oh dw, those frequencies are np for samsung b-dies.
they'll always be slightly better on ryzen as you can tighten subs quite a bit better.
the thing that matters least and will be even less important are the primaries n especially tCL as the cache is compensating for that, big time.

However there are some issues (especially in gaming) that no reviewers are really noticing.
If Zen3 didn't fix those you'll see the 8 core now topping the 5900/5950x in various application n games.

-> rig suggestion

anyways for ram:
2x16 better
3200 3600 not much of a difference if you do not tighten sub timings

This post has been edited by nrw: Nov 4 2020, 09:12 AM

for zen2 they just mirrored, so should be two ccds with 6 cores each for 5900x.
some issues with that.

effective clocks.
issues with both, synthetic and non synthetic loads.

they are lower than the displayed clock (this is no issue of the load getting split on multiple cores)
the performance impact on some dual ccd chips as a result of this is quite a bit.

this is one of the reasons aside from latency why zen2 is as much behind intel in gaming.

From what I saw so far, no. That one should be fine.
I'll be getting a single ccd chip which I can test extensively on weekend.
Can update then.

Little longer, should be 10pm for us Malaysians as the have their DST.
Lets hope at least one reviewer get it right and does extensive cache, IF testing and compares boosting and persistent clocks between single and dual ccd chips.
Counting on Luxx myself.

This post has been edited by nrw: Nov 5 2020, 09:18 AM

That one easy to do, every reviewer including our local reviewers should be able to do that properly.
Gaming with high end GPU will once more be depending on if and how they OC memory to test.

If they wanna test for the average person XMP is fine but I for myself wanna see what it can really do.

RAM OC is a major weak point on their side. Basically any OC is.
For Ampere they didn't even know they clock in 15MHz steps and compared +120 with +130 core oc which are EXACTLY the same thing after you hit apply. That's just sad for someone doing LN2 OC.

Then they even compare offsets without mentioning effective clocks, disregarding nvidias own boosting algorithms working on tdp, temp etc.
GN is great for basic reviews, that's about it.

Unfortunately the latest leaks seem to point in the direction that 3800 (1900IF) 1:1:1 might still be what we are looking at.

This is enthusiasts stuff, .. so before I get in an argument with you again I'm aware that this in not for the general audience but I still think once you touch OC and publish reviews, you should at least do it properly. (just like the Ampere clocking example up there.)

Tonight my first review to check will be by hardwareluxx (German site). They quite good when it comes to that.

Oh .. I know all the time, but still .. if you start to OC to prove a point (e.g. that this product is better than others, or that that product will deliver this or that performance) you should do it right.
If there's a flaw in your set up the results are just misleading those that don't know any better.

After all those reviewers influence others in their purchase decisions. They influence what others spend their money on.

Million dollar question. I hope some will do more than 1900 IF. But since that depends on the IOD I don't think so.
Maybe goldfires will test IF, too. Fairly quick to do that.

3900x and 3950x always had one higher and one average binned ccd yeah. ccx doesn't exist anymore for zen3.

the IOD seems to be the same (whats confirmed is that its still the same 12nm node by GF manufacturing it) ... we will see shortly, maybe some minor physical adjustments after all.

You can test the IF capabilities without adjusting ram speed.
Test with auto voltages and then if you wanna be certain can see if you can go higher by adjusting those 3 voltages connected to it.

Because of the way cache works with Ryzen for 4000c14 vs 4000c16 differences are very marginal as well.
Adjusting secondaries is much more important than adjusting tCL.
On Ryzen a well adjusted (primaries secondaries tertiaries) Micron Rev. E kit running at 3600(3733, 3800,...?)c16 is WAY better than any xmp kit you can possibly buy.

This post has been edited by nrw: Nov 5 2020, 12:08 PM

I'm famous with goldfries massive audience.



Edit:
To clarify, GN is great for review of basic items.
They are not when it comes to overclocking as they do mistakes as the one I mentioned.


This post has been edited by nrw: Nov 5 2020, 12:22 PM

Actually I like GN, if I was to buy a cooler they'd be the first site to check.



This is mostly discussed on discord servers. For RAM oc there are great communities on OCN (cstkl1 being a champ there on Intel) or hardwareluxx or computerbase.de
The latter even has some testing that can show you how much you can gain whenever you are in CPU limits.

And yes I'll do my best getting a zen3.

that's valid... should be a brazilian chap listening to it.
pretty much confirms amd's own benchmarks.

n gives some insight on difference between the zen3's he tested.

If you prefer a written review you can check this one:
https://translate.google.com/translate?sl=p...lowi-13913.html

This post has been edited by nrw: Nov 5 2020, 03:13 PM

-

This post has been edited by nrw: Nov 5 2020, 08:08 PM

3 week old chip, not bad.

so there are chips doing 4133 FCLK (booting on it) and allegedly 4066 stable

sure, what are you paying.