Friday, November 21st 2025

Dell Ship Pro Max 16 Plus Laptops with Qualcomm's Discrete NPU

If you're in the market for a new laptop and notice that the Dell Pro Max 16 Plus boasts more TOPS than usual, it's because Dell is equipping its latest models with a discrete Neural Processing Unit (NPU) from Qualcomm. The Dell Pro Max 16 Plus features a Qualcomm AI-100 Ultra NPU, delivering 870 INT8 TOPS within a 150 W TDP envelope, along with 128 GB of LPDDR5X memory. This provides enough computing power and memory capacity to run local AI models with up to 120 billion parameters on the NPU alone, without needing additional memory or computing resources.

Dell is outfitting these laptops with robust internals, including up to an Intel Core Ultra 9 285HX vPro CPU with 24 cores. Notably, Dell has removed the discrete GPU to accommodate the NPU, which offers more computing power and VRAM capacity than a traditional GPU. For local AI development, the system utilizes this NPU for all necessary tasks, paired with a CPU and 64 GB of memory. For graphics output, the system relies on the integrated GPU within the "Arrow Lake-HX" chip, which handles basic graphics tasks but isn't suitable for gaming. This design choice is logical, as the dedicated NPU is intended for AI developers and occupies the space, power, and heat dissipation capacity that a GPU would typically use.

Dell is developing laptops with dedicated NPUs to meet the demands of modern workloads that require fast, local AI processing without relying on the cloud. Tasks such as analyzing medical scans, detecting fraud in financial data, modeling complex engineering problems, and handling sensitive research demand immediate results and strict data privacy. These tasks cannot afford network latency or the risk of sending confidential information off the device. They benefit from an AI processor capable of handling large models running continuously. By incorporating powerful NPUs, Dell aims to serve professionals who need real-time insights, predictable performance, and complete control over their data, all from a portable machine that integrates seamlessly into their daily workflow.
Sources: Dell, Qualcomm
Add your own comment

21 Comments on Dell Ship Pro Max 16 Plus Laptops with Qualcomm's Discrete NPU

#1
LastDudeALive
I wonder if there's strong enough demand for it right now, or if 99% of potential customers will choose to wait for the Nvidia models coming next year.
Posted on Reply
#2
AleksandarK
News Editor
LastDudeALiveI wonder if there's strong enough demand for it right now, or if 99% of potential customers will choose to wait for the Nvidia models coming next year.
Yeah but NV GPUs won't have 128 GB of VRAM.
Posted on Reply
#3
LastDudeALive
AleksandarKYeah but NV GPUs won't have 128 GB of VRAM.
I mean the coming Nvidia ARM laptops based on the DGX Spark.
Posted on Reply
#4
TheinsanegamerN
I like the very square design of the new pro max.

What I dont like is the lack of quality. Our security guy has gone through 4 (FOUR) motherboards in 6 months because of wifi chip failure tanking the shole system causing lockups. The chip is soldered and requires entire board replacement. Our boss really REALLY want sot abandon dell right now due to the terrible quality the last few years.
Posted on Reply
#5
Six_Times
Very interesting. I'm in need of a new laptop and will check this out as a possible purchase.
Posted on Reply
#6
Wirko
I still hope someone will find a way to harness those matrix multipliers for rendering, simulations and other non-AI work.
Posted on Reply
#7
LastDudeALive
WirkoI still hope someone will find a way to harness those matrix multipliers for rendering, simulations and other non-AI work.
The problem is rendering and simulations require higher-precision calculations, while AI prefers lots of low-precision calculations. That's the entire reason why Nvidia added tensor cores to their GPUs in the first place. The regular 32-bit CUDA cores can do INT8 or FP16 or FP8 AI workloads, but dedicated Tensor cores can perform those calculations much faster. So Nvidia can optimize the CUDA cores for INT32/FP32 workloads, and the Tensor cores for faster, lower-precision AI workloads.

AMD and Nvidia actually have specialized chips for even higher-precision FP64 workloads for intense rendering, simulation, and other HPC work.
Posted on Reply
#8
bonehead123
"DUDE, you're gettin a D*E*L*L*".....whether you like it or not...

Cause we got a contract with them & we aint buyin anything else, so just sit down, STFU, and deal with it, hahahaha :D

Anyways, having used several of their business models over the past 5 years or so, as well as some HP's and Levoffvo's, I can say that, for the most part, quality-wise, they are ok and seem to work well for the office/productivity stuff.

But OTOH, I would not buy one for my personal use, as I already have a Clevo, and they are really better in almost every respect, but that's just IMHO....
Posted on Reply
#9
utmode
None of the pro or pro max or pro plus in Australia has AMD option. Dell is still doing same shenanigan.
Posted on Reply
#10
LabRat 891
utmodeNone of the pro or pro max or pro plus in Australia has AMD option. Dell is still doing same shenanigan.
They (and other OEMs) also actively 'hide' AMD-based units on their sites.
You can go into a product lineup, and every AMD unit is missing; do a websearch for the AMD model in that lineup, and get linked back to a subpage, where you just were.

"Intel... Intel never changes." :laugh:
Posted on Reply
#11
ThomasK
utmodeNone of the pro or pro max or pro plus in Australia has AMD option. Dell is still doing same shenanigan.
Intel is Dell's long time partner. Although the same amount of cash from back when Intel was the market leader isn't flowing anymore, the partnership is very much alive, and Dell's product portfolio reflects exaclty that.
Posted on Reply
#12
TumbleGeorge
This Pro Max is already being used by a lot of brands for all sorts of hardware from phones to computers. It's clearly trendy, even more so than the Ultra? Isn't it already patented?
Posted on Reply
#13
Visible Noise
ThomasKIntel is Dell's long time partner. Altough the same amount of cash from back when Intel was the market leader isn't flowing anymore, the partnership is very much alive, and Dell's product portfolio reflects exaclty that.
Dell’s product portfolio reflects what people want to buy. If Dell could make a profit selling more AMD systems then they would sell more AMD systems. It’s really very simple, and doesn’t need conspiracy fantasies to explain.
Posted on Reply
#14
lexluthermiester
LastDudeALiveI wonder if there's strong enough demand for it right now, or if 99% of potential customers will choose to wait for the Nvidia models coming next year.
OR.. we don't want them at all..
Visible NoiseDell’s product portfolio reflects what Dell thinks people want to buy.
FTFY.
Posted on Reply
#15
ThomasK
Visible NoiseDell’s product portfolio reflects what people want to buy. If Dell could make a profit selling more AMD systems then they would sell more AMD systems. It’s really very simple, and doesn’t need conspiracy fantasies to explain.
I had plenty of customers interested in buying Amd options from me, back when I worked in sales for Dell LATAM. Guess what? Dell only offered 8GB, soldered LPDDR5 entry-level Inspiron models with Ryzen in our market, made not to sell. Our sales compensation program rewarded us more for positioning Intel options, and Intel even sponsored our internal recognition program.



So please, don't claim to know or speak on behalf of "what people want". And don't get me started about conspiracy theories, when I know for a fact what goes on behind the curtains.
Posted on Reply
#16
Aquinus
Resident Wat-man
Ah, yes. The bubble grows.
Posted on Reply
#17
Visible Noise
ThomasKI had plenty of customers interested in buying Amd options from me, back when I worked in sales for Dell LATAM. Guess what? Dell only offered 8GB, soldered LPDDR5 entry-level Inspiron models with Ryzen in our market, made not to sell. Our sales compensation program rewarded us more for positioning Intel options, and Intel even sponsored our internal recognition program.



So please, don't claim to know or speak on behalf of "what people want". And don't get me started about conspiracy theories, when I know for a fact what goes on behind the curtains.
How cute. What’s Dell’s sales in Brazil?

Here’s mine, Las Vegas:
Posted on Reply
#18
ThomasK
Visible NoiseHere’s mine, Las Vegas:
So what? Many companies hold their Tech Forums annually in Vegas, including Dell.

And talking about HP, I even owned notebook from them, a couple of years back. It was the worst piece of overheating trash I ever owned. The magnesium chassis even discolored because of excessive heat. But it had Intel Inside®.

I understand you, being a newcomer to this site and also clearly ill-informed:

www.techpowerup.com/94123/eu-fines-intel-a-record-eur1-06-billion-in-antitrust-case

en.wikipedia.org/wiki/Advanced_Micro_Devices,_Inc._v._Intel_Corp.

Have a glass of water and a read, before talking bs.

Bye.
Posted on Reply
#19
remixedcat
HP = HUGE PROBLEMS

and dell has more ryzen systems these days and I might consider getting one of the pro models that do have it.
Posted on Reply
#20
SRB151
Dell does NOT even try to do AMD right. I just bought a used Alienware m18 AMD. Downloading the drivers from AMD creates a large black window that can't be closed. And before you start in on the AMD drivers suck routine, I am replacing an Asus G15 AMD advantage edition 6800m from 2022 (that I dropped, damn it), that had to co-exsist with vega CPU graphics that worked great with AMD drivers (as long as I have the samre release for vega and RDNA..
So there is no reason for this not to work except for Alienware (Dell) making life difficult for AMD owners.

On the actual topic, I saw a benchmark for the Nvidia Sparx that was 250 INT8. So this will do nearly 4 time more than that? I would hope the 250 number was a joke considering what these things cost.
Posted on Reply
#21
LastDudeALive
SRB151On the actual topic, I saw a benchmark for the Nvidia Sparx that was 250 INT8. So this will do nearly 4 time more than that? I would hope the 250 number was a joke considering what these things cost.
The DGX Spark is just a 5070 with a wider memory bus, 128GB of memory, and a CPU attached. i.e. it's plain Blackwell, meaning it's mostly CUDA cores, with only a smattering of Tensor cores.

This NPU is pure tensor cores. The DGX Spark also comes with CUDA, general compute abilities, and more TOPS if you use FP4 compute instead of INT8.
Posted on Reply
Add your own comment
Nov 15th, 2026 03:02 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts