Samsung, Hynix and Micron memory scaling on Ryzen 3000

[TheLostSwede]

I figured I'd share this here, as I think quite a few people would be interested in reading this and it's the first site I've seen that has done a comparison between different memory types on Ryzen 3000.

It would seem that there's a difference between the various memory types, even if it's rather small, even when pushed to the limit.
With B-dies it would appear you need to hit really low timings to get the best latencies, whereas Hynix is a tad more forgiving and Micron seemingly also being a bit more forgiving, but neither can hit as low latencies and Samsung if you can hit the right timings.

AMD Ryzen 3000 DDR4 Scaling – Part II – Samsung B-Die vs Micron E-Die vs Hynix CJR – English version

Samsung B-Die vs Micron E-Die vs Hynix CJR One of the main changes that came with the launch of the X570 chipset, along with the launch of the third generation Ryzen processors, was the way …

lab501.ro

 

[hzy4]

How is this possible? Hynix j-die CL16-21-20-36 has better latency then Samsung B-die CL16-16-16-36.

 

[Deleted member 50521]

How is this possible? Hynix j-die CL16-21-20-36 has better latency then Samsung B-die CL16-16-16-36.

Manufacturing improvement duh. B-die is not even in production any more

 

[hzy4]

Manufacturing improvement duh. B-die is not even in production any more

What are they making the Tridentz NEO from, old B-die stock? A-die?
How come B-die is more expensive then any other ram?

 

[Calmmo]

I've 2 CL16/3600 rated kits, 1 bdie and 1 CJR.
If I ever do manage to get a 3900x i'll test it just for the luls. ;o
F4-3600C16D-32GTZN // PVR416G360C6K

 

[TheLostSwede]

My Hynix CJR has really impressed, I can run it faster than spec and at much tighter timings at that. In fact, my timings are better than in the link above.

 

[Cowboystrekk]

Interessting, looking forward to gamingcomparisons

 

[Ripcord]

How is this possible? Hynix j-die CL16-21-20-36 has better latency then Samsung B-die CL16-16-16-36.

Hynix is running 1t the Samsung is running 2t

 

[TheLostSwede]

Interessting, looking forward to gamingcomparisons

Not sure it matters all that much in games.
However, it shows that it's kind of pointless wasting money on expensive B-dies when CJR is just as good, for a lot less money.

 

[Cowboystrekk]

Not sure it matters all that much in games.
However, it shows that it's kind of pointless wasting money on expensive B-dies when CJR is just as good, for a lot less money.

With ryzen 1 and 2-gen especially the tRFC-value made the B-die several percent faster than Hynix and Micron. In ryzencalc, the main difference between Samsung and the others is that you can run tRFC much lower and on coffee lake and previous ryzen-gens it could account for quite a bit of performance. I wonder how much tRFC impacts performance on the 3-gen. Looking at the article they were able to use 300 tRFC on B-die at 3800, but had to use 485 on CJR and 545 on E-die. I tested on my coffee lake at it alone can contribute to several percent performancedifference in games. My previous laptop (i7 6700HQ) compared 2x4 Samsung 2133cl15 vs Micron 2x8 2666c15. With XMP enabled I could set both to run at 2666cl15 With problems, but getting the tRFC under 425 on the Microns were impossible, on the Samsung I could og slightly below 300 without issues. In some games the difference alone meant 2-4% better performance With the Samsung.

 

[phanbuey]

With ryzen 1 and 2-gen especially the tRFC-value made the B-die several percent faster than Hynix and Micron. In ryzencalc, the main difference between Samsung and the others is that you can run tRFC much lower and on coffee lake and previous ryzen-gens it could account for quite a bit of performance. I wonder how much tRFC impacts performance on the 3-gen. Looking at the article they were able to use 300 tRFC on B-die at 3800, but had to use 485 on CJR and 545 on E-die. I tested on my coffee lake at it alone can contribute to several percent performancedifference in games. My previous laptop (i7 6700HQ) compared 2x4 Samsung 2133cl15 vs Micron 2x8 2666c15. With XMP enabled I could set both to run at 2666cl15 With problems, but getting the tRFC under 425 on the Microns were impossible, on the Samsung I could og slightly below 300 without issues. In some games the difference alone meant 2-4% better performance With the Samsung.

^ this is so true, also the same in my experience for skylake X - i would get a massive boost dropping TRFC from 560 to 360, much more noticeable than going from 4.6 to 4.8ghz on a 7820x. It also stopped some games from noticeably stuttering. On my 8700k TRFC also has an impact, but not as big as it was on SKL-X.

 

[TheLostSwede]

Again, what does Intel have to do with this? Why is there always someone posting in these types of threads going "My Intel platform, bla, bla, bla".

 

[phanbuey]

Again, what does Intel have to do with this? Why is there always someone posting in these types of threads going "My Intel platform, bla, bla, bla".

The importance of sub timings, specifically TRFC (also tFAW, tCKE) , on latency-starved processors (mesh/IF) - there is a Ryzen graph out there that shows the same thing.

 

[TheLostSwede]

The importance of sub timings, specifically TRFC (also tFAW, tCKE) , on latency-starved processors (mesh/IF) - there is a Ryzen graph out there that shows the same thing.

Yes and this has been discussed in plenty of Ryzen related threads already, so maybe start your own thread if you want to discuss it about Intel.
Did you even read the source link? The graph I included in my post specifically covers tweaked/optimised performance settings.

Also, I don't go into discussion threads about Intel stuff and post about my AMD rig, as it's irrelevant.

 

[hzy4]

However, it shows that it's kind of pointless wasting money on expensive B-dies when CJR is just as good, for a lot less money.

I wish I would have known this before i paid 180euro for b-die memory

 

[TheLostSwede]

I wish I would have known this before i paid 180euro for b-die memory

Well, I got lucky, as I just took a chance on something that I could afford and it has turned out to be really impressive modules.
It's also something no-one really knew from the beginning and it has taken a couple of months for my RAM to get to where it is now, in terms of UEFI support.
So early on, B-dies were the safe bet. Now it looks like affordable CJR is the way to go, unless you want to tweak your system to the edge of what's possible.

 

[Apocalypsee]

Hynix is running 1t the Samsung is running 2t

As far as I know, if 'Gear Down mode' is enabled it runs command rate at sort-of 1.5T regardless whether if its set at 1T or 2T

 

[phanbuey]

Yes and this has been discussed in plenty of Ryzen related threads already, so maybe start your own thread if you want to discuss it about Intel.
Did you even read the source link? The graph I included in my post specifically covers tweaked/optimised performance settings.

Also, I don't go into discussion threads about Intel stuff and post about my AMD rig, as it's irrelevant.

I apologize for posting my experience about my intel rig, but my points were not to be OT but:

1) to agree with a ryzen post about how the secondary timings have become even more important in some cases than frequency/core timings (i specifically quoted him)
2) your source does have them 'tweaked' but doesn't explicitly talk about them, you have to go and look at screenies of bioses to see them...

I found that post and this whole thread helpful in learning about zen 2 scaling (so thank you for that).

 

[TheLostSwede]

I apologize for posting my experience about my intel rig, but my points were not to be OT but:

1) to agree with a ryzen post about how the secondary timings have become even more important in some cases than frequency/core timings (i specifically quoted him)
2) your source does have them 'tweaked' but doesn't explicitly talk about them, you have to go and look at screenies of bioses to see them...

I found that post and this whole thread helpful in learning about zen 2 scaling (so thank you for that).

Yeah, the guy writing the article is an old overclocker and often glances over important things that most of us would find useful, like settings...
However, I haven't seen any other comparison like this and that's why I thought I'd share it.

 

[Cowboystrekk]

The graph I included in my post specifically covers tweaked/optimised performance settings.

In aida, bit gaming is a different animal Hynix "won" in aida latency in the ryzencalc-test, but samsung was often 5% ahead in gaming, mainly due to tRFC.

 

[TheLostSwede]

In aida, bit gaming is a different animal Hynix "won" in aida latency in the ryzencalc-test, but samsung was often 5% ahead in gaming, mainly due to tRFC.

But is that 5% worth twice the cost or more, as that's what we're looking at.
Obviously it might be for some, that are looking to get the most out of their system, as I pointed out above.

 

[Cowboystrekk]

But is that 5% worth twice the cost or more, as that's what we're looking at.

I'm not saying that But I'm curious to see how much of an impact it has. Where I live B-die costed only 140usd for a while, cheapest E-die costed 110usd so in that case I would consider B-die. Now the difference is larger and I would chose E-die any time but it can change again. Currently 160usd vs 95 usd where I live.

 

[nangu]

My Hynix CJR has really impressed, I can run it faster than spec and at much tighter timings at that. In fact, my timings are better than in the link above.

Same here, I'm impressed by my CJR kit (Gskill F4-3600C18D GTZRX). It's rated at 3600 18-22-22 and I'm running it at 3733 16-19-19 1T Geardown disabled, and with very little effort. Just load the XMP profile in the calc, copy the safe preset timings and that's it.

The graph shown by OP shows B-dies have 1ns better latency than CJR in 2x8 SR configuration, so CJR is not faster. Also, if you look at the Membench scores in the DRAM Calc, Bdies have better scores overall and by a decent margin, which I think they are faster in games too. In my opinion it's not worth the price difference to have B-die if you are using Ryzen 3000, but if you are after the best numbers, B-dies are still the best memory IC you can buy.