Overclock.net banner
1,061 - 1,080 of 1,783 Posts
Is it a public discord @bscool ?
 
@Ichirou

Looked and no newer bioses for MSI on discord. I did see one guy say the newest bios 192 needs higher vvdq than previous. He is on 13th gen though so not sure if it applies to 12th gen. just passing along info.
So yet another downgrade, huh? Why am I not surprised.

For 12th Gen non-12900KS, you'd just use the February BIOS, lol :p

I guess I won't get my hopes up for now. I'll try the latest BIOS for the 13900K, but if it flops, I'll test the modded BIOS.
 
Did you try with Memory Fast Boot disabled?
That worked! Running 4133g1 stable on BIOS 1.92. Here's the REAL interesting part, I did stability test, passed, and then re-enabled fast boot (put it back on Auto), and it still passes stability tests. Maybe it just trained poorly when it first booted on 1.92?

I also find it interesting how the tRTL's are very different between BIOS 1.2 and 1.92.

These are on BIOS 1.2:
Font Technology Screenshot Parallel Terrestrial plant


And these are on BIOS 1.92:

Font Technology Display device Audio equipment Multimedia
 
That worked! Running 4133g1 stable on BIOS 1.92. Here's the REAL interesting part, I did stability test, passed, and then re-enabled fast boot (put it back on Auto), and it still passes stability tests. Maybe it just trained poorly when it first booted on 1.92?

I also find it interesting how the tRTL's are very different between BIOS 1.2 and 1.92.

These are on BIOS 1.2:
View attachment 2577992

And these are on BIOS 1.92:

View attachment 2577993
You need to set these to Dynamic, only then they'll be picked correctly. Most likely it'll prevent the board from booting.
 
That worked! Running 4133g1 stable on BIOS 1.92. Here's the REAL interesting part, I did stability test, passed, and then re-enabled fast boot (put it back on Auto), and it still passes stability tests. Maybe it just trained poorly when it first booted on 1.92?

I also find it interesting how the tRTL's are very different between BIOS 1.2 and 1.92.

These are on BIOS 1.2:
View attachment 2577992

And these are on BIOS 1.92:

View attachment 2577993
They're worse, lol...
But that might mean 4,000+ MHz CL13 is doable. (I've done tests myself.)
Try it out yourself.
 
I can say with some confidence that 1.92 bios with my 13900k will outperform both my 12600k and 12700k in the memory department. My previous best bios was 1.4 and would net me 4100 15-15-15-30 with tightened sub and tertiary timings.

With the 13900k I am able to push to 4300 still G1. You can see more details in the screenshot below.
Blue Azure Font Screenshot Software
 
So. I updated the Intel Management Engine + Chipset drivers before installing the 13700k. It just works. Stock Vcore is rather high, my old 12700k was under 1.3v by default, this CPU uses 1.36v stock. I just put in the same AC_LL / DC_LL values that already worked well before (20/100). It now runs 2x56, 8x55 with less Vcore than before. Max Vcore is 1.326v.


Font Rectangle Material property Parallel Screenshot
 
Are these drivers board specific? Do you have link to these?

I've seen the ASUS boys needing to update Intel ME and stuff but haven't heard if its a thing for us MSI guys
It's not board specific. I didn't know if MSI is affected. I just read Gigabyte had the same issue so I wanted to avoid it.

 
So I found something strange that I don't have an answer to. It seems Iccmax throttles the CPU more than it should. Baseline is:

13700k
56x2
55x8
Highest V/F point +25mv
CEP disabled
AC_LL 20
DC_LL 100
LLC8

Based on this post from here:



<245 amps for daily (Cinebench R23 or lower @ <1.250v @ 245A), 307 amps for virus mode (prime95 small FFT AVX, stockfish, linpack, y-cruncher <1.182v @ 307A)
Using loadline slope 1520mv - (1.1 mohm * Amps) = Vcore target.

Of course it's your chip, do whatever you want with it, feel free to disagree with me or throw 1.38v load into it in Y-cruncher SFT test, just don't complain if it degrades..
If you look in the data sheets, Intel specifies 245-307A for iccmax. Here's Cyberpunk with 307A:
Midnight Darkness Machine Font Electricity


5.5Ghz, 1.30v, 126W. If my math is correct this is only 96A. Now I alt+tab go to Intel XTU and lower the ICCmax to 245A:
Midnight Window Electricity Space Darkness

5.1Ghz, 1.22v, 95W. This doesn't make any sense?

Tarkov with 307A:
Plant Cloud Sky Automotive tire Vegetation

5.5Ghz, 1.33v, 86W.
Plant Sky Cloud Combat vehicle Automotive tire

5.1Ghz, 1.24V, 74.W

It doesn't make any sense to me. These games for sure do not use as much power as Cinebench or some other stress testing tool, you can easily see this in terms of temperature and thermal throttling. Yet the iccmax value affects CPU throttling at stages where I think it should have no influence at all.
 
  • Rep+
Reactions: fray_bentos
Assuming this is true, it means you can hardly push the CPU past stock settings because you will run into IccMax even slightly above stock voltage.

If we give Intel credit for correctly predicting these limits in terms of CPU lifetime, it means overclocking would be severly limited. Like yes you could, but it will kill the CPU.

I am pretty sure I degraded my 12700k that I had before.
 
Personally I keep IccMax at 240A with my i7-12700K, Intel's spec value. Unlike power limits, it is actually part of the electrical specifications of the CPU, and it's not intended to be exceeded even though motherboard manufacturers tend to set it to high levels by default.

The results you reported may however also show some weaknesses in the current throttling mechanisms, in that average operating frequency will be kept low even if the current limit is likely only reached a relatively few times per second (since the average package power is low, as expected for gaming workloads).
 
I fully agree. It is not wise to exceed the electrical limits. I'm just surprised they limit my gaming performance, because this is exactly why I got this CPU. I need high clocks in low-load(ish) scenarios. Just look at Tarkov this is a 15FPS difference.
Maybe this is also because these CPUs are so new and future BIOS updates will fix it. Maybe it's also an MSI only issue. But it's strange nonetheless.
 
I fully agree. It is not wise to exceed the electrical limits. I'm just surprised they limit my gaming performance, because this is exactly why I got this CPU. I need high clocks in low-load(ish) scenarios. Just look at Tarkov this is a 15FPS difference.
Maybe this is also because these CPUs are so new and future BIOS updates will fix it. Maybe it's also an MSI only issue. But it's strange nonetheless.
So I just set ICCMax to 200 amps on my Z790 Maximus Extreme with my CPU manually set to x52 at 1.235v bios set (LLC4).
The CPU clock alternated between 4.4 to 4.5 ghz at idle, with 3600 mhz on E cores, and 4200 mhz during a Cinebench R23 run pulling 142 amps.

Translation: Don't even bother :)
 
Translation: Don't even bother :)
I am not sure what this means. Does ICCMax not work correctly in general? Or it does and we choose to ignore it? Did the reviewers set their boards to Intel standard or are all the results invalid because they ignore Intel specs?
 
It has been like this at least since Gen11 Intel processors. I had an 11700k and 11900 before my current 12700k, and IccMax also there caused average CPU current to get limited significantly lower than the configured value, although I haven't experienced severe discrepancies like you do in some games.

I think this setting is just generally misinterpreted: it works on peak current, which depends also on the instructions used (e.g. AVX instructions may cause higher peaks and cause current throttling to engage more often).

To actually limit average core current from the voltage regulators into the CPU there's "IA TDC Enable" on some motherboards (I had it on an Aorus B560).
 
1,061 - 1,080 of 1,783 Posts