Question Can anyone help me "manipulate" my non-QVL DDR5 settings?

[Paul Anderegg]

I bought a G-Skill Z5 Neo EXPO 6000MHz CL30 RAM kit that is not QVF for Asus B series MB's. I can only get it to run without errors in Windows by either setting EXPO and frequency to 5800MHz, or setting AUTO and 6000MHz, which results in 6000MHz CL50 timings.

While looking at what the memory training put into the Asus BIOS boxes, the last few timings do not match. I am wondering if this is typical, or if mismatched timings could be the reason I am having problems. I was hoping to get 1:1 IF, but for now I am 2000MHz fabric with 2900MHz memory. I tried simply raising the CAS from CL30 to CL32, but this bricked the BIOS. If I post what works and what the training puts into the boxes, could someone maybe offer some suggestions of things to alter to get the sweet spot working until I hope Asus QVLs these unpopular and unknown Z5 Neo kits? I don't want to call this tuning, as it seems, maybe manipulation is a better word?

https://www.gskill.com/product/165/393/1661410171/F5-6000J3038F16GX2-TZ5N

Paul

 

[Karadjgne]

The QVL is qualified Vendor list, it's Not a qualified Ram list. Big difference.

Ram is manufactured by only a handful of companies worldwide, like Samsung, Nanya, SkHynix, Micron etc. They make ram for every vendor. All ram is is some silicon chiplets soldered to a pcb. With a different heatsink. Like the Chevy 350 was in an el camino, the chevelle, Camero, trucks, same idea.

The model number describes several things, F5-6000J3038F16GX2-TZ5N is your ram. Broken down;
F5 = Gskill DDR5
6000 = speed
J3038 = 30-38-38-96 timing set
F16GX2 = 16Gb x 2 (32Gb) kit
TZ5N = Heatsink identification (Trident-Z Neo)

Meaning your particular kit is identical to a F5-6000J3038F16GX2-TZ5R except that ram would have a red heatsink not the Neo. Just in DDR4, there's over 3000 different models of Trident-Z , many of which are identical, using the same Samsung or SkHynix ram, the only differences being F16GX2 might be F16GX4 and be a 4 stick kit, or be a black/silver or camo or rgb or Royal designated heatsink. Which changes the model number.

There's Zero point in Asus validating 3000 seperate models, kits, speeds, colors, Cas etc when testing 5 would do. If 6400 Cas 40 16Gb 4 stick works, everything below it will work. Same ram.

Now if you take all 3000+ gskill Trident-Z, add in the value, Rip-Jaws, Aries etc, you'd be looking at closer to 10,000 individual validated tests, just for Gskill, multiplied by the 100 or so different Vendors like Crucial, Adata, Kingston, Corsair you'd have a QVL of a million different entries, thousands on pages long, costing $millions to aquire all that ram, thousands of man-hours to test and compile, and would need to be done on all 12-15 or so different Asus motherboards in the DDR4 lineup.

Asus would go broke with every release change.

So instead, they'll test a handful of different, more popular vendors, get a decent mix of kit sizes and timings, a decent amount of SkHynix and Nanya and Micron etc tested.

In the end, you get a QVL that says 'We tested a bunch of different stuff, and it seems to work with these' and that's ALL it means. It's not a guarantee, it's not a 'only this ram' will work thing, it's not a 'until I hope Asus QVLs these unpopular and unknown Z5 Neo kits?' thing. That ram most likely Is tested, under a different model, different vendor, same ram different paint job.

You bios is doubtfully bricking. It's most likely going back through memory training after you made changes and made the ram 'worse' by changing Cas from 30 to 32. It's AM5. Expo is going to find the Optimal settings for the pc, regardless of actual Primary timings. What you aren't looking at is all 40 or so Secondary and Tertiary timings that have been tweaked for better performance, and that change to 32 now changes whether or not those 40 settings are good or not.

You are only looking at a partial picture. For instance, 3200 Cas 14 is almost identical in throughput as 3600 Cas 16 ram and faster overall than 3800 Cas 20 ram. Your timings set the speed and time for data to open the door, walk into the room, cross the room, open the other door and exit. The whole picture. A speed-walking Geisha might have the fastest feet walking of any human being, but her steps are so tiny that a teenager taking long and slow strides actually covers the same distance, faster. That's what Expo is doing. Finding the right amount of strides, right speeds, right timings for data to enter and exit the fastest, which may not always be just related to the xmp speed or Fclock/mclick/Uclock of the ram.

This is why DDR5 as yet hasn't really shown much improvement over higher speed DDR4, especially for gaming which uses lots of tiny files, the actual time it takes to enter/exit the ram is relatively close, a few nanoseconds difference at best. This happens every new DDR release, not improving until the ram matures, gets faster overall. DDR5's advantage is in bandwidth, it can shove more through at the same pace as DDR4, but for tiny files waiting on the cpu to use them, that's basically moot. A data Dam, speed it fills doesn't matter as much as how wide the flood gates are.

So let Expo do its thing, you'll not see any tangible advantages to manipulation unless you think you can out-manipulate the amd engineers who figured it all out in the first place.

 

[Paul Anderegg]

I have a very rudimentary understanding of RAM but do understand that most of the various similar kits are identical save for the label and timings that have been written onto the XMP for the MB to read. My old PC had cheap Vengeance DDR4 2666. I went to the LPX page of Corsair, and simply pulled up the timing page for the 3600 version, entered the timing values into the MB, and it booted right up at 3600. That is the extent of my RAM overclocking.

When I moved to Ryzen, I opted to get the EXPO kit because I did not want to invest 4 years in college level web courses to learn AMD RAM tuning. I am hoping if I can show what the MB trains at working 5800, vs what the EXPO timings put in that gives massive errors in OCCT, maybe suggestions on specific timing to modify could be made? I have until January 31st to return this kit and try some other C32 or C36 EXPO or XMP pair, figured I would try to get this working if possible first. Asus seems to only show QVL for EXPO kits on their expensive X670E MB's.

Maybe a guru can look at the numbers and say "yeah, there is your problem right there, change 130T to 148T problem solved", or something like that. The mismatched last few timings have me thinking maybe I have a bad stick...which is why I mentioned it because I do not know if that is normal or expected? CHA vs CHB not matching numbers

Paul

The QVL is qualified Vendor list, it's Not a qualified Ram list. Big difference.

Ram is manufactured by only a handful of companies worldwide, like Samsung, Nanya, SkHynix, Micron etc. They make ram for every vendor. All ram is is some silicon chiplets soldered to a pcb. With a different heatsink. Like the Chevy 350 was in an el camino, the chevelle, Camero, trucks, same idea.

The model number describes several things, F5-6000J3038F16GX2-TZ5N is your ram. Broken down;
F5 = Gskill DDR5
6000 = speed
J3038 = 30-38-38-96 timing set
F16GX2 = 16Gb x 2 (32Gb) kit
TZ5N = Heatsink identification (Trident-Z Neo)

Meaning your particular kit is identical to a F5-6000J3038F16GX2-TZ5R except that ram would have a red heatsink not the Neo. Just in DDR4, there's over 3000 different models of Trident-Z , many of which are identical, using the same Samsung or SkHynix ram, the only differences being F16GX2 might be F16GX4 and be a 4 stick kit, or be a black/silver or camo or rgb or Royal designated heatsink. Which changes the model number.

There's Zero point in Asus validating 3000 seperate models, kits, speeds, colors, Cas etc when testing 5 would do. If 6400 Cas 40 16Gb 4 stick works, everything below it will work. Same ram.

Now if you take all 3000+ gskill Trident-Z, add in the value, Rip-Jaws, Aries etc, you'd be looking at closer to 10,000 individual validated tests, just for Gskill, multiplied by the 100 or so different Vendors like Crucial, Adata, Kingston, Corsair you'd have a QVL of a million different entries, thousands on pages long, costing $millions to aquire all that ram, thousands of man-hours to test and compile, and would need to be done on all 12-15 or so different Asus motherboards in the DDR4 lineup.

Asus would go broke with every release change.

So instead, they'll test a handful of different, more popular vendors, get a decent mix of kit sizes and timings, a decent amount of SkHynix and Nanya and Micron etc tested.

In the end, you get a QVL that says 'We tested a bunch of different stuff, and it seems to work with these' and that's ALL it means. It's not a guarantee, it's not a 'only this ram' will work thing, it's not a 'until I hope Asus QVLs these unpopular and unknown Z5 Neo kits?' thing. That ram most likely Is tested, under a different model, different vendor, same ram different paint job.

You bios is doubtfully bricking. It's most likely going back through memory training after you made changes and made the ram 'worse' by changing Cas from 30 to 32. It's AM5. Expo is going to find the Optimal settings for the pc, regardless of actual Primary timings. What you aren't looking at is all 40 or so Secondary and Tertiary timings that have been tweaked for better performance, and that change to 32 now changes whether or not those 40 settings are good or not.

You are only looking at a partial picture. For instance, 3200 Cas 14 is almost identical in throughput as 3600 Cas 16 ram and faster overall than 3800 Cas 20 ram. Your timings set the speed and time for data to open the door, walk into the room, cross the room, open the other door and exit. The whole picture. A speed-walking Geisha might have the fastest feet walking of any human being, but her steps are so tiny that a teenager taking long and slow strides actually covers the same distance, faster. That's what Expo is doing. Finding the right amount of strides, right speeds, right timings for data to enter and exit the fastest, which may not always be just related to the xmp speed or Fclock/mclick/Uclock of the ram.

This is why DDR5 as yet hasn't really shown much improvement over higher speed DDR4, especially for gaming which uses lots of tiny files, the actual time it takes to enter/exit the ram is relatively close, a few nanoseconds difference at best. This happens every new DDR release, not improving until the ram matures, gets faster overall. DDR5's advantage is in bandwidth, it can shove more through at the same pace as DDR4, but for tiny files waiting on the cpu to use them, that's basically moot. A data Dam, speed it fills doesn't matter as much as how wide the flood gates are.

So let Expo do its thing, you'll not see any tangible advantages to manipulation unless you think you can out-manipulate the amd engineers who figured it all out in the first place.

 

[Paul Anderegg]

I switched the RAM sticks between slots and tried to boot, but the orange "RAM LED is lit on my motherboard for the past 10 minutes, and it will not post. How long does "RAM training" take when installing new RAM, as I expect the BIOS thinks this is new RAM since I swapped the sticks?

My old motherboard had Q codes, am pissed I cheaped out and got a TUF that doesn't, have no idea what the damn thing is doing when it becomes unresponsive like tis.

Paul

 

[Paul Anderegg]

Cleared CMOS with a screwdriver, and continued my unedjumacated RAM manipulation. It became apparent that the EXPO would pass OCCT after applying and F10 into Windows, but upon reboot, without fail, it would not pass OCCT again. Went back to the BIOS, changed to AUTO, booted back to the BIOS, changed to EXPO, booted again, back to BIOS and disabled that memory retraining option in the timings menu. Now it reboots and has passed like 4 20 minute OCCT tests so far.

Paul