Question Max ram speed Ryzen 2600x

[bfollett]

I'm about to put together a new computer build with an Asrock B450M Pro4 motherboard and a Ryzen 2600x. The computer store had a good deal on an open package of G.Skill 3600 cl16 memory so I thought I'd future proof a bit and get the faster memory, but from what I've read the 2000 series Ryzens cpus probably won't work with ram at that high a speed, so I'm going to underclock the ram. I see lots of references of them running at 3200 cl14. Is that my best option or can someone suggest a higher setting that other's have succeeded with on 2000 series cpus. Also any recommendation for a good memory test program. I've been out of the game for awhile.

 

[AngelTech]

3200MHz is the max your board can handle.

 

[extreme_noob]

Which specific set of dimms? If its g skill ripjaws 3600 16 19 19 39, they do have a well priced 3200 16 18 18 38 kit as well. But you could try underclocking and tightening timings of the 3600mhz, just dont set xmp, boot into windows, and get thaiphoon burner and ryzen dram calculator to tell what settings would work, then tune them down.

 

[extreme_noob]

Ryzen is latency hungry, not so much frequency, so tightening 3600 c16 down to 3200c14 would help tremendously. Even if you cant hit 3200c14, you should be able to get 3200c16 with much better subtimings, which would help performance a lot.

 

[bfollett]

The ram is the 3600 c16-19-19-19-39. The manual for the motherboard lists 3200+ overclock support but I'm not sure what that "+" means. Will it allow settings of 3466 or 3600 and would the ryzen 2600+ memory controller even work at those higher speeds? If I do underclock to 3200, what cas latencies should I try? And thanks for the software ram testing suggestions.

 

[extreme_noob]

The ram is the 3600 c16-19-19-19-39. The manual for the motherboard lists 3200+ overclock support but I'm not sure what that "+" means. Will it allow settings of 3466 or 3600 and would the ryzen 2600+ memory controller even work at those higher speeds? If I do underclock to 3200, what cas latencies should I try? And thanks for the software ram testing suggestions.

Motherboard may support 3600, but your 2nd gen CPU most likely won't. Use Thaiphoon burner to find the RAM IC's, export to HTML file, then import into Ryzen DRAM calculator, click R-XMP at 3600Mhz, then specify 3200Mhz and choose fast or safe preset. It will tell you approximately which settings will work.

 

[bfollett]

Well I put my new system together. I first decided to try just lowering the frequency. I'm running stable now with the ram underclocked to 3466 c16. The computer booted to Windows when set to the full 3600 c16 but soon crashed. I wasn't surprised by that, I'd heard the Ryzen 2000 series memory controller would have problems at 3600. My question now would be if I should bother trying slower frequencies with lower cas latencies. While researching this I found someone who posted some formula Cas/Freq*1000. The lower the number(closer to 4.0) the better.
16/3466X1000=4.61 vs 14/3200X1000=4.375 By that formula it looks like the 3200 c14 would be the better option but I need to use the earlier mentioned programs to see what the ram will support, and would the results even be tangible.

 

[extreme_noob]

Well I put my new system together. I first decided to try just lowering the frequency. I'm running stable now with the ram underclocked to 3466 c16. The computer booted to Windows when set to the full 3600 c16 but soon crashed. I wasn't surprised by that, I'd heard the Ryzen 2000 series memory controller would have problems at 3600. My question now would be if I should bother trying slower frequencies with lower cas latencies. While researching this I found someone who posted some formula Cas/Freq*1000. The lower the number(closer to 4.0) the better.
16/3466X1000=4.61 vs 14/3200X1000=4.375 By that formula it looks like the 3200 c14 would be the better option but I need to use the earlier mentioned programs to see what the ram will support, and would the results even be tangible.

3200c14 is worth a shot. But what does dram calculator say? Also, you can get a lot more performance by spending time tightening subs. Heres a guide from the guy who made dram calc: https://www.techpowerup.com/review/amd-ryzen-memory-tweaking-overclocking-guide/

 

[bfollett]

Well I have a lot of learning to do before I understand exactly how to use thaiphoon burner and ryzen dram calculator. My first attempt at 3200 c14 14 14 14 34 failed to post but 3200 c14 15 15 15 36 does. I probably have too many questions to ask from here, but is there a recommended memory speed test program I can download to see whether my modified 3200 c14 is faster than the 3466 c16? Is there a tutorial on using thaiphoon burner and ryzen dram calculator together?

 

[extreme_noob]

Well I have a lot of learning to do before I understand exactly how to use thaiphoon burner and ryzen dram calculator. My first attempt at 3200 c14 14 14 14 34 failed to post but 3200 c14 15 15 15 36 does. I probably have too many questions to ask from here, but is there a recommended memory speed test program I can download to see whether my modified 3200 c14 is faster than the 3466 c16? Is there a tutorial on using thaiphoon burner and ryzen dram calculator together?

First off, learn how to use dram calc and thaiphoon. They are really helpful. Open, thaiphoon burner. It should tell you what to do to read. Them export to html file. Open ryzen dram calc and import the html. Select 3600mhz and click "R-XMP". Then set to 3200 and click safe or fast. Screenshot and print. You may need to tweak those settings a bit (I for one had to increse soc voltage to 1.1V to get it to post).

For testing ram, I suggest memtest86 for stability tests and aida64 extreme for performance testing (there is a free trial).

Also, subs will affect performance a lot, so make sure you tweak secondaries and tertiaries, not just primaries. Check the guide I linked above.

 

[bfollett]

Well lots of experimenting with the different clock speeds and latencies suggested by Dram calculator. My best stable results are at 3266 with a rather strange CL14 config of C14-17-18-28. I wonder how that compares against real CL14 3200 ram. The membench built into Dram Calculator yields a Time of ~130 on easy mode.

Thanks for the help,

Bob

 

[extreme_noob]

Whenever you have time, try tightening the subs. DRAM Calc timings are not final, so tightening them down can lead to performance gains.

 

[Crosslhs82x2]

Hi Guys
I was able to run my g.skill trident z royal gold 2x8 3600 cl16 samsung b-dies that is on my mobo qvl for a 3000 series cpu, while still running a 2600x on a Asus rog strix x470 f gaming.
I bought the ram 1st knowing it was on the qvl while waitng for the 3600x to drop in price and deciding which mobo to put the 2600x on for my son's upgrade.

The trident z royal gold of mine is Samsung b-die 16 16 16 16 36 and ran with the 2600x @ 3600 mhz @1.45v using docp or xmp and manually setting timings with
Proc 60
Rtt_nom off
Rtt_wr dynamic off
Rtt_park auto
All other settings on auto.
If i changed anything in proc or rtt's it wouldn't run.
The system ran the settings above which was confirmed with
Aida64 extreme 8-10 hrs
Memtest86 ran 3 different times with free version with pass each time.
Asus realbench stress test with 8gb tested of my 16gb. 8hrs
Also 8-10 hrs of p95 with Darkbreezes custom settings.

So either way 3200 or 3600
Stress test

 

[extreme_noob]

Hi Guys
I was able to run my g.skill trident z royal gold 2x8 3600 cl16 samsung b-dies that is on my mobo qvl for a 3000 series cpu, while still running a 2600x on a Asus rog strix x470 f gaming.
I bought the ram 1st knowing it was on the qvl while waitng for the 3600x to drop in price and deciding which mobo to put the 2600x on for my son's upgrade.

The trident z royal gold of mine is Samsung b-die 16 16 16 16 36 and ran with the 2600x @ 3600 mhz @1.45v using docp or xmp and manually setting timings with
Proc 60
Rtt_nom off
Rtt_wr dynamic off
Rtt_park auto
All other settings on auto.
If i changed anything in proc or rtt's it wouldn't run.
The system ran the settings above which was confirmed with
Aida64 extreme 8-10 hrs
Memtest86 ran 3 different times with free version with pass each time.
Asus realbench stress test with 8gb tested of my 16gb. 8hrs
Also 8-10 hrs of p95 with Darkbreezes custom settings.

So either way 3200 or 3600
Stress test

3600MT/s is definitely possible, but also not guaranteed on ryzen 2nd gen. But 60 ProcODT seems a bit high.

 

[Crosslhs82x2]

I really didn't expect it to run as well as it did, but had to get my curiosity fix.

 

[bfollett]

extreme_noob, I did as you suggested and tried to tweak a few of the main timings a bit lower.
So I went from 3266 C14-17-18-28 to C14-15-17-28. I think I'm happy with my decision to buy the 3600 CL16 ram and underclock it. Right now 3200 CL14 16GB kit ram is running $125 and up on Newegg. I got the open box 3600 ram for $67 and it seems to underclock comparably to the 3200 ram costing twice as much.

 

[extreme_noob]

extreme_noob, I did as you suggested and tried to tweak a few of the main timings a bit lower.
So I went from 3266 C14-17-18-28 to C14-15-17-28. I think I'm happy with my decision to buy the 3600 CL16 ram and underclock it. Right now 3200 CL14 16GB kit ram is running $125 and up on Newegg. I got the open box 3600 ram for $67 and it seems to underclock comparably to the 3200 ram costing twice as much.

That's great. If you can, try tightening tRAS and tRC as low as they can go. Those will impact latency the most out of the primary timings. Again, if you have time, I suggest tightening the subs, but even I am a bit too impatient for that.
The reason 3200C14 is so expensive is because those are the XMP timings and they are guaranteed B die, which should overclock to 3600C14 without much trouble. They probably also have better subs.

 

[bfollett]

Cost was a pretty big factor in my purchasing choice. My previous motherboard was sporting just a FX8320e and was really starting to show it's age, even just loading web pages was getting painful. About a month ago that motherboard started occasionally failing to post at boot up. That's when I started looking for a good buy on an upgraded motherboard/cpu/memory. Not to plug any company but I got my replacement parts at Micro Center. I was originally looking at the Ryzen 5 3600, but they currently have such a great deal on the 2600x I couldn't pass it up. Just $99.99 and if purchased with a certain motherboards the combo is discounted another $20. Fortunately they had the motherboard I was interested in as well. So for $160+ I was considering paying for the Ryzen 3600, I got the 2600x and the Asrock motherboard. The inexpensive open package ram was just the icing on the cake.

 

[Crosslhs82x2]

I Love MicroCenter

 

[bfollett]

One last question (I hope). As I tried all the various frequencies and latencies suggested by DRAM Calculator, I foolishly paid no attention to the fact that the suggested voltage was changing with the frequency changes. That may have been why I couldn't get beyond a frequency of 3266 CL14. I'd left my voltage at 1.35v during all my testing. So I upped the voltage to the 1.40v suggested for 3333 and it now runs stable. I'd like to try 3400 CL14 but the calculator suggests 1.44 and a max of 1.45 My motherboard regulator doesn't allow 1.44 so I would have to set it to 1.45v Is that pushing the safety limits for the DRAM? I don't want to burn them out prematurely.

 

[extreme_noob]

One last question (I hope). As I tried all the various frequencies and latencies suggested by DRAM Calculator, I foolishly paid no attention to the fact that the suggested voltage was changing with the frequency changes. That may have been why I couldn't get beyond a frequency of 3266 CL14. I'd left my voltage at 1.35v during all my testing. So I upped the voltage to the 1.40v suggested for 3333 and it now runs stable. I'd like to try 3400 CL14 but the calculator suggests 1.44 and a max of 1.45 My motherboard regulator doesn't allow 1.44 so I would have to set it to 1.45v Is that pushing the safety limits for the DRAM? I don't want to burn them out prematurely.

What dies are your ram? I wouldn't really suggest 1.45V for daily on most dies. Hynix AFR and Samsung B die are pretty voltage resilient, so will do it fine. The rest may or may not degrade. Technically, JEDEC specifies that all DDR4 is required to be able to run at 1.5V, but they never specify for how long. I'd stay as close to 1.4V as possible on most dies. I'd say stick to 3266 or 3333 and try tightening the timings and subtimings.
Another voltage to pay attention to is SOC voltage. Increasing SOC voltage may help with getting tighter timings or if you get just a couple errors in memtest. Generally the limit is 1.2V, but for daily, I'd suggest 1.15V limit. Also try playing around with ProcODT. On better boards, lower ProcODT will have a similar effect to increasing SOC voltage.

 

[Crosslhs82x2]

1.40v if stable is still safe for most g.skill.
For ryzen dram calculator for my 3600
With my 2600x was 1.45v to 1.48v which I ran it that way 1.45v till I got my 3600x.

 

[bfollett]

Well, I booted it at 1.45v 3400 Cas14 and ran the membench found in Dram calculator a couple of times and it didn't generate any errors. I could try it again at 1.40v but that is below the 1.43v that Dram calculator suggests as a minimum for for that frequency/cas. The memory is Hynix DJR. I guess I'll play around with a bit more. Bumping up to 3400 from 3333 reduced the runtime of membench from 128 seconds to 126 seconds. 3266 took 130 seconds.

 

[extreme_noob]

Well, I booted it at 1.45v 3400 Cas14 and ran the membench found in Dram calculator a couple of times and it didn't generate any errors. I could try it again at 1.40v but that is below the 1.43v that Dram calculator suggests as a minimum for for that frequency/cas. The memory is Hynix DJR. I guess I'll play around with a bit more. Bumping up to 3400 from 3333 reduced the runtime of membench from 128 seconds to 126 seconds. 3266 took 130 seconds.

If you want to be safe, running active cooling (a small fan) on your memory can help keep longevity when running high voltages (a lot of voltage related degradation is electromigration, which can change with temperature). So 1.45V with a 40mm fan or something will work fine for daily use.

I still suggest using AIDA64 for memory benching.

 

[Crosslhs82x2]

I didn't call it good on my 2600x or the 3600x memery tests good till I could run
Aida64 extreme, realbench stress test or prime95 for 8-10hrs each.
You can run hwinfo64 to monitor your ram temps also during.
My ram temps stayed 30 to 32c
But I do have a corsair h110i with noctua IPPC 2x140mm fans set as exhaust pull air by the ram.