3600 cl14 bargain

zx128k said:
I had the same issues, the lower the CL the more voltage I had to use. I have read 8-packs posts that my RAM is safe at 1.45volts. Just don't want to run at 1.5volts or close, to try and get cl14. The lower CL the more you have to understand what the timings do. 8-pack was right to pick b-die from samsung, it just works.
Click to expand...

8 packed, in the 8 pack ram thread said...
8 Pack said:
1.15V soc. Mems still have room 1.5V fine 24-7 on these dimms. You can also try tuning Procodt as you may be at the top of your cpu's cache or Imc capabilities.
Click to expand...

So I think, unless I'm reading wrong, that 1.5v is safe?

Also, and again I might be misunderstanding you, but, b-die is only made by Samsung.
 
Threepwood said:
8 packed, in the 8 pack ram thread said...


So I think, unless I'm reading wrong, that 1.5v is safe?

Also, and again I might be misunderstanding you, but, b-die is only made by Samsung.
Click to expand...

It's made by Samsung, the 8 pack ram also seems to run very cool at 1.45volts. Barely even warm at the moment. I have lots of air flow over the RAM area as instructed. So I took the risk of going for tightened timings at 1.45 volts.

Got 11517 CPU score in Time spy and approx. 26k in Firestrike, just from having tightened CL15 RAM timings. Firestrike https://www.3dmark.com/fs/20704966

https://www.techpowerup.com/review/...king-overclocking-guide/images/proc-odt-3.jpg

I am at 36.9 i believe. Here is a screen shot of my RAM settings. trfc 2 and 4 are the only settings I have left to auto.

 
Last edited:
Threepwood said:
@zx128k looking good man, I'm not at home so I can't really compare! :(

This Viper ram is rated at 1.45v, so to take b die to 1.5 can't be too bad, especially with 8 pack saying it is safe for 24/7 @ 1.5v, no?
Click to expand...

Yeah 8-pack knows his stuff. I have been have noticing that the clock speeds could be getting faster as I tighten the timings. I have no proof of this. It's just I have decent water cooling and I am hitting 4424-4474MHz in game and 4299MHz on heavy loads with no overclocks on the cores. My vray is 15126 approx. now, up from 14600 approx.

I have not gone for 1.5volts because I don't what to wreck what I have. Performance is decent.
 
zx128k said:
Also the L2 cache speed keeps dropping in memetest86. I can only run IF 1900 at 1.035volts SoC. Higher there is cracking with sound after running vray and lower the memory start erroring. So sitting in windows PREM_VDDR_SOC Voltage is 1.037-1.045 volts.
Click to expand...

I see!


So, I'm back at home for a few days and have put the benched stuff in the case.

The power supply is better in my case than the one for benching, the benching one is 8 years old now (ocz z series gold 850w) whereas the newer one is only a year old (evga supernova 750w P2 platinum).

This ram with the better power supply *will now do* T1 @ 14cl 3800, but it needs 1.55v to be stable.

Not only that, but if you pump the volts higher it will do 13cl as well;

w9Ul7uZ.jpg




http://www.ocztechnology.com/ocz-z-series-850w-power-supply.html
 
Nice that its CL13. The other timings can help you reduce vDRAM and keep that performance. With CL15 tight secondary etc timings. 1.44 volts vDRAM in BIOS. VDDP set to 0.88 volts. I would keep things lower than 1.5 volts vDRAM. The maximum for xmp is 1.5 volts.



I would add to above, lowering SoC will affect other voltages as well. Some can't be higher than the SoC voltage. Also in the bios there is overclocking voltages. SoC, VDDP and VDDG are there. In Ryzen master you can see their values.

They are at stock, 1.1volts for SoC, CLDO VDDP 1.09 and CLDO VDDG 1.09. Now if I set SoC to 1.05volts there, in Ryzen master that voltage is 1.05volts. Also if I set CLDO 1.009 for both CLO VDDG and VDDP. When I get restarts or issues in vray that is better that setting the bios VDDP value to 0.88volts. Note the best settings I am trying to keep things stable at IF is keep CLDO VDDP and SoC under overclocking at auto. In ryzen master this means both SoC and VDDP are 1.1 volts. I have changed VDDG to be approx. 40mV below SoC. That appears so far to get rid on sound cracking and restarts with vray. I will need to test more.

So in BIOS after updating to ABBA. OC Treaker, CPU configuration. There are two new added value, CLDO VDDP and VDDG. SoC/Uncore OC voltage was there before (see video). So IF 1900 becomes unstable with the ABBA update. Changing CLDO VDDG appears the affect the issue. I have 1.06mV set atm.

ASRock X570 Taichi BIOS Overview
https://youtu.be/hus0u-jX6nc?t=108



When you run stress tests the CPU drops to lower more stable values and you get zero issues. So the only way to make sure you are stable with the stock boost working is to just use your PC. So it takes time to get things stable and find what works best.
 
Last edited:
I have had possible issues with the ABBA bios.

1. BIOS on my motherboard sets VDDR SoC 1.1 volts, CLDO VDDP 1.0979 volts and CLDO VDDG 1.0979 volts. This causes the IF to become unstable. Setting CLDO VDDG 1.080 volts in BIOS seems to fix this and keep IF 1900 stable. Ends cracking sound and restarts in applications like vray. Also stops hard disks disappeearing in UserBenchMark.

2. There is only so much you can tune the RAM timings before performance regression. Setting RAS Active Time to say 26 will be stable but will cause Time Spy scores to regress. Example of this is that it becomes impossible to hit 11500, change from 26 to 30 and this happens https://www.3dmark.com/3dm/40441662? .

This is Aida cachemem test.




https://browser.geekbench.com/v4/cpu/14847775

https://www.3dmark.com/3dm/40443881?

So just don't set all the values to minimum, some if too low will cause performance hits.
 
Last edited:
Picked up a 4400 kit for £125 - these are brilliant fun and awesome value. Struggling to get to windows with 3800mhz 1T 14-13-8-12-27-39 @ 1.5v, but with gear down enabled its close to if not stable, a quick 100% 16 thread memtest run at 14gb passed. I think with tweaking drive strengths etc manually 1T would be possible. Mine have a hard limit of 128ns for TRFC, no post around TRFC 244T. They don't like TWR 10 or running TCWL lower than CL minus 2 at speed either so they're not quite top bin, but they're damned good and easily a match for similar bin G.Skill kits at twice the price.
 
Nope Im using knowledge... The calculator spits out some very odd and mismatched timing sets. Its clearly been done with a brute force trial and error approach and gives an odd mix of loose and tight subtimings. Props to the chap for making it, hes helped loads of people out, but its not going to give you an efficient or well matched timing set.

Ryzen can run TRCDWR down to 1/2 CL or 8 (whichever is higher) when using Samsung B or Micron E. Hynix will only go to 2/3 CL or 12. No point running it lower than the sum of TRRD_S and TRRD_L as you get no benefit. When combined with TRTP 8 it opens up a little bit more performance when you tweak turn around timings. Good tweak for SuperPi.
 
Last edited:
MrPils said:
Nope Im using knowledge... The calculator spits out some very odd and mismatched timing sets. Its clearly been done with a brute force trial and error approach and gives an odd mix of loose and tight subtimings. Props to the chap for making it, hes helped loads of people out, but its not going to give you an efficient or well matched timing set.

Ryzen can run TRCDWR down to 1/2 CL or 8 (whichever is higher) when using Samsung B or Micron E. Hynix will only go to 2/3 CL or 12. No point running it lower than the sum of TRRD_S and TRRD_L as you get no benefit. When combined with TRTP 8 it opens up a little bit more performance when you tweak turn around timings. Good tweak for SuperPi.
Click to expand...

Well, I'm looking forward to copying your finding :D
 
Just for fun... https://valid.x86.fr/9t8wal 4466mhz 1T 16-16-8-16-32 1.5v. Best £125 I've spent in a long time, these have a lot of entertainment value to them. I'll post my timings up when I have fully stable sets, bench stable is easy but passing memtest to decent coverage isn't :)

If you want to tweak your own subs then just set 1T 14-13-8-12-27-39-273 and GDM enabled, leave the rest on auto. Set a voltage you're comfortable with, start increasing memory speed and see how far you get. After that dial in your own TWR, TRTP and TWCL values. Once those are stable set TRRD_L/S and TFAW and test stability again. Finish off with Twtrs and Twtrl. Turn around timings are pretty easy, just slowly tighten from your boards auto values once you've done primaries and secondaries.
 
The heat spreader on one of the sticks fell off tonight, heard it drop in the case! Luckily no harm/short circuits happened as a result :eek:

Edit: So, the only thing that holds the ram die's side heatspreader on.. is the thermal pad strip that covers the ram.

3mmNXZ1.jpg
 
Last edited:
£25 back from the retailer and have stuck it back on, I dunno if I should just take both off now, wouldn't be surprised if they run cooler without... Which reminds me, does anyone else see the temps of this ram? I can't see memery temp in HWmonitor, so I'm wondering if they lack a temp gauge or not?
 
I might have a mess around with my RAM after reading through this, although I'm using HyperX Predator RAM and not Patriot stuff... I certainly don't like the idea of split heatspreaders being held on by the thermal pads alone!

Anyway, my Predator RAM is rated at 3200 cl16, but I bumped it to 3333 cl16 when I installed it and it's been 100% stable. That's at 1.35v with my Asrock Taichi and Ryzen 5 3600.

It uses Hynix J-die according to Thaiphoon Burner. I have no idea if J-die is good, bad or just average?
 
Threepwood said:
£25 back from the retailer and have stuck it back on, I dunno if I should just take both off now, wouldn't be surprised if they run cooler without... Which reminds me, does anyone else see the temps of this ram? I can't see memery temp in HWmonitor, so I'm wondering if they lack a temp gauge or not?
Click to expand...

might be worth adding some extra tape round the outside, double sided tape often gets weaker if the first application fails, so itll fail again :(
 
You must log in or register to reply here.
Back
Top Bottom