Intel Core Processor Variants Explained: What U, H, HX, K and Every Suffix Letter Actually Means

Every Intel processor name is a compressed spec sheet. Once you learn the code, you can judge a chip before reading a single review.
This guide decodes every letter and number — from Core 3 100U to Core i9-13900K.

Anatomy of an Intel Processor Name

Each part of the model number tells you something specific. Here’s how to read it:

Key rule: Two processors with the same generation and SKU but different suffixes are completely different products.
A Core i7-1360P (28W) and a Core i7-1370H (45W) are miles apart in sustained performance.

Intel Dropped the “i” — The 2023 Rebrand

⚠ Old and new names coexist in stores right now

Starting late 2023, Intel retired “i3/i5/i7/i9” for mainstream chips. The new brand: Core 3/5/7 for everyday use,
Core Ultra 5/7/9 for premium AI-accelerated chips with a dedicated NPU. Same underlying Raptor Lake silicon — different badge on the box.

Old Name	New Name (2023+)	Premium Tier	Segment
`Core i3`	`Core 3`	—	Entry / budget
`Core i5`	`Core 5`	`Core Ultra 5`	Mainstream
`Core i7`	`Core 7`	`Core Ultra 7`	Performance
`Core i9`	`Core 9`	`Core Ultra 9`	Enthusiast

Core Ultra chips (Meteor Lake / Arrow Lake / Lunar Lake) add a dedicated NPU for AI — up to 48 TOPS.
Mainstream Core 3/5/7 chips do not have an NPU.

The Suffix Letter — The Most Important Part of the Name

The suffix is often more important than the tier (i3 vs i5). An i5-H will outperform an i7-U in sustained tasks because it has 3× the power budget.

Ultra-low PowerThin & light laptops

TDP9–15 W

Max Turbo55 W

Typical Cores6–10

ExampleCore 3 100U

Best for all-day battery (10–14 hrs). Fan is quiet or fanless. Throttles under sustained load.

Performance MobileMainstream laptops (15″)

TDP28 W

Max Turbo64 W

Typical Cores10–14

Examplei7-1360P

Sweet spot for productivity. Noticeably more responsive than U under continuous load.

High PerformanceGaming & creator laptops

TDP45 W

Max Turbo115 W

Typical Cores12–14

Examplei7-13700H

Desktop-adjacent performance. Needs a bigger chassis for cooling. Battery: 4–7 hrs mixed use.

Extreme MobileMobile workstations

TDP55 W

Max Turbo157 W

Typical Cores16–24

Examplei9-13900HX

Desktop CPU die in a laptop. Same silicon as desktop K chips. Found in 17″+ workstation laptops.

Unlocked DesktopGaming & enthusiast desktop

TDP125 W

Max Turbo253 W

Typical Cores14–24

Examplei9-13900K

Unlocked multiplier for overclocking. Has iGPU. Needs Z-series motherboard.

Unlocked, No iGPUBudget gaming desktop

TDP125 W

iGPUNone

Examplei5-13600KF

Same chip as K, iGPU fused off. Slightly cheaper. Requires a discrete GPU — no display without one.

No Integrated GPUBudget desktop

TDP65 W

iGPUNone

Examplei5-13400F

Locked multiplier, no iGPU. Very common in prebuilt desktops that ship with a discrete GPU.

Power Optimised DesktopMini-PCs, HTPC

TDP35 W

iGPUYes

Examplei5-13400T

Desktop chip at lower voltage. Silent, small-form-factor builds. Less peak perf than non-T sibling.

Sustained Performance ↔ Battery Life

T

U

P

H

HX

K/KF/F

← Better battery, quieter fan
Higher sustained performance →

P-Cores vs E-Cores: Intel’s Hybrid Architecture (12th Gen+)

Since 12th gen (Alder Lake, 2021), Intel chips use two types of cores on the same die:

P-Core (Performance)

Hyper-Threading: 2 threads per core
High clock speed (up to 5.8 GHz on K-series)
Games, browser, UI rendering, single-threaded tasks
Consumes more power per core

E-Core (Efficiency)

No Hyper-Threading: 1 thread per core
Lower clock (up to ~4.0 GHz)
Background tasks, antivirus scans, video decode
Much lower power — better battery life

The Thread Director (hardware scheduler) automatically routes tasks to the right core type — you don’t need to configure anything.

Core Layouts by Variant

U-series

Core 3 100U

6 cores · 8 threads · 15W

PPEEEE

U-series

Core i5-1335U / Core 5 120U

10 cores · 12 threads · 15W

PPEEEEEEEE

P-series

Core i7-1360P

12 cores · 16 threads · 28W

PPPPEEEEEEEE

H-series

Core i7-13700H

14 cores · 20 threads · 45W

PPPPPPEEEEEEEE

HX-series

Core i9-13900HX

24 cores · 32 threads · 55W+

PPPPPPPPEEEEEEEEEEEEEEEE

K-series

Core i9-13900K Desktop

24 cores · 32 threads · 125W

PPPPPPPPEEEEEEEEEEEEEEEE

P-core (2 threads each)
E-core (1 thread each)

13th Gen Intel Core — Full Specification Comparison

All major 13th gen (Raptor Lake) configurations side by side. 12th gen has slightly lower clock speeds but otherwise similar structure.

Model	Suffix	Cores (P+E)	Threads	Boost Clock	Base TDP	Max Turbo	Cache	iGPU
Core 3 100U	U	2P + 4E	8	4.7 GHz	9 W	55 W	10 MB	Iris Xe 64EU
Core i3-1315U	U	2P + 4E	8	4.5 GHz	15 W	55 W	10 MB	UHD 64EU
Core i5-1335U	U	2P + 8E	12	4.6 GHz	15 W	55 W	12 MB	Iris Xe 80EU
Core i7-1365U	U	2P + 8E	12	5.2 GHz	15 W	55 W	12 MB	Iris Xe 96EU
Core i5-1340P	P	4P + 8E	16	4.6 GHz	28 W	64 W	12 MB	Iris Xe 80EU
Core i7-1360P	P	4P + 8E	16	5.0 GHz	28 W	64 W	18 MB	Iris Xe 96EU
Core i5-13500H	H	4P + 8E	16	4.7 GHz	45 W	115 W	18 MB	Iris Xe 80EU
Core i7-13700H	H	6P + 8E	20	5.0 GHz	45 W	115 W	24 MB	Iris Xe 96EU
Core i9-13900H	H	6P + 8E	20	5.4 GHz	45 W	115 W	24 MB	Iris Xe 96EU
Core i9-13900HX	HX	8P + 16E	32	5.4 GHz	55 W	157 W	36 MB	UHD 32EU
Core i5-13600K	K	6P + 8E	20	5.1 GHz	125 W	181 W	24 MB	UHD 770
Core i7-13700K	K	8P + 8E	24	5.4 GHz	125 W	253 W	30 MB	UHD 770
Core i9-13900K	K	8P + 16E	32	5.8 GHz	125 W	253 W	36 MB	UHD 770

* Base TDP is the configurable floor. Actual sustained performance varies by laptop cooling design — the same chip behaves very differently in a thin chassis versus a properly cooled one.

Deep Dive: Intel Core 3 100U (The Budget Laptop Chip)

The Core 3 100U — Intel’s new name for what used to be called the i3-U — is the most common chip in affordable thin-and-light laptops priced ₹30,000–₹55,000 in India.

Hardware Specs

ArchitectureRaptor Lake Refresh

Process nodeIntel 7 (10nm)

Cores / Threads6 cores / 8 threads

P-cores + E-cores2P + 4E

Base clock (P)0.9 GHz

Max boost clock4.7 GHz

L3 Cache10 MB Smart Cache

Power & Memory

Base TDP9 W (cTDP 9–15W)

Max Turbo Power55 W (burst)

Integrated GPUIris Xe (64 EU)

GPU max freq1.10 GHz

Memory typeLPDDR5-4800 / DDR4

Max memory64 GB

PCIe4.0 & 3.0

Real-World Workload Assessment

🌐 Web & Office

30+ browser tabs, Office suite, Teams calls, light photo editing — handles all of this effortlessly. This is where U chips shine.

💻 Coding / Dev

VS Code, Python/Node dev servers, Docker (light) — comfortable. Large builds (Android, Chromium) will take 2–4× longer than an H-series.

🎬 Video Editing

4K playback & light timeline edits: possible but sluggish. 1080p export is borderline. Not recommended for regular video work.

🎮 Gaming

Iris Xe 64EU: casual 2D games, Minecraft, CS:GO (low), GTA V (720p low) — playable. No modern AAA gaming at any meaningful settings.

🔋 Battery Life

9W base TDP = 10–14 hrs on a 50Wh battery for light use. One of the most efficient x86 laptop chips available.

🌡 Thermal Throttling

Bursts to 55W for ~30 sec, then drops to 12–18W in thin chassis. Heavy tasks will be noticeably slower in cheap plastic-build laptops.

Can These Chips Run Local LLMs?

The bottleneck for CPU inference is memory bandwidth, not clock speed. When generating tokens, model weights move from RAM → CPU cache with every token.
The faster your RAM, the more tokens/second you get. This is why a U-series chip with LPDDR5-4800 can sometimes outperform a desktop chip running slow DDR4.

Estimated CPU Inference Speed — Core 3 100U (16GB LPDDR5, Q4_K_M quantisation, llama.cpp)

Phi-3 Mini

3.8B · ~2.3 GB

~9 tok/s

✅ Good

Llama 3.2 3B

3B · ~2 GB

~11 tok/s

✅ Good

Mistral 7B

7B · ~4.7 GB

~4 tok/s

⚠ Slow

Llama 3.1 8B

8B · ~5 GB

~3.5 tok/s

⚠ Slow

Llama 3.1 13B

13B · ~8 GB

~1.8 tok/s

❌ Too slow

Llama 3.1 70B

70B · ~42 GB

Won’t fit

❌ No

* Estimates based on LPDDR5-4800 dual-channel (~65 GB/s effective). Results vary by cooling and RAM config. Speeds below ~5 tok/s feel noticeably slow for real-time chat.

LLM Capability by CPU Suffix

🔵 U-series (9–15W)

Best for: 3B–7B models at Q4.
Phi-3 Mini, Gemma 2B, Llama 3.2 3B run at ~8–12 tok/s.
7B models work but feel slow (~3–5 tok/s).
Minimum 16GB RAM; 32GB unlocks 13B at very slow speed.

🔷 P-series (28W)

Best for: 7B–13B at Q4.
Higher sustained TDP = higher sustained memory bandwidth.
7B chat is comfortable (~5–8 tok/s).
13B with 32GB is borderline usable (~3–5 tok/s).

🟣 H-series (45W)

Best for: 13B–30B at Q4.
With 32GB+ DDR5: Llama 13B is comfortable (~6–10 tok/s).
Llama 34B works with 32GB at acceptable speed.
Recommended minimum for daily LLM use.

🔴 K-series Desktop (125W)

Best for: 30B–70B at Q4 (64GB RAM).
DDR5-6000 dual channel: ~96 GB/s bandwidth.
70B at Q4 fits in 64GB, runs ~2–4 tok/s.
Full LLM workstation capability in CPU-only mode.

iGPU Offload tip: llama.cpp supports offloading model layers to the Intel Iris Xe iGPU via OpenCL/SYCL.
On a Core 3 100U with shared LPDDR5 memory, this gives a ~10–20% speed boost on small models. Use -ngl 20 in llama.cpp.

Core Ultra NPU: If you have a Core Ultra chip (Meteor Lake or newer), the dedicated NPU (11–48 TOPS) can accelerate
Phi-3 Mini and small models via OpenVINO. This is where the “Ultra” premium starts to pay off for AI work.

RAM Requirement by Model Size

Model Size	Q4_K_M Size	Min RAM	Recommended	Suitable CPU
~1–3B (Phi, Gemma)	~1–2 GB	8 GB	16 GB	Any U-series or older chip
7B (Mistral, Llama 7B)	~4.7 GB	8 GB	16 GB	U-series (slow), P/H (usable)
13B (Llama 13B)	~8 GB	16 GB	32 GB	H-series minimum for usable speed
34B (CodeLlama 34B)	~22 GB	32 GB	48 GB	H/HX laptop, K desktop
70B (Llama 70B)	~42 GB	48 GB	64 GB	Desktop K-series (64GB DDR5) only

Which Variant Should You Buy?

Choose U if: you want 10–14 hr battery, a laptop under 1.5 kg, mostly web/office/coding use, and don’t need heavy sustained load.
Avoid if: you compile large projects daily, edit video regularly, or want to run 7B+ LLMs comfortably.

Choose P if: you want the battery/performance balance in a 14–15″ laptop for demanding apps like Docker, data analysis, or multi-tasking.
The underrated sweet spot — better than U under load without the weight penalty of H.

Choose H if: you game, do content creation (video/3D/ML), or run local LLMs daily. Expect a 1.8–2.3 kg laptop that needs a charger nearby.
H-series with a discrete GPU (RTX 4060+) is the best all-rounder for LLM + gaming right now.

Choose HX if: you need desktop-class multi-core in a laptop for 3D rendering, scientific compute, or very large codebases. Expensive, heavy, battery life is poor.

Choose K if: you’re building a desktop and want maximum overclockable performance for gaming, streaming, or CPU-only LLM inference. Pair with a Z790 board.
The i5-13600K is arguably the best price/performance desktop CPU — handles 70B LLM on 64GB DDR5.

Intel Core, Iris Xe and related names are trademarks of Intel Corporation. Specifications sourced from Intel ARK and product documentation.
Performance estimates are based on community benchmarks and will vary by laptop cooling, memory configuration, and specific workload.