Overclock.net banner
Cancel
Create thread New Forums

Degrading 13th and 14th Gen CPU's

Jump to Latest

Has your Raptor Lake (13/14 Gen) degraded?

  • Yes my 13th gen stability has degraded (no overclock)

    Votes: 3 5.3%

  • Yes my 13th gen stability has degraded (overclock)

    Votes: 5 8.8%

  • No my 13th gen stability has not degraded (no overclock)

    Votes: 5 8.8%

  • No my 13th gen stability has not degraded (overclock)

    Votes: 16 28%

  • Yes my 14th gen stability has degraded (no overclock)

    Votes: 5 8.8%

  • Yes my 14th gen stability has degraded (overclock)

    Votes: 1 1.8%

  • No my 14th gen stability has not degraded (no overclock)

    Votes: 8 14%

  • No my 14th gen stability has not degraded (overclock)

    Votes: 14 25%
  • Total voters
    57
41 - 56 of 56 Posts
#41 ·
Save
Reply Quote Rep+ Rep+
#43 ·
deidian said:
Yes there is. Windows scheduler is aware of Turbo Boost Max 3.0, Intel's scaled turbo levels and also knows about the different per core PLL(P-cores can clock independently)

In a nutshell in balanced power plan will sort the cores by max clock speed. Then it will schedule tasks primarily to the fastest P-cores, if HT On both logical cores will be loaded heavily and the least priority tasks will be offloaded in small intervals to other P-cores or E-cores. Someone running one of these CPU stock in balanced will have mostly 1 of the favored cores handling the bulk of the system load at its maximum clock speed and the rest do support at small intervals to keep the 1 favored core at max clock most of the time.

What would be the point of having CPU doing 6.0Ghz in single core for best responsiveness if then the OS spreads the load around keeping most cores actives which prevents the CPU from going to 6.0Ghz?

Of course multi-core loads will use all cores but for everyday procrastination the Windows scheduler will try to take advantage of the single core boost. This behavior can be changed by having all cores boosting to the same clock speed(the scheduler will spread the load), changing to power saver plan(scheduler spreads the load and tries to keep low Vcore but sacrifices system responsiveness) or using high performance plan(scheduler attempts to get best throughput: every task is handed to any available core ASAP, but thrashes TBM 3.0)
Click to expand...
So when running Excel, it will occasionally run 5% faster? Thats great! I think everyone should risk the stability of their cpu for that extra 5% in apps like Excel and Paint.
 
#44 ·
That's why they're favored cores: Intel supposedly picks them because they're more suited to run high clocks. It's a trade-off cooling responsiveness: on an air cooler is feasible to run single core 5.9Ghz(or 6+ if cooling is that good). And CPU can handle that around 1.45V(6Ghz+)

On another hand me running all-core 6.0Ghz on a 13900KS needs to set 1.55V+ so the CPU runs stable in light loads if cooling is good enough. Don't even start speaking about how much of vulcano this CPU are when running all-core heavy loads.

The intent is getting the best of both single and multi-core worlds. It's also used for server grade CPU which can run wild clock ranges like from 2.4Ghz to 5Ghz on a 32 or 64 core package, but no way that CPU is running 5Ghz all-core without fusing itself. In the end with serious multi-core CPU you can't just set for max clock all-core like it was possible in the 4 or less cores era.

People crashing isn't running them stock as per Intel specs. They're running board 'auto' settings which do stuff like running 5.6Ghz all-core(R23 stability tested by the board manufacturer) when Intel says all-core is 5.4Ghz. The instability mess has nothing to do with TBM 3.0
 
#45 · (Edited)
persizi said:
Because for some reason 15.0.3.7 catches more instabilties than 15.0.3.8. For some reason BenchMate makes it run more stable than it really is. Everyone's free to test however they like, I just share how it is easier for stability purposes.
Click to expand...
Um....
 
Save
Reply Quote Rep+ Rep+
#46 ·
Betroz said:
That's some black magic going on there.
Click to expand...
The Benchmate problem is known. This was discussed some time back. Benchmate is a launcher and probably sets some priority or other parameters that change stability, I do know this was a topic some time ago regarding y-cruncher. But I call complete bullcrap on the .7 thing.
 
  • Rep+
Reactions: Betroz
Save
Reply Quote Rep+ Rep+
#47 ·
persizi said:
Because for some reason 15.0.3.7 catches more instabilties than 15.0.3.8. For some reason BenchMate makes it run more stable than it really is. Everyone's free to test however they like, I just share how it is easier for stability purposes.
Click to expand...
Betroz said:
That's some black magic going on there.
Click to expand...

Ok so I just checked this.
Had difficulty finding 0.3.7 that wasn't infected by a virus. Ran one version from one of those weird chinese sites and nothing happened (looked like a virus tried to get installed but failed) but apparently no damage. (everything seems clean after after checking with malwarebytes, rebooting and checking again).

(Next time dude, if you're going to post, give links okay?).

But found an official version of 0.3.7 at Guru3d.

www.guru3d.com

Cinebench 15 Download

Here you can download Cinebench 15. CINEBENCH is a real-world cross platform test suite that evaluates your computer's performance capabilities. CINEBENCH is based on MAXON's animation software CIN...
www.guru3d.com www.guru3d.com

So I checked this by installing the R15 extreme mod that was the texture pack only (rather than the full exe, which had .8).
And what the hell? This dude is right.....

at x58, 1.375v set + LLC6, v7 would not even complete a run, didn't even get 50% through one. Two app crashes and one app freeze. No WHEA was logged, it just appcrashed over and over. While .8 (with the same mod) completed four runs in a row consecutively without problems. (any lower voltage this would cause issues). I had to literally use stockfish vmin in order to get v7 to pass...literally 25mv more vcore (at least).

This guy wasn't kidding....
 
  • Rep+
Reactions: charonme, FR4GGL3, Paradigm Shifter and 2 others
Save
Reply Quote Rep+ Rep+
#49 ·
persizi said:
Next time read better, dude. The number is link :)
Click to expand...
Sorry, hyperlinks like this are amazingly difficult to notice on small screens or phones...
 
  • Rep+
Reactions: Paradigm Shifter
Save
Reply Quote Rep+ Rep+
#50 ·
Falkentyne said:
So I checked this by installing the R15 extreme mod that was the texture pack only (rather than the full exe, which had .8).
And what the hell? This dude is right.....
Click to expand...
I downloaded v7 and applied the Extreme mod. So this may be the go-to stability test for those who don't run Stockfish, OCCT or Prime95 FMA3?
 
  • Rep+
Reactions: Paradigm Shifter and Falkentyne
Save
Reply Quote Rep+ Rep+
#51 ·
@persizi
I stand corrected, CB15 v15.0.3.7 it is :)
 
  • Rep+
Reactions: persizi and Paradigm Shifter
Save
Reply Quote Rep+ Rep+
#52 ·
Hi all, is this R15 Extreme mod meant for 15.0.3.8 only? Because it is really insane hard to pass. I can pass vanilla 15.0.3.7 for 5 times in a row (maybe more, I always check 5 fast runs). But I need a ton of extra millivolts for the combination of 15.0.3.7 and Extreme mod. Are we shure that there is no bug or something? It is really strange that .3.7 is harder to pass than .3.8 both vanilla...

I even necroed a thread on Guru3D because I hoped that maybe the original Poster of this Mod might know something.
 
#54 · (Edited)
FR4GGL3 said:
Hi all, is this R15 Extreme mod meant for 15.0.3.8 only? Because it is really insane hard to pass. I can pass vanilla 15.0.3.7 for 5 times in a row (maybe more, I always check 5 fast runs). But I need a ton of extra millivolts for the combination of 15.0.3.7 and Extreme mod. Are we shure that there is no bug or something? It is really strange that .3.7 is harder to pass than .3.8 both vanilla...

I even necroed a thread on Guru3D because I hoped that maybe the original Poster of this Mod might know something.
Click to expand...
There is no bug. Your CPU is simply not as stable as you think it is. You need more voltage.

Try running Stockfish AVX2 at your vmin you tested in 15.0.3.7, with a free UCI compatible chess client like Arena or SCID or Chessbase ($$), you will get WHEA errors after some minutes, or a BSOD. 15.0.3.7 required 1.325v bios set LLC6 to pass 3 loops of R15 mod at 5.7/4.5/4.5 ghz (at just <82C), while Stockfish requires 1.355v bios set to pass hours (<87C).

The extreme mod is only a texture pack which increases the rendering time so you don't have to run the bench over and over (or use a script to run it automatically).
No it's not strange that 15.0.3.7 is harder to pass than 15.0.3.8. 15.0.3.7 came out in year 2013 (around Sandy Bridge/Ivy bridge era) and 15.0.3.8 came out in 2016 (Skylake era).

I am unable to find specific information, but there was a discussion about 15.0.3.7 triggering AVX offsets on Sandy and Haswell, even though it doesn't use AVX instructions. But I don't believe this was an issue with the program but with microcode itself, as you could trigger AVX offsets back then just loading the windows desktop! In fact, that's how the discussion about Apex Legends using AVX started happening, then the game developer posted on EA forums saying that the game did not use AVX at all, (Yet it triggered AVX offsets on a 9900k), and he explained that AVX is actually an extension of SSE3 instructions), and SSE3 can look a lot like AVX to hardware.

My guess is that there may have been a change in .3.8 that may have prevented AVX offsets from triggering incorrectly, but I don't know. I think this topic was discussed on notebookreview forums when this minor version change came out, but again I do not remember, that was 7 years ago. If the SSE instructions were slightly changed in .8 to work around this issue with AVX offsets, perhaps those instructions are easier to run than the ones in .7 (but again I have no way to prove this, I can't be bothered installing this on my Skylake laptop and seeing if the .38 version vs .37 version triggers AVX or not). But this could very well be true, just like how the BMI2 (Haswell) version of Stockfish requires more voltage than the AVX2 version of Stockfish, even though they use the same current and power.... (in fact I've seen different builds dates of just the AVX2 Stockfish have different vmins in just AVX2...I find a vmin, update the dev build 1 week later, then it generates a WHEA....)
 
  • Rep+
Reactions: persizi
Save
Reply Quote Rep+ Rep+
#55 · (Edited)
persizi said:
So after applying the extra millivolts the bug is gone? :)
Click to expand...
I haven't got to that point. Because the Temps got real high I backed up for now, but I might be close. I can run 7 with the mod 2 times in a row. The 3rd time it gets either an WHEA or the Program locks up.

@Falkentyne
I use Stockfish Chess on Arena client since the time I first read about it - from you. The thing is, that Stockfish Chess drives my CPU into the 253W limit on my settings*. So it throttles from 5.5 GHz (stock allcore boost on my 14700K) to 5.3 GHz where it can run quite comfortable around 75°C. So my "Stockfish stable Voltage" is lower than the r15 stable Voltage and doesn't help at the moment.

This is why I am interested in the combination of R15.0.3.7 and this extreme mod. This doesn't run into the Wattage Limit and seems to show me some limits. I just want to make shure that testing the Combination of the mod with the older R15 Version is something that makes sense. Because whenever I saw the Extreme Mod it was packaged with R15.0.3.8 not a single time with .3.7. I don't know if that has a specific reason. And @persizi first mentioned R15.0.3.7 without the mod, so I thought that could be an important thing. As I said: I can run 0.3.7 without the mod.


*What I try to find at the moment is a everyday Setting like Intel wants it to be with some undervoltage. So I am on AC_LL=DC_LL=VRM_LL=1.1 mOhm, CEP, TVB, eETVB etc. all set to on. But my MSI Board has some special settings like two CEP Options.



Okay, got it. I needed also 0.025 less Offset. Went from 73°C on 0.3.8 to 79°C on 0.3.7 because of this. You where right. No Bug, just more Volts. Jesus this is hard.
 
#56 ·
Falkentyne said:
So I checked this by installing the R15 extreme mod that was the texture pack only (rather than the full exe, which had .8).
And what the hell? This dude is right.....

at x58, 1.375v set + LLC6, v7 would not even complete a run, didn't even get 50% through one. Two app crashes and one app freeze. No WHEA was logged, it just appcrashed over and over. While .8 (with the same mod) completed four runs in a row consecutively without problems. (any lower voltage this would cause issues). I had to literally use stockfish vmin in order to get v7 to pass...literally 25mv more vcore (at least).
Click to expand...
Is there a way someone to check if R15.0.3.7 has the same effect on AMD CPUs?
 
41 - 56 of 56 Posts
About this Discussion
55 Replies
16 Participants
Last post:
persizi
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling