TwistedAndy

There are many factors affecting the heat pipe efficiency. The most important ones are thickness, width, amount of bends and the bend radius. Obviously, the straight round heat pipe shows best efficiency. The slim one with one or a few 90 degrees bends can be 2-5 times less efficient In terms of "cooling priority", there is no huge difference between CPU and GPU, because the liquid evaporates in a gas in the heatpipe. Yes, the GPU has a higher priority, but not that much. I've posted a link to a series of interviews with an engineer designing cooling solutions. From my perspective, the heatsink in 7670/7770 might be better, but it's not that bad. I think, it would be better to implement a thicker version of the heatsink from Precision 5770 / XPS 17 with two fans on each side, but I'm not a thermal engineer and don't know all the details. There's some heat transfer from CPU to GPU and vice-versa, but it is not that big. Gas tends to condense and extract heat on the cooler side of the heat pipe with a radiator.

I prefer keeping Turbo Boost enabled. My XPS 17 has a "lazy" fan behavior. It spins the fans after 30-50 seconds of sustained load. Then fans are spinning with 45 dBA for about 20-30 seconds, and then settle nearly 38-39 dBA with 45W PL1. Maybe, Dell Precision 7670/7770 have something similar. From my experience, their power management on modern devices is very complicated. On Alder Lake laptops it is managed by the Intel Dynamic Tuning Technology (DTT). It has a system service and a software device. It can adjust power settings (including power limits) and other stuff. It can be caused by active nVidia GPU as well. It draws 10-15W idle. You can check it using HWiNFO64. Also, in some cases, there are some exotic things like discrete TB controllers. They can easily draw 5W if the power management is disabled for them in Device Manager and heat the keyboard surface nearby.

It depends on what and where you're doing. For indoor use with the normal lightning conditions (300-500 lux) the optimal brightness level is somewhere between 100 and 150 nits. I tend to use ~100-120 nits both on the laptop and external monitor to minimize the eyes fatigue. From my perspective, the positioning of HX SKUs are weird. They start outperforming H SKUs only on higher power levels (>75W). But I don't know any laptop design, which can handle this power with acceptable amount of noise (~40 dBA). Usually, you'll get something closer to 45 dBA at 75W and ~50 dBA at 120W. It's almost impossible to productively work with the 50 dBA laptop for a noticeable period of time.

You can disable S0 in the registry, but in this case there will be no sleep options at all. If you want to enable the S3 sleep mode, you can do that in BIOS in the same way as you unlocked undervolting, but it is pretty buggy and may not work well.

Yes, correct. I was talking about Dell DOO (or Dual Opposite Outlet). This approach is applied regardless the presence of vapor chamber. It allows using fans with larger diameter in smaller casing, pressurize the inner laptop volume, and use the heatsink as an additional radiator. Actually, that's why Dell put an additional small radiator on the heatsink in Precision 7670 and 7770 🙂 You may find more details on how it works in the interview with a Dell engineer.

Dell is using DOO fans both in XPS 17 and 7670/7770. Fans are pushing air in both directions: through radiator and through the motherboard and cooling system. On your image air inlets are actually outlets. This approach works great on XPS 17, but may be not so efficient in 7670/7770 due to smaller area.

Actually, it supports hybrid CPUs pretty well. There're separate controls for P and E cores. Yes, this vulnerability is called plundervolt. Yes, I also noticed it and decided to go with XPS 17 9720 instead of 7670 three months ago.

The difference between CL38 (Fury) and regular RAM with CL40 is minimal.

Under spiked load, one or a few cores can get pretty hot. That's how the Intel Turbo Boost works. Under sustained load, the temperatures will be much lower. I don't think we will see 18" laptops in the near future. They are too big and heavy. Many large laptops are used connected to external screens. In this regard there's no significant difference between 17" and 18". In some cases even 17" is overkill. Probably, that's why Lenovo merged their P15 and P17 laptops in the one P16 device.

The situation with stock performance looks weird, because even with some QC issues with a heatsink or thermal interface, it still should handle at least 55W sustained power and get nearly 14-16,000 CBR23 sustained. 21-22,000 possible only during the first run. From my perspective it looks like some issues with Intel drivers (especially Intel DTT and IPF) or BIOS configuration. Eventually, I had a problem on my XPS 17 with PL1 locked on 20W by default because of compatibility issues between Windows and firmware.

For some reason, the CB R23 and GB5 results are pretty low. Here's what I get on XPS 17. Probably, there are some issues with power management.

A new unboxing video with Precision 7670 is out:

Samsung 990 Pro was available a few weeks ago on amazon.de on preorder. Estimated shipping date was October 11 or 12. I think, we'll see the first reviews in mid-October.

AMD 6900HX can compete with Alder Lake H/HX up to 45W sustained power limit. Alder Lake shows significantly better results on higher power levels because of better performance scaling. Actually, it can scale the performance effectively till 80W-90W. In case of AMD 6900HX there's no reason to go above 50W. For instance, on 70-80W sustained power limit, 12700H/12900HK can get ~18000-18500 while the AMD can get ~14000-14500 CB R23 MT. In other tasks the situation is nearly the same. Alder Lake is 20-30% faster. In the case of HX series, they start showing some performance advantage over regular H series on 80+W power limits So Alder Lake gives you more options here. You can have decent thermals, noise levels (<40 dBA) and slightly better performance than AMD at 45W. Or you can set the power limit to 75W, get 15-25% performance boost, and higher noise levels. As for the battery life, it is heavily depends on the particular laptop model and optimizations. Usually, laptop vendors doesn't care too much about that, but Dell did a good job in XPS 15 and 17. I get nearly 10-14 hours on my XPS 17 on light use.

A3000/3060 can use nearly 10-15W idle. I don't think A1000 is very different here. From my perspective, there's no reason always work on external GPU. The integrated one from Intel works great.

Those results were received during the first run with Turbo Boost enabled. 150W is a huge number. I think, the noise level will be above 50 dBA and incompatible with productive work. Mobile Alder Lake 6+8 and 8+8 CPUs are most effective on the 45-75W range or nearly 14500 - 18000 CB R23 MT with slight undervolting. In terms of noise it will be something between 38 and 45 dBA accordingly. Yes, you can go beyond 80W and get slightly more performance, but the fan noise...

Actually, you can undervolt the CPU with Hyper-V active. If you're planning to use ThrottleStop, you would need to disable VBS (Virtualization-Based Security). This feature locks some registers and prevents them from changing. Once you find the stable voltage offsets, you can put them directly to the BIOS settings in the same way as you unlock the undervolting. In this case, you will be able to enable VBS. On my device (XPS 9720) I was able to achieve -0.15V offset, but for some reason the system crashed while my IDE was indexing the project. Maybe, there're some JVM-specific issues there. I ended up with -0.12V offset. As for the system agent, it is unstable at low voltages, so I decided to apply -0.06V for that.

It looks like Dell XPS 17 with 3060 has some kind of a MUX switch (technically, two MUX switches). This device uses two discrete (!) Thunderbolt 4 controllers (JHL8540) and allows to switch the external display directly to the nVidia GPU. Unfortunately, you need to enter BIOS for that. The other fun fact. Each of those controllers consumes nearly 3-3.5W in the active state. It's very important to enable the power saving option for them in the Device Manager. The Intel's TB4 driver is pretty buggy at the moment and floods the Event Manager with RTD3 events, but it really helps saving power.

I don't use SupportAssist or any other software from Dell. Usually, it's pretty buggy (especially Fusion Service)

SODIMM adapters will be available later. From my perspective it is reasonable to purchase a configuration with the required amount of RAM, especially for such expensive devices like 7670/7770. Yep, Dell charges more for additional memory, but not so much. Kit with 2x32Gb Kingston Fury (KF548S38IBK2-64) now costs 485$ vs. 657$ for 64GB CAMM option. There're some options to save another $100-$150, but it doesn't really matter for $4-5K total device price. Technically, it is nice to have an opportunity to install a low-latency memory like G.Skill Ripjaws DDR5-4800 CL34, but it is limited to 16GB per module. 32GB modules from Kingston or G.Skill have a higher latency (at least CL38), which is comparable to CAMM (CL40, I assume).

I have tried that again, S3 is available and enabled, but pretty buggy: Maybe, there're some other hidden settings in the BIOS, but I don't have much time to do a further research. Switching to AHCI does not solve all the issues. For now I decided to stay with the S0 since it is much more "polished".

Here're the commands for 9520/9720 with Core i9-12900HK: setup_var CpuSetup 0x43 0x0 setup_var CpuSetup 0x10E 0x0

It works, Windows reports that S3 is available, the sleep mode is there, but there's an issue with waking up. The laptop freezes on the Dell logo. From what I've found this problem is related to the Raid mode enabled in BIOS. In this mode the system is using Intel's AHCI driver (part of IRST). Switching to AHCI in BIOS solves it, but requires some additional manipulation with the system bootloader and drivers. I'm too lazy to go that far 🙂

Memory speed does not significantly affect the Cinebench R23 score. I get nearly the same results with one or two DIMMs. As for the power limit, it makes sense to check it using the HWiNFO (CPU Package Power). Under sustained load you should see something nearly 55W. It corresponds the ~15500 score in CB R23 MT. Under 75-85W CPU Package Power there should be 18000-19000. Also, it is worth checking the Intel DTT (Dynamic Tuning Technology). It has a related telemetry service, which should be enabled for the best performance.

Disabling the modern standby (S0) and enabling the classic sleep mode (S3) is pretty tricky, but possible. First of all, you need to disable the S0 standby mode by adding a registry key. After a reboot there will be no sleep option available. That's because Dell decided to disable S3. To resolve this issue, you will need to enable the ACPI S3 Support in the BIOS. Unfortunately, this setting is hidden by default, but you can change it in a similar way as it is shown here. You need to find the "ACPI S3" setting and determine it's offset. Once you turn this feature using a command like "setup_var Setup 0x0E 0x1" (offset will be different in your case), you will see the sleep mode again, there will be S3 states available (powercfg /a), but it wouldn't work properly. I suggest switching the SSD mode from Raid On to AHCI and update drivers. It should help resolving this issue.