{"slug": "nvidia-rtx-pro-5500-blackwell-84gb", "title": "NVIDIA RTX PRO 5500 Blackwell 84GB", "summary": "NVIDIA listed the RTX PRO 5500 Blackwell, an 84GB GDDR7 ECC workstation card with 21,760 CUDA cores, 1,398 GB/s of bandwidth, 600W power draw and PCIe Gen 5, as \"Coming Soon\" on nvidia.com in September 2026 with preliminary specs and no published MSRP. The card sits between the RTX PRO 5000 and the $8,565 RTX PRO 6000, offering 2.6x the RTX 5090's memory at the same core count but 22% less bandwidth than both the RTX 5090 and RTX PRO 6000, which run at 1,792 GB/s. Tom's Hardware reads the board as 28 GDDR7 chips clamshelled on a 448-bit bus downclocked from 28 to 25 Gb/s, matching the China-exclusive RTX PRO 6000D's bandwidth, while PNY's sheet lists 416-bit.", "body_md": "# NVIDIA RTX PRO 5500 Blackwell 84GB\n\n[Find compatible models](https://tokenstead.ai/find/results?hardware_id=70&use_case=coding)\n\nNVIDIA slotted a new card between the RTX PRO 5000 and the RTX PRO 6000 in September 2026: 84GB GDDR7 ECC at 1,398 GB/s, 21,760 CUDA cores (the RTX 5090 count), 600W, PCIe Gen 5. Listed as “Coming Soon” on nvidia.com with specs marked preliminary and no MSRP published. The price field here stays empty until NVIDIA names a number.\n\n**What it does well:**\n\n- \n**84GB single-card capacity:** 70B dense at 8-bit, 235B-class MoE\nat 4-bit, with ECC GDDR7 and full CUDA. MIG can split it into two\nisolated 42GB instances.\n- \n**2.6x the RTX 5090’s memory at the same core count:** the 32GB\nconsumer flagship’s compute with the headroom it lacks - at\n1,398 GB/s instead of 1,792.\n\n**Where it falls short:** the bandwidth cut. 1,398 GB/s is 22% under\nboth the RTX 5090 and the RTX PRO 6000 (1,792 GB/s). Same core count\nas the 5090, but LLM decode is memory-bandwidth-bound, so tok/s\ntracks the bus, not the cores. Tom’s Hardware reads the board as 28\nGDDR7 chips clamshelled on a 448-bit bus, downclocked from 28 to\n25 Gb/s - the same bandwidth as the China-exclusive RTX PRO 6000D,\nthe only other 84GB card NVIDIA has shipped. (PNY’s sheet lists\n416-bit, which cannot hold 84GB clamshelled, so treat Tom’s as the\nworking read until NVIDIA publishes.) No price yet: the RTX PRO 6000\nabove it has 12GB more memory, 1,792 GB/s, and 24,064 cores at\n$8,565 MSRP, so the 5500 has to undercut that by enough to matter.\nSpecs are preliminary.\n\n**The fit case, from a Spark owner:** “it is the first single card\nthat holds my entire single spark board of open model” - the\n100B-class checkpoints @sudoingX runs (Laguna at 67GB, Ling int4 at\n72GB, Qwen 3.5 122B at 74GB) all fit with context to spare, on a\nbus 5x the Spark’s 273 GB/s. His caveat stands: “the rest is a\nbenchmark nobody has run.” And the 180B-at-FP8 and 1M-context\nbuilds still need 96GB or more, which keeps them on the desk boxes.\n\n**Run it locally:** full CUDA stack - llama.cpp, Ollama, vLLM. The\n[modeldex results](https://tokenstead.ai/find/results?hardware_id=70&use_case=coding) rank checkpoints by fit\nand estimated tok/s.\n\n- Nvidia RTX PRO 5500 Blackwell\n- 84GB vram\n- 1398 GB/s\n- 2026\n- Not specified\n\n## Top models (91 compatible)\n\n[See all 91 compatible models →](https://tokenstead.ai/find/results?hardware_id=70&use_case=coding)", "url": "https://wpnews.pro/news/nvidia-rtx-pro-5500-blackwell-84gb", "canonical_source": "https://tokenstead.ai/hardware/70", "published_at": "2026-09-19 13:36:29+00:00", "updated_at": "2026-09-19 13:52:48.449488+00:00", "lang": "en", "topics": ["ai-chips", "ai-infrastructure", "large-language-models", "ai-products"], "entities": ["NVIDIA", "RTX PRO 5500 Blackwell", "RTX PRO 6000", "RTX 5090", "RTX PRO 6000D", "Tom's Hardware", "PNY", "RTX PRO 5000"], "alternates": {"html": "https://wpnews.pro/news/nvidia-rtx-pro-5500-blackwell-84gb", "markdown": "https://wpnews.pro/news/nvidia-rtx-pro-5500-blackwell-84gb.md", "text": "https://wpnews.pro/news/nvidia-rtx-pro-5500-blackwell-84gb.txt", "jsonld": "https://wpnews.pro/news/nvidia-rtx-pro-5500-blackwell-84gb.jsonld"}}