passionne56

Tested 2nd and third options : N32 stability was worse. Up to now, I found that tRAS= 32 gives me the best result : N32 crashes after 36 min.

Sorry for the late reply : all the cores are far below 90°C : max temperature on Y-cruncher N32 or VST is 70°C (13900K delidded + direct-die watercooling). Furthermore, the RAM modules are watercooled too. tRAS = tRCD + tCWL + tWR : I tried this formula a few days ago but no improvement. DRAM_VDD(+) =1.525v & DRAM_VDDQ(-)=1.35 V : N32 crashed after 24 min. I had DRAM VDD = 1.51 V and DRAM VDDQ=1.37 V before and I crashed after 26 min. So similar result.

I lowered Vddq from 1.4 V to 1.37 V and VDD from 1.5 V to 1.49 V : y-cruncher N32 worked without any errors during 28 minutes then 1 error. I rebooted 3 times with the same parameters, memory training enabled and each time, N32 crashed in a few seconds (2 s - 15 s - 48 s) which confirms my problem : at 7800, my stability is random.

After much effort, I was able to be stable on Y-cruncher VST during 1 hour. But as I expected, after a reboot, VST is no longer stable, crashes after a few seconds. I finally understood the cause of this erratic behaviour : the memory training algorithm. To obtain the stability on VST during 1 h, I changed the memory timings in Windows with the tool Mem Tweak it, without rebooting. After finishing this optimization with success, I rebooted, changed the timings in the bios that I found stable and as expected, VST crashed after a few seconds in Windows. No memory timing, no voltages were on auto in the bios except RTLs which cannot be adjusted directly. The explanation is the following : on the reboot, after the timings modification, the memory training does not manage to find the good values for the hidden memory parameters. I rebooted 27 times, hoping that at each reboot, the memory training would find the good values for the hidden parameters but no success after 27 reboots : in most of the cases, VST crashed after a few seconds, the best training was VST without errors during 14 minutes (13th reboot). I cleared CMOS and it wasn't better. According to my experience, there are 2 solutions : - unhide all the hidden memory parameters in the bios to enable the user to enter values. And provide a Windows tool to see all the memory values without entering in the bios : with such a tool, when I was stable on VST during 1 h, I could have written all the memory values, entered them on the next reboot and disabled the memory training. I would have been stable. - in the saved bios profile, save the hidden parameters too. Only the visible ones are saved in the profiles. The Tachyon is a great motherboard in terms of hardware, but unfortunately is plagued by a bad memory training algorithm which prevents it from being used daily with very high memory frequency, in a stable manner.

OK, thank you. I came back to 7800 and I am optimizing the timings. Afterwards, I may try 8000 MT/s again with your advice.

The motherboard behaviour is erratic : I use Memtest86+ to find the voltages and the timings at 8000 MT/s on my 2x16 GB 5600 Klevv kit and I rely on the Memtest errors number to find the best parameters. Many times, with the same parameters, the result can be completely different : 4 errors on the whole Memtest86+ tests on one run and on another one, many errors. I reflashed the bios x6o since I suspected a bios corruption but same result. In order to remove this erratic behaviour, I adjusted every memory parameter except the 2 round trip latency ones : when I enter values on Round Trip Latency(CONA/DIMM0/Rank0) and Round Trip Latency(CONB/DIMM0/Rank0) parameters, I always fail to boot. The parameters I entered are the same chosen by the motherboard (64). I disabled the round trip latency training but I cannot boot. Consequently, my questions : - how to remove this erratic motherboard behaviour ? - why Round Trip Latency(CONA/DIMM0/Rank0) and Round Trip Latency(CONB/DIMM0/Rank0) prevent me to boot ?

Finally, I am stable at 7600 Mhz. I entered every timing except for the round trip latency. I would like to flash my parameters into the user XMP memory profile : I don't find where I can do that in the bios. And can Gskill memory be flashed ? I had a Kingston kit before and I was able to flash my memory profile into the XMP user one, on a MSI motherboard. I received 2 HMCG78AGBUA084N modules and the PMIC is locked at 1.425 V. Can it be unlocked ? What must I do ?

Finally, I am stable at 7600 Mhz. I entered every timing except for the round trip latency. I would like to flash my parameters into the user XMP memory profile : I don't find where I can do that in the bios. And can Gskill memory be flashed ? I had a Kingston kit before and I was able to flash my memory profile into the XMP user one, on a MSI motherboard.

I received a 32 GB Gskill DDR5-6600 Hynix A-die kit, removed the heatspreaders and cooled the RAM with a 140 mm fan. The stability is not permanent : With the X6h bios, I did not manage to obtain stability at 7200 Mhz. I flashed the motherboard with the F23a bios and managed to achieve stability at this frequency, validated with OCCT during 30 min. Then, I managed to be stable at 7600 Mhz still validated with OCCT during 46 min. OCCT is a reliable torture test and after such a long time, I am stable. After a few hours, I did OCCT again and many errors appeared after a few minutes (less than 10 min). I checked the parameters in the bios and tREFI was completely different than the value which brought me stability : above 70000 instead of 6435. I corrected the value by entering 6435, rebooted, entered in the bios again and the tREFi value did not change ! I switched off the PC, switched it on and in the bios, 6435 was applied for tREFi. I then checked stability again with OCCT but crash again after only a few minutes. Since the ambient temperature was not much higher (1-2°C higher), I flashed the F23a bios again, reloaded the 7600 profile which was validated as stable but again, OCCT fails after a few minutes whereas I was stable on this test during 46 min ! I don' t understand why I am not stable any longer : after 46 min without any errors on OCCT, I shouldn't obtain a crash after a few minutes of OCCT : this tool is reliable according to my experience (used since many years). May the training stage change values even if no parameter is modified in bios ? I see only this explanation.

Kingston M-die 5600 overclocked to 6400 1T with tight timings and 1.44 V, stable. Happy with this result. I tested 7000 with loose timings but it requested too much voltage. Now, I am waiting for a green chinese A-die kit, ordered a few days ago.

Ok, thank you.

OK, thank you.

OK, thank you. It is clearer for me. I will test F23A bios first.

Interesting : 1. Does everything work fine on a Z690 Tachyon with a Z790 Tachyon bios ? Indeed, the OC buttons on each motherboard are a little bit different and may not work properly after such a flash. 2. What are the advantages to flash the Z790 bios ? 3. The only method to flash this bios is to use Q-flash ? 4. If this bios is not satisfying, will I be able to flash back a Z690 bios ? I have just ordered the Z690 Tachyon, I did not receive it, hence my questions.