Yea @pete910 also reminded me how that works, i will, it might not be worth doing, my RAM are Micron B-Die, which are not great, but inexpensive and i wanted 32GB.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
MSI & ASUS Prep 4GHz RAM AGESA v1.2 BIOS - slow memory can fclk itself
- Thread starter philo-sofa
- Start date
Yea @pete910 also reminded me how that works, i will, it might not be worth doing, my RAM are Micron B-Die, which are not great, but inexpensive and i wanted 32GB.
But it's not just RAM, it's the fabric. And if the fabric is running faster then communication between chiplets and IO die is faster, thus reducing latency. And however good Zen 2 and Zen 3 may be, there is always going to be that latency penalty because of the MCM design, so anything that can be done to reduce that latency is good.
You're always going to get a performance boost from running faster RAM and a faster fabric, but whether that boost is tangible to your specific workload or worth the extra voltage and heat is entirely user-dependent.
Although true, sometimes it's quicker to run slow RAM\IF with tighter timings. I have debunked that in my setup so I think each system needs to be tuned.
Slow RAM with tighter timings, for sure, but a slower fabric slows comms between chiplets and IO. RAM and fabric speeds are not the same thing.
Come on, anything above 1800Mhz IF is theoretical paper exercise for benchmarketing. You won't see any considerable performance increase in any real world cases, maybe barring select one or two.
Yea that didn't turn out how i had hoped, i set it back to Auto, got my 3200Mhz ram running at 3400Mhz tho...
Like I said previously, there is always a performance boost. Whether it is actually realistic to implement and worth the tradeoffs is entirely down to your own use case. And no, 2GHz fabric isn't theoretical. Zen 2 required a lottery win to get there, but Zen 3's improvements are touted as making it more realistic. Why else would mobo vendors be pushing out BIOS updates for it?
Besides, we see Intel systems wringing the last ounce of RAM and ring speeds and marketing fluff for new boards touting ever-higher RAM clocks, why shouldn't we get the same for Ryzen? And look at it this way too, if a 2GHz fabric is realistic enough on mainstream desktop to warrant microcode and BIOS updates then it's sure as dammit going to be worth it for Threadripper and the baby-HEDT 5950X.
Why is the potential for more tweaking and performance headroom a bad thing?
It is not a bad thing, but it is completely pointless for what the sacrifice might be. BIOS updates are coming because the initial BIOS versions and AGESA is complete shambles, improved memory stability is just a byproduct of updates. AMD and mobo manufacturers time and time again show how unable they are to get stable and usable BIOSes from get go. They always need hundreds of different versions to even get a stable running system within XMP norms, let alone give us some headroom to push memory further. AMD and mobo manufacturers would never spend their time just to get you your 4ghz RAM with 2ghz IF, because you are 0.0001% of their market
Oh, so that's why AMD like talking about 2GHz fabric and shared their dismay at Ryzen 5000 not being able to hit it at launch. And how a 2GHz fabric is one of the design goals for improvements over Zen 2? Yeah, AMD really don't care about improving their offering gen on gen.
To you, maybe.
They also love to talk about better GPU availability at launch than nVidia, yet we know how that turned out to be.
So how many of you can hit that magical 2ghz IF, no no, how many of you have memory capable of running 4ghz with normal safe everyday volts?
From what I am seeing, people are RMAing their 5000 series CPUs until they get one which runs at 3600Mhz, let alone 4000mhz.
If 2ghz IF was design goal of Zen3, why didn't they change something on IO die, instead of reusing the same die as for Zen 2?
Obviously if people buy their PCs to finetune them for 6+months, then, yeah, every theoretical improvement is worth the wasted time.
And again, do you get any real world substantial performance increases going from 3600Mhz/1800Mhz to 4000Mhz/2000Mhz? Wasn't there a chart attached on this thread showing that even theoretical performance drops going from 3800/1900Mhz to 4000/2000Mhz?
- Joined
- 22 Oct 2012
- Posts
- 1,170
^^ It's interesting, the charts (copied below) did show a significant uplift in performance from increasing memory and fclk. There is a drop off above 3800; which suggest that the max speed achievable by all is about as fast as you want to go with Zen 3. The expectation/hope is that this drop-off point also increases by 200 MHz for AGESA 1.2.
1:1 fclk to speed ratio
Source: https://premiumbuilds.com/features/zen-3-ram-speeds-benchmark-analysis/
1:1 fclk to speed ratio
Source: https://premiumbuilds.com/features/zen-3-ram-speeds-benchmark-analysis/
- Joined
- 22 Oct 2012
- Posts
- 1,170
I think originally they said '4GHz is to Zen 3 as 3.6GHz is to Zen 2' when it comes to sweetspots. They then failed to fab chips which hit that goal with previous BIOSes so here's hoping....
Associate
- Joined
- 19 Dec 2017
- Posts
- 720
I found a bigger increase in performance going from 2x16gb to 4x8gb than I did from any jump in ram/IF speed.
It was only some quick back to backs in Timespy CPU, as my motivation to change wasn't really performance but I was curious nonetheless.
I've since settled on 3800/1900 CL18 as there were lots of WHEA errors at 4000/2000 and I was getting occasional memtest errors at CL16
It was only some quick back to backs in Timespy CPU, as my motivation to change wasn't really performance but I was curious nonetheless.
I've since settled on 3800/1900 CL18 as there were lots of WHEA errors at 4000/2000 and I was getting occasional memtest errors at CL16
oh yea.... ####'ing awesome, you don't need fast ram for that fast IF now.... something else for me to play with, this CPU just keep's giving and giving
Poor old humbug, so behind the times!
