RAM unstable after adding in a second pair of same type G.SKILL F4-3200C16D 16GB to 32GB

[Naelex]

Ryzen 7 2700x, MSI B450M Mortar, G.SKILL F4-3200C16D-16GTZR Trident Z RGB Series 16 GB (8 GB x 2) DDR4 3200 MHz PC4-25600

Running perfectly stable at XMP-3200 profile (voltage auto). I needed more RAM so bought another of the same kit, however with all 4 sticks on XMP-3200 I get weird browser tab crashes and eventually BSOD's.

I have tested the new pair on their own and they are also perfectly stable.

The original pair are Hynix, and I saw from Taiphoon the new pair is Samsung (supposedly better?) - could this be an issue using the pairs together?

As both pairs work fine in isolation at XMP-3200 (voltage auto), I am not sure why they do not run stably together at the same speed. Is it normal to have to lower the speed going from 2 to 4 sticks?

Is it better to set a fixed voltage (1.35v) rather than auto?

Am I better off sending the new kit back and just buying a new set? (kinda feels like a waste of the original 16GB as they are only 1.5 years old) - but i need the extra memory for WSL2/work.

Any advice much appreciated!

 

[TheoneandonlyMrK]

few issues here, first the CPU Integrated memory controller doesn't support 4 sticks at that speed.

Your four sticks are different chips.

The Xmp profile was made for two on an intel platform.

Not four, four needs different secondary and tertiary timings than two that's why they make kits of four.

Not impossible but Xmp use is, set the first four primary timing, up volts to 1.35 manually and leave other timings on auto and try it.

 

[Chrispy_]

Not impossible but Xmp use is, set the first four primary timing, up volts to 1.35 manually and leave other timings on auto and try it.

This.
It usually works when you can't use the XMP profile....

If you're still stuck after that, put both sticks from one kit into adjacent slots (ie, single-channel) so that channel of the memory controller is at least dealing with one type of RAM kit and timings. I'm not sure if there's science behind it or just luck but that voodoo worked when other options had failed and I had no choice but to mix different RAM (not even the same speed or vendor, to be fair).

 

[FinneousPJ]

Do note that two different sets of two sticks are not equivalent to one set of four sticks. That is to say, there is no manufacturer validation on two mismatched sets. If you can you should return the set and go for a 2x 16 GB or 4x 8 GB kit to ensure best functionality.

 

[Naelex]

Thanks all. I set voltage to 1.35 and primary timings manually at 3200 but only had instant crashes. As @Chrispy_ suggested I moved the sticks so each pair is in a single channel, set to 3133Mhz and 16-20-20-20-38 and it finally seems stable (20+mins of OCCT large still running). Wondering if these timings are sensible and will i lose much performance like this?

Edit: just BSOD'd after 20 mins - will try 18-20-20-38 but may well just return the kit and look at a native 32/64gb set :/ much more costly

 

[ThrashZone]

Hi,
Yeah if you can return do so
Best to buy a 2x16gb kit or 4x8gb kit and try for 3200c14 or 3600c16.

3200c16 is a very low bin if samsung.

 

[Naelex]

3200c16 is a very low bin if samsung.

Thanks for the advice, What does this mean exactly?

 

[ThrashZone]

Thanks for the advice, What does this mean exactly?

Hi,
Means it's not very good
Samsung if high bin for 3200 would be c14 with tight timings 14-14-14-34 not loose 16-18-18-38 which c16 is typical for and quite a few othe cl that also have loose timings
Goog samsung always has tight timings matching it's CL example would be c14 but c15 or c16... would be the same 15-15-... or 16-16-16...

Very high frequency may differ with timings but lower than 3600 and under are tight.

 

[Chrispy_]

3200 C16 isn't particularly great RAM but if cost is a big problem there's no real issue dialling it back to DDR4 3000 CL16. You'll be losing a bit of performance from the lower bandwidth and reduced FLCK from 1600 to 1500 but unless you're a 240Hz gamer or using bandwidth sensitive applications you're not going to be missing out on a whole lot - maybe 5% when gaming. Hell, there's a good chance you'll still be hitting the vsync on your monitor so it's irrelevant.

Ryzen vs RAM 2133 vs 2400 vs 2666 vs 3000 vs 3200MHz

What GPU/resolution are your running at anyway?

 

[thesmokingman]

The original pair are Hynix, and I saw from Taiphoon the new pair is Samsung (supposedly better?) - could this be an issue using the pairs together?

That's your issue. You cannot get to the next potential issue which is the higher density load on your IMC until you get the mixed ram issue sorted.

 

[vega22]

Could be the imc that needs more juice not the ram.

I would test the sets alone and then manually set them to the slackest sets timings and bump the imc volts.

 

[Naelex]

3200 C16 isn't particularly great RAM but if cost is a big problem there's no real issue dialling it back to DDR4 3000 CL16. You'll be losing a bit of performance from the lower bandwidth and reduced FLCK from 1600 to 1500 but unless you're a 240Hz gamer or using bandwidth sensitive applications you're not going to be missing out on a whole lot - maybe 5% when gaming. Hell, there's a good chance you'll still be hitting the vsync on your monitor so it's irrelevant.

Ryzen vs RAM 2133 vs 2400 vs 2666 vs 3000 vs 3200MHz

What GPU/resolution are your running at anyway?

Thanks, I have a Vega 56 and run 1440p/144hz. would 3000 CL16 be better than 3133/3200 CL18 then?

I've been playing valorant all night without issue at 3133Mhz, 18-20-20-38. Cost isn't necessarily an issue but 32gb for only 80 quid isn't bad vs at 200+ for a new 32gb set that isn't really longevity proof with ddr5 coming

Edit: Just had a random restart while web browsing :facepalm:

Have only a few more days to return these but it feels like an uphill battle and i dont want to have to worry about stability at all. Crucial do a 3600 C16 for 167 pounds would this be a good buy? Edit2: my Mobo says it supports up to 3466 only so would buying 3600 be pointless anyway