Ryzen - Return of the Jedi

[I.nfraR.ed]

Some interesting RAM clocks Things are getting better...

3733 is the maximum I can get at the moment, regardless of timings. Mem has XMP at 3733, not sure if there's any connection.
Going to try a different multiplier soon.

[cdawall]

Some interesting RAM clocks Things are getting better...

3733 is the maximum I can get at the moment, regardless of timings. Mem has XMP at 3733, not sure if there's any connection.
Going to try a different multiplier soon.

That is the highest I was able to boot into windows at with my 3600 kit (albeit at looser timings than yours), I was not stable for usage so didn't bother posting when I did my run through. Have you had a chance to 32m those settings? I am curious if my chip sucked and what BIOS are you on?

[drmrlordx]

That's impressive. Are you running any SATA devices at that bclk? What did you have to tweak to get it running that high?

[I.nfraR.ed]

Bios 1001. Running SATA3 SSD (Adata S511).
Turned off everything I could in bios and also had to set maxmem in windows (Just like on Kaby), otherwise it was unstable.
VGA is proven to work with really high PCI-E clocks, so no problem here. I can max out PCI-E on my 790FXB-M2RSH and the card is still stable.
This is a "known" fact that 8xxx nvidia series handle this very well. Not sure about other series, haven't tested much when there's a proven solution.
On the other hand I could not find any single secondary voltage that helps with mem clocks.

I didn't check 32M, only ran R15 and aida64.

Things are definitely getting better, Asus being ahead of competition. Hopefully it is possible to add higher mem dividers and unlock the subtimings, so we can tweak on low bclk as well.

[chew*]

My ch6 is on the way. See if i can duplicate this with DR sticks. Being dual rank im not expecting as high clocks....rated speed @ rated voltage would be nice.

[vario]

@ I.nfraR.ed

Realistically (stable) speaking , how far someone with 4x8gb B-die can expect to get ? Freq wise and Cas wise ?
@3000 Cas 14 may be doable ? Or even 3200 ?

[chew*]

Hey infra can you do me a favor. No ref clock. Baby steps and boot. Try 14-14-14-32 default ram speed. Reboot then 2666. Reboot baby step down 14-12-12-12 reboot then 14-11-11-28. Keep stock volts. Once it posts reliably baby stepping down to those timings set your cpu clocks and voltages. Lmk how you make out.

[chew*]

Ok so I did some investigating.

I am NOT CLAIMING STABILITY yet. I am just merely testing the limits of the IMC with Dual Rank dimms.









As we can see in this screenshot I have either found the limit of ram @ 1.35v or a IMC limitation.

[cdawall]

I got a new board in. Turns out I still hate it. Even using 1001 for a BIOS it is still junk IMO. I have never seen shenanigans like this even with ES chips on prerelease AM3 boards.

[chew*]

I got a new board in. Turns out I still hate it. Even using 1001 for a BIOS it is still junk IMO. I have never seen shenanigans like this even with ES chips on prerelease AM3 boards.

Ill trade you a slightly alive gaming 5

[cdawall]

Ill trade you a slightly alive gaming 5

I am up to two dead CHVI's lol

[charged3800z24]

I love my Tomahawk! :P

[chew*]

I love my Tomahawk! :P

[cdawall]

Isn't that one of the ones that can't handle full wattage of an overclocked chip?

[chew*]

Isn't that one of the ones that can't handle full wattage of an overclocked chip?

Just about any B350 is weak overclocked but I can kick peoples azz on it with memory tuning, fun stuff.

[chew*]

Ok here goes Its ugly cas latency but working is better than not working.

ONCE AGAIN IMC EXPLORATION ONLY. NOT CLAIMING STABILITY

There appears to be a CL hole between 14-17 not to mention odds are not even applying with certain dividers currently.

I see no reason why these sticks can not do 2933 at cas 14-17 other than a hole.

I skipped 18-16-16 testing as I saw no point but I can revisit it later.







once again 18-10-10 was problematic.

Based on these findings and the above findings @ 2600 speeds I had a funny feeling that AMD killed the 10 subtiming support lvl.

I had a set of corsair platinums that after recent agesa update in bios refused to load at XMP 10-12-12-28 @ 2400. Hmmmmm...

I tested the theory loaded defaults shoved them in set timings 12-12-12-28. Boom posted right up just fine.



Next up on the testing is to run 32m PI again at the best 2666 timings and the best 2933 timings to see if raw speed plus fabric beats better timings with less speed and less fabric.

Oh and pc3200 with dual rank is on the list when i can figure out how to do it.

I have this funny feeling it will require the cpu slightly sub 0 but i could be wrong.

[I.nfraR.ed]

Nice to see you got the 2933 divider working.
There are many illogical things, certain combinations not working at all, a lot of "holes" in memory overclocking, not just the one in frequency.
I still can't get a certain pattern though. Sometimes retry button on CH6 does wonders after failed training.
I also get the feeling the CPU is "learning", lol. Something that didn't work a day ago, now works.

[sdlvx]

The memory speed holes are interesting, but it doesn't surprise me. We had holes like this with Piledrivers' base clock. It's a good find, I wouldn't have expected it to show up with memory timings.

[chew*]

Pretty much right now I'm just fooling around with locked ref clock board.

Seeing as many do not have ref clock it is relevant to those with Dual Rank dimms.

I'm fairly certain with a ref clock board I can "get around" certain barriers but am unsure if getting around them will be viable for 24/7 real world usage.

I may need to take a ride this weekend, go visit an old buddy, bring back an old single stage out of retirement.

[starheap]

So that basically confirms what I went through when I was getting my dual rank 2x16g 3200 tridantZ cl14 kit setup on my 1800x. I have a gigabyte gaming 5 mobo. I ended up with the exact same cl timing hole. So anything below cl 18 wouldn't post (I only tried down to 14) but 18 was fine. Kept the timings at 18 16 16 35 instead.

[chew*]

So that basically confirms what I went through when I was getting my dual rank 2x16g 3200 tridantZ cl14 kit setup on my 1800x. I have a gigabyte gaming 5 mobo. I ended up with the exact same cl timing hole. So anything below cl 18 wouldn't post (I only tried down to 14) but 18 was fine. Kept the timings at 18 16 16 35 instead.

Cool you just answered a question for me.

The next step was to find someone with your bin and flash that SPD to my sticks to see if it was somehow XMP related.

Now I don't have to do that.

Another question for you. Did it fail post trying 3200 or did it get by memory checks but have no video?

This board has no port 80 so I went caveman and installed a speaker.

3 beeps memory failure.

1 beep it boots

0 beeps your screwed clear cmos...

With pc3200 I got 1 beep but no video....so I am close but I am missing something.

I am almost 100% positive these timings and holes effect all vendors and is not vendor specific leading me to believe it is definitely in the AGESA and not vendor bios tuning.

Anyone with a "fixed ref clock board" is going to hit these same "walls" and "holes" although a "unicorn IMC" cpu may break this trend.

[starheap]

I just remember it not posting at 3200 at pretty much any timings and I tried some pretty insanely loose timings. I don't remember the exact details atm since I haven't tried 3200 since sometime last week.

Like it wouldn't evenpost at 3200 24 24 24 50 lol

[chew*]

I just remember it not posting at 3200 at pretty much any timings and I tried some pretty insanely loose timings. I don't remember the exact details atm since I haven't tried 3200 since sometime last week.

Like it wouldn't evenpost at 3200 24 24 24 50 lol

Yah would not surprise me. I think it is an IMC limitation which is why I want to try like -30 to 0 C temps.

If you do get bored and try again stay at same timings.

Watch the 4 leds near port 80. LMK if it gets past the mem check LED and goes to boot.

Mine is definitely passing memory checks.

[starheap]

Yah would not surprise me. I think it is an IMC limitation which is why I want to try like -30 to 0 C temps.

If you do get bored and try again stay at same timings.

Watch the 4 leds near port 80. LMK if it gets past the mem check LED and goes to boot.

Mine is definitely passing memory checks.

Interesting I'll have to take a closer look at that when I get home.

[cdawall]

Just about any B350 is weak overclocked but I can kick peoples azz on it with memory tuning, fun stuff.

I was starting to get the hang of memory tweaking the board yesterday I was able to do some testing with the memory at 3800, but the chip was severely hampered for overclock. I was only able to push it to 3.7-3.8 with the memory pushed that far.