Adventures in 2700X Overclocking

it seems all teh x470 boards have crazy overkill VRMs. You want to buy samsung b-die ram
 
123Lanoix said:
it seems all teh x470 boards have crazy overkill VRMs. You want to buy samsung b-die ram
Click to expand...

Wish I had crazy overkill VRMs now. My board is what I'd call barely sufficient for OCing a 2700X. It has pretty good VRMs.

Power delivery improvements were, I suspect, a big part of the X470 push. Some X370 boards, and a lot of B350 boards, probably don't have sufficient power delivery to really extract peak performance out of this CPU.
 
123Lanoix said:
my motherboard will only do 16-15-15-15-35-51 1.35v when i set the first timing to 15 it automatically changes it to 16 and it won't post with 14-15-15-15-35-51 3400mhz 1.35v will there be a performance difference?
Click to expand...
Ah... You have one of the locked-out BIOSes then... When's the last time you updated it? What motherboard are you using currently for your 2700X?
You can easily find the BIOS version if you're using Windows 10 (I presume Windows 8 will provide this, too; can't comment on 7) by hitting the Windows key (or Start button) and typing in "System",then selecting the "System Information" app (defaults to first in the list for me). On my HP ENVY x360 laptop it's the 11th line on the main window that pops up...
Sys Info BIOS Ver.png

I mean I assume you are running the latest, but it just seems odd that it'd still be defaulting to Even CAS latencies like that. It's the exact same thing that was happening on the original Ryzen for the first few months before AMD rolled out one of the updated AGESA microcodes. (I want to say v1.0.0.4, which I think was... in May of 2017? [not to be confused with an upcoming 1.0.0.4 since they for some reason have started the numbering over])

My only guess for why, would be a really low end board that it wasn't a priority on, and maybe had redundant Odd-to-Even swapping that wasn't disabled *shrug* Hopefully this is just AMD playing it safe again by preventing it in the AGESA until everything is working well, but you'd think Zen+ would be using the same memory controller and thus this wouldn't be needed. speculation at best though... :\

With CL14 not stable at 3400 either, my only suggestion would be to try giving the RAM 1.37V. Most DDR4 ICs can handle up to 1.5V, but that is not a safe voltage to run at continuously, and while I think 1.45V is considered safe, I'd play it safe and say only 1.42V max.

Lastly, if you really want to make an attempt for higher speeds at non-crazy timings (and IIRC, allowing for Odd timings), then you can try changing the Command Rate (CR) to 2; however that usually carries with it a fair impact to speed and latency, but it can in turn stabilize higher speeds... so like 3200, 3400 or 3600... it's a trade off. 3800 at CL17 CR2 may end up providing better results than 3200 CL14. Given I have a Ryzen 1700X and not a 2700X, I can't attempt to find out :(

Addendum: While researching RAM for the last quote, I see that on G.Skill's page, their DDR4-4600 kits are specced to run at 1.5V! o_0 While I have no clue which ICs they're using, I guess some of them can be considered 24/7 safe at that voltage. I suppose it's possible that it's like DDR3, where originally they operated at 1.9V, then settled around 1.65V, then 1.5V, before having the low-voltage variants running at 1.3V. IOW, that the operational window on DDR4 may swing much higher. Sadly, DRAM prices are absurd currently so... who really wants to take that chance of burning out a DIMM... Nevermind the fact it could very well hurt the CPU's IMC as well, like with whichever models of Intel CPUs with DDR3.


Aenra said:
Finally! A person that can speak English!
Thank you oh savior! Jeesus, lol

And guys? Honestly? When someone asks a very specific, utterly technical question? For some very specific reasons, which he even goes on to state?
If you don't know the answer, just don't reply, lol
Don't be that guy!
Click to expand...
No problem :p But in fairness, I honestly just think he was misunderstanding, not that he didn't know but wanted to weigh in anyways.

I agree though, especially when you look at the Q/A on NewEgg and Amazon. People always answer with something like "I don't have it yet, but I suspect...." or "I have a different model and while it's the same X-Brand as yours, it's made by Y-Brand, but I bet it'll still work!"
When the people clearly aren't in a position to reply, but do so anyways, I simply think that they aren't capable of connecting the dots. Because, despite their heart being in the right place, they aren't actually helping. Often better to leave a Q unanswered in that instance. :p


enc0re said:
I'm considering my first AMD build in well over a decade and am a little confused as well. It sounds like with Ryzen 2700x the way to go is to leave the chip multiplier alone, rely on PB2 (even though it hertz the [H] in me), and instead run the memory as fast as possible. So I'm thinking that the Gigabyte (a brand I like anyway) Gaming 7 X470 is the way to go for me because it supports up to DDR4-3600. It also has the fancy onboard audio I want.

Looking at the QVL for the board, they list F4-3600C17Q-32GTZR (a 4x8GB TridentZ kit) for 3600MHz. I only want 16GB, because I don't need more and because LOL ram prices at the moment. But the 16GB kit isn't on the list.

Would it be safe to assume that if I picked up F4-3600C17D-16GTZR (the equivalent 2x8GB TridentZ) it would work? Also, this is G.Skill's memory for Intel and they have announced AMD memory for later, which ends in 'X' instead of 'Z'. Does that matter? I've never bought AMD/Intel specific memory before!

Thanks for any insight.
Click to expand...
With the increased load that 4 DIMMs put on the memory controller of Ryzen, I'd personally feel confident enough to buy the Dual kit with the expectation that it'd achieve the same DDR4-3600. That being said, you do need to make certain that the actual kit has the same timings; however, upon looking at G.Skill's site, you don't have to worry about additional models of the same CL but with higher/lower tRCD and tRP values like I did at 3200.

Though if you want to err on the side of caution, sticking with the 17-18-18 instead of either of the other two that are available might help: 15-15-15 and 16-16-16. Not to say they wouldn't work, since if they did work, you'd have a bit more performance. Nay, the reason to stay with that latency would be to ensure you're getting the same ICs on the modules that were tested. They could easily be something like Hynix or Micron ICs, or Samsung D.

Granted, it's most likely that the CL15 and CL16 kits are the sought after Samsung B dies, and would work just peachy. So realistically, if you were to find either of those two available at a better price, I'd dive in and get them over the CL17 kit.

As far as them being "For Intel" kits... In general and this is assuming they are not on a QVL, if they're Samsung B dies you shouldn't have much to worry about, with the exception of expecting the XMP profiles to definitively work. The timings, more specifically the sub-timings and tertiary-timings, are not necessarily going to be the same to run stable at 3600 as they may be on AMD. So while you can use them as a baseline, be prepared to have to tinker with them.
Of course anyone you see running Ryzen with Trident Z modules (at least at the time this is posted), you can be pretty sure they are the Intel kits, as I'm not entirely sure the AMD ones have hit the market yet. Mine have ran perfectly for me, from day 1, using the original shipping BIOS, where as tons of people had issues. So it's also a bit of a crap shoot either way lol

Lastly, just to be clear, they are still called Trident Z even for the AMD models (since unfortunately they've already used Trident X for DDR3 AMD kits), and it's the end of the model number where the X will be. Furthermore, these are ONLY on the stupid RGB models, which many people here have had toooo many issues with. Not just that they have problems getting them stable, but that they flat out die on them. (Speculation is due to how they program the color LEDs and something just goes wrong, causing them to no longer boot)
So I'd personally stay away from them and just try to find someone who HAS bought them, ask them to use TheStilt's RyzenTimingChecker (I think v1.03 is the latest?) and then use normal Trident Z with those sub/tertiary timings manually entered. But seeing as their kits use some really crazy timings of 18-22-22 at just 3600MHz, that kinda makes me want to ignore them on that alone, while the RGB factor just seals the deal for me.
 
my motherboard is a Asus ROG Strix x370-i gaming on latest bios ver 4009 running 2700x
 
IdiotInCharge said:
Wait, are people going for higher latency DDR4 3600 over lower latency DDR4 3200 now?
Click to expand...
Might be the case. G.Skill sure isn't helping matters by making those "AMD" Kits at 3600 with those absurd timings of 18-22-22 :\

However, I also can't totally blame G.Skill given the state of the DRAM industry. They're just kinda making the best with what they can I guess, otherwise they'd have to charge an arm and a leg.

I wouldn't be surprised if they are basically re-badging/certifying what would've been 3200 C16-18-18 kits as these 3600 C18-22-22 ones.
 
Probably.

And for those reusing RAM, I get it.

If you're not, I have no idea why anyone who is spending money on RAM not getting the DDR4 3200 C14 Flare X.
 
I've seen computerbase claim that 3600MHz is achievable on Ryzen 7 2700X , with poorer chips hitting 3466.
 
ACC said:
I've seen computerbase claim that 3600MHz is achievable on Ryzen 7 2700X , with poorer chips hitting 3466.
Click to expand...
I've heard people doing 3800, but
it is a question of timings and what is ultimately more stable.
 
Last edited:
  • Like
Reactions: N4CR
like this
IdiotInCharge said:
Probably.

And for those reusing RAM, I get it.

If you're not, I have no idea why anyone who is spending money on RAM not getting the DDR4 3200 C14 Flare X.
Click to expand...

It’s my understanding that because of Infinity Fabric, you want higher clockspeed on Ryzen. For example while 14/3200 is faster RAM than 17/3600, it is my understanding that 17/3600 will give higher performance with Ryzen.

Am I misinformed? If it’s as simple as picking up some 14/3200 and not worrying about how to get a stable, high end clockspeed; that would certainly make life easy.
 
enc0re said:
Am I misinformed?
Click to expand...

That's actually the question- has something changed?

I think there was a hard 3200 limit for Ryzen1, which meant that the C14 stuff was as fast as you could get. If relaxing the timing and increasing speeds on Ryzen2 now that >3200 is achievable is faster, then we need to see some comparison tests :).
 
enc0re said:
It’s my understanding that because of Infinity Fabric, you want higher clockspeed on Ryzen. For example while 14/3200 is faster RAM than 17/3600, it is my understanding that 17/3600 will give higher performance with Ryzen.

Am I misinformed? If it’s as simple as picking up some 14/3200 and not worrying about how to get a stable, high end clockspeed; that would certainly make life easy.
Click to expand...
Not misinformed, and you raise a valid point. So as Idiot said, we are going to need to re-evaluate the whole memory situation again and determine if the higher speed on Infinity Fabric will end up offsetting the higher latency. Ryzen IS a curious animal like that so we may be in for a surprise. Fingers crossed :D
 
IdiotInCharge said:
That's actually the question- has something changed?

I think there was a hard 3200 limit for Ryzen1, which meant that the C14 stuff was as fast as you could get. If relaxing the timing and increasing speeds on Ryzen2 now that >3200 is achievable is faster, then we need to see some comparison tests :).
Click to expand...

According to this, timings make a huge difference. The jump from 2933 to 3466 is nowhere near as big as the jump from 3466 to 3466 with optimized timings. Not sure of that means 2933 or 3200 with optimized timings would be faster than 3466 without, but it certainly raises the possibility.

https://www.computerbase.de/2018-04...rmancerating-speicheroptimierungen-frametimes
 
what do you mean by optimized timings 16-15-15-15-35-51 t1 is really tight at 3466
 
Such a massive difference with the "LL" timings. Does LL mean tight?
So leave your Ryzen+ 2700x with stock cooler on auto. Spend that cooler $ on faster RAM, Switch the focus of your [H]-OC to your RAM.
Lets see some water block OC's on those sticks boys. LOL
 
funkydmunky said:
No hard limit.
Click to expand...

Most were stuck at 2933 initially; I knew that BIOS updates had gotten them up to 3200, did they improve beyond that?

Or is it just that they didn't improve enough beyond 3200 that dropping below CAS14 at 3200 wasn't worth it?
 
funkydmunky said:
Such a massive difference with the "LL" timings. Does LL mean tight?
So leave your Ryzen+ 2700x with stock cooler on auto. Spend that cooler $ on faster RAM, Switch the focus of your [H]-OC to your RAM.
Lets see some water block OC's on those sticks boys. LOL
Click to expand...

That's what it looks like, I plan to spend some time trying to get my ram to as tight of timings as I possibly can. Looks like it can be worth 10% to do so.
 
funkydmunky said:
Such a massive difference with the "LL" timings. Does LL mean tight?
Click to expand...
That was my assumption. I took it as meaning "Low Latency", or in other words, optimized subtimings. However, given the amount of gain there, it may simply be tighter primary timings.

Alas, Shadowed did not provide a reference link, and it's an uploaded pic to boot, so I have no clue what is actually implied by it :p
 
I need to tweak the subtimings more, then. But my RAM is limited to 3000. Hynix M die, double sided, 32GB (16 GB per stick). Honestly, 3000 is pretty good for that. Can't really expect more.

But I'm stuck at CL16. And I haven't even tried playing with the subtimings. Just left at the XMP/DOCP defaults. Seems there may be some fruit left to pluck there.
 
Neapolitan6th said:
Silicon Lottery have just released their 2600X and 2700X bins and the results are as you would think. Prices for the most part aren't that extreme since, well since the bins aren't so crazy.

Link

[email protected]@1.45vcore
[email protected]@1.425 vcore
Click to expand...

Yikes... That's a whole lotta volts... Sounds like either Duron knows more about what he's doing that they do, or he ended up with a golden chip when comparing to their results >_> Especially considering they aren't using any air coolers.

DuronBurgerMan, Original Post: "Okay so for the lulz I bought a 2700X and dropped it into my Asus Prime X370 Pro board today. <snip>. Note, I am running a Noctua U12 air cooler, so I imagine folks on water will do better."

DuronBurgerMan, Second Post: "Second attempt: 42x multiplier, Vcore @ 1.365. 4.2GHz total.

Booted to Windows. Appears fully stable at this vcore. Ran Cinebench successfully (score: 1883). Will conduct burn test later to verify stability."​


Seems like their SOP is: Buy Ryzen. Insert in motherboard. Boot with voltage at 1.425V. Find max stable overclock at said voltage. Re-test in next motherboard until no boards left. Done. CONSIDER IT BINNED!
No bother of finding if it'd handle a lower voltage? :(

Frankly, if I were buying a RYZEN chip, I'm more interested in what it's going to accomplish for a max RAM speed! I don't know if Ryzen 1 was limited by design or by node process (Stilt's massive initial research seemed to indicate node process as thermals skyrocket as voltage got passed around 1.37V), but it was pretty much a case of "Read three reviews, average overclock, you're guaranteed to get that." lol Doesn't seem like much point in bothering with a binned CPU for their core clocks. On the other hand, at least know what it can do as far as max RAM speed on its particular IMC, that'd go a long ways. Particularly if they include their data in printouts
 
Yeah, RAM speed seems to be the only real variable:

And I'd still like to see testing to confirm if it's worth pursuing anything other than DDR-3200 C14, at least at 2x8GB.
 
I've had a hard time trying to find info on how memory speed and timings affect performance, most of the tests I've seen focus on one or the other and the rest have all seemed incomplete in one aspect or another. From what I've seen so far it seems like tight timings are the most important thing but there are some nice performance increases to be had up until at least 3466 mhz if you can keep the timings tight, however it sounds like with some motherboards you need to do some heavy tweaking to see much improvement because they automatically loosen up some of the advanced timings when you increase speed or tighten the regular timings.

I've had an even harder time finding is what kits will work at higher than 3200/14 or 3466/16, which BTW seems to be about a wash. I've seen a few people mention using Trident 3466/15 kits and one person claim to get 3466/14 with a 4133 mhz rated Trident kit but that's a small sample size.
 
however it sounds like with some motherboards you need to do some heavy tweaking to see much improvement because they automatically loosen up some of the advanced timings when you increase speed or tighten the regular timings.
Click to expand...

That problem is as old as time. Saw it with Asus cheating on their mobos in prepping for the 200FSB update on Athlon XP. I remember wondering why my performance dropped so much at the very same clock, then I realized I had to seriously ramp-up my OC just to get my old numbers back.
Really made me question what board makers will do to save face and claim, "Hey we can run at xyz Mhz just like everyone else!"
 
Mav451 said:
That problem is as old as time. Saw it with Asus cheating on their mobos in prepping for the 200FSB update on Athlon XP. I remember wondering why my performance dropped so much at the very same clock, then I realized I had to seriously ramp-up my OC just to get my old numbers back.
Really made me question what board makers will do to save face and claim, "Hey we can run at xyz Mhz just like everyone else!"
Click to expand...
Thankfully with the chipset now being inside the CPU (for the most part), the mobo manufacturers just have to focus on literal design, by coming up with the "quietest" traces, and delivering the most efficient power flow. Beyond that it's down to who can pack the most tweakable options into the BIOS heh

I DID try out the MSI option for RAM to auto-overclock when it had been added into the Titanium's BIOS, but it only had gotten 3400 to POST one time, ever. lol Though, my BIOS is seriously outdated and I should consider updating it to see what new performance it can unlock. I've just been so happy and stable with 3200 14-14-14 that I haven't felt any need to update it. I've been rather fortunate in that regard.

The build-your-own computing scene... it's definitely been a journey though!
 
Here's something I don't get. I got 32GB of RAM because I work with video and rendering enough that I sometimes saturate 16GB. Back when I built this box (originally with a 1700X), I found the fastest 32GB kit I could. 3000 MHz. QVL said 2666 was the fastest it would run. An Agesa update broke that down to 2133. I eventually got it working at 2933 after months of headaches, bitching, and BIOS updates.

With the 2700X, I got it up to 3000 at CL16, or 2933 at CL14. Not sure which is ultimately better - I have a lot of benchmarks to run with both.

This is dual rank RAM too (Hynix M die). Appears to be the limit.

Is dual rank actually faster than single rank? I've read this in some places, but not sure if it's bullshit. I wonder if it's worth bothering trying to get better, faster RAM for this thing.
 
DuronBurgerMan said:
With the 2700X, I got it up to 3000 at CL16, or 2933 at CL14. Not sure which is ultimately better - I have a lot of benchmarks to run with both.

Is dual rank actually faster than single rank? I've read this in some places, but not sure if it's bullshit. I wonder if it's worth bothering trying to get better, faster RAM for this thing.
Click to expand...
Given the fact there's only 66MHz difference, I think it's safe to assume that 2933 @ CL14 is going to be quite a bit faster. That's definitely what I'd be running.

As far as the rank debate, honestly, I was also under the impression that the go-to configuration were Dual-Rank modules, and I've always tried to get and recommend modules populated on both sides of the stick (since as far as I know 9 times out of 10 that means it'll be Dual Rank). Initially when I bought my kit, I was bummed to find out after getting it (but before Ryzen was released) that they were Single Rank... I mean I got so damn lucky buying the kit I did, because I purely went off of reviews for overclocking stating that the Samsung B-dies yielded some spectacular results.... on Intel chips. Thus, I did all I could to get Sammy B's and didn't really pay attention to Single or Dual Rank. So you can imagine how happy and relieved I felt when everyone started to realize that if you want to run as fast as possible and to reliably hit those speeds, then you need Single Rank Sammy B equipped RAM! *phew* lol That, and getting them for $125 after rebates a week before Ryzen launched, when comparing it to what the same kit costs now.... *whistles* I haven't been that lucky in a looong time... But my luck kept going lol I was able to hit 3200 on the shipping BIOS as well!

IF you still have your 1700X and IF you'd be willing to do this... I'd be up to match whatever CPU clock you want to set (I know mine will do 3700 at 1.25V) and then dial my RAM in with the same timings you have, main and subs, and we can compare by running a couple benchmarks. Just be aware that my internet is lackluster on top of having a monthly limit, so I dunno if I'd be too keen on anything too substantial like 3DMark :p Then agian, we're be focusing on CPU and Memory speeds, so it'd be hard to properly run games and compare, unless you also had an 8GB R9 390 non-X to throw in your machine :p As far as sharing timings, I'd say the easiest way would be posting a screenshot of RyzenTimingChecker, as that seems to display all the available options that matter much and that we have the ability to change in the BIOS.

Because it'd be interesting to find out what the difference in performance is between single and dual rank of the same IC. Baring this proposition, that'll probably be left to someone more fortunate that us, who has a bunch of DDR4 at their disposal :p
 
Formula.350 said:
Given the fact there's only 66MHz difference, I think it's safe to assume that 2933 @ CL14 is going to be quite a bit faster. That's definitely what I'd be running.

As far as the rank debate, honestly, I was also under the impression that the go-to configuration were Dual-Rank modules, and I've always tried to get and recommend modules populated on both sides of the stick (since as far as I know 9 times out of 10 that means it'll be Dual Rank). Initially when I bought my kit, I was bummed to find out after getting it (but before Ryzen was released) that they were Single Rank... I mean I got so damn lucky buying the kit I did, because I purely went off of reviews for overclocking stating that the Samsung B-dies yielded some spectacular results.... on Intel chips. Thus, I did all I could to get Sammy B's and didn't really pay attention to Single or Dual Rank. So you can imagine how happy and relieved I felt when everyone started to realize that if you want to run as fast as possible and to reliably hit those speeds, then you need Single Rank Sammy B equipped RAM! *phew* lol That, and getting them for $125 after rebates a week before Ryzen launched, when comparing it to what the same kit costs now.... *whistles* I haven't been that lucky in a looong time... But my luck kept going lol I was able to hit 3200 on the shipping BIOS as well!

IF you still have your 1700X and IF you'd be willing to do this... I'd be up to match whatever CPU clock you want to set (I know mine will do 3700 at 1.25V) and then dial my RAM in with the same timings you have, main and subs, and we can compare by running a couple benchmarks. Just be aware that my internet is lackluster on top of having a monthly limit, so I dunno if I'd be too keen on anything too substantial like 3DMark :p Then agian, we're be focusing on CPU and Memory speeds, so it'd be hard to properly run games and compare, unless you also had an 8GB R9 390 non-X to throw in your machine :p As far as sharing timings, I'd say the easiest way would be posting a screenshot of RyzenTimingChecker, as that seems to display all the available options that matter much and that we have the ability to change in the BIOS.

Because it'd be interesting to find out what the difference in performance is between single and dual rank of the same IC. Baring this proposition, that'll probably be left to someone more fortunate that us, who has a bunch of DDR4 at their disposal :p
Click to expand...

1700X was already fleabayed, unfortunately. But I'll post the Ryzen Timing Checker results later this evening.
 
Running one of my 1700x with 4 sticks of 3200 Cas 14 (32gb) no issues, ROG CH6. I've have not pushed it faster. With 16gb I could go up to 3500mhz (with BCLK), 3466 was not an issue either. Anything over 3550mhz was an issue with 16gb. I would expect the 2700x to do better in general unless one gets an orphan cpu with low end performing memory controller.
 
I think I'm getting marginally faster performance with 2933 @ CL14 than 3000 @ CL16. May stick with this instead.
 

Attachments

  • 2933.jpg
    2933.jpg
    243.5 KB · Views: 0
IdiotInCharge said:
Are you familiar with the relationship between clockrate/timings and actual latency?

Don't even have to do the math to see that 2933 CL14 is significantly lower latency ;)
Click to expand...

Zen complicates things a bit because the Infinity Fabric runs at memory clock speed. I still figured it was probably faster, but it needed to be tested.
 
DuronBurgerMan I take it they can't handle 3000 @ 14-15-15-35? Even with a small DRAM voltage bump to like 1.37V (not sure what you currently have it set to)?

I'm no RAM aficionado but if you post that RyzenTimingsChecker shot, I may be able to point out some things I've read about for you to change, if they aren't already.
 
You must log in or register to reply here.
Back
Top