GPU & Apple Silicon Comparison for LLMs
Every GPU, Apple Silicon Mac and 128 GB mini PC in our hardware catalog, side by side for local language-model inference. Compare VRAM or unified memory, vendor memory bandwidth, TDP, and the largest dense model each one holds at Q4_K_M by the site formula. This page does not list prices; check the retailer listing for the current one. Memory decides what fits, so the tier notes under the table explain how to read each group. Use the VRAM calculator to match a specific model to hardware.
| Device | VRAM | Price | Buy |
|---|---|---|---|
| Apple Mac Studio (M5 Max, 128GB) Apple Silicon After a 4 GB macOS reserve, about 124 GB is usable. 70B Q8 is 86 GB and fits. Qwen3.8-Flash-Next UD-Q4_K_XL is 111.33 GB and is tight. | 128 GB | See the listing | — |
| Apple Mac Studio (M4 Max, 128GB) Apple Silicon After a 4 GB macOS reserve, about 124 GB is usable. 70B Q8 is 86 GB and fits. Qwen3.8-Flash-Next UD-Q4_K_XL is 111.33 GB and is tight. | 128 GB | See the listing | — |
| NVIDIA DGX Spark NVIDIA | 128 GB | See the listing | As an Amazon Associate, LLM Hardware earns from qualifying purchases. Links to Amazon are affiliate links, at no extra cost to you. Amazon |
| GMKtec EVO-X2 (Ryzen AI Max+ 395, 128GB) Mini PC 128 GB at ~256 GB/s. Dense 70B Q4 (44 GB) fits and is slow. Qwen3.8-Flash-Next UD-IQ4_XS is 93.68 GB and fits. UD-Q4_K_XL is 111.33 GB and is tight. DeepSeek-V4-Flash only at an aggressive quant (~155 GB at a normal 4-bit). Not CUDA. | 128 GB | See the listing | Amazon |
| Beelink GTR9 Pro (128GB, sold out) Mini PC 128 GB at ~256 GB/s when the listing is this GTR9 Pro. UD-IQ4_XS (93.68 GB) fits. UD-Q4_K_XL (111.33 GB) is tight. Sold out on 2026-10-11. Not CUDA. | 128 GB | See the listing | — |
| Apple Mac Studio (M5 Ultra, 96GB) Apple Silicon After a 4 GB macOS reserve, about 92 GB is usable. 70B Q4 is 44 GB and fits. 70B Q8 is 86 GB and is tight. Nemotron 3.5 Bolt at Q4 is 46.4 GB and fits. Qwen3.8-Flash-Next UD-Q4_K_XL is 111.33 GB and does not fit. | 96 GB | See the listing | — |
| Apple Mac Studio (M5 Max, 64GB) Apple Silicon After a 4 GB macOS reserve, about 60 GB is usable. 70B Q4 is 44 GB and fits. 27B FP16 is 66.8 GB and does not fit. 70B Q8 is 86 GB and does not fit. | 64 GB | See the listing | — |
| Apple Mac Studio (M4 Max, 64GB) Apple Silicon After a 4 GB macOS reserve, about 60 GB is usable. 70B Q4 is 44 GB and fits. 27B FP16 is 66.8 GB and does not fit. 70B Q8 is 86 GB and does not fit. | 64 GB | See the listing | — |
| Apple Mac Studio (M5 Max, 48GB) Apple Silicon After a 4 GB macOS reserve, about 44 GB is usable. 27B Q8 is 34.4 GB and fits. 70B Q4 is 44 GB and is the edge. 70B Q8 is 86 GB and does not fit. | 48 GB | See the listing | — |
| Apple Mac mini (M4 Pro, 48GB) Apple Silicon After a 4 GB macOS reserve, about 44 GB is usable. 27B Q8 is 34.4 GB and fits. 70B Q4 is 44 GB and is the edge. 70B Q8 is 86 GB and does not fit. | 48 GB | See the listing | — |
| Apple Mac Studio (M5 Max, 36GB) Apple Silicon After a 4 GB macOS reserve, about 32 GB is usable. Qwen3.6-27B Q4 is 18.2 GB and fits. LFM2.5-24B-A2B Q4 is 16.4 GB and fits. Q8 is 30.8 GB and is the edge. 70B Q4 is 44 GB and does not fit. | 36 GB | See the listing | Amazon |
| NVIDIA GeForce RTX 5090 (32GB) NVIDIA | 32 GB | See the listing | Amazon |
| NVIDIA GeForce RTX 4090 (24GB) NVIDIA | 24 GB | See the listing | Amazon |
| AMD Radeon RX 7900 XTX (24GB) AMD | 24 GB | See the listing | Amazon |
| NVIDIA GeForce RTX 3090 (24GB) NVIDIA 24 GB fits Llama 3.1 8B Q4 (6.8 GB) and Qwen3.6-27B Q4 (18.2 GB). 70B Q4 is 44 GB and does not fit. | 24 GB | See the listing | Amazon |
| Apple Mac mini (M5 Pro, 24GB) Apple Silicon After a 4 GB macOS reserve, about 20 GB is usable. 27B Q4 is 18.2 GB and is tight. LFM2.5-24B-A2B Q4 is 16.4 GB and is tight. Q8 is 30.8 GB and does not fit. | 24 GB | See the listing | Amazon |
| Apple Mac mini (M4 Pro, 24GB) Apple Silicon After a 4 GB macOS reserve, about 20 GB is usable. 27B Q4 is 18.2 GB and is tight. LFM2.5-24B-A2B Q4 is 16.4 GB and is tight. Q8 is 30.8 GB and does not fit. | 24 GB | See the listing | — |
| Apple Mac mini (M6, 24GB) Apple Silicon After a 4 GB macOS reserve, about 20 GB is usable. 14B Q4 is 10.4 GB and fits. 27B Q4 is 18.2 GB and is tight. 32B Q4 is 21.2 GB and does not fit. | 24 GB | See the listing | Amazon |
| Apple Mac mini (M4, 24GB) Apple Silicon After a 4 GB macOS reserve, about 20 GB is usable. 14B Q4 is 10.4 GB and fits. 27B Q4 is 18.2 GB and is tight. 32B Q4 is 21.2 GB and does not fit. | 24 GB | See the listing | — |
| NVIDIA RTX 5080 (16GB) NVIDIA | 16 GB | See the listing | Amazon |
| NVIDIA RTX 5070 Ti (16GB) NVIDIA | 16 GB | See the listing | — |
| NVIDIA RTX 4080 (16GB) NVIDIA | 16 GB | See the listing | Amazon |
| NVIDIA RTX 4070 Ti Super (16GB) NVIDIA | 16 GB | See the listing | Amazon |
| PNY GeForce RTX 5060 Ti (16GB) NVIDIA | 16 GB | See the listing | Amazon |
| NVIDIA RTX 4060 Ti (16GB) NVIDIA | 16 GB | See the listing | Amazon |
| Apple Mac mini (M6, 16GB) Apple Silicon After a 4 GB macOS reserve, about 12 GB is usable. 7B Q4 is 6.2 GB and fits. 8B Q4 is 6.8 GB and fits. Gemma 3 12B Q4 is 9.2 GB and fits with little room. 14B Q4 is 10.4 GB and is tight. 27B Q4 is 18.2 GB and does not fit. | 16 GB | See the listing | Amazon |
| Apple Mac mini (M4, 16GB) Apple Silicon After a 4 GB macOS reserve, about 12 GB is usable. 7B Q4 is 6.2 GB and fits. 14B Q4 is 10.4 GB and is tight. 27B Q4 is 18.2 GB and does not fit. | 16 GB | See the listing | — |
| NVIDIA RTX 5070 (12GB) NVIDIA | 12 GB | See the listing | Amazon |
| NVIDIA GeForce RTX 4070 (12GB) NVIDIA | 12 GB | See the listing | Amazon |
| Intel Arc B580 (12GB) Intel | 12 GB | See the listing | Amazon |
| NVIDIA GeForce RTX 4060 (8GB) NVIDIA | 8 GB | See the listing | Amazon |
Max model @ Q4 is the largest dense size where params × 0.6 + 2 still fits (8 GB → 10B, 12 GB → 16B, 16 GB → 23B, 24 GB → 36B, 32 GB → 50B, 128 GB → 210B). Apple Silicon rows subtract 4 GB for macOS first, on top of that +2. The column is a capacity ceiling, not a speed. Strix Halo at 128 GB and ~256 GB/s holds a slow dense 70B Q4 and large MoE quants, not a 210B interactive model. Mixture-of-Experts models use total parameters. Full methodology.
How to read the table by memory tier
Sizes below use the site formula, parameters x 0.5 x 1.2 + 2 GB at Q4 and a short context. They are formula estimates, not measurements, and longer context needs more memory. Pick the tier by the largest model you want to run, then compare rows inside that tier on bandwidth and TDP.
- 8 GB (RTX 4060): an 8B model at Q4 is about 6.8 GB and fits. A 14B at Q4 is about 10.4 GB and does not.
- 12 GB (RTX 4070, RTX 5070, Intel Arc B580): a 14B at Q4 (10.4 GB) fits with little room left for context. A 20B at Q4 (14.0 GB) does not.
- 16 GB (RTX 4060 Ti 16 GB, RTX 5060 Ti, RTX 4070 Ti Super, RTX 5070 Ti, RTX 4080, RTX 5080): a 14B at Q4 fits with room for context, and a 20B at Q4 (14.0 GB) fits. A 27B at Q4 (18.2 GB) does not. These cards hold the same models; the Bandwidth and TDP columns are where they differ.
- 24 GB (RTX 3090, RTX 4090, Radeon RX 7900 XTX): a 27B at Q4 (18.2 GB) and a 32B at Q4 (21.2 GB) fit at a short context. A 70B at Q4 is about 44 GB and does not fit on one card.
- 32 GB (RTX 5090): the 32 GB desktop card in this table. It leaves more room for context on a 32B model, but a 70B at Q4 still does not fit.
- Apple unified memory (Mac mini and Mac Studio): this site subtracts about 4 GB for macOS before the formula, so a 16 GB Mac has about 12 GB usable, a 24 GB Mac about 20 GB and a 36 GB Mac Studio about 32 GB. For a 70B at Q4 (44 GB), 48 GB is tight and 64 GB is the safer size.
- 128 GB (DGX Spark, Strix Halo mini PCs such as the GMKtec EVO-X2, Mac Studio 128 GB): a 70B at Q4 fits with room to spare. The Strix Halo boxes run at about 256 GB/s, so a dense 70B fits but is slow; large Mixture-of-Experts quants are the more common reason to buy that much memory.
Prior-generation and watchlist rows in our catalog show a dash instead of a buy link. To size one specific model against one device, use the VRAM calculator.