WEE(K)-Log #27: Breaking CPUs and Records in Taiwan



Find out more about G.Skill Zeta R5 DDR5 RDIMM:
bit.ly/42ooOGj

———————————————————
Support me on Patreon:
https://www.patreon.com/der8auer

———————————————————

Music / Credits:

Outro:
Dylan Sitts feat. HDBeenDope – For The Record (Dylan Sitts Remix)

———————————————————
Paid content in this video:
– G.Skill Spot
– ASUS paid for my Hotel

Samples used in this video:
– All Hardware

———————————————————
Timestamps:
0:00 Intro
1:11 Public transport in Taiwan
1:50 Taking a taxi to ASUS
2:35 ASUS “Showroom
3:25 The Bench Room
4:36 Preview: WireView ecosystem
5:28 OC Impressions 1
6:24 Sapphire Rapids OC
8:09 G.Skill (Advertising)
9:00 Temperatures & FIVR at SR
11:27 Xeon w9-3495X with 4.7 GHZ All-Core
12:00 How far can we push with LN2?
13:52 5.2 GHz All-Core: The Limit?
15:19 Sapphire Rapids with RTX 4090 HOF
16:11 Elmor & Shamino: 3DMark CPU Profile Benchmark
17:40 OC Impressions 2
18:25 Evening Update 1
20:05 Elmor Labs
20:58 Prototypes at Elmor
22:40 Ryzen 7000X3D: Overclocking?
26:11 Forgot to lock the voltage?
28:15 Risk only with X3D
29:09 Evening Update 2
31:17 Sapphire Rapids & RAM
32:17 Evening Update 3
34:50 Outro

source

35 thoughts on “WEE(K)-Log #27: Breaking CPUs and Records in Taiwan”

  1. So the x3d chips are basically extreme overclocking disabled

    I guess it doesn't matter that much because they were never going to be higher clocking chips than the normal versions

    Reply
  2. The one thing I'm always concerned about with using more than one PSU is the likelihood of having current loops from the different voltages (however small the differences are) that the electronics that are being powered may not be able to handle or may cause some odd behavior or artifacts. Unexpected voltage/current differentials can cause unexpected behavior in electronic circuits. Current loops can be dangerous (to electronics) things that can cause some very strange behavior in circuits that are not protected from voltage differentials from multiple power sources. I've seen that cause a serial chip in a printer to just die before. (we were running power from two different outlets on two circuits) That's probably much more power than you might get from multiple PSUs on the same circuit. But I still am leery of introducing current loops where the electronics may not be able to handle it. I would THINK the techs at ASUS would be well aware of the risks. But I still get antsy when I see that.

    Reply
  3. i bet on fried x3d cache because 1.5 is i can say deadly to this structure because it´s constructed (a bit) different than CCX or CCD and power delivery system can be thinner but i am not sure if it is or not but it´s sensitive to temp. and voltage even if temp is fine it does not matter if voltage is too high it´s internally shorted or fried or capacitors can be damaged on intel it can be fine for a short amout of time it can survive but if we have smaller and smaller tranzistors 7nm 5nm 4nm they are more and more sensitive generally speaking so that is the main problem and if we have different structures with different power req. or litography … even worse voltage is not that big problem as amps are

    Reply
  4. AMD always said the 3D-Vcache was sensitive to voltage and temperature. So im not surprised at all it died at that 1.55v. There was a reason AMD was running the 5800X3D on such an low voltage and completely locked it. The Second gen V-Cache seen in the 7000 series saw a little increase in stock voltage and max temperatures. So they are making some improvements on that side.

    It would probably be best to not over-volt these CPU's but it would be really nice if we are able to lower the voltage. Not all boards where supporting that with the 5800X3D. Higher voltages will only kill the efficiency and will only improve performance marginally not worth it for day to day usage if you ask me.

    Reply
  5. The HPET thing is a long-standing bug with Intel CPUs. I had the same problem on my dual Xeon 8276L box – everything would jitter around and be weirdly unresponsive. There's a BCD value you can set to tell Windows to not use HPET, which saves you from disabling HPET on the CPU itself.

    Reply
  6. I think I might've figured out why the total continuous power consumption caps out around 1850W regardless of what you do with the supply voltage to the package. The control registers for the max power target are limited to 1024, and this is the max that XTU displays in amps. However, I think XTU's display of that number in amps is actually misleading. I suspect that they're actually implementing the power limit based on the power consumption you would get at the nominal package voltage of 1.8V, which would be 1.8V * 1024A = 1843W. When you increase the voltage beyond 1.8V it's actually derating the current limit and keeping you at the ~1850W power limit, rather than actually giving you a current limit of 1024A regardless of voltage. This makes sense when you consider how the FVIR architecture is set up. The onboard regulator has 320 phases per chip, 16 phases per power cell, and 25A max per power cell. Multiply that out and you get 500A per chip. Since the CPU package has four chips, that's 2000A absolute max. Given that the Vcore rail is nominally just over 1V, and accounting for some safety margin and conversion losses, a limit of 1850W total package power actually makes a lot of sense.

    Reply
  7. I think skatterbench pushed the 3d ccd with eclk, so it's possible that the voltage headroom could be useful vs locked entirely but not for most users

    Reply
  8. Urrgh, that drum and bass track 18.20 in to the video is just all kinds of filthy goodness 🥹 anyone know the track name? It played a few times. Thanks for all your high quality videos! Glad you had fun

    Reply
  9. Can you ask them, please, to fix the EXPO issue that the recent AM5 BIOS updates introduce? I cannot overclock memory at all, using EXPO or any other means, after 805 BIOS on X670E-E. On 805 the system is rock stable. So their BIOS upgrade has been a downgrade for me 🙁 Thank you!

    Reply
  10. is it possible to delid the cpu and just run no2 on the die direclty or could u build a cooling chamber without tim with direct contact of some mineral oil that isnt conductive

    Reply

Leave a Comment