Team Group 4000 CL16 - help with timings and overclocking

[korzychxp]

I have sticks like author of this post: https://www.overclockers.com/forums...ddr4-4000-cl16-tdzfd416g4000hc16cdc01.799149/

But author didn't post all of his timings to set them (zentimings program).

I can't push them more than 4400mhz cl18 with SOC 1.25v on Ryzen 5500 (AMD).

4200Mhz CL16 have not good performance. My Crucial Ballistix 3800 cl16 were faster with latencies.

 

[Woomack]

Typically I don't waste time on more than secondary timings and those are similar for most memory kits that are using the same IC. The tested Team Group kit was maybe 30+ Samsung B kit that I tested, so it wasn't anything new back then (more detailed timings were in previous reviews). If a memory kit is more interesting or wasn't built on an IC I tested multiple times before, then I run longer and more detailed tests. However, usually I have a problem with time, and testing the whole timing table is time-consuming, so I focus on things that may help a larger audience and show what they can count on buying a specific memory kit/PN. As long as I have time then I'm always open to additional tests (if I still have the tested memory kit) and answering questions.
Another thing is that most secondary and tertiary timings give you 0-0.5% performance gain, so it's a pure waste of time out of competitive benchmarking. Sadly, competitive benchmarking died a while ago.

On Samsung B, you want to stay at 3800-4000 max (4000 assuming it runs at 1:1 IF/IMC ratio). The next step would be 4800+ at tight timings, but it's hard to stabilize and on many motherboards, it won't even boot. It also works like the higher CPU, the better IMC. If you run Ryzen 5500 then probably it doesn't have the best IMC (I can be wrong as it's never guaranteed). I'm just saying it may work, but don't get too high hopes on very high memory clocks.
Your Ballistix kit had better latencies because of IMC/IF ratios. Once it moves to 1:2 (automatically at 3733+ on most motherboards) then it gets like 10-15ns hit and to drop it, you have to significantly raise the frequency, keeping quite tight timings. As I already mentioned, it won't work well on Ryzen 5000 (non-APU).

I recommend trying 3733-3866 CL14-13-13 or 14-14-14 as it gives the best results in daily usage and games. To beat it in games at a 1:2 ratio, you will need 5000+ CL18, which is barely possible on the best IMC. If you had Ryzen 5000 APU then 5000+ would be possible.

 

[korzychxp]

If you run Ryzen 5500 then probably it doesn't have the best IMC (I can be wrong as it's never guaranteed). I'm just saying it may work, but don't get too high hopes on very high memory clocks.
Your Ballistix kit had better latencies because of IMC/IF ratios. Once it moves to 1:2 (automatically at 3733+ on most motherboards) then it gets like 10-15ns hit and to drop it, you have to significantly raise the frequency, keeping quite tight timings. As I already mentioned, it won't work well on Ryzen 5000 (non-APU).

I recommend trying 3733-3866 CL14-13-13 or 14-14-14 as it gives the best results in daily usage and games. To beat it in games at a 1:2 ratio, you will need 5000+ CL18, which is barely possible on the best IMC. If you had Ryzen 5000 APU then 5000+ would be possible.

Ryzen 5000 has same chiplet as APUs.

Unfortunately, I compare Crucial and these as synchronous. I don't know why Crucials are faster. Mostly latency.

Actually i got 4200Mhz IF1:1 SOC 1,1v and CL16 and i can go 4400Mhz CL16 but with SOC 1,25v with is too much i think.

That's why I asked about Zen Timings because there are tensions and other things set up.

 

[Woomack]

A regular Ryzen 5000 has significantly worse IMC than 4000/5000 APU. On APU you can set memory clock up to 5800 and IF/IMC ratio 1:1 up to 4600. I was actually running a 5700G APU with RAM at 5400 24/7 for some months.

I have no idea what AMD has changed in Ryzen 5500, but in most cases, on a Ryzen 5600X/5800X/5900X you won't even boot past 3866 at 1:1.
The software does not always show correct values, but I'm not saying it doesn't work like you say. It just suggests that if you see worse latency at ~3600 than at 4000+, and timings are not too relaxed, it doesn't work at 1:1.

Anyway, I don't have any full-timing lists at 4000-4400 on AMD, and I don't have any DDR4 rig for tests anymore. However, Samsung B flooded the web, so I'm sure you will find something on the latest AMD or Intel chipsets (Intel timings should be similar). It doesn't have to be Team Group RAM, as many other brands are using similar Samsung B IC. Probably the most popular are G.Skill kits.
I'm at work right now, but from my head, tRRDS 4, tRRDL 6, tFAW 16, tRTP 7-8. Other things look fine or are not important. Try to set tREFI as high as you can, as it affects latency more than most other timings. It will probably have a limit of 65k, but if your motherboard has higher, then try higher. Too high may cause instability, but it's not a rule.

You can browse my reviews list, but I can't see anything that will directly help you.

 

[JLK03F150]

I have no idea what AMD has changed in Ryzen 5500, but in most cases, on a Ryzen 5600X/5800X/5900X you won't even boot past 3866 at 1:1.

The 5500 is basically a Cezanne APU (5600G), just with the GPU disabled.

 

[Woomack]

The 5500 is basically a Cezanne APU (5600G), just with the GPU disabled.

Thanks, somehow I missed this info.

 

[korzychxp]

Can't go better i think.

https://imgur.com/oKwBub1

As for tREFI on AMD it isn't settable. It's correlacted with trfc and i can't go lower without loosing stability.

 

[Woomack]

I wasn't using AM4 for a long time. I just put "quick settings" from my head when I was at work. For sure tREFI is available on AM5 and every Intel in the last years.

 

[Zerileous]

I was debating doing this (5500) a while ago for the cheep thrills to see what I could get out of my b-die kits. It seems it doesn't do as well as some of the other APUs though so that makes more sense why they decided to part them off on a cheaper SKU maybe. Of course this is just the experience of one.

I'm not sure how to help, it's been a long time since I've messed with ram timings and it is a ton of work, though I'm surprised that 3800 is running faster than 4200 if the timings are the same or similar.

My best result on 4x8GB of bdie and the system in my sig was 3600, 1.45v, 16 15 15 35 50 300. Testing was done on timespy extreme, firestrike extreme and super pi iirc. At the end of the day you're looking at fractions of a percentage in performance though so it can be fun but it's not really "worth it" aside from the challenge.