Wow! Is ddr5 very overclockable or did I just get lucky?

[iameatingjam]

Well since your not going to the moon. Either M or A can do 6400 CL34-38-38-84 with 1.5V with brute force overclocking and probably down to CL30-36-36-50. A-Die can't go as low in timings, but it can go much higher.

Another trick is to type in all loose values with high voltage, overclock to the desired freq and work your way down. All Hynix can do 40-40-40-77-117 + tRFC 840 / tRFC2 640 / tWR 96 @ 6400 1.45V. Leave the rest of the values on auto.

This is why I always suggest buying the bin you want. overclocking is very tedious because of the many unknowns. You are shooting in the dark and hoping something sticks. You don't know the maximum freq of the MB or DRAM and also need to figure out the correct CPU voltages. It's a tall order without prior experience.

Yeah I was kind of pressured into buying the lower spec kit from the store, telling me 6000 and over is unstable. What a bunch of bs. He must have been working with old information. Anyway I figure ddr5 will improve a lot in the next few years and that I probably wont have this kit for long. But at the same time, don't want to buy another one already. So if I can't have low latency, might as well have high clock speed in the meantime, that kind of makes sense, right?

Right now I'm at 6800 with 42 42 42 100 I think? ( just to be safe) passmark tells me 37 ns, pretty much the same time its been no matter the setting. Haven't done thorough testing yet but every step I've been running the benchmark and then antec777 for a few minutes just to make sure its not super unstable.

 

[ir_cow]

Yeah I was kind of pressured into buying the lower spec kit from the store, telling me 6000 and over is unstable.

I mean it absolutely can be. Ask @Dr. Dro . You are at the mercy of the CPU IMC Gods for some of it. The rest comes down to the motherboard auto voltages and such.

Right now I'm at 6800 with 42 42 42 100 I think? ( just to be safe) passmark tells me 37 ns, pretty much the same time its been no matter the setting. Haven't done thorough testing yet but every step I've been running the benchmark and then antec777 for a few minutes just to make sure its not super unstable.

Those timings are fine for testing max freq, but they are awful. Eventually you should bring timings down or if you can't, just lower the freq. On a side note, if you are putting 1.45-1.5V into the DRAM, it will need some active cooling (aka fan pointing at it). Otherwise it will error out around 60-70c. 1.4V is the sweet spot I think. You can do a lot with that voltage and still get great pref.

Here are the possible version of your memory kit. Tell us your exact part number (found on the box or ram) and you will know if its A or M die. That will help a bit in finding the limits for both freq and timings.

KF556C36BBEAK2-16 -Hynix A
KF556C36BBEK2-16 - Hynix A
KF556C36BBE-16 - Hynix M
KF556C36BBEA-16 - Hynix M
KF556C36BWEA-16 - Hynix A

 

[Dr. Dro]

This is my result with the DDR5-6800 G.Skill Trident Z5 kit that @ir_cow reviewed. It is a Hynix A kit. I wrote an in-depth reflection of my experience with the kit so far.

G.SKILL Trident Z5 RGB DDR5-6800 CL34 2x 16 GB

Got 32GB 6000C36 for $130 so I go with DDR5 instead of DDR4. What is the sweet spot for DDR5 speed?

www.techpowerup.com

I am very happy with the performance, but I have been unable to solve the bug that made sleep mode and shutdown stop working. As I mentioned in that lengthy post, once I tell Windows to shut down, it will send a halt signal to the CPU and the PC will freeze, then I have to turn it off at the power supply. Restarting the machine causes it to fully power down and retrain, but it's fully stable - I don't understand why.

If you just wanna save time and look at my timings without reading my text wall on that thread:

 

[iameatingjam]

I mean it absolutely can be. Ask @Dr. Dro . You are at the mercy of the CPU IMC Gods for some of it. The rest comes down to the motherboard auto voltages and such.

I guess I'm just bitter that I spent all this extra money on a ddr5 motherboard and ddr5 ram, when out of the box it benchmarked lower than the ddr4 I already had. Only thing is that ddr4 memory was not overclockable at all.

'Those timings are fine for testing max freq, but they are awful. Eventually you should bring timings down or if you can't, just lower the freq. On a side note, if you are putting 1.45-1.5V into the DRAM, it will need some active cooling (aka fan pointing at it). Otherwise it will error out around 60-70c. 1.4V is the sweet spot I think. You can do a lot with that voltage and still get great pref.'

Yeah, 6800 was stable at the timings I gave you. Then I went down to cl40 and did a quick test, everything seemed to be okay, went down to cl38, and wow that benchmark was the best I had ever seen 99% percentile, but the ns went down to 36 and that seems to be where trouble starts, testmem produced some errors.|

So I tried upping ddr voltage from 1.4 to 1.45 and that seemed to make things even worse. So then I had the idea to try even lower, 1.35, but that was just as bad. So I went back to 1.4 and cl40 and still got some errors so I loosened the other timings even more, and now its seems 'mostly?' stable. I ran anta777 extreme for 45 minutes and it produced 1 error. Still the overall score is higher, and latency right now is 38 ns, which is still about the same its always been. Lowest it seems to go stable, at least the way I'm doing it, is 37ns.

Here are the possible version of your memory kit. Tell us your exact part number (found on the box or ram) and you will know if its A or M die. That will help a bit in finding the limits for both freq and timings.

KF556C36BBEAK2-16 -Hynix A
KF556C36BBEK2-16 - Hynix A
KF556C36BBE-16 - Hynix M
KF556C36BBEA-16 - Hynix M
KF556C36BWEA-16 - Hynix A

Its the first one, I checked the box, except mine has a 32 after it ( for the GB I assume). Googling it, looks like only this model of ram uses it? That can't be a good sign.

This is my result with the DDR5-6800 G.Skill Trident Z5 kit that @ir_cow reviewed. It is a Hynix A kit. I wrote an in-depth reflection of my experience with the kit so far.

G.SKILL Trident Z5 RGB DDR5-6800 CL34 2x 16 GB

Got 32GB 6000C36 for $130 so I go with DDR5 instead of DDR4. What is the sweet spot for DDR5 speed?

www.techpowerup.com

I am very happy with the performance, but I have been unable to solve the bug that made sleep mode and shutdown stop working. As I mentioned in that lengthy post, once I tell Windows to shut down, it will send a halt signal to the CPU and the PC will freeze, then I have to turn it off at the power supply. Restarting the machine causes it to fully power down and retrain, but it's fully stable - I don't understand why.

If you just wanna save time and look at my timings without reading my text wall on that thread:

View attachment 299658

Thanks, I'll try it, but I have a feeling it wont boot.

 

[Dr. Dro]

I guess I'm just bitter that I spent all this extra money on a ddr5 motherboard and ddr5 ram, when out of the box it benchmarked lower than the ddr4 I already had. Only thing is that ddr4 memory was not overclockable at all.



Yeah, 6800 was stable at the timings I gave you. Then I went down to cl40 and did a quick test, everything seemed to be okay, went down to cl38, and wow that benchmark was the best I had ever seen 99% percentile, but the ns went down to 36 and that seems to be where trouble starts, testmem produced some errors.|

So I tried upping ddr voltage from 1.4 to 1.45 and that seemed to make things even worse. So then I had the idea to try even lower, 1.35, but that was just as bad. So I went back to 1.4 and cl40 and still got some errors so I loosened the other timings even more, and now its seems 'mostly?' stable. I ran anta777 extreme for 45 minutes and it produced 1 error. Still the overall score is higher, and latency right now is 38 ns, which is still about the same its always been. Lowest it seems to go stable, at least the way I'm doing it, is 37ns.





Its the first one, I checked the box, except mine has a 32 after it ( for the GB I assume). Googling it, looks like only this model of ram uses it? That can't be a good sign.



Thanks, I'll try it, but I have a feeling it wont boot.

Oh, don't copy my settings 1:1. It's probably not going to work for you, but it's a decent reference - if you're aiming for something similar. My motherboard does not like to do an inch beyond that, and the low-power state woes I have may very well indicate that it's about its breaking point. Not too upset about it considering it's an E-ATX, creator-focused board, I got a very good deal on it and that even the Z790 version isn't exactly a clocking champ (doubt the new ACE MAX is gonna change that - seems mostly adding 5GbE to one of the ports and upgrading it to Wi-Fi 7 over 6E... my router is barely Wi-Fi 6-capable so big whatever).

Remember: SA voltage is very sensitive and more is not always better. You may actually want to lower it to achieve stability. If 1.15 doesn't work, lower it to 1.1 instead of raising to 1.2 first. It might stabilize! The same goes for the VDDQ/VDD2 voltages. As for your 45-1h errors in anta777 extreme1 TM5: if you already started tweaking secondaries and tertiaries, double check your tWRRD timings, they might be bunk. Raise them by 2 or 4 clocks each, and retry it, it was how I solved that problem. Loosening it by 2 clocks from 54/42 to 56/44 resolved it. Your tRAS may also be too low or even too high, some motherboards are very particular about a range. My Z690 ACE liked doing 28, but I've been helped by someone on Discord who has a ROG Z690 Extreme that could achieve 6800 but their tRAS won't go lower than 38.

BTW, I don't know how PassMark measures its latency, but AIDA64 is still in the 55-57 range for me. With tighter timings than that, so it's a consideration

 

[ir_cow]

Its the first one, I checked the box, except mine has a 32 after it ( for the GB I assume). Googling it, looks like only this model of ram uses it? That can't be a good sign.

Congrats you are the proud owner of Hynix M Die Those generally top out at 6800 @ 1.5V, with a really good samples hitting 7600.

 

[iameatingjam]

Oh, don't copy my settings 1:1. It's probably not going to work for you, but it's a decent reference - if you're aiming for something similar. My motherboard does not like to do an inch beyond that, and the low-power state woes I have may very well indicate that it's about its breaking point. Not too upset about it considering it's an E-ATX, creator-focused board, I got a very good deal on it and that even the Z790 version isn't exactly a clocking champ (doubt the new ACE MAX is gonna change that - seems mostly adding 5GbE to one of the ports and upgrading it to Wi-Fi 7 over 6E... my router is barely Wi-Fi 6-capable so big whatever).

Remember: SA voltage is very sensitive and more is not always better. You may actually want to lower it to achieve stability. If 1.15 doesn't work, lower it to 1.1 instead of raising to 1.2 first. It might stabilize! The same goes for the VDDQ/VDD2 voltages. As for your 45-1h errors in anta777 extreme1 TM5: if you already started tweaking secondaries and tertiaries, double check your tWRRD timings, they might be bunk. Raise them by 2 or 4 clocks each, and retry it, it was how I solved that problem. Loosening it by 2 clocks from 54/42 to 56/44 resolved it. Your tRAS may also be too low or even too high, some motherboards are very particular about a range. My Z690 ACE liked doing 28, but I've been helped by someone on Discord who has a ROG Z690 Extreme that could achieve 6800 but their tRAS won't go lower than 38.

BTW, I don't know how PassMark measures its latency, but AIDA64 is still in the 55-57 range for me. With tighter timings than that, so it's a consideration

Thanks your suggestions did help me make 6800mhz stable but by the end of it, it didn't seem like it was worth it so I went back to 6400 cl38. A whole day of modifying numbers for nothing lol!

And yeah if you're having 55 ns latency at those numbers it must calculated differently.

 

[Dr. Dro]

Thanks your suggestions did help me make 6800mhz stable but by the end of it, it didn't seem like it was worth it so I went back to 6400 cl38. A whole day of modifying numbers for nothing lol!

And yeah if you're having 55 ns latency at those numbers it must calculated differently.

Well, it's not something you will see immediately, it's going to make your system snappier overall but if you're expecting an immediate benchmark crusher, sorry to disappoint

I downloaded PassMark's test - ran it quick with my stuff open and it gives me a latency score of 32, with 16KB at 16ns, 64KB at 20ns and 8MB at 60 ns. Whatever that means

 

[iameatingjam]

Congrats you are the proud owner of Hynix M Die Those generally top out at 6800 @ 1.5V, with a really good samples hitting 7600.

@Dr. Dro

Yeah 6800 was okay even pushed it to 7000 the only thing is the latencies had to be so high for it to be stable I figured it wasn't worth it. 6400 cl38 gives me nearly as good benchmarks. Thanks for your help.

This is the stable 6400 cl 38

This is the best score I got, the 6800 cl38 that unfortunately wasn't stable.

Not that much different. By the time I did get it stable with secondary and tertiaries it was about the same as above.

 

[ir_cow]

We all use different programs for benchmarks, but this one is meaningless to me. How does 4043 vs 3939 translates into real-world usage?

 

[iameatingjam]

We all use different programs for benchmarks, but this one is meaningless to me. How does 4043 vs 3939 translates into real-world usage?

Who knows? I didn't exactly spend much time using it at unstable speeds. But if the 6400 cl38 is a better result than 98% of ram kits benchmarked, it can't be that bad. The only thing I don't like is how ram amount is included in the score. That means you could get a higher score just by having more ram.

 

[ir_cow]

The only thing I don't like is how ram amount is included in the score. That means you could get a higher score just by having more ram.

Which makes it a meaningless benchmark.... Dual-Rank memory or having four DIMMs will increase the read/write speeds, but that doesn't always translate to applications.

 

[Dr. Dro]

Well, this is my score with what I'd call reasonably optimized 6400 C30. Unsure if running under Windows 10 has any real implications, but so far from my personal experience it seems that the whole thing about you must use Windows 11 to get the best of a 12th and 13th gen CPU is a bunch of hot air. 10 runs fine on my 13900KS.

 

[iameatingjam]

View attachment 299834

Well, this is my score with what I'd call reasonably optimized 6400 C30. Unsure if running under Windows 10 has any real implications, but so far from my personal experience it seems that the whole thing about you must use Windows 11 to get the best of a 12th and 13th gen CPU is a bunch of hot air. 10 runs fine on my 13900KS.

Hot damn! Wish I had those sticks!

Which makes it a meaningless benchmark.... Dual-Rank memory or having four DIMMs will increase the read/write speeds, but that doesn't always translate to applications.

I guess if your goal is solely to test the speed of ram, the total might not be relevant. But this is like a whole benchmarking suite, and I suppose it makes sense to have amount in there. If you want to benchmark the whole computer, and you only have 4gb of ram, it would make sense that would be taken into consideration because its going to hamper performance. If all you want to do is benchmark memory there's probably better tools idk I've just always used passmark and its pretty respectable.

 

[AlwaysHope]

If you want to see real world tangible differences with stuff like productivity apps & even light gaming (DX9?) run PCMark10 free edition & compare differences in scores with RAM speeds & timings. That's my go too & see if any diff in RAM OC's. The whole system gets a workout except heavy 3D.

 

[Dr. Dro]

If you want to see real world tangible differences with stuff like productivity apps & even light gaming (DX9?) run PCMark10 free edition & compare differences in scores with RAM speeds & timings. That's my go too & see if any diff in RAM OC's. The whole system gets a workout except heavy 3D.

Speaking of which, UL issues free keys for all of their old benchmarks. You can have fun with things like 3DMark 11, Vantage, etc.

Futuremark Legacy Benchmarks

Download legacy versions of 3DMark, PCMark, and other Futuremark benchmarks for free.

benchmarks.ul.com

go ham

 

[ir_cow]

DDR5-8200. If I more time, I would tweak the sub-timings. This benchmark doesn't do anything for me.

 

[Dr. Dro]

DDR5-8200. If I more time, I would tweak the sub-timings. This benchmark doesn't do anything for me.

View attachment 299842

This is pretty good, it's a nominal improvement in bandwidth and a very small one in latency compared to my 6400 C30 result. Eh, maybe I shouldn't worry as much as I do. Someday the Z790 Apex is gonna be cheap too

 

[ir_cow]

Someday the Z790 Apex is gonna be cheap too

Speaking of which, the Z790 Tachyon just came in stock. Been waiting months for this! I love my Z690 one, but that tops out at 8000 :/ hmm.. Not sure if I have $600 to spare right now...

 

[lexluthermiester]

My old ddr4 3200 kit became very unstable if I were to overclock it by even 50mhz. But I bought some 5600 CL36 kingstom fury beast ddr5 for my new motherboard and cpu (13600kf and gigabyte auros b760), even though I wanted 6000 the guy at the store convinced me 5600 would be more stable so I just went along with it. However I found out, at least on some benchmarks that it actually scored worse than my ddr4, probably because of the latency? Not happy with this I tried pushing the mts up, got all the way to 6000 without even touching timings and it performed way better than before ( on passmark, from 88th percentile to 96th, and also increased cpu performance (well obviously - but it was nice to see nontheless).

Ran a stability test for 8 hours with 0 errors. Thats crazy! It even went up to 6100 but I thought maybe I shouldn't push it and brought it back down to 6000. I tried lowering the latency from 36 to 34 without much visible effect. Possibly because I didn't touch any of the subtimings because I really don't know how they work.

Is this normal? Is it safe? Should I try pushing it even more? Or should l maybe educate myself a bit more on how timings work and take that angle instead? Or maybe I should just leave well enough alone and run at 6000 CL36. According to HWmonitor the ram isn't anywhere near hot though I understand overclocking ram can also have a small affect on cpu temps. However with my Artic Liquid Freezer II 240, I'm able to run at max boost frequency indefinitely - or so it seems. So I have a bit of headroom there. And I can't overclock the cpu frequency above what the boost would have done anyway on this motherboard (didn't really think about that before buying... hmmm).


Anyway I've just never had ram that could so easily run above it advertised clock speed. Thats awesome.

Short story, you got lucky.

 

[iameatingjam]

Wow I didn't realize how much changing the dram voltage from 1.4 to 1.5 would help. Before I couldn't get 6400 to run at anything less than cl38. But now, 36 was fine, then so was 34, and so far it looks like cl32 works fine too. Over 90 minutes into anta777, still no errors to be seen. The memory is getting kind of hot, but not to a worrisome degree I don't think, 61 degrees under load. But I do have 2 140mm fans blowing right at it and 2 140mm fans as part of the aio exhausting right there from the top as well. So they should have lots of air.

Last time I went over 1.4 was when I was trying to stabilize the 6800 cl38, when I went to 1.45 and it actually seemed to create MORE errors, though its possible that was just coincidence. Anyway I shied away from going beyond 1.4 for a while because of that. Maybe the way to go is just stay at 6400 and reduce timings as much as possible. At least I have a bit of headroom now, or so it seems.

1.5 isn't dangerous, right? As long as its cooled?

 

[lexluthermiester]

1.5 isn't dangerous, right? As long as its cooled?

Generally, electronics of any kind can tolerate voltage + or - 10%. Going from 1.4 to 1.5 volts is less than a 10% over-volt. So yeah, with proper cooling you should be fine long term. Word of caution though. Over-clocking & over-volting will generate more heat, thus it will be much more important to maintain proper ventilation for the case, keeping the case clean(dusting at least once a month) and keeping the ambient room temp down. Whatever you normally keep that room at, try to lower the temps by 4C or 5C.

But yeah, you're good!

 

[iameatingjam]

This is my result with the DDR5-6800 G.Skill Trident Z5 kit that @ir_cow reviewed. It is a Hynix A kit. I wrote an in-depth reflection of my experience with the kit so far.

G.SKILL Trident Z5 RGB DDR5-6800 CL34 2x 16 GB

Got 32GB 6000C36 for $130 so I go with DDR5 instead of DDR4. What is the sweet spot for DDR5 speed?

www.techpowerup.com

I am very happy with the performance, but I have been unable to solve the bug that made sleep mode and shutdown stop working. As I mentioned in that lengthy post, once I tell Windows to shut down, it will send a halt signal to the CPU and the PC will freeze, then I have to turn it off at the power supply. Restarting the machine causes it to fully power down and retrain, but it's fully stable - I don't understand why.

If you just wanna save time and look at my timings without reading my text wall on that thread:

View attachment 299658

So do I want tRAS to be as low as is stable? I've read conflicting things about it, like thats is supposed to be tCL + tRCD or some such. Mine started out at 80, I've gotten it down to 40 and am still completely stable. Should I keep going? Or will it be detrimental to performance at some point?

Btw by increasing trefi as you suggested a couple pages ago, I got a HUGE reduction in latency. So thanks for that! Now I'm not too far away from catching up to your score

 

[ir_cow]

The lower you go on the timings (or higher for trefi) the lower the temps needed to be. You will get errors in the 60+ C range usually if not lower. I just leave a fan pointing at mine to keep in around 45c.



Also if your memory doesn't have a thermal pad on the PMIC, I wouldn't be pushing 1.5v for long periods of time. That's just my personal opinion on the matter.

 

[iameatingjam]

The lower you go on the timings (or higher for trefi) the lower the temps needed to be. You will get errors in the 60+ C range usually if not lower. I just leave a fan pointing at mine to keep in around 45c.



Also if your memory doesn't have a thermal pad on the PMIC, I wouldn't be pushing 1.5v for long periods of time. That's just my personal opinion on the matter.

Oh really? On the ~2 hour anta777 stress test, the one module caps out at 62 and the the other one 60. But in normal operation its much lower than that, more like 40. I didn't think that was concerning, but I could be wrong.

Anyway what I was wondering about was that 'tRAS' timing, I was noticing when I changed it nothing seemed to happen, then I read some things like this on google:

"This per Raja, the Asus memory guru: tRAS is the time, with any added buffer, it takes for tCL+tRCD+tRTP, with some operations allowing, unreliably, 2 clocks less. If you set it lower, the IMC makes up its own tRAS to use.

In other words, memory kits with lesser tRAS are basically using it as a marketing point because it does no harm to fudge it. When you make it lower, you're only fudging a fudged number. The seeming gains are random noise."

So I was wondering, does changing it actually have an effect? Or if it doesn't meet the expected requirements, does the pc just make up its own more appropriate number to use in its place?

Basically, I just want to know if I'm wasting my time tinkering with this value.