Hardware

Building a 2026 Workstation: Testing Intel Core Ultra 7 270K Plus

We put the latest Intel CPU to the test across a variety of programming workflows and evaluate the performance of its integrated graphics.

Michał Dziuba

May 2, 2026 • 5 min read

Introduction

When it comes to building a high-end PC in 2026, the choice usually feels obvious. If you're building a "gaming-first" rig, you simply go with AMD and their 3D V-Cache technology. But when the focus shifts from gaming to a professional workstation, its a different story.

Recently I decided to start building a universal PC that will perform very well “in everything” - programming, video editing, CAD, gaming and local AI workflows, and Intel Core Ultra 7 270K Plus was released recently as well (ending of march 2026). It had good reception among PC building enthusiasts, and many people recommended this CPU for professional workflows while it’s still doing well in gaming.

I bought it for ~400 USD (price for polish market) and I think it’s best choice for this price category. It offers nearly identical performance (usually even beats it) to Intel Core Ultra 9 285K, Intel’s flagship CPU from previous year which currently costs ~700 USD. It doesn’t make sense to overpay for flagship Intel Core Ultra 9 285K while Intel Core Ultra 7 270K Plus exists.

But LGA 1851 is dead platform

There is one major issue, especially for people who change CPUs every 3 years or less - Intel 1851 platform is probably dead and we will not see any new CPUs for this socket. Is it really a problem? For me not really, because I plan to change CPU not earlier than 2030, so it’s very likely that DDR6 will be a thing and probably any upcoming Intel platform will be dead anyway, so I don’t really care.

Benchmarks

I made solid base for my workstation and as my CPU is very fresh, I want to make sure that I have newest kernel and drivers available and this is why I decided to use CachyOS - at least for now. We will focus on tasks where this CPU with 24 cores shines.

Testing platform specs

OS: CachyOS with 7.0.2-2-cachyos kernel
CPU: Intel Core Ultra 7 270K Plus (24 Cores / 24 Threads)
RAM: 64GB (2x32GB) DDR5 6000MHz CL28 Patriot Viper Elite 5 Ultra
Motherboard: Asus PROART Z890-CREATOR WIFI
AIO Cooling: BE QUIET! Silent Loop 3 420mm

Compiling Linux 6.15 kernel

It’s very common task to compare performance. I downloaded Linux 6.15 kernel source code and ran multiple times:

make mrproper && make defconfig && time make -j$(nproc)

On average it took 49-50s to compile kernel with defconfig settings with average CPU temperature around 70℃.

Compiling Node.js 24.15.0 LTS

It’s important task for me, because I write about my own JavaScript runtime for bachelor thesis and most painful part about this project is compiling v8 engine as it takes forever on my gaming laptop (around 30-40 minutes).

But here? It takes 6 minutes on average to compile whole Node.js:

make distclean && ./configure && time make -j$(nproc)

CPU during compilation had 70℃ temperature on average and 85℃ highest (105℃ is max safe temperature).

7z compression & decompression

7z has built-in command that allows to simply benchmark your computer. My CPU achieved 186 246 MIPS in total (average of compressions and decompression speeds). It’s very good score and even comparable with some Intel Xeons and AMD Threadrippers.

❯ 7z b

7-Zip 26.01 (x64) : Copyright (c) 1999-2026 Igor Pavlov : 2026-04-27
 64-bit locale=pl_PL.UTF-8 Threads:24 OPEN_MAX:4096, ASM

Compiler:  ver:15.2.1 20260209 GCC 15.2.1 : AVX2
Linux : 7.0.2-2-cachyos : #1 SMP PREEMPT Wed, 29 Apr 2026 19:06:32 +0000 : x86_64
PageSize:4KB HPS:2MB THPS:2MB THP:always hwcap:2 hwcap2:2
Intel(R) Core(TM) Ultra 7 270K Plus
(C0662)  (LP)

1T CPU Freq (MHz):  4573  4692  4847  5391  5382  5384  5390
12T CPU Freq (MHz): 1113% 4634   1114% 4761
24T CPU Freq (MHz): 2054% 4215   2181% 4502

RAM size:   63554 MB (LP),  # CPU hardware threads:  24
RAM usage:   5339 MB (LP),  # Benchmark threads:     24

                       Compressing  |                  Decompressing
Dict     Speed Usage    R/U Rating  |      Speed Usage    R/U Rating
         KiB/s     %   MIPS   MIPS  |      KiB/s     %   MIPS   MIPS

22:     163288  1602   9917 158847  |    2448011  2146   9726 208720
23:     142120  1513   9568 144803  |    2385675  2132   9681 206368
24:     171832  1959   9432 184754  |    2302521  2119   9535 202029
25:     166183  1955   9707 189742  |    2188419  2086   9333 194708
----------------------------------  | ------------------------------
Avr:    160856  1757   9656 169536  |    2331156  2121   9569 202956
Tot:            1939   9612 186246

Running LLMs on CPU

We will also check how local LLMs will work on this CPU - to be realistic, it makes sense to only check smaller models.

I used llama.cpp downloaded from homebrew (I didn’t compile it from source). I asked two small Qwen models to implement linked list in C++.

Here is a speed for text generation:

Qwen 3.5 4b Q4_K_XL - 9-10 t/s
Qwen 3.5 9b Q4_K_XL - 6 t/s

Gaming benchmark?

I don’t have dedicated GPU yet, so obviously iGPU will be huge bottleneck to make reasonable tests.

But…what if we test just for fun how this iGPU performs in modern games?

Gaming on integrated graphics

There is one important thing to be aware, when it comes to testing modern Intel iGPU on Linux. There are two drivers: legacy i915 and modern Xe. For some reason, my OS by default was using i915 which offers worse performance but probably is more stable.

I had to force Xe drivers in limine boot loader config, by adding i915.force_probe=!7d67 xe.force_probe=7d67 to cmdline, where 7d67 is specific PCI Device ID for this CPU.

ESP_PATH="/boot"
KERNEL_CMDLINE[default]+="quiet nowatchdog splash rw rootflags=subvol=/@ root=UUID=7394a9b8-08cd-4a14-a1d5-97eb1e9e5e07 i915.force_probe=!7d67 xe.force_probe=7d67"
BOOT_ORDER="*, *lts, *fallback, Snapshots"

DOOM Eternal

Doom Eternal is one of the best optimized modern games, and with mixed medium/high settings (mostly high) I got 40-55 FPS in FULL HD.

Cyberpunk 2077

Cyberpunk is probably one of the most demanding games even today and with mixed low/medium settings, FULL HD, Intel XeSS enabled (similar technology to Nvidia’s DLSS) I was getting 35-50 FPS, but it wasn’t good experience as everything were very blurry and input lag was noticable.

Minecraft

I was using newest Minecraft via Prism Launcher from Flatpak with Sodium & Lithium mods installed for optimizations. I was getting 250-500 FPS with 16 chunks render distance and 12 chunks simulation distance.

I also installed Iris mod and with BSL shaders enabled I had 30-60 FPS, 40 on average.

Other games

The WItcher 3: steam deck profile FULL HD (55-60 FPS)
Half-Life 2: high settings FULL HD (150-400 FPS)
Counter-Strike 2: low settings FULL HD (75-100 FPS)
Need For Speed: Unbound: medium, rainy weather FULL HD (25-30 FPS)

Summary

WIth this release, Intel “did AMD” and made excellent CPU for good price. Recent years were pretty bad for Intel and AMD domination in this space is not a good thing. Competition is good for market, technological advancement and us - a customers.