# The best Macs and mini PCs for running LLMs locally (2026)

> The **Mac Studio M5 Max 128GB** (about $5,100) is the best all-round pick: its 614 GB/s is more than twice what its PC rivals offer, and it runs gpt-oss-120b at ~64 tok/s (estimated). A **Ryzen AI Max+ 395** mini PC runs the same models at about half the speed for about $3,300, and **DGX Spark** is the one to buy if you need CUDA. All of them shine on mixture-of-experts models; for dense models up to 32B, a GPU desktop is faster.

A Mac Studio, an NVIDIA [DGX Spark](/gpus/dgx-spark) or a mini PC built on AMD's [Ryzen AI Max+ 395](/gpus/ryzen-ai-max-plus-395) can give its GPU 96GB or more of memory, several times the 24GB to 32GB on a consumer graphics card. That lets one quiet box run models like [gpt-oss-120b](/models/gpt-oss-120b) and [Qwen3.5 122B-A10B](/models/qwen3-5-122b-a10b), which would otherwise need several GPUs or a $16,000 workstation card.

The catch is speed. These machines share LPDDR5X memory between the CPU and GPU, and it is slower than the GDDR memory on a graphics card. So they are excellent for mixture-of-experts (MoE) models, which read only a small part of their weights for each token, and slow on large dense models. If the models you want fit in 24GB to 32GB, a GPU desktop is the better buy; our [GPU buying guide](/best/gpus-for-local-llms) covers those.

> **About these prices:** Prices are October 2026 street or list prices. The 2026 memory shortage hit these machines too: 128GB Strix Halo mini PCs went from about $2,000 in 2025 to $3,300 or more, and DGX Spark went from $3,999 at launch to about $6,950. Apple withdrew the [512GB M3 Ultra](/gpus/mac-studio-m3-ultra-512gb) configuration in March 2026, and it had not priced the [512GB M5 Ultra](/gpus/mac-studio-m5-ultra-512gb) by mid-October.

## The trade-off: capacity versus bandwidth

To write each token, a model reads its active weights from memory, so generation speed is roughly memory bandwidth divided by the bytes read per token. A dense model reads all of its weights every time. An MoE model reads only the experts it routes to: gpt-oss-120b has 117B parameters but only 5.1B active, so each token touches a small slice of its 65GB of weights.

That is why the same machine can feel fast or slow depending on the model. Estimated, a Ryzen AI Max+ 395 runs gpt-oss-120b at ~35 tok/s but the dense [Qwen3.8 27B](/models/qwen3-8-27b) at Q4_K_M at only ~11 tok/s, even though the second model is a quarter of the size. The [Mac Studio M5 Max](/gpus/mac-studio-m5-max-128gb) has 614 GB/s, about 2.4 times the Strix Halo's 256 GB/s, and runs gpt-oss-120b at ~64 tok/s. The [M5 Ultra](/gpus/mac-studio-m5-ultra-256gb) doubles that bandwidth again.

Estimated generation speed for gpt-oss-120b at MXFP4:

- RTX PRO 6000: ~170 tok/s
- M5 Ultra 256GB: ~93 tok/s
- M3 Ultra 512GB: ~70 tok/s
- M5 Max 128GB: ~64 tok/s
- M4 Max 128GB: ~58 tok/s
- DGX Spark: ~37 tok/s
- Ryzen AI Max+ PRO 495: ~37 tok/s
- Ryzen AI Max+ 395: ~35 tok/s

The models that suit these machines are MoE designs: gpt-oss-120b (about 68GB with 8K of context), Qwen3.5 122B-A10B (about 77GB at Q4_K_M), [GLM-4.5-Air](/models/glm-4-5-air) (about 68GB at Q4_K_M) and [Qwen3.6 35B-A3B](/models/qwen3-6-35b-a3b) at high quality. With 256GB or more you can step up to [DeepSeek V4 Flash](/models/deepseek-v4-flash) (about 168GB at its native FP4) and [GLM-5.3-Flash](/models/glm-5-3-flash) (about 161GB at Q3_K_M). Speeds in this guide are estimates from our [estimate model](/methodology), which assumes 8K of context and one user; we have not tested these machines ourselves.

## Prompt processing is the other half

Generation speed is only half of the wait. Before a model writes anything, it has to read your prompt, and that step depends on raw compute rather than memory bandwidth. For a short chat message the difference barely registers. For a coding agent that sends a large chunk of your codebase with every request, or a long PDF, it can mean a long pause before the first word appears.

Here the machines differ more than their bandwidth suggests. Fast NVIDIA graphics cards are quicker than all of them. Among the small machines, DGX Spark is the strongest, much faster than Strix Halo. Apple says the M5 generation adds a Neural Accelerator to each GPU core, which speeds up a step where earlier Macs were notably slow. Our estimates cover generation only, so check measured prompt speeds on our [benchmarks page](/benchmarks) if long inputs are your main workload.

## How much memory can the GPU actually use?

None of these machines hands all of its memory to the GPU by default. The operating system needs its share, and each platform sets the limit differently.

| System | Memory | Bandwidth | GPU can use |
| --- | --- | --- | --- |
| Ryzen AI Max+ 395 | 128GB | 256 GB/s | Up to 96GB on Windows, about 120GB on Linux |
| DGX Spark | 128GB | 273 GB/s | About 119GB, shared with the OS |
| [Ryzen AI Max+ PRO 495](/gpus/ryzen-ai-max-plus-pro-495) | 192GB | 273 GB/s | Not documented yet |
| Mac Studio M5 Max | 128GB | 614 GB/s | About 96GB by default |
| Mac Studio M5 Ultra | 256GB or 512GB | About 1,228 GB/s | About 192GB or 384GB by default |

**Macs.** macOS lets the GPU use about 75 percent of unified memory by default. You can raise that with the `iogpu.wired_limit_mb` setting (for example, `sudo sysctl iogpu.wired_limit_mb=110000` on a 128GB machine), which lasts until you restart. Leave enough for macOS and your apps, or the system can slow to a crawl or become unstable.

**Strix Halo.** On Windows you set the GPU's share in the BIOS or AMD's driver software, up to 96GB. On Linux the GPU can use about 120GB after you change a kernel setting, which is enough for Qwen3.5 122B-A10B at Q6_K (about 105GB) or the 3-bit DeepSeek V4 Flash builds of about 103GB to 110GB. Our estimates assume the 96GB Windows limit.

**DGX Spark.** Linux reports about 119GB, and the CPU and GPU draw from the same pool.

**Ryzen AI Max+ PRO 495.** AMD hasn't said how much of the 192GB the GPU can use. Our estimates assume 75 percent, about 144GB, until it does.

## Software: MLX, CUDA, ROCm and Vulkan

On a Mac, llama.cpp runs through Apple's Metal API, and MLX is Apple's own machine learning framework, with MLX builds of most popular models published on Hugging Face. LM Studio can use either, and Ollama works too. What you can't run is anything that needs CUDA, which rules out some fine-tuning libraries and serving tools.

DGX Spark ships with DGX OS, NVIDIA's Ubuntu-based Linux, and the full CUDA stack, so vLLM, TensorRT-LLM, PyTorch and llama.cpp all run. Its CPU is Arm-based, so the occasional x86-only tool or container image needs a workaround.

Strix Halo machines run Windows or Linux like any PC. llama.cpp's Vulkan backend works with standard AMD drivers and is the simplest route, including in LM Studio on Windows; ROCm is the other option on Linux. Whatever you use, keep runtimes current: newer models such as Qwen3.5 and DeepSeek V4 Flash use attention designs that need recent llama.cpp, MLX or vLLM builds.

## Best overall: Mac Studio M5 Max 128GB

**Best overall: Apple Mac Studio M5 Max 128GB** (128GB, 614 GB/s, ~$5,100). A quiet 128GB machine with more than twice the bandwidth of Strix Halo or DGX Spark. Details: https://llamabuilds.ai/gpus/mac-studio-m5-max-128gb

The [Mac Studio M5 Max](/gpus/mac-studio-m5-max-128gb) with the 40-core GPU and 128GB has 614 GB/s of bandwidth, more than twice what Strix Halo or DGX Spark offer, and about 96GB of it is available to the GPU by default. It's the best balance of capacity and speed in this class, in a quiet desktop you can also use as your everyday computer.

Estimated, it runs gpt-oss-120b at ~64 tok/s, Qwen3.5 122B-A10B at Q4_K_M at ~34 tok/s and GLM-4.5-Air at Q5_K_M at ~25 tok/s. Qwen3.6 35B-A3B runs at Q8_0 at ~56 tok/s. Dense models are slower: Qwen3.8 27B at Q4_K_M runs at ~23 tok/s, about 60 percent of a used [RTX 3090](/gpus/rtx-3090)'s speed.

It costs about $5,100 as configured (the M5 Max Mac Studio starts at $2,499 with 36GB), the memory can't be upgraded later, and there is no CUDA. Order the 40-core GPU, because the 614 GB/s figure applies to that version. The [M4 Max 128GB](/gpus/mac-studio-m4-max-128gb) it replaced has 546 GB/s, about 11 percent less, and runs gpt-oss-120b at ~58 tok/s; it launched at $3,499 in this configuration, so a refurbished or used one is a good buy if it sells well below the M5 Max. For head-to-heads, see [DGX Spark vs M5 Max](/compare/dgx-spark-vs-mac-studio-m5-max-128gb) and [Ryzen AI Max+ 395 vs M5 Max](/compare/ryzen-ai-max-plus-395-vs-mac-studio-m5-max-128gb).

## Best value: Ryzen AI Max+ 395 mini PCs

**Best value: AMD Ryzen AI Max+ 395 (128GB)** (128GB, 256 GB/s, ~$3,300). The cheapest way to run 100B-class MoE models in one quiet box, though prices have climbed past $3,300. Details: https://llamabuilds.ai/gpus/ryzen-ai-max-plus-395

The [Ryzen AI Max+ 395](/gpus/ryzen-ai-max-plus-395), known as Strix Halo, pairs 16 Zen 5 cores with a 40-CU Radeon 8060S GPU and up to 128GB of LPDDR5X-8000 at 256 GB/s. Several brands sell it in small desktops, and the chip runs at 45 to 120W, so these machines stay quiet. It's also a complete x86 PC, which makes it the easiest of these machines to fit into an existing Windows or Linux setup.

Estimated, gpt-oss-120b runs at ~35 tok/s, Qwen3.5 122B-A10B at Q4_K_M at ~18 tok/s and GLM-4.5-Air at Q5_K_M at ~13 tok/s. Qwen3.6 35B-A3B at Q4_K_M runs at ~49 tok/s, which is quick enough for interactive coding help.

Dense models are where it struggles: Qwen3.8 27B at Q4_K_M runs at ~11 tok/s. Prompt processing is slower than on DGX Spark, so long inputs take a while. And the price has climbed: 128GB systems sold for about $3,300 to $4,350 in October 2026, up from about $2,000 in 2025. It is still the cheapest new machine that runs 100B-class models entirely in GPU-accessible memory.

## Best for CUDA developers: NVIDIA DGX Spark

**Best for CUDA developers: NVIDIA DGX Spark** (128GB, 273 GB/s, ~$6,950). A quiet CUDA desktop with 128GB, but 273 GB/s limits speed and the price has climbed to about $6,950. Details: https://llamabuilds.ai/gpus/dgx-spark

[DGX Spark](/gpus/dgx-spark) is NVIDIA's compact desktop built on the GB10 Grace Blackwell chip, with a 20-core Arm CPU, a Blackwell GPU and 128GB of LPDDR5X at 273 GB/s. Linux reports about 119GB usable, more than a 128GB Mac gives its GPU by default. Two units can be linked through the built-in ConnectX-7 networking.

Generation speed is in the same class as Strix Halo. Estimated, gpt-oss-120b runs at ~37 tok/s, Qwen3.5 122B-A10B at Q4_K_M at ~19 tok/s (or at Q6_K, which fits in the extra memory, at ~14 tok/s), and Qwen3.6 35B-A3B at Q8_0 at ~32 tok/s. Where the Spark pulls ahead is prompt processing, which is much faster than on Strix Halo, and software. It runs the same CUDA stack as NVIDIA's datacenter systems, so work you prototype on it carries over to servers.

The problem is price. It launched at $3,999 in October 2025, rose to $4,699 in February 2026 and listed at about $6,950 in October 2026, more than twice a Strix Halo mini PC for similar generation speed. A 64GB OEM version is due from $4,999 on October 23, but 64GB is too small for gpt-oss-120b, which needs about 68GB. Buy the Spark if CUDA matters to your work; otherwise, see [DGX Spark vs Ryzen AI Max+ 395](/compare/dgx-spark-vs-ryzen-ai-max-plus-395).

## Best for the biggest models: Mac Studio M5 Ultra 256GB

**Best for the biggest models: Apple Mac Studio M5 Ultra 256GB** (256GB, 1,228 GB/s, ~$9,499). Most of what the M5 Ultra offers, with 256GB, shipping now at $9,499. Details: https://llamabuilds.ai/gpus/mac-studio-m5-ultra-256gb

The [Mac Studio M5 Ultra with 256GB](/gpus/mac-studio-m5-ultra-256gb) has about 1.2TB/s of bandwidth, by Apple's figure, and about 192GB of GPU memory by default. At $9,499 it costs nearly twice the M5 Max, but it doubles both the memory and the bandwidth, and that opens up a class of models that 128GB machines can't hold at full quality.

Estimated, it runs DeepSeek V4 Flash at its native FP4 (about 168GB) at ~45 tok/s and GLM-5.3-Flash at Q3_K_M (about 161GB) at ~39 tok/s. Smaller models get fast: gpt-oss-120b at ~93 tok/s and Qwen3.5 122B-A10B at Q8_0 at ~33 tok/s. Even dense Qwen3.8 27B at Q4_K_M reaches ~39 tok/s, the same as a used RTX 3090. For comparison, a pair of [RTX PRO 6000](/gpus/rtx-pro-6000-blackwell) cards with 192GB runs DeepSeek V4 Flash at ~72 tok/s, for about $32,000 in cards alone.

The [512GB M5 Ultra](/gpus/mac-studio-m5-ultra-512gb) ships in late October, and Apple hadn't announced its price as of mid-October. It is the only current desktop that holds [GLM-5.3](/models/glm-5-3), at Q3_K_M in about 375GB, which only just fits under the default GPU limit (~19 tok/s, estimated), or GLM-5.3-Flash at Q6_K (~25 tok/s). If you don't need that much, the M5 Ultra also starts at $5,499 with 96GB. A used [M3 Ultra 512GB](/gpus/mac-studio-m3-ultra-512gb) has 819 GB/s and runs gpt-oss-120b at ~70 tok/s, but used prices are volatile, so check completed sales first.

## How they compare

| GPU | Memory | Bandwidth | Price |
| --- | --- | --- | --- |
| M5 Max 128GB | 128GB | 614 GB/s | ~$5,100 |
| Ryzen AI Max+ 395 | 128GB | 256 GB/s | ~$3,300 |
| DGX Spark | 128GB | 273 GB/s | ~$6,950 |
| M5 Ultra 256GB | 256GB | 1,228 GB/s | ~$9,499 |
| M5 Ultra 512GB | 512GB | 1,228 GB/s | n/a |
| Ryzen AI Max+ PRO 495 | 192GB | 273 GB/s | ~$6,800 |
| RTX 3090 | 24GB | 936 GB/s | ~$1,050 used |
| RTX 5090 | 32GB | 1,792 GB/s | ~$4,300 |

## Also considered: Ryzen AI Max+ PRO 495

The [Ryzen AI Max+ PRO 495](/gpus/ryzen-ai-max-plus-pro-495), known as Gorgon Halo, raises memory to 192GB but keeps the same 256-bit bus, so bandwidth only rises to 273 GB/s. It runs the same models as Strix Halo at about the same speed (Qwen3.5 122B-A10B at Q6_K at ~14 tok/s, estimated). The extra memory buys higher-quality quantizations, or room for 200B-class models at around 3 bit, but nothing in speed. At $6,799 for the Framework Desktop (shipping in November) or $7,399 for the Minisforum MS-S1 MAX-P495, the $9,499 M5 Ultra 256GB costs 40 percent more and has about 4.5 times the bandwidth. We'd wait until AMD documents how much of the memory the GPU can use.

## When a GPU desktop is the better buy

Unified memory wins when you want 100B-class MoE models, low noise and one simple box. A GPU desktop wins almost everywhere else.

Estimated generation speed for Qwen3.8 27B at Q4_K_M:

- RTX 5090: ~80 tok/s
- RTX 3090: ~39 tok/s
- M5 Ultra 256GB: ~39 tok/s
- M5 Max 128GB: ~23 tok/s
- DGX Spark: ~12 tok/s
- Ryzen AI Max+ 395: ~11 tok/s

**If your models fit in 24GB to 32GB.** The strongest models that fit on one card, such as Qwen3.8 27B and [Gemma 4 31B](/models/gemma-4-31b), are dense. A used RTX 3090 at about $1,050 runs Qwen3.8 27B at ~39 tok/s (estimated), against ~23 tok/s on the M5 Max and ~11 tok/s on Strix Halo. An [RTX 5090](/gpus/rtx-5090) reaches ~80 tok/s.

**If you work with long prompts.** NVIDIA GPUs read prompts far faster, which matters for coding agents and document work.

**If you need CUDA or plan to fine-tune.** Training tools and many serving stacks assume an NVIDIA GPU. DGX Spark covers CUDA, but for models that fit in a graphics card's memory, a desktop GPU is faster for less money.

**If you want to upgrade later.** You can add a second card to a desktop. Unified memory is fixed the day you buy it.

Use the [VRAM calculator](/tools/vram-calculator) to check what your target models need, and our guide to [how much VRAM you need](/guides/how-much-vram-do-you-need) to decide how much headroom to buy.

## Frequently asked questions

### Is a Mac good for running local LLMs?

Yes, especially for large mixture-of-experts models. A Mac Studio M5 Max with 128GB gives its GPU about 96GB by default at 614 GB/s, enough to run gpt-oss-120b at about 64 tok/s (estimated) with MLX or llama.cpp. Macs are slower than NVIDIA GPUs at processing long prompts and can't run CUDA-only tools, so a GPU desktop is the better buy for dense models up to 32B.

### How much of a Mac's memory can the GPU use for LLMs?

By default, macOS lets the GPU use about 75 percent of unified memory: about 96GB on a 128GB Mac, 192GB on 256GB and 384GB on 512GB. You can raise the limit with the `iogpu.wired_limit_mb` setting, for example `sudo sysctl iogpu.wired_limit_mb=110000` on a 128GB machine, as long as you leave enough for macOS itself. The change lasts until you restart.

### Is DGX Spark or a Ryzen AI Max+ 395 mini PC better for local AI?

They generate text at about the same speed because both have roughly 256 to 273 GB/s of memory bandwidth: gpt-oss-120b runs at about 37 tok/s on DGX Spark and 35 tok/s on a Ryzen AI Max+ 395 (estimates). DGX Spark processes long prompts much faster and runs CUDA, but it cost about $6,950 in October 2026, against about $3,300 for a 128GB Strix Halo mini PC. Buy the Spark for CUDA development and the AMD box for value.

### Can a 128GB mini PC run 70B models?

It can load them, but dense 70B models run slowly. Llama 3.3 70B at Q4_K_M needs about 47GB and generates at about 4.5 tok/s on a Ryzen AI Max+ 395 and about 9 tok/s on a Mac Studio M5 Max (estimates), because every token reads all 70B weights. Mixture-of-experts models of similar or larger size, such as gpt-oss-120b, run several times faster on the same machines.

### What is the cheapest way to run gpt-oss-120b locally?

A 128GB Ryzen AI Max+ 395 mini PC, at about $3,300 to $4,350 in October 2026, runs gpt-oss-120b entirely in memory at about 35 tok/s (estimated). If you already own a desktop with a 24GB GPU and 64GB of system RAM, it can run the model with part of it offloaded at about 21 tok/s. The model needs about 68GB with 8K of context.

---
Source: https://llamabuilds.ai/best/macs-and-mini-pcs-for-local-llms · Updated October 14, 2026 · Llama Builds. Speeds marked ~ are estimates from our bandwidth model (https://llamabuilds.ai/methodology); measured results are labeled as such.