memory timings benching

[spiritedandy]

i dont know lazyness or forgetfullnes but ive not touched memory really looking back i used to but now well no . witheverything else to think about ive took the shot it in xmp2 and increase to 4000/2000 and all has been fine .
but now im thinking should i have pushed it . ijust had quick play and i can do better lol think im going to have to have rethink

ballistick rgb 32gb 2x16 3600Mhz 16-18-18-36 is what i have

 

[MaddMutt]

IIRC - the AMD chips are best between 3600 - 3800 depending on the IMC. You also want to keep it in a 1 to 1 ratio (fsb to mem) for best performance. I will let ED who has ~ your setup explain it better. I can give help on the Intel side though.

 

[JLK03F150]

MadMutt, FSB is 100.

It is the DRAM to Infinity Fabric that needs to be 1:1 & it is difficult to get the IF past 3800. Once the DRAM:IF switches to 1:2 you will lose a fair amount of performance.

 

[Johan45]

Looks like you have one sweet 5950X there. From what I see you're running 4000 MHz at 1:1 which is rare for one of those. If anything you can try tightening the timings a bit for competition.

 

[spiritedandy]

cheers for replys running 1:1 ive done ran few tests with time spy just to see . johan ill try tighter timing at 3600 or 4000 or both lol
here few scribblings good luck understanding

- - - Auto-Merged Double Post - - -

ps is it worth going over 4000

 

[spiritedandy]

latest best score with memory oc 14-18-18-18-34-80 T1 4000/2000 gain of 350 points from start with just memory i can see 19623 best score ive done

 

[Woomack]

You should be able to set CL14-17-15-15 or 14-17-16-15 and something like ~70 tRC but the timings you should work on are for sure tRFC. I bet that all are at auto and they're maxing out. Check something like 620/390/280. If you make it work then check something like 580/360/260.
One more thing, check RTTNom at RZQ/6 or 7, RTTWr at 3, RTTPArk at RZQ/6 or 7. It may help with latency or not. It may boot or not so have to play with that a bit. Latency can check with AIDA64

You may also check how high it will go at 1:2 ratio and tweak RAM and other things at that clock so manually 2000MHz FCLK/UCLK and 2400MHz+ RAM clock. This RAM should run up to 5200+ if the IMC and motherboard can handle it. On X570 Unify (which is about the same) I could set 5000 for benching with my Ryzen 5900X. 4800-5000 should work at CL18-21-19 or 18-22-20 what in some benchmarks may give better results than 1:1. You won't really know until you test it. Generally tests based on games work better at 1:1, other tests are sometimes better at 1:2 but 4800+ memory clock.

 

[spiritedandy]

You should be able to set CL14-17-15-15 or 14-17-16-15 and something like ~70 tRC but the timings you should work on are for sure tRFC. I bet that all are at auto and they're maxing out. Check something like 620/390/280. If you make it work then check something like 580/360/260.
One more thing, check RTTNom at RZQ/6 or 7, RTTWr at 3, RTTPArk at RZQ/6 or 7. It may help with latency or not. It may boot or not so have to play with that a bit. Latency can check with AIDA64

You may also check how high it will go at 1:2 ratio and tweak RAM and other things at that clock so manually 2000MHz FCLK/UCLK and 2400MHz+ RAM clock. This RAM should run up to 5200+ if the IMC and motherboard can handle it. On X570 Unify (which is about the same) I could set 5000 for benching with my Ryzen 5900X. 4800-5000 should work at CL18-21-19 or 18-22-20 what in some benchmarks may give better results than 1:1. You won't really know until you test it. Generally tests based on games work better at 1:1, other tests are sometimes better at 1:2 but 4800+ memory clock.

I could not go less than 18 -18-18 1.5 v at 4000 well lot better i beat timespy score memory def helped cheers woomack everyone .
i could not set RTTnom RTTWr as it was in Ohms beyond me but im happy with my tighter timings . will see if i can get 5000 tomorrow

 

[Woomack]

I know it's a different CPU but 4933 1:1 on 5700G and the same RAM as you have - Ballistix 2x16GB DDR4-3600 CL16 / Micron B
Quite relaxed timings as I was testing max 1:1 clock, not max RAM clock.



This kit can make 5400 CL18-26-25 or 5200 18-24-22 but at 5000+ it will require 1.55-1.65V. It's not scaling past 1.7V. It literally OC worse or crash at a too high voltage. Try 1.55V+ at 4800+. It's still safe and should be stable up to ~65°C. Worse is that cheaper Ballistix modules don't have thermal sensors so you won't check that. Anyway, if there is an open space rig or even any airflow, then it should keep low enough temps.

It's weird it can't work at tighter timings than 18-18-18. If it's Micron B then it should or you are missing something in additional timings. Usually, tRP can be 1-2 lower than the tRCD. On the other hand, it doesn't change much and it's better is to focus on sub-timings.

2 more screenshots - Ballistix MAX and MAX RGB. All are using the same Micron B. MAX is generally the best but a standard Ballistix (used for the 1st screenshot, above) is overclocking better than the MAX RGB.





All of them are scaling up to ~1.65V and can make CL18 up to 5200+.

 

[spiritedandy]

Cheers Woomack wow on youre memory lol erm ok will give ago tomorrow . And the timings again will check 18-18-18
That aida is different to mine could not see that type of bench will look into that too .
Think hardware monitor cpuid has temp of ram my second post shows temp . And deffo ill have to up voltage , and thanx for info again Â….

 

[Woomack]

Hmm, weird, 2 of my kits were not showing temps in diagnostic software while the MAX series does. But looks like your kit has thermal sensors.
In AIDA64 click on the tools tab and next there are stability tests, diagnostic, and bandwidth/latency benchmarks. The free version won't show everything but I think it shows latency. It's about the same as in the #8 post but in the square window. AIDA64 cpuid can replace cpu-z. It shows about the same info. It's simply not so popular and hwbot requires cpu-z.
I'm using hwinfo64 for all the sensor reading as it has more options and usually works the best with various motherboards and graphics cards.

Btw. check memtryit and other options in BIOS. I guess there are most profiles for Samsung but I think that for Micron there is something too.
Higher X570 ASUS ROG motherboards have 4133 CL14-13-13 profiles for benching. It's for Samsung but suggests that some benchmarks run faster at these settings and 1:2 than ~3600 and 1:1. Of course, it highly depends on the test. Because Micron runs at more relaxed sub-timings then I recommend higher frequency as sub-timings at 4266, 4600 and 4800 are not so much different. Bigger steps are above 5000.

For me the next thing that I want to test is passive/semi-passive ITX build and how high it can run at 1:1. I don't know if I do anything this week as I will be busy at work with expected over hours.

 

[spiritedandy]

Ok i used AIDA bench and everything crashed lol i increased vcore cpu running bench now just at 4000 still just want that out of way then try higher .ive used mem try but found hands on better its good though . Cheers woomack youre loosing me at end itmmax ish thingy sorry . First time i have really bothered with memory . I put it in let it go no internal settings a justed but now omg every time i will ..

Ps thanx upgraded Aida

PPS mesed up had gpu fpu cpu cache mem on stability

PPPS is this erm ive lost it . Ok overclock cpu and memory test or just memory

PPPPs its ok lol aghhh

 

[Woomack]

I was trying to reply 3 times and each time the forums were forcing me a captcha screen so here is a short version.
You can use any other stability test. For shorter test you can use even the Final Fantasy XV benchmark as it crashes quick when there is any RAM problem. In AIDA64 use memory and cache tests. These two load RAM and memory controller and are enough. I recommend about 3h test as recently it was crashing after about 2-2.5h when I was testing max 1:1. For benching only it doesn't matter, as long as it passes the benchmark.

 

[spiritedandy]

Woomack im sorry ive tried everything . Voltage timings cpu speed ram lowering ratio several options i can not get above 4200 just will not post .
Even took ram out reseated . Also all auto settings and increased mhz slowly nothing . You have helped me loads and im very happy at my new timings at 4000/2000 all day long . So thank you perhaps sign was 18-18-18 could not go lower . Only thing not done is flash bios i cant see that helping . Could even mess up the good things i have now . I do find it strange that it is that stubborn

 

[Woomack]

Maybe your RAM is using other IC than mine but it's single rank in two 16GB modules so should be Micron B. 4200+ will work only at 1:2 ratio so you have to adjust that manually, sometimes in the CPU/AMD OC options. I don't think it will give you much better results but I thought it's worth to try.

 

[spiritedandy]

No i give in im happy whith my timings thank you for all youre help

Ps mine was only 3600mhz sticks