AMD Ryzen 7000 "Zen 4" Processors Have DDR5 Memory Overclocking Design-Focus

[Deleted member 24505]

Hm, I haven't used much time for RAM overclocking since the legendary BH-5 DDR1 chips..

Best i have ever had was this DDR2
https://www.techpowerup.com/forums/posts/293068

Was good enough for my near world record E6300 OC

 

[Bloax]

Overclocked memory that is famous for giving users only a few percent real-world speed advantage, if any at all, is not gonna make a "big splash" with me.

Depends on what you need in terms of "real-world speed" - most things are perfectly good with a locked 12 ms frametime on variable-refresh displays, which is super easily achievable on Zen3/ADL.
Other things are more low latency focused, which makes them very stutter-sensitive. They want a stable, high framerate, without any massive distracting stutters (very low minimum FPS).

6.7% Increase in average FPS? Pretty meaningless.
17.4% Increase in low-FPS scenarios? Very nice.
68.9% Increase in minimum FPS? Incredible! (visible in a much less "pretty" graph)

 

[FeelinFroggy]

Wow haven’t seen a true troll in awhile. I thought they finally left the internet a year or so ago.

Every now and then you will see one out in the wild again.

 

[Bloax]

It is a prayer to the great Trickster God, each time you speak the truth while simultaneously giving people pie to the face.

And it is an entirely serious post, if delivered comedically - power users are important for helping regular users not misconfigure their system, and suffer.
Unfortunately "misconfiguring" has been extremely easy on AM4, so I hope it will get harder to mess up - and easier to correct - on AM5.

 

[Jism]

I hope this is'nt a Fury repeat of how that was supposed to be a OC dream.

 

[thesmokingman]

I hope this is'nt a Fury repeat of how that was supposed to be a OC dream.

That guy in charge was let go remember. He's with Intel now lolz. But to your point, crosses fingers...

 

[AusWolf]

View attachment 243580
Depends on what you need in terms of "real-world speed" - most things are perfectly good with a locked 12 ms frametime on variable-refresh displays, which is super easily achievable on Zen3/ADL.
Other things are more low latency focused, which makes them very stutter-sensitive. They want a stable, high framerate, without any massive distracting stutters (very low minimum FPS).

6.7% Increase in average FPS? Pretty meaningless.
17.4% Increase in low-FPS scenarios? Very nice.
68.9% Increase in minimum FPS? Incredible! (visible in a much less "pretty" graph)

I still think it's meaningless because
1. You don't see 0.2 percentile values much. It can mean just one minor stutter that you don't even notice.
2. The data shown is way above 200 fps with all configurations. I can't even tell the difference between 60 and 80 fps, let alone 200.
3. This is one single game - a hand-picked example from among all the games, the majority of which are GPU-limited unless you have an extremely slow CPU.
4. (Edit) The data for this graph was taken with a mid-/high-tier graphics card at probably low game settings. No one in their right mind would play a game like that. You're always going to be GPU-limited at higher settings and/or with a more affordable card.

Edit 2: Just to illustrate what I mean: A 100% difference between 25 and 50 fps is significant, but a 100% difference between 100 and 200 fps is nothing because you can't see it. At that 25 to 50 fps range, you're GPU-limited 99% of the time. The remaining 1% is a hard CPU limit. You can't pair a fast, modern CPU with such slow RAM that will bottleneck it in a perceptible way. Ever. Unless you're playing on your iGPU that is.

according to Joseph Tao who is a Memory Enabling Manager at AMD.

I never knew such a job existed. Learn something everyday.

I just find it funny that nearly everybody is a manager of something nowadays.

If you won't give 'em more money, just give 'em a title, I guess.

 

[Mussels]

Good - XMP is flawed like this


I saved my overclocked SPD timings to XMP2 on my RAM (yeah, it's a thing. we have a thread on it) and some of the values simply dont get read, and rely on automatic timings. Huge latency difference with manual tweaking, huge waste of time, too.


If they can fix that situation up from the start with better translation from XMP and better automated settings (tight/loose default options) it could make some very big changes for user friendliness (loose option would make 4x sticks a lot easier to run) and performance (tight auto settings for the common 2xsticks of ram setup)

 

[Eskimonster]

I mean, thats a good thing since Ryzen based processors really like higher speed memory. I will be curious what gains you get on Raphael with higher frequencies and where the performance gains start dropping off.

Are you trolling me with infinity fab

I would love to see them doing higher clocks on mem.