readygpu

NVIDIA DGX Spark 128GB vs NVIDIA RTX PRO 5000 Blackwell 48GB

For running models locally, priced in the UK. At 32k context the first runs 79 models fully in memory, the second 65.

In short

The NVIDIA DGX Spark 128GB has more memory for models: 128 GB against 48 GB. At 32k context the NVIDIA DGX Spark 128GB runs 79 of the models we track fully in memory and the NVIDIA RTX PRO 5000 Blackwell 48GB 65; only the NVIDIA DGX Spark 128GB holds DeepSeek V4 Flash. On the models both run fully in memory, the NVIDIA RTX PRO 5000 Blackwell 48GB is about 4.9× as fast as the NVIDIA DGX Spark 128GB (estimated, median across models). In the UK the NVIDIA DGX Spark 128GB costs £5,400 at eBay and the NVIDIA RTX PRO 5000 Blackwell 48GB £5,087 at eBay: the NVIDIA RTX PRO 5000 Blackwell 48GB is £313 cheaper.

Updated . Speeds are estimates; how they are worked out.

NVIDIA DGX Spark 128GB

£5,400

eBay

Add to build
NVIDIA RTX PRO 5000 Blackwell 48GB

£5,087

eBay

Add to build

Specifications

SpecNVIDIA DGX Spark 128GBNVIDIA RTX PRO 5000 Blackwell 48GB
ChipGB10 Grace Blackwell–
Memory bus256-bit384-bit
Memory bandwidth273 GB/s1,344 GB/s
FP16 compute62 TFLOPS73.69 TFLOPS
SoftwareCUDACUDA
Unified memory128 GB–
GPU can use by default119 GB–
GPU can use, tuned122 GB–
Operating systemdgx-os–
Processor20-core Arm (GB10)–
Storage4000 GB–
GPU–RTX PRO 5000 Blackwell 48GB
Memory–48 GB GDDR7
Board power–300 W
PCIe generation–Gen 5
PCIe lanes–x16

What each runs

Estimated decode speed with the most precise quant that fits each one est.

ModelNVIDIA DGX Spark 128GBNVIDIA RTX PRO 5000 Blackwell 48GB
DeepSeek V4 Flash11tok/sUD-Q2_K_XLwon't run
GLM 5.3 Flash11tok/sUD-Q2_K_XLwon't run
Qwen 3.8 Flash Next2tok/sQ4_K_Mwon't run
MiniMax M2.78tok/sUD-Q3_K_Mwon't run
Qwen 3.8 27B5tok/sQ8_027tok/sQ8_0
Qwen 3.5 122B-A10B11tok/sQ6_K116tok/sUD-Q2_K_XL
Qwen 3.6 27B6tok/sQ8_027tok/sQ8_0
gpt-oss 120B23tok/sMXFP4offload
Qwen3 Coder Next20tok/sQ8_0154tok/sQ4_K_M
Nemotron 3 Super 120B-A12B9tok/sUD-Q6_Koffload
Gemma 4 31B5tok/sQ8_023tok/sQ8_0
Mistral Small 4 119B15tok/sUD-Q6_K160tok/sUD-Q2_K_XL
Qwen 3.6 35B-A3B27tok/sQ8_0131tok/sQ8_0
Qwen 3.5 27B6tok/sQ8_027tok/sQ8_0
Gemma 4 26B-A4B20tok/sQ8_099tok/sQ8_0
GLM 4.5 Air5tok/sQ8_0offload
Qwen 3.5 35B-A3B27tok/sQ8_0131tok/sQ8_0
Qwen3 Next 80B-A3B25tok/sQ8_0187tok/sQ4_K_M
GLM 4.7 Flash14tok/sQ8_071tok/sQ8_0
Devstral Small 2 24B6tok/sQ8_027tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B28tok/sUD-Q6_K_XL140tok/sUD-Q6_K_XL
gpt-oss 20B31tok/sMXFP4155tok/sMXFP4
Kimi Linear 48B-A3B29tok/sQ8_0180tok/sQ6_K
Qwen3 Coder 30B-A3B16tok/sQ8_078tok/sQ8_0
Qwen3 32B4tok/sQ8_019tok/sQ8_0
Gemma 4 12B13tok/sQ8_062tok/sQ8_0
Qwen 3.5 9B16tok/sQ8_078tok/sQ8_0
Mistral Small 3.2 24B6tok/sQ8_027tok/sQ8_0
Gemma 3 27B5tok/sQ8_026tok/sQ8_0
DeepSeek R1 Distill Qwen 32B4tok/sQ8_019tok/sQ8_0
Llama 3.3 70B2tok/sQ8_0offload
Ministral 3 14B9tok/sQ8_042tok/sQ8_0
Qwen3 14B8tok/sQ8_040tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B13tok/sQ8_062tok/sQ8_0
Qwen3 8B13tok/sQ8_062tok/sQ8_0
Phi-4 14B8tok/sQ8_037tok/sQ8_0
Qwen 3.5 4B30tok/sQ8_0149tok/sQ8_0
Ministral 3 8B13tok/sQ8_061tok/sQ8_0
Granite 4.2 8B12tok/sQ8_057tok/sQ8_0
Gemma 4 E4B19tok/sQ8_091tok/sQ8_0
LFM2.5 8B-A1B38tok/sQ8_0187tok/sQ8_0
Llama 3.1 8B16tok/sQ6_K77tok/sQ6_K
Qwen2.5 0.5B181tok/sQ8_0892tok/sQ8_0
Qwen3.5 0.8B137tok/sQ8_0676tok/sQ8_0
Qwen3.5 2B91tok/sQ5_K_M449tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B60tok/sQ8_0294tok/sQ8_0
MiniCPM5 2B41tok/sQ8_0204tok/sQ8_0
Spark X2.5 4B39tok/smlx-4bit194tok/smlx-4bit
LFM2.5 2.6B35tok/sQ8_0171tok/sQ8_0
LFM2.5 VL 3B35tok/sQ8_0171tok/sQ8_0
Mellum2.1 12B A2.5B Thinking32tok/sQ8_0158tok/sQ8_0
FrogNano 4B 260930tok/sQ8_0145tok/sQ8_0
granite vision 4.1 4b27tok/sQ8_0132tok/sQ8_0
granite 4.2 3b26tok/sQ8_0127tok/sQ8_0
Kimi VL A3B Thinking 250625tok/sQ8_0121tok/sQ8_0
Kimi VL A3B24tok/smlx-8bit117tok/smlx-8bit
Ministral 3 3B Instruct 251224tok/sQ8_0117tok/sQ8_0
LFM2.5 24B A2B21tok/sQ8_0104tok/sQ8_0
GLM 4 9B 041418tok/smlx-5bit87tok/smlx-5bit
DeepSeek R1 Distill Qwen 7B17tok/sQ8_084tok/sQ8_0
GLM 4.6V Flash15tok/sQ8_074tok/sQ8_0
DeepSeek R1 Distill Llama 8B13tok/sQ8_065tok/sQ8_0
GLM 4.1V 9B Thinking13tok/smlx-8bit64tok/smlx-8bit
Ministral 3 8B Reasoning 251213tok/sQ8_061tok/sQ8_0
Ministral 3 14B Reasoning 25129tok/sQ8_042tok/sQ8_0
DeepSeek R1 Distill Qwen 14B8tok/sQ8_038tok/sQ8_0
Qwen3.8 27B OBLITERATED7tok/smlx-4bit36tok/smlx-4bit
Devstral Small 25076tok/sQ8_027tok/sQ8_0
Magistral Small 25066tok/sQ8_027tok/sQ8_0
Magistral Small 25076tok/sQ8_027tok/sQ8_0
Magistral Small 25096tok/sQ8_027tok/sQ8_0
Muse Glimmer 30B5tok/sQ8_025tok/sQ8_0
GLM 4.5V5tok/sQ8_0offload
GLM 4.6V5tok/sQ8_0offload
granite 4.2 30b4tok/sQ8_021tok/sQ8_0
DeepSeek R1 Distill Llama 70B2tok/sQ8_0offload
Kimi Dev 72B2tok/sQ8_0offload
Devstral 2 123B Instruct 25122tok/sQ6_Kwon't run
Mistral Medium 3.5 128B2tok/sQ6_Kwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Zerowon't runwon't run
DeepSeek V3won't runwon't run
DeepSeek V3 0324won't runwon't run
DeepSeek V3.1won't runwon't run
DeepSeek V3.1 Terminuswon't runwon't run
DeepSeek V3.2won't runwon't run
DeepSeek V4 Flash 0731won't runwon't run
DeepSeek V4 Flash Vision Expwon't runwon't run
GLM 4.5won't runwon't run
GLM 4.6won't runwon't run
GLM 4.7won't runwon't run
GLM 5won't runwon't run
GLM 5.1won't runwon't run
GLM 5.2won't runwon't run
GLM 5.3won't runwon't run
Kimi K2won't runwon't run
Kimi K2 Instruct 0905won't runwon't run
Kimi K2 Thinkingwon't runwon't run
Kimi K2.5won't runwon't run
MiniMax M2won't runwon't run
MiniMax M2.1won't runwon't run
MiniMax M2.5won't runwon't run
MiniMax M3won't runwon't run

Questions

Which is better for local AI, the NVIDIA DGX Spark 128GB or the NVIDIA RTX PRO 5000 Blackwell 48GB?
Choose the NVIDIA DGX Spark 128GB to run bigger models. Choose the NVIDIA RTX PRO 5000 Blackwell 48GB for speed. Choose the NVIDIA RTX PRO 5000 Blackwell 48GB to spend less.
Which can run bigger models, the NVIDIA DGX Spark 128GB or the NVIDIA RTX PRO 5000 Blackwell 48GB?
At 32k context the NVIDIA DGX Spark 128GB runs 79 of the models we track fully in memory and the NVIDIA RTX PRO 5000 Blackwell 48GB 65; only the NVIDIA DGX Spark 128GB holds DeepSeek V4 Flash.
Which is faster for local AI, the NVIDIA DGX Spark 128GB or the NVIDIA RTX PRO 5000 Blackwell 48GB?
On the models both run fully in memory, the NVIDIA RTX PRO 5000 Blackwell 48GB is about 4.9× as fast as the NVIDIA DGX Spark 128GB (estimated, median across models).
What is the price difference between the NVIDIA DGX Spark 128GB and the NVIDIA RTX PRO 5000 Blackwell 48GB?
In the UK the NVIDIA DGX Spark 128GB costs £5,400 at eBay and the NVIDIA RTX PRO 5000 Blackwell 48GB £5,087 at eBay: the NVIDIA RTX PRO 5000 Blackwell 48GB is £313 cheaper.

More comparisons

More hardware for local AI