← All machines

NVIDIA workstation · Dell

Dell Pro Precision with GB300

Dell's DGX Station-class tower with NVIDIA's GB300 superchip, 252GB of HBM3e on the GPU and 496GB of CPU memory. It is priced for labs and companies that want frontier-size open models on a desk rather than a rack.

Memory for models

252 GB

of 748 GB in the largest config

Memory bandwidth

7,100 GB/s

Maker price

~$159,283

Largest at 4-bit

MiniMax M3

244.8 GB at 8,192 tokens of context

gpt-oss-20b, 4-bit

~1,678–2,143 tok/sestimate

no measurement on this hardware yet

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.

MemoryPriceMemory the GPU can use
748 GB~$159,283
252 GB gpu-usable memory

252GB HBM3e on the GPU (NVIDIA spec). The 496GB LPDDR5X attached to the CPU is shared with the GPU over NVLink-C2C but is far slower than the HBM, so models are sized to the HBM for full speed. 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. 35 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.

ModelNeedsFitGeneration speed
Kimi K32.8T, 104B active · UD-Q4_K_XL1406.8 GBcontext memory estimatedDoes not fit
DeepSeek V4 Pro1.6T, 49B active · Q4_K_M890.5 GBcontext memory not countedDoes not fit
Kimi K2.61T, 32B active · UD-Q4_K_XL545.7 GBcontext memory estimatedDoes not fit
DeepSeek V4.1 Flash552B, 16B active · Q4 (DwarfStar)484.5 GBDoes not fit
GLM-5.3753.3B · UD-Q4_K_XL437.4 GBcontext memory estimatedDoes not fit
DeepSeek V3.2671B, 37B active · Q4_K_M379.6 GBcontext memory estimatedDoes not fit
MiniMax M3428B, 23B active · Q4_K_M244.8 GBcontext memory not countedFits~238–304 tok/sestimate
Qwen3.5 397B-A17B397B, 17B active · Q4_K_M229.1 GBcontext memory estimatedFits~320–409 tok/sestimate
GLM-5.3 Flash320B, 18B active · UD-Q4_K_XL187.6 GBcontext memory estimatedFits~298–380 tok/sestimate
Hy3295B, 21B active · Q4_K_M173.6 GBFits~258–329 tok/sestimate
MiniMax M2.7228.7B · Q4_K_M132.7 GBFits–
Qwen3.8 Flash-Next125B, 6B active · Q4_K_M113.1 GBcontext memory estimatedFits~583–744 tok/sestimate
Nemotron 3 Super 120B-A12B120B, 12B active · Q4_K_M82.6 GBcontext memory estimatedFits~385–491 tok/sestimate
Qwen3.5 122B-A10B122B, 10B active · Q4_K_M73.0 GBcontext memory estimatedFits~533–681 tok/sestimate
Mistral Medium 3.5 128B128B · Q4_K_M74.0 GBFits~64–77 tok/sestimate
GLM-4.5-Air106B, 12B active · Q4_K_M70.9 GBFits~405–517 tok/sestimate
Mistral Small 4 119B (2603)119B, 6.5B active · Q4_K_M69.3 GBcontext memory estimatedFits~843–1,076 tok/sestimate
Llama 4 Scout 17B-16E Instruct109B, 17B active · Q4_K_M62.4 GBcontext memory not countedFits~328–419 tok/sestimate
gpt-oss-120b117B, 5.1B active · MXFP4 (native)60.2 GBcontext memory estimatedFits~1,222–1,561 tok/sestimate
Qwen3 Coder Next80B, 3B active · Q4_K_M46.9 GBcontext memory estimatedFits~1,838–2,347 tok/sestimate
Llama 3.3 70B Instruct70B · Q4_K_M43.6 GBFits~113–136 tok/sestimate
Nemotron 3.5 Lightning 30B-A3B30B, 3B active · Q4_K_M25.3 GBcontext memory estimatedFits~1,313–1,676 tok/sestimate
Qwen3.6 35B-A3B35B, 3B active · Q4_K_M22.4 GBcontext memory estimatedFits~1,751–2,236 tok/sestimate
Ornith 1.5 35B-A3B35B, 3B active · Q4_K_M21.9 GBcontext memory estimatedFits~1,797–2,295 tok/sestimate
Qwen3 Coder 30B-A3B Instruct30.5B, 3.3B active · Q4_K_M19.5 GBFits~1,666–2,127 tok/sestimate
Gemma 4 31B IT30.7B · Q4_K_M18.9 GBcontext memory estimatedFits~263–315 tok/sestimate
Granite 4.2 30B29.3B · Q4_K_M20.3 GBFits~267–320 tok/sestimate
Qwen3.8 27B27B · Q4_K_M18.2 GBcontext memory estimatedFits~276–331 tok/sestimate
Gemma 4 26B-A4B IT25.2B, 3.8B active · Q4_K_M17.4 GBcontext memory estimatedFits~1,302–1,663 tok/sestimate
Qwen3.6 27B27B · Q4_K_M17.7 GBcontext memory estimatedFits~287–343 tok/sestimate
Devstral Small 2 24B (2512)24B · Q4_K_M16.1 GBFits~336–403 tok/sestimate
gpt-oss-20b21B, 3.6B active · MXFP4 (native)12.5 GBcontext memory estimatedFits~1,678–2,143 tok/sestimate
Phi-414.7B · Q4_K_M11.3 GBFits~542–650 tok/sestimate
Ministral 3 14B Reasoning (2512)13.9B · Q4_K_M10.4 GBFits~585–701 tok/sestimate
Gemma 4 12B IT11.95B · Q4_K_M8.2 GBcontext memory estimatedFits~677–811 tok/sestimate
Qwen3.5 9B9B · Q4_K_M7.0 GBcontext memory estimatedFits~849–1,017 tok/sestimate
Granite 4.2 8B8.8B · Q4_K_M7.9 GBFits~870–1,043 tok/sestimate
Gemma 4 E4B IT8B · Q4_K_M6.5 GBcontext memory not countedFits~1,585–1,900 tok/sestimate
Qwen3 8B8.2B · Q4_K_M7.3 GBFits~959–1,149 tok/sestimate
Llama 3.1 8B Instruct8B · Q4_K_M7.1 GBFits~980–1,174 tok/sestimate
Qwen3.5 4B4B · Q4_K_M4.3 GBcontext memory estimatedFits~1,759–2,107 tok/sestimate

Measured speeds

Nobody has published a run on this chip with a stated model, quant and backend that we could find. The speeds above are estimates.

Specifications

Chip
GB300 Grace Blackwell Ultra (72-core Grace CPU + Blackwell Ultra GPU)
Memory
252GB HBM3e (GPU) + 496GB LPDDR5X (CPU), coherent
Memory bandwidth
7,100 GB/s
FP16 compute
NVIDIA DGX Station spec: FP16/BF16 Tensor 5 PFLOPS and FP4 20 PFLOPS, both with sparsity; no dense FP16 figure (FP4 15 PFLOPS dense)
Software
CUDA
Ships with
Ubuntu 24.04 LTS with NVIDIA AI Developer Tools
Power
1600 W (DGX Station total system power 1,600W, 20A circuit required (NVIDIA spec))
Form factor
Tower
Networking
ConnectX-8 SuperNIC, 2x QSFP112 (400Gb/s each, up to 800Gb/s), 10GbE
Maker page
dell.com

Sources

  • dell.com: name, $159,283 price, memory, Grace 72-core with ConnectX-8, Ubuntu 24.04 with NVIDIA AI Developer Tools
  • nvidia.com: 252GB HBM3e 7.1 TB/s, 496GB LPDDR5X 396 GB/s, NVLink-C2C 900 GB/s, tensor rates (sparse), 1,600W, networking ports

Similar machines