Head to head · NVIDIA vs NVIDIA
DGX Spark vs RTX PRO 6000 for local LLMs
The RTX PRO 6000 is the faster card: 96 GB of VRAM at 1,792 GB/s, so models that fit run roughly 3.2–6.8× faster than on a DGX Spark (our estimate). The Spark is a complete 128 GB computer from $5,859; the PRO 6000 is $17,899 for the card alone.
How do the specs compare?
| DGX Spark / GB10 | RTX PRO 6000 | |
|---|---|---|
| Price | $5,859cheapest GB10 box (Lenovo ThinkStation PGX), complete system | $17,899card only, needs a host PC |
| Stock | Out of stock | Out of stock |
| Memory | 128 GB unified (~109 GB usable) | 96 GB VRAM (~89 GB usable) |
| Bandwidth | 273 GB/s | 1,792 GB/s |
| Models that fit (4-bit) | 20 / 22 | 20 / 22 |
Which models fit, and how fast (4-bit)?
| Model | DGX Spark | RTX PRO 6000 |
|---|---|---|
| Qwen3.5 9B~7 GB at 4-bit | Fits~39 tok/s | Fits~214 tok/s |
| Gemma 4 12B~9 GB at 4-bit | Fits~29 tok/s | Fits~170 tok/s |
| GPT-OSS 20B~13 GB at 4-bit | Fits~80 tok/s | Fits~264 tok/s |
| Gemma 4 26B-A4B~18 GB at 4-bit | Fits~70 tok/s | Fits~247 tok/s |
| Qwen3.8 27B~19 GB at 4-bit | Fits~13 tok/s | Fits~83 tok/s |
| Gemma 3 27B~19 GB at 4-bit | Fits~13 tok/s | Fits~82 tok/s |
| GLM-4.7-Flash 30B-A3B~21 GB at 4-bit | Fits~83 tok/s | Fits~269 tok/s |
| Qwen3 30B-A3B~21 GB at 4-bit | Fits~77 tok/s | Fits~260 tok/s |
| Gemma 4 31B~21 GB at 4-bit | Fits~11 tok/s | Fits~73 tok/s |
| Qwen3 32B~22 GB at 4-bit | Fits~11 tok/s | Fits~70 tok/s |
| DeepSeek R1 Distill Qwen 32B~22 GB at 4-bit | Fits~11 tok/s | Fits~70 tok/s |
| Qwen3.6 35B-A3B~24 GB at 4-bit | Fits~83 tok/s | Fits~269 tok/s |
| Llama 3.3 70B~46 GB at 4-bit | Fits~5 tok/s | Fits~34 tok/s |
| DeepSeek R1 Distill Llama 70B~46 GB at 4-bit | Fits~5 tok/s | Fits~34 tok/s |
| Qwen3-Next 80B-A3B~52 GB at 4-bit | Fits~83 tok/s | Fits~269 tok/s |
| Llama 4 Scout 109B-A17B~71 GB at 4-bit | Fits~19 tok/s | Fits~106 tok/s |
| GPT-OSS 120B~67 GB at 4-bit | Fits~61 tok/s | Fits~231 tok/s |
| Mistral Small 4 119B-A6B~78 GB at 4-bit | Fits~45 tok/s | Fits~194 tok/s |
| Nemotron 3 Super 120B-A12B~78 GB at 4-bit | Fits~26 tok/s | Fits~135 tok/s |
| Qwen3.5 122B-A10B~79 GB at 4-bit | Fits~31 tok/s | Fits~152 tok/s |
Estimated single-user decode at ~8K context; GPU-card figures assume the model is fully in VRAM. How we estimate
How do measured runs compare with our estimates?
Public single-user decode results we calibrate against. Runs differ in build and settings, so compare like with like.
| Hardware | Model / run | Measured tok/s | Our formula | Source |
|---|---|---|---|---|
| DGX Spark273 GB/s | Qwen2.5-Coder 7B Q8_0llama.cpp tg32 | 29.4 | 26.0 | source |
| DGX Spark273 GB/s | Gemma 3 4B Q4_0llama.cpp tg32 | 81.0 | 83.4 | source |
| DGX Spark273 GB/s | gpt-oss-120b MXFP4llama.cpp tg32 | 58.7 | 61.4 | source |
| DGX Spark273 GB/s | gpt-oss-20b MXFP4llama.cpp tg32 | 83.4 | 80.0 | source |
| DGX Spark273 GB/s | Qwen3-Coder 30B-A3B Q8_0llama.cpp tg32 | 61.1 | 49.8 | source |
| DGX Spark273 GB/s | GLM-4.7-Flash Q8_0llama.cpp tg32 | 48.7 | 53.9 | source |
| DGX Spark273 GB/s | Qwen3.5-122B-A10B Q5_K_Sllama.cpp tg128 | 23.0 | 27.7 | source |
| RTX PRO 60001,792 GB/s | Llama-2 7B Q4_0llama.cpp tg128 | 274.2 | 280.4 | source |
| RTX PRO 60001,792 GB/s | Llama 3.3 70B Q4_K_Mllama.cpp tg128 | 34.0 | 33.8 | source |
| RTX PRO 60001,792 GB/s | gpt-oss-120b MXFP4llama.cpp tg128 | 196.3 | 231.4 | source |
| RTX PRO 60001,792 GB/s | gpt-oss-20b MXFP4llama.cpp tg128 | 286.9 | 264.1 | source |
| RTX PRO 60001,792 GB/s | gpt-oss-120b MXFP4llama.cpp 4K ctx (newer build) | 221.3 | 231.4 | source |
| Dell Pro Max GB10273 GB/s | Llama 3.1 70B Q4_K_MOllama (Geerling) | 4.7 | 4.9 | source |
Which is better value?
On GPT-OSS 120B, the Lenovo ThinkStation PGX costs about $95.44 per estimated tok/s and the PRO 6000 card about $77.35, before you add a host PC for the card. The Spark is the cheaper way to get ~128 GB; the PRO 6000 is the faster one.
What do they cost today?
PNY NVIDIA RTX PRO 6000 Blackwell
96GB card
96 GB · 1,792 GB/s (card only, needs a host PC)
Check priceFrequently asked questions
Is the RTX PRO 6000 faster than DGX Spark?
Yes. It has 1,792 GB/s vs 273 GB/s, so on models both can hold we estimate 3.2–6.8× faster decode. Measured on GPT-OSS 120B: 59 tok/s on the Spark and 196–221 tok/s on the PRO 6000.
Do they fit the same models?
Nearly: ~109 GB usable on the Spark vs ~89 GB on the PRO 6000. Every model that fits one fits the other at 4-bit.
Can both run Llama 70B?
Yes at 4-bit (~46 GB): we estimate ~5 tok/s on the Spark and ~34 tok/s on the PRO 6000.
What does each cost?
The cheapest GB10 box we track is the Lenovo ThinkStation PGX at $5,859, a complete system. The RTX PRO 6000 Blackwell is $17,899 for the card; you still need a workstation to put it in, which we don’t price.