Post your Ryzen memory speeds!

There's only so many updates I can permit myself in one post :p

New day, new tests. I bumped the voltage one notch last night to 1.125V on SoC, but since it was already late I just put the system to sleep. Been doing stress test today, and while I don't know what period of time is needed in AIDA to qualify as having tested a system's total capacity... it's been 3hr exactly so far. Note: This is also with ProcODT at 48Ohm. If this ends up being stable in FO76 as well, at least I'll know and can then go back and raise the ODT to 53.3 to determine what (if any) impact it has.
 
There's only so many updates I can permit myself in one post :p

New day, new tests.
I know the feeling. I'm awaiting the rest of the watercooling replacements and a new case for the 2600 box and really start to eek out some goodness
 
Well I feel like 4hrs of AIDA64 Memory Stress Test and 4hrs of uninterrupted FO76 are pretty conclusive (with GPU fans below what I'd normally have it at, so it's probably 83C instead of 65-70C).

MSI X370 Titanium w/ BIOS 1.J0 and a Ry 7 1700X, 2x8GB kit of originally 15-15-15-35
CPU @ 38.5x 1.35V, LLC Mode 1
CPU-NB @ 1.125V, LLC Mode 1
DRAM @ 3200 13-13-13-30 43 CR 1T, ProcODT 48Ohm 1.41V, GearDownMode Disable, BankGroupSwap Disable (all other timings, see below image)
DRAM VREF Ch.A and Ch.B @ 0.720V

I think that's all the relevant info.... If it's not mentioned, it's Auto or Default. Even Spread Spectrum is enabled, for that matter. [All boxes at top with a Yellow Rectangle are Auto]

DDR4-3200 13-13-13-30.png



March 9 UPDATE: Just a small one to mention I once again was able to play FO76 for about 4.5hrs straight without a crash :D Yay!
(That's even with increasing my GPU's memory speed 50MHz! [200Mhz effective]) I think this may not have been stable after all. No big deal.


March 22 UPDATE: Many days with many hours of gaming and I've experienced zero instability (I only sleep my computer, so this is important). Super happy, and further surprised by this!
 
Last edited:
Well, I upgraded from my c6h/2700X to an x399m/1920X threadripper build, added two more sticks of the corsair lpx3000 ram from the OP and set xmp profile and it ran fine. To test stability I ran about 7 hours of the 24h of nurburgring in iracing straight in VR (I drove 3.5 hours of it), though that doesn't fully load the cpu, but is a good combined load. I'll do some photogrammetry in 3df zephyr which will load the CPU and GPU 100% to really test stability later this week.

That's quad channel on hynix ram on 1st gen ryzen, supposedly worst case scenario, and it works fine out of the box at 3000mhz cl15. Not sure what bios revision I'm even running yet, will probably spend some time this week tweaking it to see where the limits are.

Now to wait for core madness with zen2.
 
Last edited:
Well, I upgraded from my c6h/2700X to an x399m/1920X threadripper build, added two more sticks of the corsair lpx3000 ram from the OP and set xmp profile and it ran fine. To test stability I ran about 7 hours of the 24h of nurburgring in iracing straight in VR (I drove 3.5 hours of it), though that doesn't fully load the cpu, but is a good combined load. I'll do some photogrammetry in 3df zephyr which will load the CPU and GPU 100% to really test stability later this week.

That's quad channel on hynix ram on 1st gen ryzen, supposedly worst case scenario, and it works fine out of the box at 3000mhz cl15. Not sure what bios revision I'm even running yet, will probably spend some time this week tweaking it to see where the limits are.

Now to wait for core madness with zen2.
That sounds like a meh upgrade at best
 
Well, I upgraded from my c6h/2700X to an x399m/1920X threadripper build, added two more sticks of the corsair lpx3000 ram from the OP and set xmp profile and it ran fine. To test stability I ran about 7 hours of the 24h of nurburgring in iracing straight in VR (I drove 3.5 hours of it), though that doesn't fully load the cpu, but is a good combined load. I'll do some photogrammetry in 3df zephyr which will load the CPU and GPU 100% to really test stability later this week.

That's quad channel on hynix ram on 1st gen ryzen, supposedly worst case scenario, and it works fine out of the box at 3000mhz cl15. Not sure what bios revision I'm even running yet, will probably spend some time this week tweaking it to see where the limits are.

Now to wait for core madness with zen2.

Tr is not ryzen
 
Tr is not ryzen

Uh, what's that say in huge letters in the middle of the heat spreader? I can't quite read it. Plus, TR can be pretty finnicky due to being quad channel, especially first gen...More so than am4 with 2 ram sticks (which is what I was running before).
 

Attachments

  • 20190504_145747.jpg
    20190504_145747.jpg
    477.7 KB · Views: 0
Last edited:
That sounds like a meh upgrade at best

For gaming it is, but for photogrammetry it's not. Plus, at $299 it's basically the same cost as a 2700X, but you get an extra 4 cores/8 threads and only around 5% lower single core performance. It runs 3.7 all core and 4.2 on up to 4 cores, so pretty similar in lower threaded apps to the 2700X which ran 4.35 single and down to around 4.2 on 4 cores, 4.0 all cores. In highly threaded stuff it's ~35% faster. I sold the 2700X and board for what the chip and hsf cost, so the main expense was the board - and now I have the ability to go 32c in the future (or 64c, depending on what zen2 brings, which is why I didn't go for a higher end tr, this may be temporary).
 
Anybody have a cold boot fix?

I have a Crosshair VI, 2700x, and 2x 8GB B-die.

I have issues where as long as I keep the computer running, it is stable, but getting a boot after a shut down can be a real pain.

I am using the Stilt's safe 3200 mhz timings, and 1.4vdimm. I had been running 3200mhz ram, but recently can't get a boot above 3133 for the ram. Like I said, if I can get the computer to post it is stable for months (I never shut down). but my 3 year old likes to press illuminated buttons......
 
Anybody have a cold boot fix?

I have a Crosshair VI, 2700x, and 2x 8GB B-die.

I have issues where as long as I keep the computer running, it is stable, but getting a boot after a shut down can be a real pain.

I am using the Stilt's safe 3200 mhz timings, and 1.4vdimm. I had been running 3200mhz ram, but recently can't get a boot above 3133 for the ram. Like I said, if I can get the computer to post it is stable for months (I never shut down). but my 3 year old likes to press illuminated buttons......
Doesn't the Crosshair VI have voltage for DRAM during POSTing? (Training Voltage I think it's called?)
If so, bump that up to like 1.43V.

It'll raise the DRAM voltage to that during POST but then bring it back to the 1.4V afterwards.

Alternatively, have you tried increasing the CPU_NB (aka SOC) voltage a tiny bit? It may be stable, but help in that instance.
 
Anybody have a cold boot fix?

I have a Crosshair VI, 2700x, and 2x 8GB B-die.

I have issues where as long as I keep the computer running, it is stable, but getting a boot after a shut down can be a real pain.

I am using the Stilt's safe 3200 mhz timings, and 1.4vdimm. I had been running 3200mhz ram, but recently can't get a boot above 3133 for the ram. Like I said, if I can get the computer to post it is stable for months (I never shut down). but my 3 year old likes to press illuminated buttons......

Post the rest of your voltages.
 
X470 Gaming Pro Carbon with 2x8gb F4-3733C17D and Ryzen 1600 I can get 3466 14-14-14-34

X470 Gaming M7 with 2x8gb F4-3600C15D and Ryzen 2600x I can get 3533 14-15-14-36, this board even manages to run my XMP profile of 3600c15 with a slight bump in voltage
 
Currently at G.SKILL Trident Neo b-die 32GTZN OC 1:1 3800 15 15 15 28 1.42v. I basically used dram calculator to get it up to 3800 then tightened the timings from there with my limited understanding of ram OC.

I can run at these speeds 1.4v forHours in memtest86, but sometimes when my pc wake's from sleep it breaks IF so I tried bumping the volts up to 1.41 then 1.42v a bit for stability. This seems ok so far.
 
Last edited:
Real men run 3866 J/K LOL
DDR4 CL16 3866Mhz Single Rank RAM
1933Mhz Fabric clock (FCLK)
1933Mhz Memory Controller(UCLK=MEMCLK)
gUzgLFQ.png
 
Hynix AFR 3000C15 @ 2933. The settings are basically the DRam Calculator SafePreset with one modification from the Fast preset.
 

Attachments

  • a.png
    a.png
    72.4 KB · Views: 0
32gb of 3200 DDR4, 4 sticks rated for CAS 14, 8gb/stick Samsung B-Die. I only use the mem calc as a basic guideline, usually I get poorer memory speeds if I set the settings exactly as the mem calc prescribes. SOC voltage is at 1.15 (needed for stability with 4 sticks), mem voltage is at 1.43v. Did a 400% MemTestPro test -> No errors. As a note this configuration went over 1000% without errors for MemTestPro and Prime95 blend.

Mem3800.jpg
Just bought 64gb of DDR4, 4 sticks, 3466 Samsung B-Die -> Seems like the only B-Die I could find for such a low price. Everything indicates they are B-Die, speeds, MSI memory compatibility for MSI PRO TRX40 PRO 10G sTRX4 memory list, plus B-Die finder:
Edit: Now I am not so sure these are B-Die, the price indicates otherwise, GSkill can be very sneaky on changing chips while having a different color scheme. It could be E-Die, I guess I just have to see what arrives. If B-Die I got a great deal if not -> They should work nonetheless.
 
Last edited:
I really need to get some faster RAM.
Translating memory speeds into performance is really per application and use case. For Monero mining, using RandomX, it translates well with speed and faster timings -> significantly. For games, zero playing experience difference for me due to GPU and what resolution I am using, 3d rendering -> zero difference. I don't play games at 1080p and low settings just to show that my faster ram has some difference on 1% lows, those tests are really meaningless, playing on normal settings and resolution should be how it is done. 3d rendering is virtually just as fast at 2100mhz as 3800mhz in Cinebench. Large data sets, as in video editing in real time etc. could see some good benefits if you have enough cores. So when I build my Threadripper 3 rig, fast enough ram will be fine, faster would just be more for epeen or just experimentation -> Not going to waste $600+ for known B-Die ram, now I did spend $350 for 64gb which just might be B-Die but if it does 3466 and above and not - not a big deal.
 
Translating memory speeds into performance is really per application and use case. For Monero mining, using RandomX, it translates well with speed and faster timings -> significantly. For games, zero playing experience difference for me due to GPU and what resolution I am using, 3d rendering -> zero difference. I don't play games at 1080p and low settings just to show that my faster ram has some difference on 1% lows, those tests are really meaningless, playing on normal settings and resolution should be how it is done. 3d rendering is virtually just as fast at 2100mhz as 3800mhz in Cinebench. Large data sets, as in video editing in real time etc. could see some good benefits if you have enough cores. So when I build my Threadripper 3 rig, fast enough ram will be fine, faster would just be more for epeen or just experimentation -> Not going to waste $600+ for known B-Die ram, now I did spend $350 for 64gb which just might be B-Die but if it does 3466 and above and not - not a big deal.

Yeah the only way I am buying new Ram is if I was to build an entirely new system, or for some stupid reason needed 64GB which I don't. I did not want to go overboard when I bought it and got lucky that it runs great at 3800 instead of 3600.
 
Yeah the only way I am buying new Ram is if I was to build an entirely new system, or for some stupid reason needed 64GB which I don't. I did not want to go overboard when I bought it and got lucky that it runs great at 3800 instead of 3600.
I am planning to build a ThreadRipper system and have the same type of view. It was way smarter or more economical to get the 3200mhz Cas 14 -> B-Die and overclock it then buying the 4000mhz uber expensive ram, at least in my experience. The Uber fast ram costs are way beyond the benefits in my opinion but everyone is different.
 
I don't think my results are anything special for Samsung B-dies.

But I'm satisfied with my results on this cheap motherboard.

1582058321229.png
1582058339770.png


Timings as shown in Ryzen DRAM calculator 1.7.0

1582058941260.png
 
Last edited:
32gb of 3200 DDR4, 4 sticks rated for CAS 14, 8gb/stick Samsung B-Die. I only use the mem calc as a basic guideline, usually I get poorer memory speeds if I set the settings exactly as the mem calc prescribes. SOC voltage is at 1.15 (needed for stability with 4 sticks), mem voltage is at 1.43v. Did a 400% MemTestPro test -> No errors. As a note this configuration went over 1000% without errors for MemTestPro and Prime95 blend.

Just bought 64gb of DDR4, 4 sticks, 3466 Samsung B-Die -> Seems like the only B-Die I could find for such a low price. Everything indicates they are B-Die, speeds, MSI memory compatibility for MSI PRO TRX40 PRO 10G sTRX4 memory list, plus B-Die finder:
Edit: Now I am not so sure these are B-Die, the price indicates otherwise, GSkill can be very sneaky on changing chips while having a different color scheme. It could be E-Die, I guess I just have to see what arrives. If B-Die I got a great deal if not -> They should work nonetheless.
My 1900 FCLK settings on CH VI, 3900x, 4x8gb b-die.....


Capture.PNG
everything auto except 16-16-16-32 auto-tunes almost exactly like your settings except my tRFC=666. Did you key in manually 364?
 
My 1900 FCLK settings on CH VI, 3900x, 4x8gb b-die.....


View attachment 225294 everything auto except 16-16-16-32 auto-tunes almost exactly like your settings except my tRFC=666. Did you key in manually 364?
Yes, I keyed tRFC in. Nice to see that this board does very good with 4 dims, mine and yours. 3900x too, cool. Currently testing out the 64gb ram kit in the CHVI, they turned out to be Hynix ram and not Samsung, will show results later. First set I was able to get them to pass Memtest for over 400% XMP timings, 3466mhz. Testing second set now.
 
Yes, I keyed tRFC in. Nice to see that this board does very good with 4 dims, mine and yours. 3900x too, cool. Currently testing out the 64gb ram kit in the CHVI, they turned out to be Hynix ram and not Samsung, will show results later. First set I was able to get them to pass Memtest for over 400% XMP timings, 3466mhz. Testing second set now.

1900/3800 is much slower than 1800 or even 1600 in Cinebench R20.
 
1900/3800 is much slower than 1800 or even 1600 in Cinebench R20.
I will test that out, what do you mean much slower? How many percent? That was not the case before but believe you maybe right due to cache speed increasing. I will have to check this out.
 
I will test that out, what do you mean much slower? How many percent? That was not the case before but believe you maybe right due to cache speed increasing. I will have to check this out.
With all core at 4250/1.3V, I get 7627 at 1800mhz. I get in the 7500s at 1600mhz. Topped out in the 7200s/7300s across multiple runs at 1900.
 
With all core at 4250/1.3V, I get 7627 at 1800mhz. I get in the 7500s at 1600mhz. Topped out in the 7200s/7300s across multiple runs at 1900.
That is 1%-2% difference, 7550 1%, 7500 1.5% or so. I would call that insignificant myself.
 
MemKit.jpg

Quick rundown on 64gb GSkill kit testing. Only tested two modules at a time to ensure they would do XMP, 1.35v at 3466mhz by installing them into an X370 motherboard, ASUS CHVI in sig. Ram Kit tested:
https://www.newegg.com/g-skill-64gb-288-pin-ddr4-sdram/p/N82E16820232424?Item=N82E16820232424
First two sticks when installed would not go into the bios using memory slots A1/B1, this is per motherboard instruction for using two dims. ASUS safeboot, taking them out and in etc. When I traded slots to A2/B2, it booted into the bios without issue. Interesting since this is a T-Topology board it should really make a difference, not sure what the reason was.
In the bios with stock CPU settings with precision boost, configured memory to XMP profile, 3466mhz, 1.35v rated volts. It booted right into windows. Used Typhoon Burner to determine who made the DRAM and it turned out to be Hynix MRF. The Dram Calculator had some recommended weird unusable 0 settings seen below when setting to the ram type -> So I ignored those settings, the settings in the blue are the actual memory settings which are the XMP and board memory tuning.

ThaiphonDramCalc.png

Did MemTestPro test and after 800% had 2 errors. Adjusted procODT from 53.3ohms to 48ohms, then 60ohms and the error rate was worst each time, so the default value was the best for this motherboard. Went to CAD_BUS and set it to 24, 24, 30, 30, (slightly higher than default values). Memtest Pro then passed the first set of modules. I removed and installed the second set of modules and since memory training was already done, bios already configured, tuned memory set, it booted right into Windows. It too passed MemTestPro for over 400%
MemTest.png
The performance actually surprised me, better than I expected for 3466mhz with sorta loose timings, latency was higher than I would want:
XMR32gDualHynixDim.png

Since these are for a future Threadripper build, I did not spend time optimizing/OCing them, that from my past experience this can take days if not a few weeks to ensure the OC is good. Sitting on my shelf for now. I just wanted to make sure the ram was good so I had time to return them if not.
 
That is 1%-2% difference, 7550 1%, 7500 1.5% or so. I would call that insignificant myself.

7300 to 7627 is a 4.3% difference, which is in line with the latency difference between 1800 (67.2) and 1900 (64.2) as reported by AIDA. However it's going in the opposite direction, which may indicate instability.
Aren't we here because we want to get that extra 4%? :)
 
First two sticks when installed would not go into the bios using memory slots A1/B1, this is per motherboard instruction for using two dims. ASUS safeboot, taking them out and in etc. When I traded slots to A2/B2, it booted into the bios without issue. Interesting since this is a T-Topology board it should really make a difference, not sure what the reason was.

This board is well well well known to have compatibility and performance issues with non B Die memory. I spent another 100 to upgrade from E die dual rank modules to B die single rank to get past 2933. Happy to report it was successful :)
 
This board is well well well known to have compatibility and performance issues with non B Die memory. I spent another 100 to upgrade from E die dual rank modules to B die single rank to get past 2933. Happy to report it was successful :)
With the same board model as you CH6 Hero, my first set of dims were Hynix 3200mhz rated ram, single rank -> It would not go over 2666mhz -> Got the B-Die and it would do 3500mhz on the 1700x if OCing the fsb, that was with Ryzen 1, 1700x. Interestingly, same year, 2017, the BioStar B350 board (still in use) would do 3200mhz with 1.45v Hynix ram with a 1700x (I bought two 1700x, the same 1700x that could not do over 2666mhz on the CH6 was able to do 3200mhz on the Biostar board).

Ryzen had a shaky first year. Anyways this same board with Ryzen 3, newest bios (bios's have much improved in the last 3 years for this board) now can do 3800mhz, 4 dims vice 2, without OCing the fsb, with the same ram. I have not pushed it further. Also now Hynix ram on this board, dual rank, not sure if the same dram chips, does 3466mhz at XMP settings and voltages. Frankly this board is a champ, ASUS software was pure crap and very detrimental when first launched, other issue with this board is shakey temperature sensors which for some caused all sorts of havoc with fan speeds, mine is OK there.

I also have a B450 MSI board which has 3000mhz Corsair Hynix dims doing 3200mhz without issue with a Ryzen 2700 XMP profile. Ryzen 3 memory controller is really good and can handle much better different ram manufacturers.
 
Last edited:
With the same board model as you CH6 Hero, my first set of dims were Hynix 3200mhz rated ram, single rank -> It would not go over 2666mhz -> Got the B-Die and it would do 3500mhz on the 1700x if OCing the fsb, that was with Ryzen 1, 1700x. Interestingly, same year, 2017

There was a BIOS released some months after the March release date that got me from 2666 to 2933 on the dual rank Samsung E die 3200 C16 memory, but no faster, either on a 1700X or 2700.
I can't OC the FSB due to NVME drive.

I refused to be content with that and upgraded to single rank B die in advance of the 3900X.
I'm very happy the board can supply enough power to OC a 3900x to 4.25 and not to have to drop the 350 on a CH VIII.
 
There was a BIOS released some months after the March release date that got me from 2666 to 2933 on the dual rank Samsung E die 3200 C16 memory, but no faster, either on a 1700X or 2700.
I can't OC the FSB due to NVME drive.

I refused to be content with that and upgraded to single rank B die in advance of the 3900X.
I'm very happy the board can supply enough power to OC a 3900x to 4.25 and not to have to drop the 350 on a CH VIII.
Well at least the board supports dual rank, 32gb (two dims) at 3466mhz with Hynix ram, surprised that the E die dims did not punch past 2933mhz -> I would have made the same call. I am not sure how much further I could push the Hynix ram but not going to waste my time there since they will be going into another motherboard.
 
Since switching OS drives my latency increased from being just a fresh windows install with drivers to my long-lived intel OS I used in my 8700k gaming rig, still decent though as my latency only increased by 1.4ns.

bandicam 0403.jpg
bandicam 0404.jpg
 
I've never had an issue running 3600mhz, on both a Ryzen 3700x and a 3600x, both using Corsair LPX and Vengeance RGB. The 3600x hainvg 2x8 config and the 3700x using 4x8 @3600mhz CL18. That being said, I'm not sure if I hit the Ryzen's wall, because I could never get either to take a 1900mhz fclk, nor run the memory at 3800, even when loosening the timings.
 
Back
Top