14th Gen Intel Core - Meteor Lake

sbawe64 je 25. apr 2024 ob 09:51 izjavil:

Samo Intel i9-14900K je omenjen v novici, in sicer padec hitrosti med 8 in 11%.
So kot kaže tovarniško preveč navili vse skupaj in silicij več ne zdrži kot na začetku.

Pri AMD s CPB (core performance boost) in pbo (Precision Boost Overdrive) se zna situacija ponoviti, sploh če ga 24/7 ženeš na 95°C.

Kolikor je meni znano, je bil problem pri x3d amd modelih.

Pri intu pa je 100°C, unlimited Power unlimited, voltage da dosezes te frekvence..
Tako se namesto 10+ let cpu degredacije, doseze v 2 letih...

Sicer pa kaj pricakovati od gen 12 intel cpu, ki se prepakira ++ do gen 14. Na koncu pa je to dosezeno z unlimited power.

Tisti razumni uporabniki, ki so v BIOS-u nastavili omejitev porabe za cpu, imo kakšnih pretresov ne bodo doživeli.
Zmogljivost pri utilizaciji nekaj jeder tudi ne bi smela trpeti.
Se bo pa poznalo pri določenih testih.
Me zanima, če bosta GN in TPU imela jajca, da bi določene zadeve naknadno testirala. HU prav gotovo.

Glede

Pri AMD s CPB (core performance boost) in pbo (Precision Boost Overdrive) se zna situacija ponoviti, sploh če ga 24/7 ženeš na 95°C.

PBO (curve optimizer) ni problematičen, ker lahko dejansko undervoltaš cpu (offset krivulje v minus), CPB je druga zgodba - tega imam na disabled.

Jarno je 25. apr 2024 ob 14:52 izjavil:

Tisti razumni uporabniki, ki so v BIOS-u nastavili omejitev porabe za cpu, imo kakšnih pretresov ne bodo doživeli.
Zmogljivost pri utilizaciji nekaj jeder tudi ne bi smela trpeti.
Se bo pa poznalo pri določenih testih.
Me zanima, če bosta GN in TPU imela jajca, da bi določene zadeve naknadno testirala. HU prav gotovo.

i9-14900K je best, kar ima intel ponuditi, če hočeš top hitrost in taki uporabniki ne bodo šli omejevati cpu.
Ćista psihologija.

Mr.B je 25. apr 2024 ob 12:26 izjavil:

sbawe64 je 25. apr 2024 ob 09:51 izjavil:

Samo Intel i9-14900K je omenjen v novici, in sicer padec hitrosti med 8 in 11%.
So kot kaže tovarniško preveč navili vse skupaj in silicij več ne zdrži kot na začetku.

Pri AMD s CPB (core performance boost) in pbo (Precision Boost Overdrive) se zna situacija ponoviti, sploh če ga 24/7 ženeš na 95°C.

Kolikor je meni znano, je bil problem pri x3d amd modelih.

Pri intu pa je 100°C, unlimited Power unlimited, voltage da dosezes te frekvence..
Tako se namesto 10+ let cpu degredacije, doseze v 2 letih...

Sicer pa kaj pricakovati od gen 12 intel cpu, ki se prepakira ++ do gen 14. Na koncu pa je to dosezeno z unlimited power.

Pri x3d amd modelih je bil problem v kombinaciji z Asus ploščami, kjer je bila privzeto previsoka Vsoc voltaža (več kot 1.300V, kar naj bi ibla zgornja meja), kar je trajalo nekje konca aprila 2023 (torej od začetka 09/2022), da so končno pri asusu naredili primeren bios.To tem fiasku so iz njihove strani izbrisali/odstranili vse stare biose. Gigabyte, msi, biostart in asrock tega problema niso imeli pri am5 ploščah.
https://www.tomshardware.com/news/amd-r...

Za koliko časa mislite, da bo 12400F cpu v redu? Jaz sem reko kolegu približno 6-7let.

Če se razpase uporaba AI gimmickov preko v cpu vgrajenih "NPU" pospeševalnikov, se zna zgodit, da bodo vsi trenutni cpuji "neuporabni" v nekaj letih. Zdi se, da si industrija tega zelo želi ...

Jaz itak mislim enkrat naslednje leto menjati komp. Počakal bom na novi platformi Intela in AMDja.

Han je 26. apr 2024 ob 12:44 izjavil:

Meni X3370 (=Q9650) po več kot 15 letih še vedno dobro služi.

C2D E6600 sem jaz mel leta 2006, pol sem presedo na i7 930 in sedaj sem na i9 10900X.

Intel cilja na velike uspehe v 2025. Roko na srce, Krzanich je bil kr'zanič CEO, kravatar, Gelsinger je podedoval hudo stagnacijo na področju litografij.
Vsaj za okolje bo dobro, če bodo za potrošnike dobavljali "K" robo s porabo tam 125W.

V tem trenutku se mi zdi, da pri AMD-ju nekako vemo, kaj lahko pričakujemo z zen5 in kasnejšimi X3D izvedenkami.

Pri Intelu tega načeloma ne vemo, kar pa je lahko tudi povod za nov Hype-Train (procesorji na Intel 3 procesu...). Vsakomur svoje.

Ob tako nizki ceni Intel delnice bi jih pa veljalo malo naložiti. Se moram dogovoriti, da mi zopet aktivirajo degiro račun.
Zdaj, ko je padlo, mtosev še omenja, da je delničar? Lima zaslonske slike?

Intel issues statement about CPU crashes, blames motherboard makers -- BIOSes disable thermal and power protection, causing issues

Intel has observed 600/700 Series chipset boards often set BIOS defaults to disable thermal and power delivery safeguards designed to limit processor exposure to sustained periods of high voltage and frequency, for example:
- Disabling Current Excursion Protection (CEP)
- Enabling the IccMax Unlimited bit
- Disabling Thermal Velocity Boost (TVB) and/or Enhanced Thermal Velocity Boost (eTVB)
- Additional settings which may increase the risk of system instability:
- Disabling C-states
- Using Windows Ultimate Performance mode
- Increasing PL1 and PL2 beyond Intel(R) recommended limits

Intel requests system and motherboard manufacturers to provide end users with a default BIOS profile that matches Intel recommended settings.

Osebno mi je že 170W 2-chiplet zen4 procesorjev preveč, Intelov uradni PL2 za "i9" znaša 253W..."No way, Hose!", da to dam v ohišje.

Nič ni zamolčal.. MOBO proizvajalcem so samo predlagali osnovne nastavitve, niso jih pa strogo forsirali. S tem so si zavarovali rit, hkrati pa upali, da bodo proizvajalci plošč v želji po dobičku "overclockali" nazivne settingse preko priporočenih do čiste skrajnosti.. In točno to se je tudi zgodilo, rezultat pa ta afera. Neizogibno, saj gre že v osnovi za požrešno arhitekturo.

Intel mora samo "preživet" do konca leta, do Arrow Lakea.. Takrat bo vse to pozabljeno. Sledi popoln reset. Napredna litografija, povsem nove tehnologije, novo poimenovanje.. Nobene 15. generacije, ampak 1. oziroma 2. generacija.

Trenutno se pač šlepajo.. Takorekoč prodajajo tehnologijo iz leta 2021.. 12. generaciji so zlepili nekaj dodatnih E-jedrc in malo predpomnilnika.. To je to.. To sta 13. in 14. generacija. :S

toše, tole ni AMD tema.
Ravno ti, ki se rad igraš moderatorja ga mimo sekaš.

Da pa ne bo off topic: če (tebi tako omiljeni) Intel nekaj zaserje ga to, da se s prstom pokaže na AMD, ne opere krivde. Da, tako preprosto je to.

Vidim, da so pri asusu izdali BIOS update za B760M-Plus WIFI D4, ki sem jo predlagal znancu. Darwin je tukaj reko, da alder lake ni affected samo bom vseeno updajtal, da bo plošča podpirala 13th in 14th gen CPUje.

FireSnake je 27. apr 2024 ob 21:39 izjavil:

toše, tole ni AMD tema.
Ravno ti, ki se rad igraš moderatorja ga mimo sekaš.

Da pa ne bo off topic: če (tebi tako omiljeni) Intel nekaj zaserje ga to, da se s prstom pokaže na AMD, ne opere krivde. Da, tako preprosto je to.

$AMD delnica spet pada, huh ?

Jarno je 25. apr 2024 ob 14:52 izjavil:

Tisti razumni uporabniki, ki so v BIOS-u nastavili omejitev porabe za cpu, imo kakšnih pretresov ne bodo doživeli.
Zmogljivost pri utilizaciji nekaj jeder tudi ne bi smela trpeti.
Se bo pa poznalo pri določenih testih.
Me zanima, če bosta GN in TPU imela jajca, da bi določene zadeve naknadno testirala. HU prav gotovo.

No, pri HU so res po-testirali i9 14900K z baseline profili:

chatgpt 3.5 povzetek zgornjega videa (25 minut) https://slo-tech.com/forum/t812629/p820... :
- Intel CPUs are experiencing crashes, potentially due to silicon degradation, particularly with 13th and 14th gen KCU processors.
- The issue is linked to Intel's power specifications, with many Z790 motherboards not adhering to the recommended PL1 and PL2 limits of 253 watts.
- While some boards, like MSI, allow CPUs to run without limits, others, like Gigabyte, enforce lower power limits, impacting performance.
- Intel has attempted to address the issue with a new Intel Baseline Profile, but there's confusion and variation in how motherboard manufacturers implement it.
- Intel seems to shift blame onto board partners, despite knowing about the issue for years and acknowledging that running CPUs without power limits is technically within spec.


claude-3 opus
Here are 5 key points summarizing the video:
1. Intel has had loosely defined power specifications for years, causing inconsistencies in how motherboards handle power limits for Intel CPUs out of the box. This has led to varying performance between boards.
2. Recent reports suggest some 13th and 14th gen Intel CPUs are crashing during gaming, likely due to silicon degradation caused by motherboards allowing the CPUs to run at high power levels without enforcing Intel's power limits.
3. Intel has introduced a new "Intel Baseline Profile" for motherboard BIOSes to address the stability issues. Asus and Gigabyte have released BIOS updates with this profile, though Gigabyte's implementation results in more significant performance drops.
4. Benchmarks comparing the Intel Baseline Profile to default non-power limited settings show performance drops ranging from 2-11% on the Asus board and 5-25% on the Gigabyte board, depending on the game and settings.
5. The video argues Intel has been aware of motherboards running CPUs without power limits for years and considered it "in-spec" behavior, but is now trying to avoid blame for the resulting stability issues by introducing the Baseline Profile and putting responsibility on board partners.


gemini-1.5-pro-api:
## Intel CPUs Crashing: 5 Key Takeaways
* **Intel's Loose Power Specs:** Intel CPUs, particularly 13th and 14th gen, are experiencing crashes possibly due to silicon degradation. This is attributed to Intel's loosely defined power specifications, which have been a long-standing issue.
* **Motherboard Manufacturers Implement Varied Power Limits:** Motherboard manufacturers like ASUS, MSI, Gigabyte, and ASRock interpret Intel's power specs differently, leading to inconsistent performance. Some boards have no limits, pushing CPUs to their maximum potential, while others enforce stricter limits.
* **Performance vs. Stability:** Running CPUs without power limits can result in higher performance but at the cost of increased power consumption, heat generation, and potential silicon degradation.
* **Intel Baseline Profile Introduced:** To address stability concerns, Intel introduced the "Baseline Profile", aiming to standardize power limits and ensure safe operation. However, implementation varies across manufacturers, leading to performance discrepancies.
* **Intel's Role & Responsibility:** While motherboard manufacturers share some responsibility for the inconsistencies, the core issue lies with Intel's lack of clear and enforced power specifications. This has created confusion and potentially contributed to CPU degradation issues.



Iz yt video komentarjev:
#1
Gotta love how the "intel baseline" isn't the same across motherboard vendors. Almost like the intel baseline isn't provided by intel but cooked up by each vendor on their own.
Also it's worth noting that SVID set to intel fail safe runs more Vcore than the ASUS default settings for any scenario that doesn't hit the power limit. Gigabyte's intel baseline profile does the same. Most gigabyte boards use AC_LL and DC_LL of less than 1.1mOhms on default but with the baseline profile the AC and DC_LL get set to 1.7mOhm which raise the core voltage for any workload that doesn't hit power limits.

#2
1.7 mOhm? Why is it so high? I'm running 0.20 mOhm. And SVID scenario set to "typical".

#3
1.7mOhm is absolutely insane and above the Intel Spec for the 8 core SKUs. 1.7mohms is only acceptable 6 cores and lower.

Tega zdaj ni potreboval Intel.. Že tako jim prodajne številke pešajo, predvidevam.. Vsekakor na DIY trgu.. Zdaj pa še ta blamaža.

Nekaj je jasno. Arrow Lake bo izšel takoj ko bo možno.. Mogoče še prej, v obliki paper launcha. :S

Pol pa upajmo, da bo res revolucionaren izdelek, kot se obljublja. Čeprav po govoricah bi naj bili samo I5 (6+8 core) modeli delani na 2nm procesu, preostali pa na 3nm od TSMCja.. Dokaj logično, da Intel še nekaj časa ne bo imel dovolj 2nm kapacitet. Kakorkoli, tudi 3nm ni od muh in glede na vsa vlaganja Intel tokrat takorekoč ne more razočarat. Edino porodni krči. Jasno, ker gre za povsem novo platformo. Največji tehnološki preskok v zadnjih ~20 letih?

Glede na vse navedeno, bom počakal na 15to generacijo Intela, sme hotel en DEV PC zamenjati, samo me malo skrbi, če ne bo HT:
https://semiwiki.com/forum/index.php?th...

Vsaj do sedaj je HT bil zelo koristen.

DarwiN je 29. apr 2024 ob 17:10 izjavil:

So različne teorije o umaknitvi HTja.. Moja teorija je, da ker HT zahteva določen kos silicija, so ta kos preprosto raje namenili e-jedrom.. Ta bodo zato zmogljivejša in več jih bodo skupaj "zlepili". Očitno jih ta prva gen. ne bo imela več, ampak sčasoma 40 in več e-jeder. Nima smisla komplicirat s fizičnimi jedri, virtualnimi in še e-jedri.

Ampak glede na ves napredek bi moralo tudi teh 8+16 ARL jeder občutno prekositi to skropucalo v obliki 14. generacije.

Mimogrede, 15. generacije ne bo.. Meteor Lake (samo laptopi) je 1. generacija, Arrow Lake bo 2. generacija, itd..

Intel želi to nočno moro s starimi "10nm" izdelki čim prej pozabiti. Like it never even happened. :)

Ko bo deliver, se pa slišimo, do takrat... bo vsaj eno generacijo zadaj...

sbawe64 je 28. apr 2024 ob 21:27 izjavil:

chatgpt 3.5 povzetek zgornjega videa (25 minut) https://slo-tech.com/forum/t812629/p820... :
- Intel CPUs are experiencing crashes, potentially due to silicon degradation, particularly with 13th and 14th gen KCU processors.
- The issue is linked to Intel's power specifications, with many Z790 motherboards not adhering to the recommended PL1 and PL2 limits of 253 watts.
- While some boards, like MSI, allow CPUs to run without limits, others, like Gigabyte, enforce lower power limits, impacting performance.
- Intel has attempted to address the issue with a new Intel Baseline Profile, but there's confusion and variation in how motherboard manufacturers implement it.
- Intel seems to shift blame onto board partners, despite knowing about the issue for years and acknowledging that running CPUs without power limits is technically within spec.


claude-3 opus
Here are 5 key points summarizing the video:
1. Intel has had loosely defined power specifications for years, causing inconsistencies in how motherboards handle power limits for Intel CPUs out of the box. This has led to varying performance between boards.
2. Recent reports suggest some 13th and 14th gen Intel CPUs are crashing during gaming, likely due to silicon degradation caused by motherboards allowing the CPUs to run at high power levels without enforcing Intel's power limits.
3. Intel has introduced a new "Intel Baseline Profile" for motherboard BIOSes to address the stability issues. Asus and Gigabyte have released BIOS updates with this profile, though Gigabyte's implementation results in more significant performance drops.
4. Benchmarks comparing the Intel Baseline Profile to default non-power limited settings show performance drops ranging from 2-11% on the Asus board and 5-25% on the Gigabyte board, depending on the game and settings.
5. The video argues Intel has been aware of motherboards running CPUs without power limits for years and considered it "in-spec" behavior, but is now trying to avoid blame for the resulting stability issues by introducing the Baseline Profile and putting responsibility on board partners.


gemini-1.5-pro-api:
## Intel CPUs Crashing: 5 Key Takeaways
* **Intel's Loose Power Specs:** Intel CPUs, particularly 13th and 14th gen, are experiencing crashes possibly due to silicon degradation. This is attributed to Intel's loosely defined power specifications, which have been a long-standing issue.
* **Motherboard Manufacturers Implement Varied Power Limits:** Motherboard manufacturers like ASUS, MSI, Gigabyte, and ASRock interpret Intel's power specs differently, leading to inconsistent performance. Some boards have no limits, pushing CPUs to their maximum potential, while others enforce stricter limits.
* **Performance vs. Stability:** Running CPUs without power limits can result in higher performance but at the cost of increased power consumption, heat generation, and potential silicon degradation.
* **Intel Baseline Profile Introduced:** To address stability concerns, Intel introduced the "Baseline Profile", aiming to standardize power limits and ensure safe operation. However, implementation varies across manufacturers, leading to performance discrepancies.
* **Intel's Role & Responsibility:** While motherboard manufacturers share some responsibility for the inconsistencies, the core issue lies with Intel's lack of clear and enforced power specifications. This has created confusion and potentially contributed to CPU degradation issues.



Iz yt video komentarjev:
#1
Gotta love how the "intel baseline" isn't the same across motherboard vendors. Almost like the intel baseline isn't provided by intel but cooked up by each vendor on their own.
Also it's worth noting that SVID set to intel fail safe runs more Vcore than the ASUS default settings for any scenario that doesn't hit the power limit. Gigabyte's intel baseline profile does the same. Most gigabyte boards use AC_LL and DC_LL of less than 1.1mOhms on default but with the baseline profile the AC and DC_LL get set to 1.7mOhm which raise the core voltage for any workload that doesn't hit power limits.

#2
1.7 mOhm? Why is it so high? I'm running 0.20 mOhm. And SVID scenario set to "typical".

#3
1.7mOhm is absolutely insane and above the Intel Spec for the 8 core SKUs. 1.7mohms is only acceptable 6 cores and lower.

Članek na to temo na anandtech:
https://www.anandtech.com/show/21374/in...

Kdaj pride arrow lake? Še letos ne?

DarwiN je 29. apr 2024 ob 17:10 izjavil:

So različne teorije o umaknitvi HTja.. Moja teorija je, da ker HT zahteva določen kos silicija, so ta kos preprosto raje namenili e-jedrom.. Ta bodo zato zmogljivejša in več jih bodo skupaj "zlepili". Očitno jih ta prva gen. ne bo imela več, ampak sčasoma 40 in več e-jeder. Nima smisla komplicirat s fizičnimi jedri, virtualnimi in še e-jedri.

Ampak glede na ves napredek bi moralo tudi teh 8+16 ARL jeder občutno prekositi to skropucalo v obliki 14. generacije.

Mimogrede, 15. generacije ne bo.. Meteor Lake (samo laptopi) je 1. generacija, Arrow Lake bo 2. generacija, itd..

Intel želi to nočno moro s starimi "10nm" izdelki čim prej pozabiti. Like it never even happened. :)

HT ne vzame veliko prostora. Okoli 5 %. Kar ni veliko, v zameno pa dobiš do 30 % več zmogljivosti. Aritmetičn-logične enote so precej bolje izkoriščene, je pa poraba potem tudi bolj skoncentrirana na manjši del in posledično so lahko problemi s temperaturo. Skupaj z out of order execution pa zgleda, da imajo probleme z varnostjo. Pa še dve niti si delita predpomnilnik.
Neke okvirne številke bi pomenile, da bi prihranili za 40 % enega P jedra, če bi umaknili HT iz 8 jeder. V prostor pa bi lahko dali npr. dva E jedra. 8+2 se ne sliši tako dobro kot 8/16+2. Ceneje pustiti HT in vreči ven celo P jedro in ga nadomestiti s 4 E jedri.
Sicer bomo videli, kako bodo nadomestili zmogljivost, če ven vržejo HT in ga nadomestijo s čim drugim.

Intel ukinja HT, ker je P/E core kobinacija bolj BS. In se jo izogibajo...
Sedaj je in ti. Rentable Units, kjer se ka razbije thted glede na zahtevnost.... No pa se vec cacha zahteva... iscite technical deep dive 16th/17th gem

bbbbbb2015 je 30. apr 2024 ob 11:58 izjavil:

HT komplicira več stvari. Upravljanje s kešem, prefetch execution, virtualizacijo - imaš recimo zahtevo VMWARE, ORACLE VIRTUALBOX, da alociraš toliko jeder na virtualno mašino, koliko jih imaš, ne virtualno število. Precej se poslabša out-of-order ekskucija, single thread performance je vedno slabši.

HT je dober za en idealen primer, kjer ena nit dela intenzivno I/O obdelavo (=prenos iz pomnilnika), druga nit pa dela kalkulacije, kjer ima podatke/programe v kešu (večinoma). Začetni Pentiumi so imeli precej počasen pomnilnik, tam je to prišlo do izraza.

ARM nikoli ni imel HT, niti AMD. Z več-nitnim pristopom je še eksperimentiral SPARC:
SPARC @ Wikipedia

kjer so število niti še povečali na recimo osem niti na jedro. Na SPARCU to deluje drugače, ker ima čisto RISC arhitekturo in ima recimo registrsko okno, t.j. določen del registrov je na voljo (eni niti), drug del pa rugi, tretji tretji, ipd.

HT se da povsem bom rekel povsem ukiniti, če se potem poveča out-of-order eksekucija, keš.
Recimo IBM z15 dela samo na tem:
https://en.wikichip.org/wiki/ibm/microa...

Tako da verjetno se da doseči enak ali boljši rezultat brez HT.

Komplicira že, ampak spet ne toliko, da ga ne bi implementirali. Bi rekel, da upravljanje s predpomnilnikom niti ni kaj zahtevnejše, namreč z out-of-order execution lahko tudi ena nit zateva več dostopov "naenkrat", prefetch enako. Single performance je lahko za odtenek slabši, ampak danes večinoma tako ali drugače uporabljamo večnitne OS in programe. Če eno jedro brez SMT zmore na eno časovno enoto sprocecirati eno nit, jedro s SMT pa eno nit in še malo dodatne, je prednost jedra s SMT jasna.
VirtualBox priporoča sicer uporabo fizičnih jeder, ampak pusti tudi čez.
Singlethread zmogljivost je praktično enaka. vir: Anandtech.
Multithread prinaša večinoma koristi, izjemoma slabosti. V povprečju 22 % pri AMD 5950X. V nekaterih primerih tudi po 60 % več.

Arm Cortex-A65 ima SMT. 2 niti na jedro.
Ene arhitekture morda res ne potrebujejo SMT, imajo druge koristi, namene, so optimizirani za druge vrste operacij kot x64 ali pa kaj tretjega. Ne bi vedel točno.
AMD pa Intel vseeno precej prostora na siliciju namenita za zadnjih par 100 MHZ več in širšim podatkovnim ukazom, bi bilo škoda, da bi bile te enote slabše izkoriščene. Zanimivo bo videti, kaj bo Intel naumil in kako bo nadomestil izpad HT.

Upam, da čimprej rešijo zadevo.