← All machines

Single GPU · DIY build

1x RTX PRO 6000 Blackwell

96 GB on one card, so 70B dense at 8-bit or ~120B-class MoE models at 4-bit with no multi-GPU split. The GPU price has nearly doubled since launch.

Memory for models

96 GB

VRAM

Memory bandwidth

1,792 GB/s

Parts

~$20,833

sum of parts

Largest at 4-bit

Nemotron 3 Super 120B-A12B

82.6 GB at 8,192 tokens of context

gpt-oss-20b, 4-bit

287 tok/smedian of 3 runs

Parts

PartQtyEach
GPUNVIDIA RTX PRO 6000 Blackwell Workstation Edition1~$16,000
CPUAMD Ryzen 9 9950X (16-core, 170 W)1~$649
MotherboardASUS TUF GAMING X870E-PLUS WIFI71~$325
RAMCorsair Vengeance 128GB (2x64GB) DDR5-6400 CL42 + HX1200i 1200 W PSU bundle1~$3,084
PSUCorsair HX1200i (1200 W), included in the RAM bundle above1~$0
CaseCorsair 7000D AIRFLOW full tower (450 mm GPU clearance, 8 slots)1~$290
StorageCorsair MP600 CORE XT 2TB PCIe 4.0 NVMe1~$485
Total$20,833

New parts are priced at the maker’s own store or list price. Used cards have no price here yet, and a total with a missing GPU would understate the build, so none is shown.

What it runs at 4-bit

Every model in the catalog at its 4-bit file with 8,192 tokens of context. 29 of 41 fit. Speeds are the median of measured runs on the same GPUs 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 countedDoes not fit
Qwen3.5 397B-A17B397B, 17B active · Q4_K_M229.1 GBcontext memory estimatedDoes not fit
GLM-5.3 Flash320B, 18B active · UD-Q4_K_XL187.6 GBcontext memory estimatedDoes not fit
Hy3295B, 21B active · Q4_K_M173.6 GBDoes not fit
MiniMax M2.7228.7B · Q4_K_M132.7 GBDoes not fit
Qwen3.8 Flash-Next125B, 6B active · Q4_K_M113.1 GBcontext memory estimatedDoes not fit
Nemotron 3 Super 120B-A12B120B, 12B active · Q4_K_M82.6 GBcontext memory estimatedFits~51–75 tok/sestimate
Qwen3.5 122B-A10B122B, 10B active · Q4_K_M73.0 GBcontext memory estimatedFits98 tok/smedian of 1 run
Mistral Medium 3.5 128B128B · Q4_K_M74.0 GBFits~12–18 tok/sestimate
GLM-4.5-Air106B, 12B active · Q4_K_M70.9 GBFits101 tok/smedian of 1 run
Mistral Small 4 119B (2603)119B, 6.5B active · Q4_K_M69.3 GBcontext memory estimatedFits~112–164 tok/sestimate
Llama 4 Scout 17B-16E Instruct109B, 17B active · Q4_K_M62.4 GBcontext memory not countedFits~44–64 tok/sestimate
gpt-oss-120b117B, 5.1B active · MXFP4 (native)60.2 GBcontext memory estimatedFits202 tok/smedian of 4 runs
Qwen3 Coder Next80B, 3B active · Q4_K_M46.9 GBcontext memory estimatedFits91 tok/smedian of 1 run
Llama 3.3 70B Instruct70B · Q4_K_M43.6 GBFits32 tok/smedian of 1 run
Nemotron 3.5 Lightning 30B-A3B30B, 3B active · Q4_K_M25.3 GBcontext memory estimatedFits~175–256 tok/sestimate
Qwen3.6 35B-A3B35B, 3B active · Q4_K_M22.4 GBcontext memory estimatedFits~233–341 tok/sestimate
Ornith 1.5 35B-A3B35B, 3B active · Q4_K_M21.9 GBcontext memory estimatedFits~240–350 tok/sestimate
Qwen3 Coder 30B-A3B Instruct30.5B, 3.3B active · Q4_K_M19.5 GBFits~222–324 tok/sestimate
Gemma 4 31B IT30.7B · Q4_K_M18.9 GBcontext memory estimatedFits62 tok/smedian of 1 run
Granite 4.2 30B29.3B · Q4_K_M20.3 GBFits~49–77 tok/sestimate
Qwen3.8 27B27B · Q4_K_M18.2 GBcontext memory estimatedFits~50–79 tok/sestimate
Gemma 4 26B-A4B IT25.2B, 3.8B active · Q4_K_M17.4 GBcontext memory estimatedFits197 tok/smedian of 1 run
Qwen3.6 27B27B · Q4_K_M17.7 GBcontext memory estimatedFits~52–82 tok/sestimate
Devstral Small 2 24B (2512)24B · Q4_K_M16.1 GBFits~61–96 tok/sestimate
gpt-oss-20b21B, 3.6B active · MXFP4 (native)12.5 GBcontext memory estimatedFits287 tok/smedian of 3 runs
Phi-414.7B · Q4_K_M11.3 GBFits~99–155 tok/sestimate
Ministral 3 14B Reasoning (2512)13.9B · Q4_K_M10.4 GBFits~107–167 tok/sestimate
Gemma 4 12B IT11.95B · Q4_K_M8.2 GBcontext memory estimatedFits~123–194 tok/sestimate
Qwen3.5 9B9B · Q4_K_M7.0 GBcontext memory estimatedFits~155–243 tok/sestimate
Granite 4.2 8B8.8B · Q4_K_M7.9 GBFits~159–249 tok/sestimate
Gemma 4 E4B IT8B · Q4_K_M6.5 GBcontext memory not countedFits~289–454 tok/sestimate
Qwen3 8B8.2B · Q4_K_M7.3 GBFits174 tok/smedian of 1 run
Llama 3.1 8B Instruct8B · Q4_K_M7.1 GBFits144 tok/smedian of 2 runs
Qwen3.5 4B4B · Q4_K_M4.3 GBcontext memory estimatedFits~321–503 tok/sestimate

Measured speeds

23 published runs with one user, plain decoding and at most 4,096 tokens already in context, on RTX PRO 6000 Blackwell Workstation Edition. Runs on Dell Pro Max Tower T2 with NVIDIA RTX PRO 6000 Blackwell with the same GPUs count here too, and say so.

Plain single-user runs
ModelQuantBackendGenerationPromptContextMeasured onSource
Gemma 4 26B-A4B ITQ4_Kllama.cpp (build not stated)197 tok/s9,437 tok/s4,096-token prompt4,096RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Gemma 4 31B ITQ4_Kllama.cpp (build not stated)62 tok/s3,750 tok/s4,096-token prompt4,096RTX PRO 6000 Blackwell Workstation Editionreview2026-03
GLM-4.5-AirQ4_Kllama.cpp (build not stated)101 tok/s2,795 tok/s4,096-token prompt4,096RTX PRO 6000 Blackwell Workstation Editionreview2026-03
gpt-oss-120bMXFP4llama.cpp (build not stated)210 tok/s4,579 tok/s4,096-token prompt4,096RTX PRO 6000 Blackwell Workstation Editionreview2026-03
gpt-oss-120bMXFP4SGLang (lmsysorg/sglang:spark image)208 tok/s949 tok/s2,048-token prompt0RTX PRO 6000 Blackwell Workstation Editionreview2025-10
gpt-oss-120bMXFP4 MoEllama.cpp b6205 CUDA196 tok/s5,518 tok/s2,048-token prompt0RTX PRO 6000 Blackwell Workstation Editionllama.cpp benchmark discussion2025-08
gpt-oss-120bMXFP4Ollama (version not stated)153 tok/s3,409 tok/s0RTX PRO 6000 Blackwell Workstation Editionreview2025-10
gpt-oss-120bQ8_0llama.cpp b8288221 tok/s6,513 tok/s4,096-token prompt4,096RTX PRO 6000 Blackwell Workstation Editionreview2026-03
gpt-oss-20bMXFP4SGLang (lmsysorg/sglang:spark image)300 tok/s2,481 tok/s2,048-token prompt0RTX PRO 6000 Blackwell Workstation Editionreview2025-10
gpt-oss-20bMXFP4 MoEllama.cpp b6205 CUDA287 tok/s11,522 tok/s2,048-token prompt0RTX PRO 6000 Blackwell Workstation Editionllama.cpp benchmark discussion2025-08
gpt-oss-20bMXFP4Ollama (version not stated)215 tok/s10,108 tok/s0RTX PRO 6000 Blackwell Workstation Editionreview2025-10
Llama 3.1 8B InstructQ4_K_MOllama (version not stated)202 tok/s38,864 tok/s0RTX PRO 6000 Blackwell Workstation Editionreview2025-10
Llama 3.1 8B InstructQ4_K_MLocalScore (llamafile; version not stated on leaderboard)86 tok/s6,346 tok/s0RTX PRO 6000 Blackwell Workstation EditionLocalScore2026-10
Llama 3.1 8B InstructQ8_0llama.cpp b7278165 tok/s14,040 tok/s512-token prompt0RTX PRO 6000 Blackwell Workstation Editionowner report2025-12
Llama 3.1 8B InstructQ8_0llama.cpp b7278156 tok/s10,093 tok/s512-token prompt0RTX PRO 6000 Blackwell Workstation Editionowner report2025-12
Llama 3.1 8B InstructFP8SGLang (lmsysorg/sglang:spark image)144 tok/s38,744 tok/s2,048-token prompt0RTX PRO 6000 Blackwell Workstation Editionreview2025-10
Llama 3.1 8B InstructQ8_0Ollama (version not stated)144 tok/s40,037 tok/s0RTX PRO 6000 Blackwell Workstation Editionreview2025-10
Llama 3.3 70B InstructQ4_Kllama.cpp (build not stated)32 tok/s1,689 tok/s4,096-token prompt4,096RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Llama 3.3 70B InstructQ8_0llama.cpp b727821 tok/s1,735 tok/s512-token prompt0RTX PRO 6000 Blackwell Workstation Editionowner report2025-12
Llama 3.3 70B InstructQ8_0llama.cpp b727819 tok/s907 tok/s512-token prompt0RTX PRO 6000 Blackwell Workstation Editionowner report2025-12
Qwen3 8BQ4_Kllama.cpp (build not stated)174 tok/s10,964 tok/s4,096-token prompt4,096RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Qwen3 Coder NextINT4 experts / HQQ4 attention / k4v4 KVKrasis (MoE runtime; version not stated in table)91 tok/s11,211 tok/s0RTX PRO 6000 Blackwell Workstation Editionindependent benchmark2026-10
Qwen3.5 122B-A10BQ4_Kllama.cpp b828898 tok/s3,055 tok/s4,096-token prompt4,096RTX PRO 6000 Blackwell Workstation Editionreview2026-03

Deep into a long context

Measured with more than 4,096 tokens already in context, where every new token also reads the context memory.

Runs at long context
ModelQuantBackendGenerationPromptContextMeasured onSource
Gemma 4 26B-A4B ITQ4_Kllama.cpp (build not stated)173 tok/s8,454 tok/s16,384-token prompt16,384RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Gemma 4 26B-A4B ITQ4_Kllama.cpp (build not stated)170 tok/s7,108 tok/s32,768-token prompt32,768RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Gemma 4 31B ITQ4_Kllama.cpp (build not stated)60 tok/s3,062 tok/s16,384-token prompt16,384RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Gemma 4 31B ITQ4_Kllama.cpp (build not stated)56 tok/s2,087 tok/s32,768-token prompt32,768RTX PRO 6000 Blackwell Workstation Editionreview2026-03
GLM-4.5-AirQ4_Kllama.cpp (build not stated)83 tok/s2,107 tok/s16,384-token prompt16,384RTX PRO 6000 Blackwell Workstation Editionreview2026-03
GLM-4.5-AirQ4_Kllama.cpp (build not stated)64 tok/s1,450 tok/s32,768-token prompt32,768RTX PRO 6000 Blackwell Workstation Editionreview2026-03
gpt-oss-120bMXFP4llama.cpp (build not stated)182 tok/s4,061 tok/s16,384-token prompt16,384RTX PRO 6000 Blackwell Workstation Editionreview2026-03
gpt-oss-120bMXFP4llama.cpp (build not stated)162 tok/s3,368 tok/s32,768-token prompt32,768RTX PRO 6000 Blackwell Workstation Editionreview2026-03
gpt-oss-120bQ8_0llama.cpp b8288193 tok/s5,722 tok/s16,384-token prompt16,384RTX PRO 6000 Blackwell Workstation Editionreview2026-03
gpt-oss-120bQ8_0llama.cpp b8288179 tok/s5,000 tok/s32,768-token prompt32,768RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Llama 3.3 70B InstructQ4_Kllama.cpp (build not stated)28 tok/s1,355 tok/s16,384-token prompt16,384RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Llama 3.3 70B InstructQ4_Kllama.cpp (build not stated)26 tok/s1,008 tok/s32,768-token prompt32,768RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Qwen3 8BQ4_Kllama.cpp (build not stated)141 tok/s7,588 tok/s16,384-token prompt16,384RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Qwen3 8BQ4_Kllama.cpp (build not stated)111 tok/s5,301 tok/s32,768-token prompt32,768RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Qwen3.5 122B-A10BQ4_Kllama.cpp b828894 tok/s2,836 tok/s16,384-token prompt16,384RTX PRO 6000 Blackwell Workstation Editionreview2026-03
Qwen3.5 122B-A10BQ4_Kllama.cpp b828891 tok/s2,583 tok/s32,768-token prompt32,768RTX PRO 6000 Blackwell Workstation Editionreview2026-03

Specifications

GPUs
1 × NVIDIA RTX PRO 6000 Blackwell Workstation Edition
VRAM
96 GB
System RAM
128 GB
RAM bandwidth
2 channels x 6400 MT/s x 8 bytes = 102.4 GB/s (dual-channel AM5)
Power supply
1200 W
Estimated load
RTX PRO 6000 600 W + Ryzen 9 9950X 170 W + 100 W = 870 W
Software
CUDA, Vulkan

Before you build it

  • DDR5 prices are at shortage highs: Corsair lists a 64 GB (2x32) DDR5-6000 kit at $1,274.99, so RAM can cost more than a used GPU.
  • The Workstation Edition's 600 W double-flow-through cooler exhausts into the case; NVIDIA's 300 W Max-Q edition is the version meant for stacking several cards.

Sources

Compared with

Similar machines