for the 6000 CL30 kit in this instance. 

[RONOTHAN##]

3 minutes ago, Schnoz said:

Though, I'd recommend first running it at 6800 MHz and running benchmarks to see if it's faster than 6000, which I'd assume it almost certainly is.

It won't be. The 2:1 mode required to get 6800 to work has such a massive performance disadvantage that unless you're at DDR5 7600+, it's just slower than 1:1 mode in basically everything, and by the time you get to 7600+ it gets really hard to keep stable without quite a bit of effort. 

 

6 minutes ago, Schnoz said:

Memory sticks have several different timings for different frequencies stored on them, so yes, you'll be able to run it at 6000 MHz.

It's not that common to have multiple XMP profiles like you're implying here. Yeah, you can run 6000 CL30 on that kit without much trouble, but you're almost certainly going to have to manually input those settings rather than just hit "XMP-6000"

[Somerandomtechyboi]

Id go for the 6400c32 kit

Should most likely be hynix a die and if the cpu doesnt do 6400 1:1 just downclock to ~6200

 

 

6000 has a higher chance of being hynix m die

 

6800 is guaranteed a die but under the assumption of not wanting to copy paste whatever hynix a die 6200-6400 profile you find (destroying xmp kits in the process if subs tightened) and do a few hours of stress tests to ensure stability (which will problably be stable anyways since this is a rather slow speed oc) i wouldnt reccomend it

 

Hence 6400 being the middle ground of (almost) guaranteed hynix a die so you can clock to 8000+ when the cpus start requiring those freqs but you can be lazy enough to only have to change freq (literally 1 value in bios) if cpu cant handle it or do nothing at all if the cpu can handle it in 1:1 which some chips are able to do

[MiszS]

6 minutes ago, Somerandomtechyboi said:

Id go for the 6400c32 kit

Should most likely be hynix a die and if the cpu doesnt do 6400 1:1 just downclock to ~6200

 

 

6000 has a higher chance of being hynix m die

 

6800 is guaranteed a die but under the assumption of not wanting to copy paste whatever hynix a die 6200-6400 profile you find (destroying xmp kits in the process if subs tightened) and do a few hours of stress tests to ensure stability (which is problably will be stable anyways since this is a rather slow speed oc) i wouldnt reccomend it

 

Hence 6400 being the middle ground of (almost) guaranteed hynix a die so you can clock to 8000+ when the cpus start requiring those freqs but you can be lazy enough to only have to change freq (literally 1 value in bios) if cpu cant handle it or do nothing at all if the cpu can handle it in 1:1 which some chips are able to do

im a bit confused, why is the 6800Mhz kit worse? also what do you mean by destroying xmp kits, and if its A die, does that mean i will be able to run it at more than 6800MHz?

[Somerandomtechyboi]

3 hours ago, Schnoz said:

Oh, just adding on to this because I can...I'd like to point out that guy's bio:

 

"Idiot 16 year old that does basically everything to save money (diying and fabricating shit that looks like a fire waiting to happen or will cut me if i dare touch it), generally does not give a shit about safe voltages or theory or any of that crap when overclocking, and prefers to do his own research and use educated assumptions rather than blindly listening to others and wasting money and performance as a result"

 

"not gonna be able to take off the damn waterblock (superglued) without nuking the cpu"

 

"1gb x3 team extreem dark 1066 6-6-6-18, clocks upto 1520mhz on p5q w 7-9-6-9 (nice) 2.38v" (For reference, the safe operating voltage of DDR3 is anywhere from 1.2-1.5V.)

 

Suffice to say...if you build a PC following his advice, I would probably need to call the local military, because congratulations, you just built an IED.

Yea thats well outdated

I should problably put in a note saying the bio is irrelevant

 

I think that was written when i was in 8th grade, just never bothered to update it at all and i still wont bother aside from adding a disclaimer, i have updated it from time to time but most of it is from 8th grade me

 

4 hours ago, Schnoz said:

Memory overclocking is insanely difficult to do correctly, and that person is...uh...not doing it correctly

Define correctly

Tightening cl? Tightening primaries? Literally ignoring the secondaries and tertiaries and expecting it to perform any better than an xmp kit?

 

4 hours ago, Schnoz said:

I don't overclock memory given how insanely difficult it is

Thats literally where all the fun is at nowadays when bclk doesnt even exist, im not even entertained with bclk overclocking let alone cpu overclocking where its just set your vcore set your multi and done

 

Slow speeds within reach of xmp are generally pretty easy aswell so as long as you are sticking to slower speeds it wont be that hard, most of the effort here is just tightening subs and primaries, and you can just find an oc profile for your ics online to just plug and play for a free perf boost

 

Only reason i can get such high freq with ease is because ddr3 is just plain easy to overclock, just set vdimm, imc volt, primaries, and done. I do have a feeling that i may struggle abit more trying to get ddr4 to high freqs when i get a platform for it, or it may just be the same thing as ddr3 except a few more voltages and values to set. Properly finishing off the overclock and tightening the subs so the performance is actually there and its not just a bloody frequency validation is more tedious than hard but oc profiles exist, ram oc only gets hard when you are trying to push to that bleeding edge, 95% is a breeze but getting that last 5% ranges from hard to absolute hell

 

Spoiler

Heres my freq validation just for fun which is what i assume you are referring too which is absolutely not a dailyable overclock mainly due to it literally not being stable, p95 large ffts just insta crashes if i try to even run it

 

Highest stable ddr3 oc so far has been 3000 so only 100 below max freq with those same gdie sticks, i honestly forgot the settings as i was just screwing around seeing whats the max freq i can get prime95 large ffts stable and not crashing but i think i was using about 2v vdimm to get it stable at 10-15-15-31 trfc 150, the rest of the subtimings are trash though as i just took a set of subtimings from the hwbot gdie thread and loosened them to ensure stability when running high freq and never bothered to actually tighten them as i dont think i stand a chance of even getting any decent perf when my board has a trash 3rd channel, no way id get any decent perf at only 2600 running the uncore at the same freq as the rams as opposed to 2100 ddr3 + 4200mhz uncore. I might aswell do a rerun of that overclock and properly tighten the subs and primaries just to get some performance data off it and compare vs whatever slow 1200-1600 and max uncore 2100-2200, maybe spi/wprime would do for a benchmark

 

Though if i did any perf testing itd problably be warped because im only running 2 sticks but i might aswell since ive got nothing better to do

 

3 hours ago, Schnoz said:

"1gb x3 team extreem dark 1066 6-6-6-18, clocks upto 1520mhz on p5q w 7-9-6-9 (nice) 2.38v" (For reference, the safe operating voltage of DDR3 is anywhere from 1.2-1.5V.)

Just noticed this but uhh that is not ddr3 but ddr2 so that is a pretty normal voltage

 

That was a particularly hard overclock to stabilize, the sticks did 1470 with ease, just set the primaries to 7-9-9-10 vdimm 2.2v and done, but 1520 was absolute hell to stabilize, it needed to stay really cool and needed a fan over the rams otherwise crash, vnb at 1.58v instead of 1.56v, and a bunch of subtimings needing to be loosened, i think that overclock took like 2 weeks to have some semblance of stability and even then it seems to last a day and a half before crashing in prime95 largeffts and i just gave up on absolute stability cause oh hell no i would rather not go insane so i think thats an example of an improper oc and more of a freq validation than an actual performance oc cause alot of those subs i had to loosen over the mobo auto set subs, seems like the last few mhz at the very bleeding edge are kinda like a diminishing returns zone on some ics and perf completely drops because of it so 1470 > 1520 cause subs are not nuked, read speed went from 8000 all the way down to 5000 iirc, now that i look at it this overclock was a complete waste of time and had basically nothing to show for it aside from the mostly stable high freq but hey i guess i had some fun even if i was being driven insane for the most part

 

 

p5q can set upto 2.9v but i simply dont need that much voltage as the sticks im using dont actually need high volt hence 2.38v instead of whatever 2.6v+ if i got an ic that actually scales with volt and i wanted to take ddr2 oc seriously again which i dont intend on doing cause theres like no info on ics so i have no idea what other ics aside from micron d9 are good

 

the x58a ud3r i am using does allow me to set 2.6v but ive only run that on my garbage f die stick just for shits and giggles and only a bios run at that since ive already tested a 2.46v overclock on that same stick with a p6t deluxe v2 (prime95 largeffts for 8 hours with no signs of degradation on both the cpu and ram stick), only other time ive gotten close was those hynix cfrs now running 2800c11 when trying to stabillize 3000, yea 2.3-2.5v did not help at all and they seem to have degraded slightly because of that 4 hour ish attempt as they were doing 2800 11-14-13 2.06v but now 11-14-15 at 2.1v, but running the things <2.2v seem to be fine as ive hammered them with p95 largeffts for a few hours without any sign of degradation so looks like my theory of ram ics only degrading when theyre pushed well out of their max volt they scale to is somewhat accurate but need more sticks to validate

 

 

I push voltage simply because i need it and i actually prefer lower voltage when possible, why the hell would i push 2v just for 2000c8 when i can run 1.5v at 2000c10 and lose basically no performance since its just cl being loosened, aside from being lazy cause i run the 6x4 d9qbj on my still yet to be sold gaming pc at 1900-2000 9-12-12 at 2v, im sure i could loosen to something like 11-12-12 or 11-15-15 and run at 1.5v or lower but i am just too lazy to bother screwing around with something that already works flawlessly, especially if i have to retune the damn subs cause ive already tightened those and i refuse to do all that again

 

4 hours ago, MiszS said:

im a bit confused, why is the 6800Mhz kit worse? also what do you mean by destroying xmp kits, and if its A die, does that mean i will be able to run it at more than 6800MHz?

Cant run 1:1 uclk or whatevers the terminology, performance will be shit because the rams are too slow to make up for non 1:1 uclk, youll only see benifits when you are pushing for 8000+ but that is just uneccesary pain and youd only do that for fun, id totally push and daily those freqs but only for fun, for a more daily oc you are best off sticking with xmp speeds unless you enjoy the fun and pain of ram overcloxking like me which most ppl dont (problably because it can drive you insane and takes absolutely forever to manually dial in a high freq oc especially without a reference oc profile, ask me how i know)

 

Reason i reccomend 6400c32 is so you can lazy set freq to ~6200 if the cpu cant run 6400 and when the time comes to upgrade cpu and that cpu requires fast rams you simply overclock the rams to whatever the new cpu requires, by then imcs and problably bioses have improved and what once was an annoying asf 8000 ddr5 overclock is now relatively easy since its within range of what xmp kits can do

 

There are actually a few instances of ppl on this forum looking for a ram upgrade only in frequency even though their current rams are still fine, like an old 2133-2666 ddr4 kit that can simply be overclocked to 3200-3800 (depending on ic) instead of having to go out and buy a 3200/3600 kit of the same capacity, hence why i reccomend hynix a die kits just so you dont have to upgrade purely for higher ram freq in the future and youll only need to copy paste an oc profile to enjoy your new cpu at most of its potential, think this is especially important now since i doubt most ppl will need more than 32gb even a few years from now unlike ddr4 where some ppl do find a need for 32gb of ram over 16gb