Apple M3 Ultra 28 Core — Benchmarks & Specs
Short answer: the Apple M3 Ultra 28 Core generates 18.1 tokens/sec on deepseek-r1:671b (q4_0) under mlx in a single cited run, per Hostbor; there are too few comparable runs on file for a median.
Bottom line: how fast is the Apple M3 Ultra 28 Core?
For local LLM inference it generates 18.1 tokens/sec running deepseek-r1:671b at q4_0 under mlx, per Hostbor. In PassMark CPU Mark it scores 69,044 points, per PassMark Software.
Every figure above is a row in the tables below, and each row links out to the review or public benchmark database the number was taken from. SpecPicks aggregates published measurements; it does not report first-party benchmark runs.
The Apple M3 Ultra 28 Core is a processor from Apple. Data on this page draws on 23 synthetic benchmark results (top 10 shown), 16 community AI inference reports (top 12 shown), 9 measured game frame-rate results, compiled from MacRumors Forums, ComputerBase, Hostbor, llama.cpp GitHub Discussions, TBreak, Olares Blog, PassMark Software, CpuTronic, Digital Trends, EveryMac, Geekbench Browser, TurboVS, XDA Developers; each table row links to its source. Read this page when shopping the Apple M3 Ultra 28 Core, comparing it against other processors in your build, or sizing it for a specific workload (gaming at 1080p/1440p/4K, synthetic benchmark scores, or local LLM inference).
Gaming Performance (measured FPS)
Average and 1% low frame rates by game, resolution, and quality preset. Bars are scaled against the fastest result on this page.
| Game | Resolution | Settings | Relative | Avg FPS | 1% low | Test GPU | Source |
|---|---|---|---|---|---|---|---|
| Cyberpunk 2077 | 1080p | Medium FSR Balanced | 136 fps | — | Not recorded | ComputerBase 2025-08 | |
| Baldur's Gate 3 | 1440p | Ultra | 135 fps | — | Not recorded | Hostbor 2025-03-12 | |
| Cyberpunk 2077 | 1440p | Medium FSR Balanced | 118 fps | — | Not recorded | ComputerBase 2025-08 | |
| Cyberpunk 2077 | 1440p | Ultra FSR Balanced | 90 fps | — | Not recorded | ComputerBase 2025-08 | |
| Cyberpunk 2077 | 1080p | RT Ultra RT on FSR Performance | 81 fps | — | Not recorded | ComputerBase 2025-08 | |
| Cyberpunk 2077 | 4K | Custom MetalFX Balanced, no RT | 71 fps | — | Not recorded | TBreak 2025-07-24 | |
| Cyberpunk 2077 | 4K | For This Mac MetalFX Standard | 64 fps | — | Not recorded | TBreak 2025-07-24 | |
| Cyberpunk 2077 | 1440p | RT Ultra RT on FSR Performance | 62 fps | — | Not recorded | ComputerBase 2025-08 | |
| Cyberpunk 2077 | 4K | RT Ultra RT on MetalFX Standard | 32 fps | — | Not recorded | TBreak 2025-07-24 |
AI Inference Performance
Tokens per second under each model + quantization. Higher = faster generation. Bars compare runs across the same model.
| Model | Quantization | Relative | Tokens/sec | VRAM used | Source |
|---|---|---|---|---|---|
| deepseek-r1:7b | q4_K_M LM Studio (MLX) | 128.6 tok/s | 4.3 GB | MacRumors Forums 2025-05-05 | |
| qwen3:8b | q4_K_M LM Studio (MLX) | 114.9 tok/s | 4.6 GB | MacRumors Forums 2025-05-05 | |
| llama2:7b | q4_0 llama.cpp | 88.4 tok/s | — | llama.cpp GitHub perf thread (Apple Silicon) 2025-03-15 | |
| llama3.1:7b | q4_0 llama.cpp | 88.4 tok/s | — | llama.cpp GitHub Discussions 2023-11-21 | |
| qwen3:30b-a3b | — ollama | 84.1 tok/s | — | Olares Blog 2025-11-04 | |
| qwen3:14b | q4_K_M LM Studio (MLX) | 70.3 tok/s | 8.3 GB | MacRumors Forums 2025-05-05 | |
| llama2:7b | q8_0 llama.cpp | 63.5 tok/s | — | llama.cpp GitHub perf thread (Apple Silicon) 2025-03-15 | |
| llama2:7b | FP16 llama.cpp | 42.2 tok/s | — | llama.cpp GitHub perf thread (Apple Silicon) 2025-03-15 | |
| qwen3:32b | q4_K_M mlx | 33.9 tok/s | — | MacRumors Forums 2025-05-05 | |
| deepseek-r1:32b | q4_K_M mlx | 33.3 tok/s | — | MacRumors Forums 2025-05-05 | |
| gemma3:27b | — ollama | 27.5 tok/s | — | Olares Blog 2025-11-04 | |
| deepseek-r1:671b | q4_0 mlx | 18.1 tok/s | — | Hostbor 2025-03-12 |
Synthetic Benchmarks
Higher is better. Bars are scaled within each benchmark family (multi-thread, single-thread, etc.) so you can compare like-with-like at a glance.
| Benchmark | Relative | Score | Source |
|---|---|---|---|
| PassMark CPU Mark | 69,044 points | PassMark Software retrieved 2026-06-01 | |
| Cinebench R23 multi | 37,200 points | CpuTronic retrieved 2026-08-03 | |
| Geekbench 6 multiapple mac studio m3 ultra review | 28,778 points | XDA Developers 2025-04-02 | |
| Geekbench 6 multi | 28,417 points | Hostbor 2025-03-12 | |
| Geekbench 6 multi | 27,749 points | TurboVS 2025-03 | |
| Geekbench 6 multi | 27,551 points | MacRumors Forums 2025-03-21 | |
| Geekbench 6 multiapple mac studio 2025 review | 27,404 points | Digital Trends 2025-03-11 | |
| Geekbench 6 multi | 26,610 points | Geekbench Browser retrieved 2026-09-04 | |
| Geekbench 6 multi | 26,530 points | EveryMac (Geekbench Browser averages) 2026-09-27 | |
| PassMark CPU Mark (single-thread) | 5,139 points | PassMark Software retrieved 2026-06-01 |
Apple M3 Ultra 28 Core — Frequently Asked Questions
What is the Apple M3 Ultra 28 Core best used for?
Where do the Apple M3 Ultra 28 Core benchmark numbers come from?
How many cores does the Apple M3 Ultra 28 Core have, and how does it benchmark?
Where can I buy the Apple M3 Ultra 28 Core?
Buying guides that rank the Apple M3 Ultra 28 Core's class
This page is the raw performance data. The guides below turn it into a ranked pick for a specific build.
Editorial guides covering the Apple M3 Ultra 28 Core
In-depth SpecPicks reviews, build guides, and head-to-heads referencing this processor.
- 4K Gaming PC Build 2026: Cost, Parts, FPS
- 1440p Gaming PC Build 2026: Cost, Parts, FPS
- 1080p Gaming PC Build 2026: Cost, Parts, FPS
- 9800X3D + RTX 5070 Ti PC Build 2026: 1440p High-Refresh
- Ryzen CPU Deals 2026: 5600X, 5800X3D and 9800X3D Price Lines
- Prebuilt Gaming PCs 2026: RTX 5060 vs 5060 Ti vs 5070
- Gaming Desk Deals 2026: 5 L-Shaped and Standing Desks for a Tower + Multi-Monitor Setup
- Arc B580 12GB vs RTX 3060 12GB: Which 12GB Card for 1440p in 2026?
- Browse all SpecPicks reviews →
More guides & deep dives from the SpecPicks archive
Browse all articles & guides →- RTX 4070 Super vs RX 7800 XT — Which to Buy in 2026
- How to Build a Windows 98 Retro PC in 2026
- Best 1440p Gaming GPUs in 2026
- The Complete Voodoo5 5500 AGP Driver Guide (2026 Edition)
- Best Budget Gaming PC Build 2026 — ~$1,000 ($800 on Sale)
- Emulation Hardware in 2026: FPGA, Software, and Cart-Reader Ecosystems
- Best Retro Handhelds in 2026 — From $35 to $500
More reviews from the SpecPicks archive
Browse all reviews →- Ryzen 9800X3D vs 7800X3D: Best CPU for an RTX 5070 Build
- GameNative Unlocks Up to 100 FPS PC Gaming on Android
- 16% of Freelance Jobs Are Now AI-Doable: The Local Agent Rig That Runs Them
- Self-Host Jellyfin on a Ryzen 5 5600G Mini Build in 2026: Measured Power and Transcoding
- Sound BlasterX G6 for Retro PC Gaming Audio: An External Card for Win98/XP and DOSBox Builds
- Can a 12GB RTX 3060 Still Run 2026's Local LLMs?
- Best GPU for Training CNNs at Home in 2026: The RTX 3060 12GB Case
- AMD Ryzen AI Max 400 'Gorgon Halo': 192GB Unified Memory for Local LLMs
- Best Budget Gaming CPU: Ryzen 5 5600G vs 5700X vs i7-9700K
- Samsung Winds Down Chip Production Six Days Before Strike: What Memory Buyers Should Do Right Now
- Local LLM in VSCode: The 2026 Setup Guide
- Best Single-Board Computer Project Kits for Beginners (2026)
- OpenAI Names Its Biggest Data Center Yet, With NVIDIA Backing: What It Means
- Gemma 4 31B-IT on a 12GB RTX 3060: What Fits, What Offloads, How Fast
- Blue Yeti vs HyperX QuadCast 2 S: Best USB Mic for Streaming in 2026
- Best Parts for a Budget Ryzen + RTX 3060 Gaming PC Build in 2026
- Best SATA SSD for Gaming and Console Upgrades in 2026
- Running Gemma 4 31B Finetunes Locally: Dual RTX 3060 12GB vs Single 24GB Card
- Best Plug-and-Play Retro & Budget Consoles in 2026: 5 Picks Ranked
- A DIY Wireless Fingerprint Authorization Device Turns Heads on Hackaday
- Glide vs OpenGL vs Direct3D: The API War That DirectX 9 Ended
- Ryzen 9 3900X vs Ryzen 5 5600G for CPU-Only Local LLM Inference (2026)
- Corsair's New Gaming Mouse Has a Dedicated Stream Deck Launch Button
- Intel Arc AI Inference: 2025 Performance & Setup Guide
More buying guides from SpecPicks
Browse all buying guides →- Best GPUs for Running Local LLMs in 2026
- Best GPUs for 4K Gaming in 2026
- Best Gaming Mice for 2026
- Best CPUs for Gaming in 2026
- Best NVMe External Enclosures for 2026
- Best 4K Monitors for Content Creators in 2026
- Best NVMe SSDs for Gaming in 2026
- Best CPUs for Content Creators in 2026
- Best CPU Coolers for 2026
- Best AM5 Motherboards for 2026
- Best Tools for Building and Repairing Retro PCs in 2026
- Best Retro Gaming Consoles & Handhelds for 2026
- Best Controllers for PC Gaming in 2026
- Best 1440p 240Hz Gaming Monitors in 2026
- Best Gaming Monitors for 2026
- Best GPU for Running 27B-32B Local LLMs in 2026
- Best Mechanical Keyboards for Gaming in 2026
- Best Graphics Cards for Gaming in 2026
- Best External SSDs for Content Creators in 2026
- Best PC Cases for Building in 2026
- Best DDR5 RAM for Gaming PCs in 2026
Hardware benchmark data on SpecPicks
All benchmarks →- AMD Ryzen 9 7900 — benchmarks & specs
- Ryzen 5 3500U — benchmarks & specs
- Radeon RX 6300M — benchmarks & specs
- Radeon RX 7900 XTX — benchmarks & specs
- Ryzen 7 5700X — benchmarks & specs
- AMD Ryzen 7 PRO 5755G — benchmarks & specs
- Radeon RX 7700S — benchmarks & specs
- AMD Radeon Pro W7900 48GB — benchmarks & specs
- Radeon RX 9060 — benchmarks & specs
- Radeon RX 6450M — benchmarks & specs
- AMD Ryzen 3 7320C — benchmarks & specs
- NVIDIA Jetson Orin Nano Super — benchmarks & specs
- GeForce RTX 6070 — benchmarks & specs
- Radeon RX 5500 XT — benchmarks & specs
- Radeon RX 6900 XT — benchmarks & specs
- RTX5000-Ada-4Q — benchmarks & specs
- GeForce RTX 4080 — benchmarks & specs
- Ryzen 7 9850X3D — benchmarks & specs
- AMD Ryzen Threadripper 7970X — benchmarks & specs
- AMD Ryzen 9 5900HX — benchmarks & specs
- Apple M1 Max 10 Core 3200 MHz — benchmarks & specs
- AMD Ryzen 5 7645HX — benchmarks & specs
- AMD Ryzen 3 5300GE — benchmarks & specs
- Intel Arc A750 — benchmarks & specs