Dell Pro Max 16/18 Plus (2025 model) pre-release discussion — MB16250, MB18250

[AL123]

5 hours ago, Aaron44126 said:

The GPU card fits in the top-right area (viewing from the bottom), a custom form factor DGFF card like they've been doing for several years now. It can be seen in the manual. (Tossed some pictures from the manual in the spoiler block below.)

 

The only new thing here to me is that it appears to be double-stacked over part of the motherboard (but not directly above the CPU). You can see the double-stacking best in the second-to-last image in @AL123's post above. Stacking the dGPU card on top of the motherboard is not that new, it was standard design when Dell was using MXM cards (2017 systems and earlier), but this is the first time that I have seen it since they switched to DGFF.

 

When the Qualcomm AI thing is available, my understanding is that it will also be a DGFF card fitting in the same space, so you'll have to pick between it or an NVIDIA GPU.

 

  Reveal hidden contents

 

 

I’d certainly like a better idea of how this stacking is arranged, not brave enough to take these apart any further though! 
 

I can try and take some more pictures around that area when I am back in the office though. 

[AL123]

5 hours ago, yslalan said:

Well, it’s kind of hard to say. Those AI Chips are not a new model (advertised as dNPUs). In the Pro Max-Plus series, Dell is using the solution from Qualcomm shipped in H1 2021 (a pair of AI 100 Pro chips), but four years have already passed, and there hasn’t been much presence in industrial usage.

ah didn’t realise it was that old,  perhaps the first gen might be a bit limited but hopefully a sign dell are looking to innovate.

 

i think the NVIDIA spark / Dell Pro Max GB10 will make more sense, 128GB or 256 if you join two together 

https://www.nvidia.com/en-gb/products/workstations/dgx-spark/
 

so small you could jsut pop in your bag,

 

5 hours ago, yslalan said:

Well, it’s kind of hard to say. Those AI Chips are not a new model (advertised as dNPUs). In the Pro Max-Plus series, Dell is using the solution from Qualcomm shipped in H1 2021 (a pair of AI 100 Pro chips), but four years have already passed, and there hasn’t been much presence in industrial usage.

 

[MyPC8MyBrain]

40 minutes ago, win32asmguy said:

they made a decision of function over form here

I’m not sold on the idea that this form will hold up under sustained heavy loads. Mobile chassis designs already have minimal wiggle room, and if the trade-off here is improve latency over optimizing thermals, I’d take optimizing thermals—because once temps spike, latency will happen anyway. Lower latency isn’t much of a win if the rest of the system is cooking itself. That’s not just counterproductive, it’s self-defeating in the long run.

[yslalan]

7 hours ago, AL123 said:

small you could jsut pop in your bag

Yea, I am awaiting news about the Nvidia N1x SoC, which is based on the same architecture as GB10 but targeted towards the laptop market. Hopefully, it's Linux and Windows on Arm compatible. 

[AL123]

14 hours ago, yslalan said:

Yea, I am awaiting news about the Nvidia N1x SoC, which is based on the same architecture as GB10 but targeted towards the laptop market. Hopefully, it's Linux and Windows on Arm compatible. 

Ah I hadn’t heard about that one, would be more aligned with the dominance NVIDIA have in that market along with partnership Dell have with them.

 

perhaps the focus on AI will also mean we will see a mid life increase in graphics memory per ampere mobile where they doubled many of the capacities of the range IIRC. just not sure if the memory density will allow for that in such a small space to cram them in and cool the chips. 

[MyPC8MyBrain]

While SoCs do offer real engineering advantages in terms of performance per watt, the business model they enable is a major contributor to the growing e-waste problem and the erosion of consumer freedom.
 

SoCs in personal computing are, in my view, an abomination—a step in the wrong direction for the average user. They exist not because they serve the long-term interests of consumers, but because they maximize profit, streamline manufacturing, and reinforce planned obsolescence.
 

The trade-off is clear: ownership, control, and product longevity are sacrificed for short-term efficiency and corporate margins.

[AL123]

15 hours ago, MyPC8MyBrain said:

While SoCs do offer real engineering advantages in terms of performance per watt, the business model they enable is a major contributor to the growing e-waste problem and the erosion of consumer freedom.
 

SoCs in personal computing are, in my view, an abomination—a step in the wrong direction for the average user. They exist not because they serve the long-term interests of consumers, but because they maximize profit, streamline manufacturing, and reinforce planned obsolescence.
 

The trade-off is clear: ownership, control, and product longevity are sacrificed for short-term efficiency and corporate margins.

 Certainly not ideal, but what did surprise me is what NVIDIA is doing with local AI  is making it more affordable for once. With GB10 it’s  ~$3k for 128GB (2bn parameter models)  or $6k for 256GB if you link two together.  won’t be as fast as a system with one or more 96GB RTX 6000 Blackwell but less costly and more capacity. Perhaps I’ll give them a pass on using a SoC in that case. Storage is user upgradable, but the memory is soldered 

 

I hope that CAMM memory will see wider adoption and make it less necessary to have soldered ram. Sometimes it’s the chip makers decision e.g new AMD strix point/ halo have apus with soldered memory In the AMD versions of the base Pro Max 14/16. I believe tthat is the only option due to AMD choices, but more often it’s the manufacturer, Dell, HP, Lenovo….. choice. 
 

I really don’t understand why for instance Dell are using LP-CAMM on their entry level Pro Max 14 but soldered memory on their Pro Max 14 & 16 Premium devices. They really aren’t that much thinner to make be believe soldered is necessary, given the pioneered CAMM modules this is very disappointing.

 

Apple seem to have gained some advantages from their SoC approach for and I’m fine with a combined APU, if it’s genuinely better, but would be nice to have upgradable memory if CAMM was fast enough and no excuse for non upgradable storage. Not sure apple will ever do this thouge,  even user upgradable storage has gone by the way for no justification I can see just greed, just add gen 5 m.2 pcie slot .

 

we have to use windows due to certain pro applications, but for teams like our video/ marketing macs would make sense, the snag other than supporting 2 different OS,  is either paying through the nose for storage or having loads of external drives that are harder to control from a IT perspective 

[SvenC]

Still no review from the new Dell Pro Max Plus 16/18.

 

We'll order a Pro Max Plus 18" tomorrow. 265HX, iGPU only, CSoDimm (1x 16GB to upgrade myself to 2x 32GB). Should be delivered in 3 weeks, according to our sales rep.

[AL123]

2 hours ago, SvenC said:

Still no review from the new Dell Pro Max Plus 16/18.

 

We'll order a Pro Max Plus 18" tomorrow. 265HX, iGPU only, CSoDimm (1x 16GB to upgrade myself to 2x 32GB). Should be delivered in 3 weeks, according to our sales rep.

Will be interesting to see how the iGPU version performs,  from the manual it looks like the heatsink is different. The Pro Max 18 with RTX 4000 I have runs the Ultra 9 CPU at sustained power of around 100W (PL1 = 98W) 

 

 

[SvenC]

2 hours ago, AL123 said:

Will be interesting to see how the iGPU version performs,  from the manual it looks like the heatsink is different. The Pro Max 18 with RTX 4000 I have runs the Ultra 9 CPU at sustained power of around 100W (PL1 = 98W) 

 

 

How long does it run on battery on low use? Does it get hot on medium or high load?

[AL123]

3 minutes ago, SvenC said:

 

How long does it run on battery on low use? Does it get hot on medium or high load?

Early days for testing not  used it much for day to day work jsut a bit of benchmarking; but had More time with the lower pro max 16 with intel h series and that certainly seems much better than earlier core i7/i9 

 

On the pro max 18 plus so far pleased with the temps under high load, chassis doesn’t seem to heat up the same way as previous gen, 

[MyPC8MyBrain]

our sales rep called me last week to let me know they are running specials on the new 16/18 pro max's internally for limited time, it seem they proactively trying to push these out.
 

[AL123]

back in the office today, much to catch up on but couldn’t resist a bit of testing while doing dull admin tasks 

 

Focused on Pro Max 18 as its most popular with our team so far (massive screen!) specs-

Ultra 9, RTX 4000, 64GB CSODIMM, 2TB gen 5 SSD

 

This is now a full production unit with latest BIOS, drivers etc and my impressions very positive particularly for the new cooling system excellent performance and the chassis is remarkably cool ( just  warm at top left where the fans/ heatsink are) 

 

The CPU still seems to be limited to around 100W for sustained workloads (98w PL1)

 

I tried more synthetic test cinebench  2024 + a real world CFD simulation which is heavily multi threaded. the power draw hovered around this 100W mark on both, with the average a little higher in the real world CFD  test I suspect because it doesn’t load cores 100% all the time.
 

From looking at others tests online with u9 275HX gaming laptops this seems to be the sweet spot for performance and you have to go to crazy power levels to get much any more performance vs 100W it just plateaus there for a bit so may be a sensible limit from Dell. 
 

 

the RTX Pro 4000 on the other hand i am seeing higher power than my last test unit whether it is the newer bios, drivers or jsut because i am using hw info now. It is capable of sitting at a sustained 170-175W for a GPU rendering task  lasting 10 mins + 

 

I then tried GPU+CPU concurrent

 

the rendering on GPU 100% load and ran a multi threaded cpu test on cinebench 2024 for a concurrent torture test!
 

 I think dell are under claiming if anything!
 

. GPU ran at 140-145W and cpu around 60-65 (63W average)  so yes the cpu does throttle back if the GPU is under heavy load but still performant and above the claimed 200W concurrent.
 

It was reporting 250-260W total system power in hw info. Saw a brief spike to 300W so it used a bit of battery for a while.

Would be nice if there was some utility we could choose to allocate a bit more power budget to the cpu, but right now in the real world the tools we use that benefit from GPU aren’t simultaneously loading the cpu 100%

 

Haven’t tried it on the dock yet to see if it behaves any different. 

 

[MyPC8MyBrain]

19 minutes ago, AL123 said:

Would be nice if there was some utility we could choose to allocate a bit more power budget to the cpu

did you look into changing power plan settings in bios to performance for low level settings,
and than higher in windows power plan settings?

there's also settings in NVIDIA to prefer more performance etc. in control panel setting as well as additional tweaks via external third party tools.

with my 7770 i undervolted my cpu, run bios on cool profile, and windows profile in balanced mode, i created a powershell script that changed bios power plan to performance from within windows, and switched windows to maximum performance power plan (old classic power schema), the script would toggle everything back if i run it again, i think i posted it somewhere in the 7770 owners thread you could possibly adopt it to this unit easily with your preferred settings to quickly switch to pedal to the metal profile.

[Aaron44126]

The power profiles (Balanced / Quiet / Cool / High performance) will impact power draw behavior. Setting it to "Cool" might be a way to allocate more power budget to the CPU, for example; in past systems that I have messed with, this power setting put a cap on the GPU power draw but not on the CPU power draw.

 

You used to adjust this in the Dell Power Manager app, but that isn't a thing anymore, I guess maybe it is in Dell Optmizer?

[AL123]

1 hour ago, MyPC8MyBrain said:

did you look into changing power plan settings in bios to performance for low level settings,
and than higher in windows power plan settings?

there's also settings in NVIDIA to prefer more performance etc. in control panel setting as well as additional tweaks via external third party tools.

with my 7770 i undervolted my cpu, run bios on cool profile, and windows profile in balanced mode, i created a powershell script that changed bios power plan to performance from within windows, and switched windows to maximum performance power plan (old classic power schema), the script would toggle everything back if i run it again, i think i posted it somewhere in the 7770 owners thread you could possibly adopt it to this unit easily with your preferred settings to quickly switch to pedal to the metal profile.

So far this is just Windows Balanced power Plan and then under settings > power, for plugged in it is set to “Best Performance” 

 

Genrally found that on more recent versions of windows 11 balanced  power plan has been fine  rather than needing high performance/ ultimate power plans, but I’ll certainly test to see if they make a difference. 

 

These will have a company image on them, no undervolting/

other tuning  just what we can do out of the box that gives us the best balance for our users

 

we only tend to include dell command update, no optimiser or support assist, been burnt too many times by that hogging resource, I heard they might be combining command update and support assist, I hope the end result it is more like command update if so! 

 

I’ll take a look at the settings in BIOS/ Dell Optimizer see if they make a difference to power usage  far it just seems like the NVIDIA cards have more headroom pier limit wise this laptop looks like being higher power limit  than a desktop RTX PRO 4000 Blackwell. Although a fairer comparison is really the 4500/5000 desktop card as the product stack is still a little different mobile vs desktop. 
 

the other impressive thing is I was working on the laptop unplugged by mistake for a while and it wasn’t until I ran a benchmark test that I noticed the diffeeence, can’t say the same for older gen laptops, still not worked on this laptop for long on battery but a promising sign 
 

the windows power settings for unplugged were balanced rather than best performance 

[AL123]

1 hour ago, Aaron44126 said:

The power profiles (Balanced / Quiet / Cool / High performance) will impact power draw behavior. Setting it to "Cool" might be a way to allocate more power budget to the CPU, for example; in past systems that I have messed with, this power setting put a cap on the GPU power draw but not on the CPU power draw.

 

You used to adjust this in the Dell Power Manager app, but that isn't a thing anymore, I guess maybe it is in Dell Optmizer?

Yeh worth an experiment even though I think dell have got the cpu power balance about right, be good to have choice for users who aren’t needing high GPU power. 

[Aaron44126]

5 minutes ago, AL123 said:

Yeh worth an experiment even though I think dell have got the cpu power balance about right, be good to have choice for users who aren’t needing high GPU power. 

 

You can also adjust this setting in the BIOS if you don't have Dell Optimizer installed. (You just have to reboot in order to access it...) It is independent from the Windows power settings.

 

I think you may be able to update it from PowerShell as well, using the tools that Dell has to update system settings from PowerShell (for purposes of mass deployment / standardization), but I have not ever tried to do that myself.

 

I agree, though, in the event of a combo load I would prefer to see it prioritize the GPU.

[MyPC8MyBrain]

1 hour ago, Aaron44126 said:

I think you may be able to update it from PowerShell as well

most of these setting can definitely be changes via powershell,
porb why Dell retired most of the older resource hogging tools,

@AL123 have a looksee here
https://notebooktalk.net/topic/632-dell-precision-7670-dell-precision-7770-owners-thread/?do=findComment&comment=21688

 

[win32asmguy]

On 8/21/2025 at 12:28 PM, AL123 said:

Focused on Pro Max 18 as its most popular with our team so far (massive screen!) specs-

Ultra 9, RTX 4000, 64GB CSODIMM, 2TB gen 5 SSD

 

Thanks for your continued review / information! By chance would you be able to check HWInfo64 memory information page for detailed specs on the timings for both the CSODIMMs and CAMM modules? It looks like this:

 

 

That is from a non-CSODIMM Kingston Fury kit with tREFI tuned and at 1.25v. With SA Fabric and D2D tuned its around 90ns latency which while far behind a tuned desktop 285K is much better than most stock OEM shipping laptops. Some guys with Lenovo Legion 7i's have also hit close to 80ns as well. In theory a CAMM XMP module could do really well here if they offered single rank Hynix modules with desktop-grade timings.