intel 13/14th gen fast RAM experiences: >7,000 MT/s is XMP stable!

[fusseli]

Welp I'm back to square 1. I updated Aorus Elite X Wifi7 bios to F5g over F5e and jesus, nothing is stable. Freeze, crash, dies in sleep. I'm reverted to slowass 7200 for now. Sigh. Gigabyte is messing with a lot of stuff behind the scenes in the memory timings settings. The bios descriptions are changing and so are the options, and the stability has fallen on its face.

 

[ShrimpBrime]

Welp I'm back to square 1. I updated Aorus Elite X Wifi7 bios to F5g over F5e and jesus, nothing is stable. Freeze, crash, dies in sleep. I'm reverted to slowass 7200 for now. Sigh. Gigabyte is messing with a lot of stuff behind the scenes in the memory timings settings. The bios descriptions are changing and so are the options, and the stability has fallen on its face.

Well if it's XMP 7200, why don't you increase the dram frequency/divider a little at a time manually. See how far the timings go, but a little at a time. So 7266mhz, test, then try 7300mhz, wash rinse repeat until it gets unstable.

I would fair to say at least 1.35v System Agent though. Maybe as low as 1.30v. I'd do 1.5v v-mem. Maybe the board will train better at smaller increments but with a little extra wattage to prevent droop.

So I'm saying start the manual OC over the XMP 7200mhz profile. Might just be the cpu prefers this particular divider. Could be it wants Cas a couple clicks looser.

Gotta put in some effort to get a result. My cheap corsair 3600 sticks happily clock 4500mhz.

 

[ir_cow]

Why not go back for BIOS version?

 

[Upgrayedd]

Welp I'm back to square 1. I updated Aorus Elite X Wifi7 bios to F5g over F5e and jesus, nothing is stable. Freeze, crash, dies in sleep. I'm reverted to slowass 7200 for now. Sigh. Gigabyte is messing with a lot of stuff behind the scenes in the memory timings settings. The bios descriptions are changing and so are the options, and the stability has fallen on its face.

Do you have fast boot off and training on?

My biggest problem was this thermal grizzly contact frame. Then disabling fast boot helped a lot. Then lowering the cache target helped too. I have 7600 stable without lowering cache ratio, 7800 needed a drop to 48. I'm honestly still messing with a few things. I've found stability though. I've got my cache ratio at 49 right now from 50.

 

[fusseli]

Do you have fast boot off and training on?

My biggest problem was this thermal grizzly contact frame. Then disabling fast boot helped a lot. Then lowering the cache target helped too. I have 7600 stable without lowering cache ratio, 7800 needed a drop to 48. I'm honestly still messing with a few things. I've found stability though. I've got my cache ratio at 49 right now from 50.

Yep training on and all of the many settings are on Auto and cpu is stock/Auto. The update to F5g bios is significantly unstable whereas before I was XMP stable with Spec Enhance.

Well if it's XMP 7200, why don't you increase the dram frequency/divider a little at a time manually. See how far the timings go, but a little at a time. So 7266mhz, test, then try 7300mhz, wash rinse repeat until it gets unstable.

I would fair to say at least 1.35v System Agent though. Maybe as low as 1.30v. I'd do 1.5v v-mem. Maybe the board will train better at smaller increments but with a little extra wattage to prevent droop.

So I'm saying start the manual OC over the XMP 7200mhz profile. Might just be the cpu prefers this particular divider. Could be it wants Cas a couple clicks looser.

Gotta put in some effort to get a result. My cheap corsair 3600 sticks happily clock 4500mhz.

This kit is gskill 7800 C36 46-46-125-171, it passed stress tests stock/XMP prior to the bios update, right now I'm 7200 c38 and it's freezing overnight idle

My previous ram kit was corsair 7200 C34 44-44-96-140

Why not go back for BIOS version?

Gigabyte doesn't have all versions available for download, unless there's another way that I'm unaware of.

 

[ShrimpBrime]

Yep training on and all of the many settings are on Auto and cpu is stock/Auto. The update to F5g bios is significantly unstable whereas before I was XMP stable with Spec Enhance.


This kit is gskill 7800 C36 46-46-125-171, it passed stress tests stock/XMP prior to the bios update, right now I'm 7200 c38 and it's freezing overnight idle

My previous ram kit was corsair 7200 C34 44-44-96-140


Gigabyte doesn't have all versions available for download, unless there's another way that I'm unaware of.

Hate when bios flash wrecks stability and overclocking.

Have you tried only loosen Cas to 36 or 38 maybe? 1T vs 2T command rate? Ect.

 

[ir_cow]

Have you tried only loosen Cas to 36 or 38 maybe? 1T vs 2T command rate? Ect.

Command Rate 1 does not work on DDR5 that I am aware of.

 

[ShrimpBrime]

Command Rate 1 does not work on DDR5 that I am aware of.

It may not. I don't have my ddr5 platform up atm. Got some 6000mhz Gskill Z5 I need to test soon though.

 

[fusseli]

Hate when bios flash wrecks stability and overclocking.

Have you tried only loosen Cas to 36 or 38 maybe? 1T vs 2T command rate? Ect.

Yeah presently loose to 7200 c38 and it froze idle overnight. Sigh.

 

[Upgrayedd]

Yeah presently loose to 7200 c38 and it froze idle overnight. Sigh.

I just passed memtest86 with 7800 using XMP Tweak. XMP1 was considerably faster than XMP2 and now Tweak is faster than XMP1.
Currently testing 8000 Tweak.

 

[chrcoluk]

Having seen in person and done an install myself, the contact frames will not damage a board unless you decide to sneeze all over everything or had way too much alcohol. They're VERY easy to install and the 14900k desktop with the frame on that my wife uses has zero issues. Actually that rig does everything and more perfectly.

If anything they protect a board as the default socket frame requires quite a lot of force which risks warp.

Okay the gloves are off. I’ve added an absolutely necessary Noctua fan to my ram

I did run maxed trefi for some months on my 8600k, when I got annoyed the cpu+board combo couldnt do 3200mhz, I then setup an uber 3000CL12 configuration which included max trefi, it had a eye watering effect on one particular game final fantasy 13-2 which was prone to a stutter when a monster was about so spawn in the field. Bizarrely this totally fixed it. For a while I thought it was primarily the tighter timings, but when I thought about it some more it didnt make a lot of sense. So on my 9900k which was running on the looser timed 3200CL14 configuration, the stutters were back so I kept the timings and memory clocks the same but maxed trefi, and wow, ff13-2 was able to spawn in monsters again without a stutter.

However my dimms have no temp sensors and high trefi is vulnerable to high temperatures so I backed out of it.

Good idea you got them cooled if you boosting trefi.

7200MT XMP just works, and using Buildzoid timings (which HUB showed in their video) and I spent less than 2mins to get the most out of my RAM kit
View attachment 327054

I use Phantek T30 on the top as intake to blow air over the DIMM, PMIC temp only goes to 42C in game

Nice share, for reference the timings are at this part of the video, your part shows the results, so I will post a new link.

Goes to show it seems even on DDR5 same as DDR4 tuned timings mean more than faster dimms at XMP.

 

[Toothless]

If anything they protect a board as the default socket frame requires quite a lot of force which risks warp.

The stock frames are built well enough, with the backplate to where the board isn't going to warp. Literally never heard of that happening.

If a NH-D15 isn't going to bend a board, the stock retention bracket isn't.

 

[ir_cow]

@fusseli have you tried to manually set the CPU voltages for mem? Sometimes between BIOS updates someone goofs and breaks the auto stuff.

 

[fusseli]

Yep, can confirm, Gigabyte fucked up the F5g bios bad. Literally nothing is stable, not even 5600. So I'm running JEDEC atm

@fusseli have you tried to manually set the CPU voltages for mem? Sometimes between BIOS updates someone goofs and breaks the auto stuff.

Great idea. On my z490 board I always had to set dram voltage because auto/xmp didn't actually apply the 1.45v.

 

[ir_cow]

What about going down to F4?

 

[fusseli]

What about going down to F4?

Wow I just checked and there's already a new F5 out, released monday, Christmas day. Guess I'll try that.

 

[dgianstefani]

I didn't waste money on a $60 Thermal grizzly frame. Paid $7 for the Thermalright one. Bought 4 in total (yes all are being used). All the same. X pattern until finger tight. I've probably done it 30-40 times now at least. I'm Nov it was about 10 trying different 14900Ks and motherboards.

Never once did "overtightened" happen. The same can't be said for waterblocks. Those monoblocks that screw in from the back give me trouble all the time.

Unrelated to the contact frame, I personally wouldn't go past those voltages you have already for long term use. Truthfully, no one knows how long it will last. Could be 1 year or 10 years. Some of those crazy XoC guys go with 1.4+ on the SA all day long, but they aren't planning on keeping it past the next socket either.

I've noticed what I call a roll over voltage (aka playing dead). For example if I put in 1.4v SA after a few of hours I'll start to get errors no matter if the freq is 6000 or 8400. Same goes with the other two. 1.45 and it will get mad at me eventually. Combine all three and it's like 30 minutes.

For most of the 12th gen I thought it was the MB and memory giving me grief. Nope, just the CPU.

I think the all these CPUs just like the IMC have different limits. no one can agree what's "safe" or produces errors.

Yup, too much voltage can lead to instability not just not enough voltage. Not heat related either. Even if you aren't having instability, especially on X3D chips, too much voltage will lead to reduced CPU power budget which also hurts performance.

 

[iameatingjam]

I was able to get my hynix m kit stable at 6800 but the timings had to be pretty high, ended up going down to 6400 cl30 and ran that for quite a while on my 13600kf and then my first 14700kf.

Then when my replacement 14700kf came in, that same config was no longer stable. It was really close, but anta777 would spit out 1-3 errors for the entire 3 hour thing. I'm not sure if it was the new cpu or the new bios (also a gigabyte board) but I don't really care. I went down to 6000 cl32, that way I can decrease voltage from 1.5 to 1.35 and I don't have to change any of the cpu-side voltages (having one cpu die within a month has made me pretty apprehensive about fucking with things).

Brought trefi down from 60,000 to 40,000 too ( @ir_cow tells me there's not much performance benefit past that anyway) and everything seems stable now and still pretty fast, in passmark my score dropped from 4500 to 4400. w/e.

 

[ir_cow]

There is zero performance to be gained past 65K tREFi. Well... dgianstefani did the math for me once, its like 0.3% more time between cell refreshes if you go from 65k to 262K and that is not worth it.

 

[InVasMani]

I'd recommend having x9 tREFI amount divide evenly with tREFI value so it syncs evenly with refresh intervals. Something like (x9 tREFI 126) x 504 = tREFI 63504 though if you've got stability issues you can drop things lower like (x9 tREFI 105) x 504 = tREFI 52920 or (x9 tREFI 105) x 441 = tREFI 46305. The lower you drop x9 tREFI and tREFI the easier it'll be on stability, but worse on performance. It'll keep things more consistant if it subdivides syncronizing evenly much like beats, bars, and measures with music.

If you're on the verge of stability and instability you might be able to fine tune it some if you aren't you'll need to be much more aggressive on lower both most likely though especially tREFI to reach a stability point and even then it's no guarentee based on a number of other variables.

I have things semi stable at about 1.35 MC and VID the SA is about 0.925v doesn't seem to have any overly "obvious" impact on stability for me with 4x16GB 7000MT/s CL34. Browsing and video streaming seems stable while gaming not 100%, but did play ROR2 for about 30-40minutes until the game crashed. The system didn't blue screen though just a game crash. I haven't really had much time to try to pinpoint the instability, but probably MC or VID just needs a nudge more voltage.

I haven't done a over abundance of gaming testing so far with iGPU nor discrete, but the little I did with iGPU left me moderately impressed by it. It's better than I anticipated out of it. I don't really plan to use it, but the option of it is nice and it could be put to use for some lower resolution co-op or side action gaming very conceivably which is nice actually.

 

[ir_cow]

I have things semi stable at about 1.35 MC and VID the SA is about 0.925v doesn't seem to have any overly "obvious" impact on stability for me with 4x16GB 7000MT/s CL34. Browsing and video streaming seems stable while gaming not 100%, but did play ROR2 for about 30-40minutes until the game crashed. The system didn't blue screen though just a game crash. I haven't really had much time to try to pinpoint the instability, but probably MC or VID just needs a nudge more voltage.

The official name for that type of instability is called "4DIMM and Not enough voltage" lol.

 

[Upgrayedd]

XMP1 8000 passed . I've got a few more things to test but I believe I'm about there.

 

[InVasMani]

The official name for that type of instability is called "4DIMM and Not enough voltage" lol.

It's SR at least so might not be so bad. I'm hoping pushing things up a bit further to 1.375v to 1.4v will do the trick.

 

[fusseli]

I'm back in business after another bios update, thank god! I'm also manually forcing cpu VDDQ/VDD2 to 1.35v and ram VDD/VDDQ to 1.45v. I know Gigabyte is rolling in new changes to the memory timing systems on this board so I will kinda start over on the tuning process. It was a little touch and go there for a minute. But I also willingly signed up for this by getting a brand new z790 Elite X Wifi7 board, when I installed win11 I had to manually install drivers on the command line in win11 setup because it is so new.... that was definitely a sign.

I'd recommend having x9 tREFI amount divide evenly with tREFI value so it syncs evenly with refresh intervals. Something like (x9 tREFI 126) x 504 = tREFI 63504 though if you've got stability issues you can drop things lower like (x9 tREFI 105) x 504 = tREFI 52920 or (x9 tREFI 105) x 441 = tREFI 46305. The lower you drop x9 tREFI and tREFI the easier it'll be on stability, but worse on performance. It'll keep things more consistant if it subdivides syncronizing evenly much like beats, bars, and measures with music.

If you're on the verge of stability and instability you might be able to fine tune it some if you aren't you'll need to be much more aggressive on lower both most likely though especially tREFI to reach a stability point and even then it's no guarentee based on a number of other variables.

I have things semi stable at about 1.35 MC and VID the SA is about 0.925v doesn't seem to have any overly "obvious" impact on stability for me with 4x16GB 7000MT/s CL34. Browsing and video streaming seems stable while gaming not 100%, but did play ROR2 for about 30-40minutes until the game crashed. The system didn't blue screen though just a game crash. I haven't really had much time to try to pinpoint the instability, but probably MC or VID just needs a nudge more voltage.

I haven't done a over abundance of gaming testing so far with iGPU nor discrete, but the little I did with iGPU left me moderately impressed by it. It's better than I anticipated out of it. I don't really plan to use it, but the option of it is nice and it could be put to use for some lower resolution co-op or side action gaming very conceivably which is nice actually.

Thanks for the suggestion I'm going to try this first

 

[Upgrayedd]

I was getting a weird consistent frametime spike every 1 second in metro exodus, even in the intro video. Changing to XMP1 from XMP2 solved it. It also doesn't spike when using XMP Tweak.
I didn't try any other games but I can if anybody wants to know if it's just metro or if it's global.