GPU comparison
RTX PRO 5000 Blackwell (72GB) vs RTX PRO 6000 Blackwell
Compared on the specs that decide a purchase: memory, bandwidth, power, and which workloads each card actually fits — gaming, creative, or local models when that is the job.
Specifications verified September 26, 2026. Launch MSRP is not today's price — in September 2026 the memory shortage pushed street prices 1.2×–3× above it; the notes row below carries a dated lowest UK/US price where we have one.
NVIDIA
RTX PRO 5000 Blackwell (72GB)
72GB VRAM · 1344 GB/s
NVIDIA
RTX PRO 6000 Blackwell
96GB VRAM · 1792 GB/s
Amazon UK search links are provided for current availability; no affiliate programme is currently configured, and no live price is claimed here. Affiliate disclosure.
| Specification | RTX PRO 5000 Blackwell (72GB) | RTX PRO 6000 Blackwell |
|---|---|---|
| Vendor | ||
| Vendor | NVIDIA | NVIDIA |
| Memory | ||
| Memory | 72GB | Winner: 96GB |
| Usable for a model | ||
| Usable for a model | 72GB | Winner: 96GB |
| Bandwidth | ||
| Bandwidth | 1344 GB/s | Winner: 1792 GB/s |
| Segment | ||
| Segment | workstation | workstation |
| Released | ||
| Released | Not verifiedUnverified | March 18, 2025 |
| Launch MSRP | ||
| Launch MSRP | No launch MSRP published | ~$8,565~£6,344 |
| Notes | ||
| Notes | 14,080 CUDA cores, 72GB GDDR7 on a 384-bit bus at 1,344 GB/s in a 300W card (a 48GB version also exists). Fills the gap between 48GB Ada and the 96GB PRO 6000: holds 70B at Q4 with context room. Retail from ~$9,209 (Tom's Hardware, Sep 2026); no launch MSRP is shown. | Workstation Edition: 96GB GDDR7 ECC, ~1.8 TB/s, 600W dual-flow cooler. Launch MSRP stays $8,565 in the DualPrice column. NVIDIA marketplace listed $16,000 (VideoCardz 2026-08-12); Tom's Hardware put retail at $15,599 on 2026-09-14. Treat marketplace/street as the buy number, not launch MSRP. RunPod rents it from $1.69/hr (Community) / $2.09 (Secure) — the rent-first alternative. This is the single-card path for ~70B+ local models that consumer 32GB cards cannot host. |
| UK street price | ||
| UK street price | Search Amazon (opens in new tab) | Search Amazon (opens in new tab) |
Local models
What each card can actually run
Computed from published VRAM and the memory each model needs at Q4, including runtime overhead.
| Model | RTX PRO 5000 Blackwell (72GB) | RTX PRO 6000 Blackwell |
|---|---|---|
| Qwen 3 4B4B parameters | ||
| Qwen 3 4B4B parameters | Fits | Fits |
| Qwen 3.8 27B27B parameters | ||
| Qwen 3.8 27B27B parameters | Fits | Fits |
| Gemma 4 31B31B parameters | ||
| Gemma 4 31B31B parameters | Fits | Fits |
| Llama 3.1 8B Instruct8B parameters | ||
| Llama 3.1 8B Instruct8B parameters | Fits | Fits |
| Qwen 3 14B14B parameters | ||
| Qwen 3 14B14B parameters | Fits | Fits |
| Gemma 3 27B27B parameters | ||
| Gemma 3 27B27B parameters | Fits | Fits |
| Qwen 3 30B-A3B30B parameters | ||
| Qwen 3 30B-A3B30B parameters | Fits | Fits |
| Qwen 3 32B32B parameters | ||
| Qwen 3 32B32B parameters | Fits | Fits |
| Llama 3.3 70B70B parameters | ||
| Llama 3.3 70B70B parameters | Fits | Fits |
| Muse Glimmer 30B30B parameters | ||
| Muse Glimmer 30B30B parameters | Fits | Fits |
| Mixtral 8x22B141B parameters | ||
| Mixtral 8x22B141B parameters | Does not fit | Tight |
| Qwen 3 235B-A22B235B parameters | ||
| Qwen 3 235B-A22B235B parameters | Does not fit | Does not fit |
| Qwen 3 Coder 480B-A35B480B parameters | ||
| Qwen 3 Coder 480B-A35B480B parameters | Does not fit | Does not fit |
| DeepSeek V4-Flash284B parameters | ||
| DeepSeek V4-Flash284B parameters | Does not fit | Does not fit |
Common questions
RTX PRO 5000 Blackwell (72GB) vs RTX PRO 6000 Blackwell
Answered from the verified figures on this page rather than general guidance.
Which is better for high-refresh gaming, RTX PRO 5000 Blackwell (72GB) or RTX PRO 6000 Blackwell?
Which has more VRAM for gaming and creative work, RTX PRO 5000 Blackwell (72GB) or RTX PRO 6000 Blackwell?
For local inference only — which holds larger models, RTX PRO 5000 Blackwell (72GB) or RTX PRO 6000 Blackwell?
For local inference only — is RTX PRO 5000 Blackwell (72GB) or RTX PRO 6000 Blackwell faster at generating tokens?
For local inference only — can either run a 70B model?
If one card holds a model the other cannot, that capacity difference usually outweighs a bandwidth advantage. A model that spills to system memory can be substantially slower, but the actual impact depends on the backend, transfer path, context, and workload.