Infinity Fabric question.

AVATARAT

Limp Gawd
Joined
Jun 16, 2020
Messages
272
I wonder and I can't find information on how bad would be to use IF @ 2000MHz for 24/7

The IF on my 5600x can run at 2066MHz, but there are errors in Windows in HWinfo, at 2033MHz I have errors, but rarely. Therefore, I set it to 2000Mhz, I have no errors and the system is stable at the moment and there is no problem even with OCCT Small FFT for 20 minutes and 1 hour MemTest64 (I will test it more in the next few days).

However, I wonder if it is healthy for the CPU to work like this permanently?
 
Did you have to crank up voltage to get it to work at 2000mhz, or is everything just stock?
 
Last edited:
Stock, Auto as voltages for CPU and SOC.
Then I don’t see any reason why it wouldn’t be ok long term. You’re also quite lucky - my 5950 won’t go over 1866 IF with stock voltage. I’m still tuning mine and haven’t tried to boost the SOC voltage yet.
 
OK, I am at home and did few screenshots in the BIOS.

210112201059.png
210112200249.png
210112200334.png
210112201902.png
210112200418.png
210112200534.png

testOCCT.png
 
I've been running 24/7 @ 1866, but that's more the limit of my ram than anything. To keep hw errors down, I had to crank up soc voltage on mine. So far I've had this up and running 1.5 months now with only reboots to test new bios updates/graphics drivers.
 
Please add ZenTimings screen, it summarizes all voltages and memory settings pretty well. Unless your voltages are way off you can run it this way, I must say you're pretty lucky if it works out of the box ;)
 
oh that's indeed Asus default :) great results, I may need to check this new BIOS as I have very similar motherboard (just full ATX version)
if FCLK is stable you can try and lower some other timings, get Dram Calculator for Ryzen, check some tutorials and slowly start tightening these secondary timings
 
Unfortunately Dram Calc got me timings that didn't work, and I will need to do it by the slow way :(
But I am happy with the result that I had now, I want just a bit longer test with MemTest64 (TPU) :)
 
Oh you can't apply them all at once, and you have to test what's working and what's not ;) some of the timings are connected, so just leave everything you don't change on auto.
 
Not to hijack this thread, but i read on the asus rog forum that reducing SOC voltage could help booting up with higher IF speed. I set mine to 1.025 and it worked great - I managed to boot and compete cinebench with ram and IF set to 2000 (haven’t figured out where to set uclk yet). Unfortunately, moving to 2000 dropped my cinebench score about 300 (29504 -> 29212).

what’s really annoying in the crosshair hero viii bios is most of these settings are in there twice. I have no idea what happens if you set one value, but not the other...
 
If you can run 2000 then she should be good. Also its the same with my bios too, same stuff is in a different section. Seems like a waste.. the could have uclk settings there instead of the extra code for that lol.
 
Downloaded zen timings and it showed I only had IF at 1866. Found another place in the bios to set it to 2000 - it boots, but that’s about all it does before crashing.

i also had to use the bios search feature to find the UCLK divider setting. Set it to match memory speed and it’s working great at 1866.

I’m now playing with curve optimizer more before trying to tackle IF again.
 
I've been on 2000 for a couple of weeks now, no issues so far. I had my first system crash yesterday, but I was undervolting the GPU which may have caused it :)
 
GB 5 is not so strong but in CB 15 / CB 20 the single core is good. I can't get both strong or maybe I need better cooling than current water loop, I am not sure.

GeekBench 5

cinebench-15-2019.png

Cinebench-20-4729.png


BCLK Frequency [99.6875]
ASUS Performance Enhancement [Enabled]
Max CPU Boost Clock Override [525]
Platform Thermal Throttle Limit [79]
VDDCR CPU Load Line Calibration [Level 2]
VDDCR CPU Current Capability [110%]
VDDCR CPU Switching Frequency [350]
VDDCR CPU Power Phase Control [Extreme]
VDDCR CPU Power Duty Control [Extreme]
VDDCR CPU Voltage [Offset mode]
VDDCR CPU Offset Mode Sign [-]
VDDCR CPU Offset Voltage [Auto]

Curve Optimizer [Per Core] [Negative]
Core 0 Curve Optimizer Magnitude [7]
Core 1 Curve Optimizer Magnitude [1]
Core 2 Curve Optimizer Magnitude [10]
Core 3 Curve Optimizer Magnitude [2]
Core 4 Curve Optimizer Magnitude [12]
Core 5 Curve Optimizer Magnitude [8]
 
That's as low as you can go on CO? As all the numbers are different I assume you put in the insane effort to test each core individually?
 
When you set CO Per Core you can optimize max boost on each core to boost a 25-75 MHz more than if you set it just on auto.
It depend what you want, more single or more multi thread.

You can control it by:
Negative CO per core,
LLC 1 or 2 or Auto
Negative Offset Voltage


boost.PNG


boost-2.PNG
 
Last edited:
Yes I know what CO does, just seems like a lot of work to individually tune each core. And your magnitude is quite low; I have my best cores at -10 and the rest at -20. Depends on silicon quality.
 
You can set -10 or -20 when you set maximum boost to +200 up to +300.

If you try to go up (as me +525) you need to set it lower.
And yeah you need to test how much can go each core.

But the base is following:
Weak core can go -8 to 12.

Medium can go to -6 to 8.

Strong core must be lower so -0 to -2

And there is one more trick and it is BLCK frequency, I drop it with few steps, these is the last thing to be "unlocked" ~25MHz more.
My best core were up to 5125 in the beginning now it go to 5150MHz and few times I saw it to 5175MHz.
 
Wow, that is very impressive. And benchmarks confirm you're getting better performance, not clock stretching?

Yes, I'm at +200 boost.
 
Here is the difference, clocked by multiplayer manual 47.25
You lose some multi to gain some single.

cinebench-15-OC-4725.png
 
What's your sustained single core? I can only get 4950 sustained, transient peaks are like 5125 but not sustained. Cpu-z sc 683 but I'm only running 3266 ram, so with 3600 or 3800 I'm guessing it'd be 700ish. My cpu can handle pretty hefty CO undervolt at -30 for all but the 4 best cores, which are at -20. I get around 4.6ghz all core but edc limited (190A). Temps aren't bad, top out around 85C under long stress tests.

I'm thinking higher end boards with hefty vrm are actually gong to be worth it this gen as if I had better cooling I could push this a little more, but pretty well matched atm. Pulling around 215W.
 
You can try Super Pi for single core test or BoostTester.
About sustained megahertz, they depend from the core load but my are similar to yours.
For better result you need 17-20C in the water loop and B-die RAM

Here tested for you 1600 IF / 3200Mhz RAM.

1612889698051.png
 
Finally I got to the moment to clock my RAM. The bad is that I back to 100 BCLK because of hardware errors, the good is that I optimize my RAM timings :)
DRAM Calculator for Ryzen was helpful but maximum there is 1933MHz IF so I just used it and edit a bit the result from it.
I test it to 500% here is just to 340% but ye, no errors with this timings:

memtest_106_min.PNG
 
Finally I got to the moment to clock my RAM. The bad is that I back to 100 BCLK because of hardware errors, the good is that I optimize my RAM timings :)
DRAM Calculator for Ryzen was helpful but maximum there is 1933MHz IF so I just used it and edit a bit the result from it.
I test it to 500% here is just to 340% but ye, no errors with this timings:

View attachment 329045
That's pretty impressive clocks on the cpu and infinity fabric. Looks like stock voltage on the soc too? If not can you share your voltage settings and how you got the cpu clocked that high? Pbo with curve optimizer?
 
Thanks :)
SoC is on auto voltage it goes higher if I move to more than 2000 IF.
All BIOS settings are in post #20 I have few numbers changed only in CO and BCLK and 0.5V for RAM (1.44 > 1.44500) :)
 
2000 IF is nuts. I can't get over 1866 with newest agesa at CL14 mem. I had bought fast memory to try and see how hard I can push the IF, but it was not meant to be on my x570 msi board. Not sure why AMD said 2000 would be the sweet spot when it was impossible to acheive on so many boards. That's like saying '5.8ghz seems to be where the ryzen 5xxx series really comes into its own'
 
Did you try 2000 ?

I have errors in HWinfo on 1966 and 1933 but it work fine on 1900 and 2000 IF, so you can just try ;)
And do not try at CL14, you better start with high CL and if it work then try to lower it.

I use AGESA 1.1.9.0 because with the newest AGESA my max boost is locked to 4850MHz.
 
Oddly enough, 1900 won’t even boot for me, but I can run for a day (ish) on 1933 without errors, and a few hours at 1966 and 2000. Only 1866 is long term stable for me though.
 
The main problem is with the RAM it need to work 1:1 with the Infinity Fabric, so you need to match them to work together.
You need to look how much voltage wants your IF to work over 1900. As much as I know over 1.10v it is not recommended.
So you need to look did your motherboard set enough (or too much) voltage for SoC and did your RAM have enough voltage too (it's depend of the quality of the die (A-die/B-die etc) on the stick / revision).
The RAM timings are very important for the stability too, so you must try with high in the beginning.

_____________
The memory that I use is:
TeamGroup Xtreem 16GB, DDR4-4500, CL18-20-20-44 / 1.45v
 
I’m running 2x 32gb Gskill 4000. Have tried roughly 30 different variants on voltages, and started with super relaxed RAM timings. Synced CLDO and CCD voltage seems to work a little better than different, but no change seems to stabilize 1933 for 48 hours of prime grid.

1616862430859.png
 
Back
Top