just use bandwidth to compensate with the latency and test it with games to see which gives you better performance. around 50ns is damn good if you can reach 70+K.

No point going very low latency but bandwidth also low. Need to find the balance. iinm if you playing with gear2, better go for 4500mhz+ else best you stay at gear 1 and tighten 3866mhz to 15 or 14.

Yea, sacrifice latency in Gear2. After I’m done with 4266, then I’ll try Gear1. From my quick try previously, max I can go with Gear1 is 3733. Likely not very good IMC.
So I feel that 4266 tight timing Gear2 can beat 3733 Gear1 or till I'm proven incorrect.

you have to run your own gaming benchmark liao yeah.... IMC luck, mine also not that good. need quite alot of SA/IO voltage to go further, albeit my 3200c14 seems to be a pretty decent kit.

I don’t game so much and even if I do, RAM definitely not the bottleneck because I run 4K resolution

oooooohhhh it does much more than you'd think. This is why despite AMD being the top in 1080P gaming, it actually kena beat down in 1440P and 4K due to bandwidth and latency.

I'm using 1st gen RTX 2080 only wor sure bottleneck 9 9

the ram can squeeze about another 1-5% wan... even in a gpu bottleneck situation. That's why it is quite significant, especially those competitive 3dmark port royal guys.

Hmm, maybe I should check 3dmark Port Royal score later.
And compare with 3733 Gear1 result.

port royal doesnt require high ram low latency

only timespy does.

4k ram influence is not much except in min frames and with dlss. intel favor bandwidth first.

There you go llaw sifu has spoken , dont have to crack your head so much on chasing low latencies. Prioritize bandwidth first.

This post has been edited by Bonchi: Jul 9 2021, 02:24 PM

LOL... if I want absolute bandwidth, then I should switch back to my previous Hynix DJR -- almost able to achieve 4800MHz. If I spend more time tweak it, should be doable.

Anyway, on B-die, I'm able to get 4266 17-17-17-37 @ 1.5V stable, passing few rounds of TM5 and also GSAT. The trick is to enable DIMM ODT training to find the best RTT_* value. After that, disable DIMM ODT training and manually fix the value so it doesn't change.

Only 1 small problem remain... upon cold boot when temperature is around 30C, will fail memtest. Need to warm up the RAM first before everything passing. So, I guess I need to reduce the tREFI, cannot use the max value.

errr thats not what it does and the odt doesnt change doing that..

Hmm... I’m consistently getting error with RTT_WR/PARK/NOM = 80/48/40. This is stock.
Once changed to 120/48/40, no more error. Got this value after enabling ODT training.
Strange thing is that only ChannelA change RTT_WR=120.
Perhaps my RAM is not perfectly identical.

hmmm odd.. not sure how much work sham did for hero but afaik not much..

most of it is carry over from extreme.. not individually tuned like apex.. but then again 80% of apex tuning has atleast 60% common sets from extreme...

here the other thing u can try

but it has to be when training ram at first time so it will affect dram vref.. (disable mrc fast)

set vtt to 0.725 but eventual 0.8-0.85v

see that helps or not with default odt.. also bro ure timings are kindda whack...

Here's my current stable config:



With both RTL training and ODT training enabled.
May I know which timing is whack?

Will try your suggestion later.
My actual goal is to boot 4533 with this kit. So far always fail training - stuck at postcode 55.

use memtweakit..

u will notice something

Hmm... this is what I have:



Care to tell which timing is out of place?

the public service i do...



sorry typo.. its "setting 0 doesnt mean zero, the min is 4"

This post has been edited by cstkl1: Jul 10 2021, 09:37 PM

Ah, okay. Thank you for pointing it out. Appreciate your tips
Time to study some specs document.

your values are a little off so the training result on the RTL is kinda extreme which gave you the stability issues and the reason you need to pump so much vdimm. Best is really just copy some of the values by owikh and ctskl and test it out first then tweak.