DGX Spark 128GB vs M5 Ultra 256GB for local LLMs
- DGX Spark 128GB: Uses its largest memory option, 128 GB, with 8k context and an f16 KV cache.
- M5 Ultra 256GB: Uses its largest memory option, 256 GB, with 8k context and an f16 KV cache.
What to know before choosing
This page compares one 128GB DGX Spark with the M5 Ultra in its 256GB configuration, the chip in the Mac Studio (2026). CanRun counts about 70% of the Mac's memory for its GPU, 179.2 GB, against 116.9 GB on the DGX Spark. That decides the biggest files: DeepSeek V4 Flash 0731 runs great on the M5 Ultra, a theoretical estimate, and does not fit one DGX Spark. Two linked 128GB units or four 64GB units run it well, also theoretical. GLM-5.3-Flash at Q4_K_M does not fit the Mac's GPU share, so CanRun calculates it CPU-only and caps it at Runs slowly, although its theoretical speed estimate alone would rate higher. The macOS default limit is higher than CanRun's figure, so check ollama ps there.
Below those sizes the M5 Ultra's bandwidth leads. Gemma 4 31B runs great on the M5 Ultra and slowly on the DGX Spark, a public benchmark. For reading long prompts, CanRun's rough estimates (±40%) put the M5 Ultra almost twice as fast on MoE models such as gpt-oss-120b and the two close on dense ones such as Gemma 4 31B.
The DGX Spark's 64GB configuration and the M5 Ultra's 96GB one are not on this page; see the NVIDIA DGX Spark and Apple M5 Ultra pages and our DGX Spark guide.
Speed
The M5 Ultra 256GB generates text about 3.3× faster than the DGX Spark 128GB (median of our estimates over the 56 models that fit entirely in memory on both at Q4_K_M or the closest available quant).
- Reference model
- Ling-3.0 Flash at Q4_K_M, the largest model that fits entirely in memory on both at Q4_K_M or the closest available quant
- DGX Spark 128GB
- est. 52.8 tok/s (36.9–68.6, theoretical estimate ±30%)
- M5 Ultra 256GB
- est. 106.3 tok/s (74.4–138.2, theoretical estimate ±30%)
- Prompt processing on Ling-3.0 Flash
- DGX Spark 128GB: about 1,129 tok/s (678–1,581, rough estimate ±40%); M5 Ultra 256GB: about 2,064 tok/s (1,238–2,889, rough estimate ±40%)
Specs side by side
| Spec | DGX Spark 128GB | M5 Ultra 256GB | Difference |
|---|---|---|---|
| Memory | 128 GB of unified memory | 256 GB of unified memory | +128 GB on the M5 Ultra 256GB |
| Memory bandwidth | 273.0 GB/s | 1,228.0 GB/s | 4.5× on the M5 Ultra 256GB |
| FP16 compute | 62.5 TFLOPS | 33.2 TFLOPS | 1.88× on the DGX Spark 128GB |
| Launch year | 2025 | 2026 | — |
| Price | $3,999 launch MSRP (128 GB) | — | — |
| Public benchmark results | 14 results | 1 result | — |
Verdicts at a glance
Out of 59 models at Q4_K_M (or the closest available quant) with 8k context.
- DGX Spark 128GBRuns great34 modelsRuns well12 modelsRuns slowly10 modelsWon't run3 models
- M5 Ultra 256GBRuns great57 modelsRuns well1 modelRuns slowly1 modelWon't run0 models
How to read the verdicts
- Runs great
- Fully on the GPU at 20 tok/s or more
- Runs well
- Fully on the GPU at 8–20 tok/s; MoE experts in system RAM at 20 tok/s or more; or at least 90% on the GPU at 8 tok/s or more
- Runs slowly
- 2–8 tok/s; CPU-only; less than 90% on the GPU; or MoE experts in system RAM below 20 tok/s
- Won't run
- Does not fit, or under 2 tok/s
Where the verdicts differ
25 models get a different verdict on the two.
Scroll sideways to see every column.
| Model | DGX Spark 128GB verdict | DGX Spark 128GB speed | M5 Ultra 256GB verdict | M5 Ultra 256GB speed |
|---|---|---|---|---|
| DeepSeek V4 Flash 0731MXFP4 | Won't run | — | Runs great | est. 52.5 tok/s36.7–68.2theoretical estimate ±30% |
| Solar Open 2 250BIQ4_XS | Won't runTry Q2_K (heavy quality loss): Runs well | est. 22.0 tok/s17.6–26.3calibrated estimate ±20% | Runs great | est. 50.3 tok/s40.3–60.4calibrated estimate ±20% |
| Qwen3.5 122B-A10BQ4_K_M | Runs well | 29.0 tok/s8k estimate 24.0 tok/s (19.2–28.8, calibrated estimate ±20%)measured (1 run, 4k context) | Runs great | est. 58.6 tok/s46.9–70.4calibrated estimate ±20% |
| Laguna S 2.1Q4_K_M | Runs well | est. 28.4 tok/s22.7–34.1calibrated estimate ±20% | Runs great | est. 61.9 tok/s49.5–74.3calibrated estimate ±20% |
| Solar Open 100BQ4_K_M | Runs well | est. 22.8 tok/s18.3–27.4calibrated estimate ±20% | Runs great | est. 56.5 tok/s45.2–67.8calibrated estimate ±20% |
| EXAONE 4.5 33BQ4_K_M | Runs slowly | est. 10.3 tok/s8.3–12.4calibrated estimate ±20% | Runs great | est. 34.0 tok/s27.2–40.8calibrated estimate ±20% |
| Nanbeige4.3 33BQ4_K_M | Runs slowly | est. 9.4 tok/s7.6–11.3calibrated estimate ±20% | Runs great | est. 31.1 tok/s24.8–37.3calibrated estimate ±20% |
| DeepSeek R1 Distill Qwen 32BQ4_K_M | Runs slowly | est. 9.7 tok/s7.7–11.6calibrated estimate ±20% | Runs great | est. 31.8 tok/s25.5–38.2calibrated estimate ±20% |
| Qwen3 32BQ4_K_M | Runs slowly | 10.7 tok/s8k estimate 9.7 tok/s (7.8–11.7, calibrated estimate ±20%)measured (2 runs, 512 context) | Runs great | est. 32.0 tok/s25.6–38.3calibrated estimate ±20% |
| EXAONE 4.0 32BQ4_K_M | Runs slowly | est. 10.7 tok/s8.6–12.9calibrated estimate ±20% | Runs great | est. 35.2 tok/s28.2–42.3calibrated estimate ±20% |
| Gemma 4 31BQ4_K_M | Runs slowly | 10.1 tok/s8k estimate 11.2 tok/s (9.0–13.5, calibrated estimate ±20%)measured (1 run, 512 context) | Runs great | est. 36.9 tok/s29.5–44.2calibrated estimate ±20% |
| Muse Glimmer 30BQ4_K_M | Runs well | est. 12.2 tok/s9.8–14.7calibrated estimate ±20% | Runs great | est. 40.2 tok/s32.1–48.2calibrated estimate ±20% |
| Granite 4.2 30BQ4_K_M | Runs slowly | est. 10.7 tok/s8.6–12.9calibrated estimate ±20% | Runs great | est. 35.2 tok/s28.2–42.3calibrated estimate ±20% |
| Qwen3.5 27BQ4_K_M | Runs slowly | 11.3 tok/s8k estimate 12.1 tok/s (9.7–14.5, calibrated estimate ±20%)measured (1 run, 4k context) | Runs great | est. 39.7 tok/s31.7–47.6calibrated estimate ±20% |
| Qwen3.6 27BQ4_K_M | Runs well | est. 12.3 tok/s9.8–14.7calibrated estimate ±20% | Runs great | est. 40.3 tok/s32.3–48.4calibrated estimate ±20% |
| Qwen3.8 27BQ4_K_M | Runs slowly | est. 11.8 tok/s9.5–14.2calibrated estimate ±20% | Runs great | est. 38.9 tok/s31.1–46.7calibrated estimate ±20% |
| Gemma 3 27BQ4_K_M | Runs well | est. 12.4 tok/s9.9–14.8calibrated estimate ±20% | Runs great | est. 40.6 tok/s32.5–48.8calibrated estimate ±20% |
| Mistral Small 3.2 24BQ4_K_M | Runs well | est. 13.6 tok/s10.9–16.3calibrated estimate ±20% | Runs great | est. 44.7 tok/s35.7–53.6calibrated estimate ±20% |
| Qwen3 14BQ4_K_M | Runs well | est. 20.6 tok/s16.5–24.7calibrated estimate ±20% | Runs great | est. 67.7 tok/s54.1–81.2calibrated estimate ±20% |
| DeepSeek R1 Distill Qwen 14BQ4_K_M | Runs well | est. 20.1 tok/s16.1–24.1calibrated estimate ±20% | Runs great | est. 66.0 tok/s52.8–79.2calibrated estimate ±20% |
| HyperCLOVA X SEED Think 14BQ4_K_M | Runs well | est. 20.9 tok/s14.6–27.2theoretical estimate ±30% | Runs great | est. 68.7 tok/s48.1–89.3theoretical estimate ±30% |
| Phi-4Q4_K_M | Runs well | est. 19.8 tok/s15.9–23.8calibrated estimate ±20% | Runs great | est. 65.2 tok/s52.2–78.3calibrated estimate ±20% |
| Gemma 4 12BQ4_K_M | Runs well | est. 29.4 tok/s23.5–35.2calibrated estimate ±20% | Runs great | est. 96.5 tok/s77.2–115.8calibrated estimate ±20% |
| Llama 3.3 70BQ4_K_M | Runs slowly | 4.7 tok/s8k estimate 4.7 tok/s (3.8–5.7, calibrated estimate ±20%)measured (2 runs, 512 context) | Runs well | est. 15.5 tok/s12.4–18.6calibrated estimate ±20% |
| GLM-5.3 FlashQ4_K_M | Won't run | — | Runs slowly | est. 39.5 tok/s27.6–51.3theoretical estimate ±30% |
Every model on both
Scroll sideways to see every column.
| Model | DGX Spark 128GB verdict | DGX Spark 128GB speed | M5 Ultra 256GB verdict | M5 Ultra 256GB speed |
|---|---|---|---|---|
| DeepSeek V4 Flash 0731MXFP4 | Won't run | — | Runs great | est. 52.5 tok/s36.7–68.2theoretical estimate ±30% |
| Solar Open 2 250BIQ4_XS | Won't runTry Q2_K (heavy quality loss): Runs well | est. 22.0 tok/s17.6–26.3calibrated estimate ±20% | Runs great | est. 50.3 tok/s40.3–60.4calibrated estimate ±20% |
| Ling-3.0 FlashQ4_K_M | Runs great | est. 52.8 tok/s36.9–68.6theoretical estimate ±30% | Runs great | est. 106.3 tok/s74.4–138.2theoretical estimate ±30% |
| Qwen3.5 122B-A10BQ4_K_M | Runs well | 29.0 tok/s8k estimate 24.0 tok/s (19.2–28.8, calibrated estimate ±20%)measured (1 run, 4k context) | Runs great | est. 58.6 tok/s46.9–70.4calibrated estimate ±20% |
| Laguna S 2.1Q4_K_M | Runs well | est. 28.4 tok/s22.7–34.1calibrated estimate ±20% | Runs great | est. 61.9 tok/s49.5–74.3calibrated estimate ±20% |
| gpt-oss-120bMXFP4 | Runs great | 58.7 tok/s8k estimate 50.8 tok/s (40.6–60.9, calibrated estimate ±20%)measured (1 run, 2k context) | Runs great | est. 131.0 tok/s104.8–157.2calibrated estimate ±20% |
| Solar Open 100BQ4_K_M | Runs well | est. 22.8 tok/s18.3–27.4calibrated estimate ±20% | Runs great | est. 56.5 tok/s45.2–67.8calibrated estimate ±20% |
| Kolibri 1Q4_K_M | Runs great | est. 52.2 tok/s41.7–62.6calibrated estimate ±20% | Runs great | est. 108.1 tok/s86.5–129.8calibrated estimate ±20% |
| K2-Horizon MoVA 36B-A4BQ4_K_M | Runs great | est. 37.7 tok/s26.4–49.1theoretical estimate ±30% | Runs great | est. 82.3 tok/s57.6–106.9theoretical estimate ±30% |
| Qwen3.5 35B-A3BQ4_K_M | Runs great | est. 61.9 tok/s49.5–74.2calibrated estimate ±20% | Runs great | est. 116.8 tok/s93.5–140.2calibrated estimate ±20% |
| Qwen3.6 35B-A3BQ4_K_M | Runs great | est. 70.5 tok/s56.4–84.6calibrated estimate ±20% | Runs great | est. 125.7 tok/s100.5–150.8calibrated estimate ±20% |
| MiniCPM-V 4.7 35B-A3BQ4_K_M | Runs great | est. 76.0 tok/s60.8–91.2calibrated estimate ±20% | Runs great | est. 130.8 tok/s104.7–157.0calibrated estimate ±20% |
| EXAONE 4.5 33BQ4_K_M | Runs slowly | est. 10.3 tok/s8.3–12.4calibrated estimate ±20% | Runs great | est. 34.0 tok/s27.2–40.8calibrated estimate ±20% |
| Nanbeige4.3 33BQ4_K_M | Runs slowly | est. 9.4 tok/s7.6–11.3calibrated estimate ±20% | Runs great | est. 31.1 tok/s24.8–37.3calibrated estimate ±20% |
| DeepSeek R1 Distill Qwen 32BQ4_K_M | Runs slowly | est. 9.7 tok/s7.7–11.6calibrated estimate ±20% | Runs great | est. 31.8 tok/s25.5–38.2calibrated estimate ±20% |
| Qwen3 32BQ4_K_M | Runs slowly | 10.7 tok/s8k estimate 9.7 tok/s (7.8–11.7, calibrated estimate ±20%)measured (2 runs, 512 context) | Runs great | est. 32.0 tok/s25.6–38.3calibrated estimate ±20% |
| EXAONE 4.0 32BQ4_K_M | Runs slowly | est. 10.7 tok/s8.6–12.9calibrated estimate ±20% | Runs great | est. 35.2 tok/s28.2–42.3calibrated estimate ±20% |
| Nemotron 3.5 Lightning 30B-A3BQ4_K_M | Runs great | est. 69.8 tok/s48.8–90.7theoretical estimate ±30% | Runs great | est. 168.6 tok/s118.0–219.2theoretical estimate ±30% |
| Gemma 4 31BQ4_K_M | Runs slowly | 10.1 tok/s8k estimate 11.2 tok/s (9.0–13.5, calibrated estimate ±20%)measured (1 run, 512 context) | Runs great | est. 36.9 tok/s29.5–44.2calibrated estimate ±20% |
| Qwen3 30B-A3B (2507)Q4_K_M | Runs great | est. 60.7 tok/s48.6–72.9calibrated estimate ±20% | Runs great | est. 106.6 tok/s85.3–127.9calibrated estimate ±20% |
| Muse Glimmer 30BQ4_K_M | Runs well | est. 12.2 tok/s9.8–14.7calibrated estimate ±20% | Runs great | est. 40.2 tok/s32.1–48.2calibrated estimate ±20% |
| Granite 4.2 30BQ4_K_M | Runs slowly | est. 10.7 tok/s8.6–12.9calibrated estimate ±20% | Runs great | est. 35.2 tok/s28.2–42.3calibrated estimate ±20% |
| Qwen3.5 27BQ4_K_M | Runs slowly | 11.3 tok/s8k estimate 12.1 tok/s (9.7–14.5, calibrated estimate ±20%)measured (1 run, 4k context) | Runs great | est. 39.7 tok/s31.7–47.6calibrated estimate ±20% |
| Qwen3.6 27BQ4_K_M | Runs well | est. 12.3 tok/s9.8–14.7calibrated estimate ±20% | Runs great | est. 40.3 tok/s32.3–48.4calibrated estimate ±20% |
| Qwen3.8 27BQ4_K_M | Runs slowly | est. 11.8 tok/s9.5–14.2calibrated estimate ±20% | Runs great | est. 38.9 tok/s31.1–46.7calibrated estimate ±20% |
| Gemma 3 27BQ4_K_M | Runs well | est. 12.4 tok/s9.9–14.8calibrated estimate ±20% | Runs great | est. 40.6 tok/s32.5–48.8calibrated estimate ±20% |
| Gemma 4 26B-A4BQ4_K_M | Runs great | 47.0 tok/s8k estimate 51.5 tok/s (41.2–61.8, calibrated estimate ±20%)measured (1 run, 32k context) | Runs great | est. 115.8 tok/s92.6–138.9calibrated estimate ±20% |
| Mistral Small 3.2 24BQ4_K_M | Runs well | est. 13.6 tok/s10.9–16.3calibrated estimate ±20% | Runs great | est. 44.7 tok/s35.7–53.6calibrated estimate ±20% |
| gpt-oss-20bMXFP4 | Runs great | est. 71.7 tok/s57.4–86.0calibrated estimate ±20% | Runs great | est. 187.4 tok/s149.9–224.9calibrated estimate ±20% |
| Kanana 1.5 15.7B-A3BQ4_K_M | Runs great | est. 62.2 tok/s49.7–74.6calibrated estimate ±20% | Runs great | est. 128.1 tok/s102.5–153.7calibrated estimate ±20% |
| Qwen3 14BQ4_K_M | Runs well | est. 20.6 tok/s16.5–24.7calibrated estimate ±20% | Runs great | est. 67.7 tok/s54.1–81.2calibrated estimate ±20% |
| DeepSeek R1 Distill Qwen 14BQ4_K_M | Runs well | est. 20.1 tok/s16.1–24.1calibrated estimate ±20% | Runs great | est. 66.0 tok/s52.8–79.2calibrated estimate ±20% |
| HyperCLOVA X SEED Think 14BQ4_K_M | Runs well | est. 20.9 tok/s14.6–27.2theoretical estimate ±30% | Runs great | est. 68.7 tok/s48.1–89.3theoretical estimate ±30% |
| Phi-4Q4_K_M | Runs well | est. 19.8 tok/s15.9–23.8calibrated estimate ±20% | Runs great | est. 65.2 tok/s52.2–78.3calibrated estimate ±20% |
| Gemma 3 12BQ4_K_M | Runs great | est. 27.2 tok/s21.7–32.6calibrated estimate ±20% | Runs great | est. 89.3 tok/s71.5–107.2calibrated estimate ±20% |
| Mellum2.1 12B-A2.5B ThinkingQ4_K_M | Runs great | est. 104.4 tok/s83.5–125.2calibrated estimate ±20% | Runs great | est. 183.0 tok/s146.4–219.6calibrated estimate ±20% |
| Gemma 4 12BQ4_K_M | Runs well | est. 29.4 tok/s23.5–35.2calibrated estimate ±20% | Runs great | est. 96.5 tok/s77.2–115.8calibrated estimate ±20% |
| Qwen3.5 9BQ4_K_M | Runs great | 35.2 tok/s8k estimate 34.7 tok/s (27.8–41.6, calibrated estimate ±20%)measured (1 run, 4k context) | Runs great | est. 114.0 tok/s91.2–136.8calibrated estimate ±20% |
| K2-Horizon 7BQ4_K_M | Runs great | est. 31.3 tok/s25.1–37.6calibrated estimate ±20% | Runs great | est. 103.0 tok/s82.4–123.6calibrated estimate ±20% |
| Qwen3 8BQ4_K_M | Runs great | 43.9 tok/s8k estimate 34.1 tok/s (27.3–41.0, calibrated estimate ±20%)measured (1 run, 512 context) | Runs great | est. 112.2 tok/s89.8–134.7calibrated estimate ±20% |
| DeepSeek R1 0528 Qwen3 8BQ4_K_M | Runs great | est. 34.1 tok/s27.3–41.0calibrated estimate ±20% | Runs great | est. 112.2 tok/s89.8–134.7calibrated estimate ±20% |
| DeepSeek R1 Distill Llama 8BQ4_K_M | Runs great | est. 35.5 tok/s28.4–42.6calibrated estimate ±20% | Runs great | est. 116.8 tok/s93.4–140.1calibrated estimate ±20% |
| Kanana 1.5 8BQ4_K_M | Runs great | est. 35.5 tok/s28.4–42.6calibrated estimate ±20% | Runs great | est. 116.8 tok/s93.4–140.1calibrated estimate ±20% |
| Llama 3.1 8BQ4_K_M | Runs great | est. 35.5 tok/s28.4–42.6calibrated estimate ±20% | Runs great | est. 116.8 tok/s93.4–140.1calibrated estimate ±20% |
| Gemma 4 E4BQ4_K_M | Runs great | est. 61.5 tok/s43.1–80.0theoretical estimate ±30% | Runs great | est. 159.1 tok/s111.4–206.8theoretical estimate ±30% |
| Ling-3.0 TinyQ4_K_M | Runs great | est. 179.8 tok/s125.8–233.7theoretical estimate ±30% | Runs great | est. 264.1 tok/s184.9–343.3theoretical estimate ±30% |
| Gemma 4 E2BQ4_K_M | Runs great | est. 115.9 tok/s81.1–150.6theoretical estimate ±30% | Runs great | est. 252.2 tok/s176.5–327.9theoretical estimate ±30% |
| Qwen3.5 4BQ4_K_M | Runs great | est. 68.7 tok/s55.0–82.5calibrated estimate ±20% | Runs great | est. 225.9 tok/s180.7–271.1calibrated estimate ±20% |
| Nanbeige4.2 3BQ4_K_M | Runs great | est. 34.8 tok/s24.3–45.2theoretical estimate ±30% | Runs great | est. 114.3 tok/s80.0–148.5theoretical estimate ±30% |
| Spark-X2.5 4BQ4_K_M | Runs great | est. 73.4 tok/s58.7–88.1calibrated estimate ±20% | Runs great | est. 241.2 tok/s193.0–289.4calibrated estimate ±20% |
| Llama 3.2 3BQ4_K_M | Runs great | est. 72.0 tok/s57.6–86.3calibrated estimate ±20% | Runs great | est. 236.5 tok/s189.2–283.8calibrated estimate ±20% |
| LFM2.5 2.6BQ4_K_M | Runs great | est. 118.0 tok/s82.6–153.4theoretical estimate ±30% | Runs great | est. 388.0 tok/s271.6–504.3theoretical estimate ±30% |
| MiniCPM5 2BQ4_K_M | Runs great | est. 111.4 tok/s89.1–133.6calibrated estimate ±20% | Runs great | est. 366.0 tok/s292.8–439.2calibrated estimate ±20% |
| HyperCLOVA X SEED 1.5BQ4_K_M | Runs great | est. 117.3 tok/s93.8–140.8calibrated estimate ±20% | Runs great | est. 385.6 tok/s308.5–462.7calibrated estimate ±20% |
| EXAONE 4.0 1.2BQ4_K_M | Runs great | est. 162.1 tok/s129.7–194.6calibrated estimate ±20% | Runs great | est. 533.0 tok/s426.4–639.6calibrated estimate ±20% |
| Llama 3.2 1BQ4_K_M | Runs great | est. 197.5 tok/s158.0–236.9calibrated estimate ±20% | Runs great | est. 649.1 tok/s519.2–778.9calibrated estimate ±20% |
| MiniCPM5 1BQ4_K_M | Runs great | est. 238.9 tok/s191.1–286.7calibrated estimate ±20% | Runs great | est. 785.3 tok/s628.2–942.4calibrated estimate ±20% |
| Llama 3.3 70BQ4_K_M | Runs slowly | 4.7 tok/s8k estimate 4.7 tok/s (3.8–5.7, calibrated estimate ±20%)measured (2 runs, 512 context) | Runs well | est. 15.5 tok/s12.4–18.6calibrated estimate ±20% |
| GLM-5.3 FlashQ4_K_M | Won't run | — | Runs slowly | est. 39.5 tok/s27.6–51.3theoretical estimate ±30% |
Models with a thinking mode spend extra tokens thinking, so their speed thresholds are 1.5× stricter (30 / 12 / 3 tok/s). With thinking turned off, a model may do one grade better.
Public benchmarks
Measured results on the DGX Spark 128GB
| Model | Quant | Backend | Context | Prompt (tok/s) | Generation (tok/s) | Flags | Source | Measured |
|---|---|---|---|---|---|---|---|---|
| gpt-oss-120b | MXFP4 | llama.cpp | 2k | 2,444.0 | 58.7 | CUDA, no mmap | github.com | 2026-02-05 |
| Gemma 4 31B | Q4_K_M | llama.cpp | 512 | 687.0 | 10.1 | CUDA | github.com | 2026-04-14 |
| Gemma 4 26B-A4B | Q4_K_M | llama.cpp | 32k | — | 47.0 | CUDA, depth 32768 | github.com | 2026-06-11 |
| Qwen3.5 35B-A3B | Q8_0 | llama.cpp | 2k | 2,572.0 | 49.2 | CUDA | github.com | 2026-03-12 |
| Qwen3.5 9B | Q4_K_M | llama.cpp | 4k | — | 35.2 | — | github.com | 2026-08-13 |
| Qwen3.5 27B | Q4_K_M | llama.cpp | 4k | — | 11.3 | — | github.com | 2026-08-13 |
| Qwen3.5 122B-A10B | Q4_K_M | llama.cpp | 4k | — | 29.0 | — | github.com | 2026-08-13 |
| Llama 3.3 70B | Q4_K_M | llama.cpp | 512 | 284.0 | 4.7 | CUDA, Vulkan | github.com | 2025-11-22 |
| Llama 3.3 70B | Q4_K_M | llama.cpp | 512 | — | 4.6 | CUDA | github.com | 2026-08-12 |
| Llama 3.3 70B | Q8_0 | llama.cpp | 512 | — | 2.9 | CUDA | github.com | 2026-08-12 |
| Llama 3.3 70B | Q2_K | llama.cpp | 512 | — | 7.2 | CUDA | github.com | 2026-08-12 |
| Qwen3 32B | Q4_K_M | llama.cpp | 512 | 630.0 | 10.5 | CUDA, Vulkan | github.com | 2025-11-22 |
| Qwen3 32B | Q4_K_M | llama.cpp | 512 | 762.0 | 10.8 | CUDA | github.com | 2025-12-05 |
| Qwen3 8B | Q4_K_M | llama.cpp | 512 | 3,168.0 | 43.9 | CUDA | github.com | 2025-12-05 |
Measured results on the M5 Ultra 256GB
| Model | Quant | Backend | Context | Prompt (tok/s) | Generation (tok/s) | Flags | Source | Measured |
|---|---|---|---|---|---|---|---|---|
| llama-2-7b | Q4_0 | llama.cpp | 512 | 4,945.0 | 179.1 | Metal, FA | github.com | 2026-09-30 |
Public benchmarks we calibrate against. Their conditions (context, backend, flags) can differ from the estimates above.
Frequently asked questions
Which is faster for local LLMs, the DGX Spark 128GB or the M5 Ultra 256GB?
The M5 Ultra 256GB generates text about 3.3× faster than the DGX Spark 128GB (median of our estimates over the 56 models that fit entirely in memory on both at Q4_K_M or the closest available quant). Estimated speed on Ling-3.0 Flash at Q4_K_M: the DGX Spark 128GB est. 52.8 tok/s (36.9–68.6, theoretical estimate ±30%); the M5 Ultra 256GB est. 106.3 tok/s (74.4–138.2, theoretical estimate ±30%).
What is the largest model each can run entirely in memory?
The DGX Spark 128GB fits Ling-3.0 Flash at Q4_K_M (77.01 GB file) entirely in 128 GB of unified memory, at est. 52.8 tok/s (36.9–68.6, theoretical estimate ±30%). The M5 Ultra 256GB fits DeepSeek V4 Flash 0731 at MXFP4 (154.99 GB file) entirely in 256 GB of unified memory, at est. 52.5 tok/s (36.7–68.2, theoretical estimate ±30%).
How many local LLMs run well on each?
At Q4_K_M with 8k context, out of 59 tracked models: the DGX Spark 128GB — 34 Runs great, 12 Runs well, 10 Runs slowly, 3 Won't run; the M5 Ultra 256GB — 57 Runs great, 1 Runs well, 1 Runs slowly, 0 Won't run. 25 models get a different verdict.
Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.