RTX PRO 5500 Blackwell: specs, release status and where the 84 GB card fits

RTX PRO 5500 release date and launch status

NVIDIA put the product page online in early September 2026, with specifications marked “Preliminary” and a “Coming Soon” notice with a “Notify Me” sign-up. As of 5 November 2026 the page shows “Where to Buy” with the line “Offered through NVIDIA OEM and distribution partners”, marks no specification as preliminary and links the datasheet of the “NVIDIA RTX PRO 5500 Blackwell Workstation Edition” (document 5513311, October 2026). NVIDIA states neither a release date nor a price on the page. NVIDIA describes it as “designed for rack-mounted workstation deployment with a choice of air- or liquid-cooled thermal solutions”. NVIDIA’s CUDA GPU list, in its version of 16 October 2026, places the “NVIDIA RTX PRO 5500 Blackwell Workstation Edition” under compute capability 12.0, alongside the RTX PRO 6000 and the other RTX PRO Blackwell cards. CUDA code compiled for that architecture therefore also targets the 5500; the datasheet lists CUDA 12.8 among its compute APIs. When we compared the RTX PRO family on 14 September, the card appeared only in a footnote; this article collects what NVIDIA has published about it. We supply the RTX PRO 5500; it is available to order, with quantities and lead time confirmed in writing with the quote. Describe your workload and our team will help you choose between the 5000, 5500 and 6000.

RTX PRO 5500 specifications NVIDIA has published

NVIDIA RTX PRO 5500 Blackwell Workstation Edition datasheet (5513311, October 2026) and product page, read 5 November 2026; MIG user guide, supported GPUs (updated 4 November 2026); vGPU supported-GPU list (2 October 2026) and CUDA GPU list (16 October 2026). Performance figures are NVIDIA’s peak rates at boost clock.

A few details are still open. NVIDIA defines “RXM” neither on the product page nor in the datasheet, and the card is not yet in the supported-GPU list of NVIDIA’s MIG user guide (update of 4 November 2026), so the MIG profile names are still to come. The datasheet gives peak rates but no clock speeds.

RTX PRO 5500 vs RTX PRO 5000 and 6000

NVIDIA datasheets (April 2025 to October 2026), product pages, MIG user guide and vGPU documentation; RTX PRO 5500 from its datasheet of October 2026, read 5 November 2026. FP32 and AI TOPS are peak rates at boost clock, AI TOPS in FP4 with sparsity. The RTX PRO 5000 details are in our 48 or 72 GB comparison.

Compared with the 72 GB RTX PRO 5000, the 5500 has 16.7 per cent more memory, 4.0 per cent more bandwidth, 62 per cent more FP32, 15 per cent more AI TOPS and twice the power ceiling. Compared with the RTX PRO 6000 Workstation Edition, which has the same 600 W envelope, it has 87.5 per cent of the memory, 78.0 per cent of the bandwidth, 84 per cent of the FP32 rate and 59 per cent of the AI TOPS.

LLM inference: token rates and memory capacity

Generating a token reads the active weights once, so on a model that fits both cards the 5500 should generate at almost exactly the rate of a 5000: its bandwidth is only 4 per cent higher. The RTX PRO 6000 should be about 28 per cent faster than the 5500 and a third faster than the 5000. Prompt processing, batch throughput, image generation and fine-tuning depend on compute. There the datasheet places the 5500 between the 5000 and the Max-Q: 105 TFLOPS FP32 and 2,376 AI TOPS, against 65 and 2,064 on the 5000 and 110 and 3,511 on the Max-Q. This is our reading of the specifications, not a measurement; we found no independent test of the card.

The 5500’s advantage is its 84 GB, as the arithmetic for a 70B model shows. We size it with the rule of our users-per-card guide: Llama 3.3 70B takes 67.7 GiB in NVIDIA’s FP8 checkpoint and 39.8 GiB in its NVFP4 checkpoint, one 8,192-token conversation needs 1.25 GiB of FP8 cache, and about a tenth of the card plus 3 GiB goes to the runtime.

Our arithmetic: 90 per cent of the memory the driver reports (95.6 GiB on the 96 GB card, scaled in proportion for the 72 and 84 GB cards), less about 3 GiB for activations and the runtime, less the weights; FP8 KV cache at 160 KiB per token; an FP16 cache halves the conversation counts. The 5500 figures use the 84 GB of its datasheet. Production servers keep more in reserve.

The FP8 row shows the main difference from the 5000: on the 5500 a 70B model fits at 8-bit precision, with cache for about three 8K conversations. The same logic applies to gpt-oss-120b at 60.8 GiB, which leaves under 1 GiB on a 72 GB card and about 11 GiB on the 5500. In MIG mode each 42 GB half takes a 32B model in FP8, or gpt-oss-20b with long contexts, but not a 70B model in 4-bit with any useful context.

If you are sizing a 70B model or several smaller ones, tell us the model and the number of concurrent users, and we say which of the RTX PRO 5000, 5500 and 6000 fits.

Video engines, rendering and rack-mounted workstations

The 5500 carries three NVENC and three NVDEC engines, like the 5000, where the 6000 has four of each, and it encodes and decodes H.264 and HEVC in 4:2:2 like the other RTX PRO Blackwell cards. NVIDIA’s page repeats the family claims, Tensor cores “up to 3x” and RT cores “up to 2x” the previous generation, without test conditions; rendering benchmarks will follow once independent testers have cards.

NVIDIA says the card is “built for rack-mounted workstation deployment”, where IT teams “centralize GPUs into managed racks, share them across users”. This is the remote-workstation model, with the GPU in a rack and the user at a desk elsewhere. That puts it next to the RTX PRO 6000 Server Edition, which is passive, runs at 1,597 GB/s and supports vGPU. Unlike it, the 5500 is actively cooled and is not on NVIDIA’s vGPU list; both list MIG and four DisplayPort outputs. At up to 600 W it needs the same power supply headroom as a 6000 Workstation Edition, through one PCIe CEM5 16-pin connector according to the datasheet.

If the card is going into a workstation or server you already run, we check the platform, power and cooling before the quote. Write to us with its model and its power supply, and we say in advance if something will not work.

RTX PRO 5000, 5500 or 6000: which to choose

If the model fits in 72 GB, the RTX PRO 5000 72 GB gives almost the same token rate at a 300 W board power; the 5500 adds 12 GB and 62 per cent more FP32 at up to twice the power.

If you need 96 GB or the most bandwidth, buy an RTX PRO 6000. For several cards in one machine, the Max-Q is the stronger choice on paper: more memory, 28 per cent more bandwidth and half the power ceiling of the 5500. Its datasheet shows what the 5500’s 600 W ceiling buys: 105 TFLOPS FP32 and 2,376 AI TOPS, below the Max-Q’s 110 TFLOPS and 3,511 AI TOPS at 300 W. How the two compare for language models, MIG and four-card builds is in our RTX PRO 5500 vs RTX PRO 6000 Max-Q comparison.

If you need 84 GB on one card, for a 70B model in FP8 or large render scenes, compare the 5500 with a 6000 in the same quote: it has 87.5 per cent of the memory and 84 per cent of the FP32 rate of the Workstation Edition.

If you need vGPU, look at the RTX PRO 6000 Server Edition, or, for graphics vGPU on Red Hat KVM 9.6, at the 5000 72 GB.

What we supply

We supply the RTX PRO Blackwell cards in our range across the EU, on one EU contract and invoice, with manufacturer warranty, including the RTX PRO 6000 in all three editions: Workstation, Max-Q and Server Edition. We can put the RTX PRO 5500 in the same quote as the RTX PRO 5000 and 6000, so the three can be compared side by side for your workload. Quantities and lead time are confirmed in writing with the quote.

FAQ