← All machines
NVIDIA GB10 · Dell
Dell Pro Precision with GB10
Dell's GB10 box (model FCM1253, first sold as Dell Pro Max with GB10), with 128GB, DGX OS and Dell support options. It runs the same software as the DGX Spark and differs in price, storage choices, power adapter and warranty.
Memory for models
120 GBassumption
The maker publishes no GPU limit for this coherent memory pool; 8 GB is assumed for the operating system.
Memory bandwidth
273 GB/s
Maker price
~$5,523
Largest at 4-bit
113.1 GB at 8,192 tokens of context (an assumed limit, see Memory for models)
gpt-oss-20b, 4-bit
83 tok/smedian of 13 runs
Memory configurations
One page covers every memory option the maker sells. Prices are the maker’s list price for the cheapest configuration with that much memory.
| Memory | Price | Memory the GPU can use |
|---|---|---|
| 128 GB | ~$5,523 | 120 GB gpu-usable memoryassumption |
The maker publishes no GPU limit for this coherent memory pool; 8 GB is assumed for the operating system. How GPU memory limits work on unified-memory machines.
What it runs at 4-bit
Every model in the catalog at its 4-bit file with 8,192 tokens of context. 30 of 41 fit. Speeds are the median of measured runs on the same chip where any exist; otherwise an estimate from memory bandwidth. Try other quants and context lengths.
| Model | Needs | Fit | Generation speed |
|---|---|---|---|
| Kimi K32.8T, 104B active · UD-Q4_K_XL | 1406.8 GBcontext memory estimated | Does not fit | |
| DeepSeek V4 Pro1.6T, 49B active · Q4_K_M | 890.5 GBcontext memory not counted | Does not fit | |
| Kimi K2.61T, 32B active · UD-Q4_K_XL | 545.7 GBcontext memory estimated | Does not fit | |
| DeepSeek V4.1 Flash552B, 16B active · Q4 (DwarfStar) | 484.5 GB | Does not fit | |
| GLM-5.3753.3B · UD-Q4_K_XL | 437.4 GBcontext memory estimated | Does not fit | |
| DeepSeek V3.2671B, 37B active · Q4_K_M | 379.6 GBcontext memory estimated | Does not fit | |
| MiniMax M3428B, 23B active · Q4_K_M | 244.8 GBcontext memory not counted | Does not fit | |
| Qwen3.5 397B-A17B397B, 17B active · Q4_K_M | 229.1 GBcontext memory estimated | Does not fit | |
| GLM-5.3 Flash320B, 18B active · UD-Q4_K_XL | 187.6 GBcontext memory estimated | Does not fit | |
| Hy3295B, 21B active · Q4_K_M | 173.6 GB | Does not fit | |
| MiniMax M2.7228.7B · Q4_K_M | 132.7 GB | Does not fit | |
| Qwen3.8 Flash-Next125B, 6B active · Q4_K_M | 113.1 GBcontext memory estimated | Fitsassumed limit | ~22–29 tok/sestimate |
| Nemotron 3 Super 120B-A12B120B, 12B active · Q4_K_M | 82.6 GBcontext memory estimated | Fitsassumed limit | 17 tok/smedian of 2 runs |
| Qwen3.5 122B-A10B122B, 10B active · Q4_K_M | 73.0 GBcontext memory estimated | Fitsassumed limit | 21 tok/smedian of 2 runs |
| Mistral Medium 3.5 128B128B · Q4_K_M | 74.0 GB | Fitsassumed limit | ~2.5–3.0 tok/sestimate |
| GLM-4.5-Air106B, 12B active · Q4_K_M | 70.9 GB | Fitsassumed limit | ~16–20 tok/sestimate |
| Mistral Small 4 119B (2603)119B, 6.5B active · Q4_K_M | 69.3 GBcontext memory estimated | Fitsassumed limit | ~32–41 tok/sestimate |
| Llama 4 Scout 17B-16E Instruct109B, 17B active · Q4_K_M | 62.4 GBcontext memory not counted | Fitsassumed limit | ~13–16 tok/sestimate |
| gpt-oss-120b117B, 5.1B active · MXFP4 (native) | 60.2 GBcontext memory estimated | Fitsassumed limit | 53 tok/smedian of 23 runs |
| Qwen3 Coder Next80B, 3B active · Q4_K_M | 46.9 GBcontext memory estimated | Fitsassumed limit | 49 tok/smedian of 2 runs |
| Llama 3.3 70B Instruct70B · Q4_K_M | 43.6 GB | Fitsassumed limit | ~4.4–5.2 tok/sestimate |
| Nemotron 3.5 Lightning 30B-A3B30B, 3B active · Q4_K_M | 25.3 GBcontext memory estimated | Fitsassumed limit | ~50–64 tok/sestimate |
| Qwen3.6 35B-A3B35B, 3B active · Q4_K_M | 22.4 GBcontext memory estimated | Fitsassumed limit | ~67–86 tok/sestimate |
| Ornith 1.5 35B-A3B35B, 3B active · Q4_K_M | 21.9 GBcontext memory estimated | Fitsassumed limit | ~69–88 tok/sestimate |
| Qwen3 Coder 30B-A3B Instruct30.5B, 3.3B active · Q4_K_M | 19.5 GB | Fitsassumed limit | ~64–82 tok/sestimate |
| Gemma 4 31B IT30.7B · Q4_K_M | 18.9 GBcontext memory estimated | Fitsassumed limit | ~10–12 tok/sestimate |
| Granite 4.2 30B29.3B · Q4_K_M | 20.3 GB | Fitsassumed limit | ~10–12 tok/sestimate |
| Qwen3.8 27B27B · Q4_K_M | 18.2 GBcontext memory estimated | Fitsassumed limit | ~11–13 tok/sestimate |
| Gemma 4 26B-A4B IT25.2B, 3.8B active · Q4_K_M | 17.4 GBcontext memory estimated | Fitsassumed limit | ~50–64 tok/sestimate |
| Qwen3.6 27B27B · Q4_K_M | 17.7 GBcontext memory estimated | Fitsassumed limit | ~11–13 tok/sestimate |
| Devstral Small 2 24B (2512)24B · Q4_K_M | 16.1 GB | Fitsassumed limit | ~13–15 tok/sestimate |
| gpt-oss-20b21B, 3.6B active · MXFP4 (native) | 12.5 GBcontext memory estimated | Fitsassumed limit | 83 tok/smedian of 13 runs |
| Phi-414.7B · Q4_K_M | 11.3 GB | Fitsassumed limit | ~21–25 tok/sestimate |
| Ministral 3 14B Reasoning (2512)13.9B · Q4_K_M | 10.4 GB | Fitsassumed limit | ~22–27 tok/sestimate |
| Gemma 4 12B IT11.95B · Q4_K_M | 8.2 GBcontext memory estimated | Fitsassumed limit | ~26–31 tok/sestimate |
| Qwen3.5 9B9B · Q4_K_M | 7.0 GBcontext memory estimated | Fitsassumed limit | ~33–39 tok/sestimate |
| Granite 4.2 8B8.8B · Q4_K_M | 7.9 GB | Fitsassumed limit | ~33–40 tok/sestimate |
| Gemma 4 E4B IT8B · Q4_K_M | 6.5 GBcontext memory not counted | Fitsassumed limit | ~61–73 tok/sestimate |
| Qwen3 8B8.2B · Q4_K_M | 7.3 GB | Fitsassumed limit | ~37–44 tok/sestimate |
| Llama 3.1 8B Instruct8B · Q4_K_M | 7.1 GB | Fitsassumed limit | 38 tok/smedian of 5 runs |
| Qwen3.5 4B4B · Q4_K_M | 4.3 GBcontext memory estimated | Fitsassumed limit | ~68–81 tok/sestimate |
Measured speeds
53 published runs with one user, plain decoding and at most 4,096 tokens already in context, on GB10 Grace Blackwell. Runs on NVIDIA DGX Spark with the same chip count here too, and say so.
| Model | Quant | Backend | Generation | Prompt | Context | Measured on | Source |
|---|---|---|---|---|---|---|---|
| gpt-oss-120b | Q4_K - Medium | llama.cpp b8580 CUDA | 62 tok/s | 2,444 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-120b | Q4_K - Medium | llama.cpp b8580 CUDA | 59 tok/s | 2,315 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7941 CUDA | 59 tok/s | 2,444 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7535 CUDA | 58 tok/s | 2,438 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp (build not stated) CUDA | 57 tok/s | 1,816 tok/s2,048-token prompt | 0 | a machine with the same chip | llama.cpp benchmark discussion2025-11 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 56 tok/s | 1,939 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 | llama.cpp b7070 | 56 tok/s | 966 tok/s2,048-token prompt | 0 | this machine | owner report2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7941 CUDA | 56 tok/s | 2,310 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-120b | MXFP4 | llama.cpp (build not stated) | 55 tok/s | 1,725 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | maker2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 55 tok/s | 1,738 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7535 CUDA | 55 tok/s | 2,294 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 53 tok/s | 1,832 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 | llama.cpp (build not stated) | 53 tok/s | 1,729 tok/s2,048-token prompt | 0 | a machine with the same chip | owner report2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 51 tok/s | 1,778 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 | SGLang (lmsysorg/sglang:spark image) | 51 tok/s | 676 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| gpt-oss-120b | MXFP4 | llama.cpp (build not stated) | 49 tok/s | 1,608 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | owner report2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6816 CUDA | 46 tok/s | 1,737 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b8322 CUDA | 45 tok/s | 2,047 tok/s4,096-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6816 CUDA | 43 tok/s | 1,778 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 | Ollama (version not stated) | 42 tok/s | 95 tok/s | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp (build not stated) CUDA | 42 tok/s | 1,864 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 | Ollama v0.12.6 | 41 tok/s | 1,169 tok/s | 0 | NVIDIA DGX Spark (same chip) | Ollama blog2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp (build not stated) CUDA | 38 tok/s | 1,731 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp b7030 Metal,BLAS | 118 tok/s | – | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-11 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp b7067 CUDA | 86 tok/s | 3,798 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-11 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp b7030 CUDA | 86 tok/s | – | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-11 |
| gpt-oss-20b | Q4_K - Medium | llama.cpp b8580 CUDA | 86 tok/s | 4,453 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp (build not stated) CUDA | 84 tok/s | 3,730 tok/s2,048-token prompt | 0 | a machine with the same chip | llama.cpp benchmark discussion2025-11 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp b7941 CUDA | 83 tok/s | 4,506 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-20b | MXFP4 | llama.cpp (build not stated) | 83 tok/s | 3,670 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | maker2025-10 |
| gpt-oss-20b | Q4_K - Medium | llama.cpp b8580 CUDA | 82 tok/s | 4,230 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp b7941 CUDA | 79 tok/s | 4,158 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-20b | MXFP4 | llama.cpp b7070 | 79 tok/s | 1,646 tok/s2,048-token prompt | 0 | this machine | owner report2025-12 |
| gpt-oss-20b | MXFP4 | SGLang (lmsysorg/sglang:spark image) | 70 tok/s | 434 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| gpt-oss-20b | MXFP4 | Ollama (version not stated) | 61 tok/s | 2,054 tok/s | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| gpt-oss-20b | MXFP4 | Ollama v0.12.6 | 58 tok/s | 3,224 tok/s | 0 | NVIDIA DGX Spark (same chip) | Ollama blog2025-10 |
| Llama 3.1 8B Instruct | Q4_K_M | Ollama (version not stated) | 43 tok/s | 23,170 tok/s | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| Llama 3.1 8B Instruct | NVFP4 | TensorRT-LLM (version not stated) | 39 tok/s | 10,257 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | maker2025-10 |
| Llama 3.1 8B Instruct | Q4_K_M | Ollama v0.12.6 | 38 tok/s | 7,614 tok/s | 0 | NVIDIA DGX Spark (same chip) | Ollama blog2025-10 |
| Llama 3.1 8B Instruct | Q4_K_M | LocalScore (llamafile; version not stated) | 35 tok/s | 2,010 tok/s | 0 | a machine with the same chip | LocalScore2026-10 |
| Llama 3.1 8B Instruct | Q4_K_M | LocalScore (llamafile; version not stated) | 34 tok/s | 1,986 tok/s | 0 | a machine with the same chip | LocalScore2026-10 |
| Llama 3.1 8B Instruct | Q8_0 | Ollama (version not stated) | 29 tok/s | 19,826 tok/s | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| Llama 3.1 8B Instruct | Q8_0 | Ollama v0.12.6 | 25 tok/s | 6,110 tok/s | 0 | NVIDIA DGX Spark (same chip) | Ollama blog2025-10 |
| Llama 3.1 8B Instruct | FP8 | SGLang (lmsysorg/sglang:spark image) | 21 tok/s | 7,991 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| Nemotron 3 Super 120B-A12B | MXFP4 MoE | llama.cpp b8580 CUDA | 17 tok/s | 730 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Nemotron 3 Super 120B-A12B | MXFP4 MoE | llama.cpp b8580 CUDA | 17 tok/s | 725 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3 Coder 30B-A3B Instruct | Q8_0 | llama.cpp b7941 CUDA | 61 tok/s | 2,987 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| Qwen3 Coder 30B-A3B Instruct | Q8_0 | llama.cpp b7941 CUDA | 55 tok/s | 2,633 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| Qwen3 Coder 30B-A3B Instruct | Q8_0 | llama.cpp (build not stated) | 55 tok/s | 2,705 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | owner report2025-12 |
| Qwen3 Coder Next | MXFP4 MoE | llama.cpp b8580 CUDA | 49 tok/s | 1,743 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3 Coder Next | MXFP4 MoE | llama.cpp b8580 CUDA | 49 tok/s | 1,708 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3.5 122B-A10B | MXFP4_MOE (unsloth) | llama.cpp b8580 CUDA | 21 tok/s | 1,075 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3.5 122B-A10B | MXFP4_MOE (unsloth) | llama.cpp b8580 CUDA | 21 tok/s | 1,050 tok/s2,048-token prompt | 4,096 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
With speculative decoding
A small draft model or extra prediction heads guess several tokens ahead. Generation is faster on text the guesses get right, so these are kept out of the medians above.
| Model | Quant | Backend | Generation | Prompt | Context | Measured on | Source |
|---|---|---|---|---|---|---|---|
| Qwen3.8 27B | Q4_K_M | llama.cpp (ghcr.io full-cuda Docker image, build not stated) | 18 tok/s | 489 tok/s6,231-token prompt | 6,231 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-08 |
| Qwen3.8 27B | UD-Q4_K_XL | llama.cpp (ghcr.io full-cuda Docker image, build not stated) | 15 tok/s | – | – | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-08 |
Many users at once
Total throughput with several requests in flight. One person chatting sees far less than this.
| Model | Quant | Backend | Generation | Prompt | Context | Measured on | Source |
|---|---|---|---|---|---|---|---|
| gpt-oss-120b | MXFP4 | SGLang (lmsysorg/sglang:spark image)32 requests at once | 203 tok/s | 307 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| gpt-oss-120b | MXFP4 | SGLang (lmsysorg/sglang:spark image)8 requests at once | 117 tok/s | 302 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| gpt-oss-20b | MXFP4 | SGLang (lmsysorg/sglang:spark image)32 requests at once | 522 tok/s | 546 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| gpt-oss-20b | MXFP4 | SGLang (lmsysorg/sglang:spark image)8 requests at once | 219 tok/s | 553 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| Llama 3.1 8B Instruct | FP8 | SGLang (lmsysorg/sglang:spark image)32 requests at once | 368 tok/s | 7,950 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
| Llama 3.1 8B Instruct | FP8 | SGLang (lmsysorg/sglang:spark image)8 requests at once | 144 tok/s | 7,744 tok/s2,048-token prompt | 0 | NVIDIA DGX Spark (same chip) | review2025-10 |
Deep into a long context
Measured with more than 4,096 tokens already in context, where every new token also reads the context memory.
| Model | Quant | Backend | Generation | Prompt | Context | Measured on | Source |
|---|---|---|---|---|---|---|---|
| gpt-oss-120b | Q4_K - Medium | llama.cpp b8580 CUDA | 56 tok/s | 2,222 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7941 CUDA | 53 tok/s | 2,217 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7535 CUDA | 52 tok/s | 2,149 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-12 |
| gpt-oss-120b | Q4_K - Medium | llama.cpp b8580 CUDA | 51 tok/s | 1,948 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7941 CUDA | 49 tok/s | 1,956 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 49 tok/s | 1,738 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7535 CUDA | 48 tok/s | 1,824 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 48 tok/s | 1,714 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 45 tok/s | 1,526 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 44 tok/s | 1,533 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | Q4_K - Medium | llama.cpp b8580 CUDA | 44 tok/s | 1,561 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-120b | MXFP4 | llama.cpp (build not stated) | 43 tok/s | 1,357 tok/s2,048-token prompt | 16,384 | a machine with the same chip | owner report2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7941 CUDA | 43 tok/s | 1,567 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-120b | MXFP4 | llama.cpp b7070 | 42 tok/s | 305 tok/s2,048-token prompt | 16,384 | this machine | owner report2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6816 CUDA | 42 tok/s | 1,720 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b7535 CUDA | 41 tok/s | 1,416 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-12 |
| gpt-oss-120b | MXFP4 | llama.cpp (build not stated) | 40 tok/s | 1,308 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | owner report2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 39 tok/s | 1,242 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6817 CUDA | 39 tok/s | 1,238 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6816 CUDA | 38 tok/s | 1,512 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 | llama.cpp (build not stated) | 37 tok/s | 1,074 tok/s2,048-token prompt | 32,768 | a machine with the same chip | owner report2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b8322 CUDA | 37 tok/s | 1,676 tok/s4,096-token prompt | 20,000 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp (build not stated) CUDA | 36 tok/s | 1,628 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 | llama.cpp (build not stated) | 34 tok/s | 1,061 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | owner report2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b6816 CUDA | 34 tok/s | 1,232 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp (build not stated) CUDA | 34 tok/s | 1,395 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2025-10 |
| gpt-oss-120b | MXFP4 | llama.cpp b7070 | 33 tok/s | 181 tok/s2,048-token prompt | 32,768 | this machine | owner report2025-12 |
| gpt-oss-120b | MXFP4 MoE | llama.cpp b8322 CUDA | 30 tok/s | 1,225 tok/s4,096-token prompt | 48,000 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-20b | Q4_K - Medium | llama.cpp b8580 CUDA | 77 tok/s | 3,982 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp b7941 CUDA | 75 tok/s | 3,994 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-20b | Q4_K - Medium | llama.cpp b8580 CUDA | 72 tok/s | 3,366 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp b7941 CUDA | 70 tok/s | 3,450 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-20b | Q4_K - Medium | llama.cpp b8580 CUDA | 62 tok/s | 2,616 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| gpt-oss-20b | MXFP4 MoE | llama.cpp b7941 CUDA | 62 tok/s | 2,689 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| gpt-oss-20b | MXFP4 | llama.cpp b7070 | 61 tok/s | 474 tok/s2,048-token prompt | 16,384 | this machine | owner report2025-12 |
| gpt-oss-20b | MXFP4 | llama.cpp b7070 | 48 tok/s | 271 tok/s2,048-token prompt | 32,768 | this machine | owner report2025-12 |
| Nemotron 3 Super 120B-A12B | MXFP4 MoE | llama.cpp b8580 CUDA | 17 tok/s | 721 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Nemotron 3 Super 120B-A12B | MXFP4 MoE | llama.cpp b8580 CUDA | 16 tok/s | 714 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Nemotron 3 Super 120B-A12B | MXFP4 MoE | llama.cpp b8580 CUDA | 16 tok/s | 696 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3 Coder 30B-A3B Instruct | Q8_0 | llama.cpp b7941 CUDA | 48 tok/s | 2,354 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| Qwen3 Coder 30B-A3B Instruct | Q8_0 | llama.cpp b7941 CUDA | 40 tok/s | 1,909 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| Qwen3 Coder 30B-A3B Instruct | Q8_0 | llama.cpp (build not stated) | 35 tok/s | 1,631 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | owner report2025-12 |
| Qwen3 Coder 30B-A3B Instruct | Q8_0 | llama.cpp b7941 CUDA | 30 tok/s | 1,348 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-02 |
| Qwen3 Coder 30B-A3B Instruct | Q8_0 | llama.cpp (build not stated) | 26 tok/s | 1,169 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | owner report2025-12 |
| Qwen3 Coder Next | MXFP4 MoE | llama.cpp b8580 CUDA | 47 tok/s | 1,678 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3 Coder Next | MXFP4 MoE | llama.cpp b8580 CUDA | 45 tok/s | 1,607 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3 Coder Next | MXFP4 MoE | llama.cpp b8580 CUDA | 41 tok/s | 1,493 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3.5 122B-A10B | MXFP4_MOE (unsloth) | llama.cpp b8580 CUDA | 21 tok/s | 1,026 tok/s2,048-token prompt | 8,192 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3.5 122B-A10B | MXFP4_MOE (unsloth) | llama.cpp b8580 CUDA | 20 tok/s | 976 tok/s2,048-token prompt | 16,384 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
| Qwen3.5 122B-A10B | MXFP4_MOE (unsloth) | llama.cpp b8580 CUDA | 20 tok/s | 897 tok/s2,048-token prompt | 32,768 | NVIDIA DGX Spark (same chip) | llama.cpp benchmark discussion2026-03 |
Specifications
- Chip
- GB10 Grace Blackwell
- Memory
- LPDDR5X, 256-bit, coherent unified
- Memory bandwidth
- 273 GB/s
- FP16 compute
- NVIDIA quotes up to 1 PFLOP FP4 with sparsity only; no dense FP16 figure
- Software
- CUDA, Vulkan
- Ships with
- NVIDIA DGX OS 7
- Power
- 280 W (280W USB-C AC adapter (Dell configurator))
- Form factor
- Mini PC
- Networking
- 10GbE; ConnectX-7 SuperNIC (200Gbps per NVIDIA GB10 spec)
- Maker page
- dell.com
Sources
- dell.com: name, $5,903 price, 1TB -$380 option, 128GB LPDDR5x up to 8533 MT/s, ConnectX-7 SuperNIC, DGX OS 7, 280W USB-C adapter
- nvidia.com: GB10 specs: 20-core Arm CPU, 128GB LPDDR5x 256-bit, 273 GB/s, 1 PFLOP FP4 (sparse), 10GbE + ConnectX-7 200Gbps, 240W PSU, 140W GB10 TDP, DGX OS, cluster up to 4; 64GB config only via OEMs
Compared with
Similar machines
- NVIDIA DGX Spark~$4,699
- Apple Mac Studio (M5 Ultra, 64-core GPU)~$5,499–$9,499
- 2x Radeon RX 7900 XTX~$5,637