ddr4 silicon lottery win ?

[scruffy1]

i know the proof is in the benching, and i still can't drive zen2

but i got the itch to see what chips i got dealt for my set of G.skill F4-3200C16D-16GVKB, and lo and behold thaiphoon tells me this :

i believe the app reads my personal sticks to tell me their manufacturer, and so, is that a generic "good thing" as i understand it (samsung b-die seems revered by the current players)?

because as i have read the tealeaves it would appear i should be able to punish them to run at 3600 without having to loosen the timings much

but by all means educate me if you have experience with this platform, because i sure don't, although i have begun a cursory tinker with dram-calc and am still endeavouring to comprehend infinity fabric as a concept

i'm waiting on the extra bits of the hyper 212 black edition to arrive (the backplate was shipped to another ocau buyer still attached to its old motherboard ) and i have discovered a pair of unused gamer storm 4 pin fans that i can run push-pull on the heatsink to encourage the boost speed to throttle higher

it's exciting, like way back when i "discovered" that my corsair 2x256mb ddr was sporting bh5 chips    - thanks to the people here for enlightening me all those years ago; here's hoping i can lean on you for help this decade

trust life is travelling okay for you people in the current situation; going to work here seems to be less risky by the day thankfully, but it ain't over yet any time soon i reckon

[Bones]

Some of it will depend as always on the IMC of the chip, the 3xxx series Ryzens can do it with ease.

Anything older may or may not have trouble just getting there regardless of the sticks. Since yours are PC3200, could be a stretch but overall still highly possible to get it without changing much.

Yes, B die are the ones to look for overall, I can't say anything about the newest sticks.

[Macsbeach98]

All depends on the board and CPU as well as the memory but 3600 shouldnt be too much of a stretch.

[georgekokovinis]

First generation B-die chips on A0 /A1 pcb layout.
B grade means average overclocking potential.
A+ , A, B+ , B, e.t.c.

Sorry, but not a silicon lottery set of mem sticks.
Maybe good for Ryzen at 3600 / 14-14-14-28-1T at around 1.55 to 1.60V.
For latest Intel platforms it can be considered a solid daily set but nothing more.

[Macsbeach98]

Hey Ash your Corsair 2x 256mb sticks would the ones I just put into my NF7S for a run tomorrow or Sunday.

[scruffy1]

go the bh5 !

i can't believe what a great set of chips those were as my first ever "name brand" lucky dip

georgekokovinis : if they'll do 3600 at their rated cas i'm still ahead on points
this is my daily desktop, not a benching machine, so any free grunt is welcome

given they are paired with a ryzen 5 3600 (not an x), i feel they are a solid win for my needs

here's how i rate bog stock, according to this site (not that it's the uber insight site)

www.userbenchmark.com/UserRun/26810903

edit :  sorry i have been slack at digging out the dddr2 stash, pete

work is a bit exhausting presently, but give me time and i'll go hunting, and nudge me if i'm too slow for you

edit #2 : my post count here is pathetic; i need to overclock my contributing i think

[Macsbeach98]

No rush Ash on the digging stuff out at your leisure.

[alpi]

Not really need to force c14 on with zen2 for 24/7. Set at least 3600 mhz but should try 3733 - 3800, set c16 and focus on subs. Better performance can achieved that way.
Ofc with 1:1 Infinity clocks. Use my A0 Bdie stick at 3733 C16 daily. Pretty good performer.

[scruffy1]

late return to the party

that ram was disinclined to really go past the rated specs, and despite best effort i was stuck there albeit on 1.34vdimm

low and behold, my careful loitering over at ocau got me a pair of these siblings for $120aud delivered (that's like 87 greenbacks)

minimal adjustment later and they are doing 3600 16-16-16-32 1T as suggested by dram calc, at the same 1.34vdimm, 

it tells me 14-14-15-28  has potential at that speed, or up to 4110 at c16

i have been eyeing a 5600x too, but the bios upgrades supporting it aren't likely till january, and realistically i don't need to go any faster; the zen 2 is already crazy capable

regards to all you crazy clockers

it's great to see team amd kicking arse like the days of yore; i enjoyed intel for a long time without moving past the second series, but amd is the one i love to see going faster

[Mr.Scott]

Holy shit, it's Ash. What's goin on man?

[scruffy1]

hey scott

just mooching around down here in the onset of spring, and playing with zen2, and having just gotten the "new" b.die sticks my mind of course drifted to the heady days of amd superiority back when my t.bred was a baby

i send telepathy to you people quite often as you probably realise, but rarely get on the keyboard, being intrinsically lazy



most of my recent playing around has been with radio control toys, building / flashing / tinkering, and the occasional upgrade of household hardware, most of which is parked on 2nd gen intel with fast vintage bits

that and buying too many hifi components (nice headphones, a selection of dacs)


you people will always be family, only in many ways better than family - i can drift in and out with no significant responsibility, and still feel the love



best wishes to all the classic members of warp 9

[scruffy1]

oh boy oh boy

valid.x86.fr/btzjmz

3600 at 14-14-14-28 1T  with 1.424vdimm

so now the question is, am i safe at that voltage as a 24/7 setting ?

even if i have to revert to 16-16-16-36 1T at 1.34vdimm i am still a happy camper

edit - the c16 settings are rock stable, but the c14 will need tweaking and volts to pass all 13 tests after a failed run... so i reckon i'll tighten the subtimings at c16 and see how low the vdimm can sit and still be stable

after all it's putatively my "work" platform so stable beats e-peen

[Bones]

Normally anything under 1.50v's is OK for 24/7 use but note degregation could set in over time.
If you have active cooling (Fan) on them that certainly helps.

You could try 3600 15-15-15-31 @1.40v's if you want and see how it does, again with a small fan on them.

[georgekokovinis]

Nov 10, 2020 4:24:13 GMT -5 Bones said:

Normally anything under 1.50v's is OK for 24/7 use but note degregation could set in over time.
If you have active cooling (Fan) on them that certainly helps.

You could try 3600 15-15-15-31 @1.40v's if you want and see how it does, again with a small fan on them.

It is a lot easier to degrade or even kill the memory controller of any chip, than the sticks themselves.

Since industry presented ddr4 chips single side ( B-die ), three versions evolved.

A0 pcb layout, A1 and A2.

Currently most are A2 layout.

Here is my personal experience, stretching three sets of mems over any logical limit.

1) Galax Hall of Fame ( dual channel, A1 layout ), 4133 / 19-21-21-41-1T at 1.40V from the factory. These are the 2018 model, bright silver.

4200 / 12-11-11-28-1T ultra tight 2.03V used on Apex IX and X, for prolonged benching sessions of at least 5-6 hours each. IMC under cold.

These sticks have operated more than 500 hours under these conditions. They are alive and as new and have not lost their potential one bit.

2) G.Skill ( dual channel, A2 layout ), 4133 / 17-17-37-2T at 1.40V. Model year 2019.

4600 / 17-17-37-2T with 1.50V ( 24/7 ), on Apex XI.

4933 / 14-13-13-24-2T with 2.04V on Apex XI. Again more than 500 hours of benching. As new.

3) Quad channel ( 4 X 8gb ), G.Skill F4-4133GTZ ( 19-21-21-41-1T ).

Rampage VI extreme and Apex VI. 4000 / 15-16-16-36-1T with 1.50V rock stable 24/7. My main rig with I9-7980XE tuned for utmost stability.

4080 / 12-11-11-28-1T with 1.92V for benching. As new. The pc I am currently working on.


Conclusion : Be careful of VccIO and VccSA voltages that degrade and kill memory controllers and actually the chips themselves.

Except for some very cheap low quality mem sets, do not be afraid of voltages on the mem sticks.   

[scruffy1]

i dialled in all the sub settings recommended by 1usmus at c16 and i can wind the volts even lower

my long term strategy is always max clock at a minimal undervolt

does the vdimm on ryzen have any reactive changes in the cpu internal voltages ?   i like playing with settings, but the mortar max interface is clumsy, and most of the online tutorials seem to favour gigabyte or asus, and i haven't had an msi motherboard since the one i quickly exchanged for the asus a7n8x deluxe  on the good advice of my the bespoke builder

i liked the colour of the msi (bright red back them), but no soundstorm, and a very limited bios adjust

the mortar max however got great press as the one to get for a lot of features (components and cooling), but sadly the bios isn't its greatest asset, even though the no cpu flashing ability and 32mb bios chip are a big plus

my case is big and well cooled :

current build is ryzen 5 3600 / hyper 212 / msi b450m mortar / samsung 970 evo plus 500gb nvme/ wd 2tb / 2x8gb G.Skill ripjawsV @ 3600 / sonata hd pro dac + klipsch promedia 2.1 + alo pan am + focal elear + AKG Q701 / Lian Li PC-A17B & corsair hx750i & noctua fans / gigabyte gtx1060 oc 6g / dell u2515h

[scruffy1]

wrong place probably, but a quick question re the above setup


why does the zen2 3600 idle by default at ~ 3525 when the specs suggest 3600 nominally ?

like it's of no consequence as the thing dynamically throttles constantly, but.... why is it so ?

[Vinster]

is your freq stepping doen 2mhz?

Vin

[scruffy1]

Nov 11, 2020 21:37:35 GMT -5 Vinster said:

is your freq stepping doen 2mhz?

Vin

do what ?

i actually upped the bclk to 100.1 (actually reads 100.05) because i felt robbed when the ram read 1799

but the base speed is 3525 - or close enough even with the older 3200 ram and the newer 3600

it's just a mystery to me

[Vinster]

ah... so it's rounding off 1mhz... it's got to be a stepping thing then, and you're riding a threshold.

Vin

[alpi]

It's a C-state, depends on win actual energy scheme and the selected cpu min. If 100% there is untouched ratio but it's usually 0.25 lower then spec (36x = > 35.75x at You). If it's 99% 0,75 lower multi given so 35.25 at You. Thats why Your cpu idling at 3525.  

[alpi]

But whatever the freq shows, sleeping cores, C6 state, is still can be active.

[scruffy1]

mine isn't constantly at full revs, it's running whatever they call the current iteration of cool 'n' quiet

edit :  i get it !  i can set minimum at 100% and 3.6 is the minimum active clock

does 99% make it significantly cooler ? is the only benefit seeing the magic 36 in my system tray ?

on another note :

my g.skills run happily at 3600 in tightened sub-timings courtesy of 1usmus' ram calculator, at 16-16-16-32

they will boot and load windoze at c14, but are not memtest stable even at 1.45 in the bios, and i'm not keen on upping the ram voltage further; a little bit of tweaking of sub-voltages for the ram hasn't been any better, and i simply can't be arsed

so, two questions :

if i spend up on better chips, will i see any meaningful gains in performance at that speed but cl14 ? i am no hard core gamer, and benchmarking just for e-peen is no longer stimulating, as it were

if the ram runs memtest stable at 3600 c16, is there any downside from lowering voltage till it fails, given ddr4 is by default 1.2v, and currently (see what i did there) runs happily at 1.32v in the bios, and probably can go lower

tested : runs memtest without error at 1.296v in hwmonitor (1.27 in the msi bios)for c16

any downside given the volts for c14 unstable are waaaaaaaay higher ?



thanks for your counsel, team warp; the interweb seems to be full of opinions, none of which impress me with their knowledge of this subject

everyone wants to go big and fast and cooking; no-one seems to take this route less travelled

ps : it's been around 40c here the last 2 days (southerly change just dropped it to 24c) - not the best ambient temp to stress test really

[ShrimpBrime]

Hey Scruffy!

To answer your question about gains getting better or faster memory would be = No. You won't really notice any substantial gains from CL14 to CL16 unless the programming really leans on memory performance.

In some cases, I found through benching 3D earlier this year, a more relaxed timing set actually faired better than tight timings. At a certain point, the bandwidth is great enough the timings do little to help performance. I'm talking well and above 4000mhz memory frequency, CL 16 to CL 19.

If you lower the Ram voltage, that's ok so long as it's stable I think. Most of the time, the stability gets lost from running memory warmer, so the lower voltage is good for keeping temps lower.

AnyHoot - CL16 seems to be a pretty standard Cas latency and seems to fair well from 3200mhz to 3600mhz both in bandwidth and latency. Stick with what's stable I always say! Unless benching, then that's obviously different.

Running 1467mhz CL 14-16-16-36 on Gen 1 Zen Athlon 220ge. I think the voltage is on auto, so probably at 1.2v but I'd really have to double check that. Been stable for quite a while now, test OCCT stable. Good enough for me!

[scruffy1]

thanks my man

yeah, for my desktop machine i always have the urge to make it the best stable settings for what i've got, as fast as it will go at the least volts - temperatures here in summer mean less power = less sweaty in the study

[Vinster]

Nov 29, 2020 13:05:01 GMT -5 ShrimpBrime said:

Hey Scruffy!

To answer your question about gains getting better or faster memory would be = No. You won't really notice any substantial gains from CL14 to CL16 unless the programming really leans on memory performance.

In some cases, I found through benching 3D earlier this year, a more relaxed timing set actually faired better than tight timings. At a certain point, the bandwidth is great enough the timings do little to help performance. I'm talking well and above 4000mhz memory frequency, CL 16 to CL 19.

If you lower the Ram voltage, that's ok so long as it's stable I think. Most of the time, the stability gets lost from running memory warmer, so the lower voltage is good for keeping temps lower.

AnyHoot - CL16 seems to be a pretty standard Cas latency and seems to fair well from 3200mhz to 3600mhz both in bandwidth and latency. Stick with what's stable I always say! Unless benching, then that's obviously different.

Running 1467mhz CL 14-16-16-36 on Gen 1 Zen Athlon 220ge. I think the voltage is on auto, so probably at 1.2v but I'd really have to double check that. Been stable for quite a while now, test OCCT stable. Good enough for me!

I can confirm this a little. before my current bios bricked my ability to change timings on my new build (new bios released. hope to fix that now) I found that gaming at a higher frequency on memory with better timings yielded a difference in FPS that were negligible while in game play.

But I myself still want to go through the process to get the best I can get stable because in my head. it's not as good as it could be...

What I may do is get as high as I can go for bench it a bit and get reference scores.. then back them off a touch for 24/7 as I don't want to have to be a scientist every time I turn on my PC. But that's me on this build in particular. everyone else's mileage will vary

Vin

[ShrimpBrime]

For Zen, get the IF and MeM frequency as high as possible. Timings loose, best gains on that 1:1:1 ratio.