英伟达悄然扩充了其 Blackwell 架构工作站显卡产品线，新增 RTX Pro 5500。该卡定位介于 RTX Pro 5000 与旗舰级 RTX Pro 6000 之间，但在显存容量和最大功耗等多项指标上更接近后者，主要原因在于其庞大的显存池。官方目前仅标注为“即将推出”，未公布具体上市时间。
PNY 公布的部分规格披露了更多细节。RTX Pro 5500 采用面积 750 mm² 的 Blackwell GPU（很可能是削减版的 GB202），拥有 922 亿个晶体管和 21760 个 CUDA 核心，核心数量与 GeForce RTX 5090 相同。作为对比，RTX Pro 6000 拥有 24064 个 CUDA 核心，而 RTX Pro 5000 则为 14080 个。
更值得关注的是显存配置。PNY 称 RTX Pro 5500 配备 84GB 支持 ECC 的 GDDR7 显存，带宽标称 1398 GB/s，同时列出 416-bit 显存位宽——但这两项规格本身并不兼容。416-bit 对应 13 个常规 32-bit 显存通道，很难凑出 84GB 这种容量。
反过来看，14 颗 6GB 的 GDDR7 颗粒正好组成 84GB，对应 448-bit 位宽；若等效速率约为 25 Gbps，带宽也将接近 1400 GB/s，几乎与英伟达标称的 1398 GB/s 完全吻合。这意味着目前公布的 416-bit 规格很可能存在笔误。
在其他规格方面，RTX Pro 5500 支持 PCIe 5.0 x16，最多提供四个 DisplayPort 2.1b 输出，内置三个第九代 NVENC 编码器和三个第六代 NVDEC 解码器，并可通过 MIG 技术作为单个 84GB 实例或两个互相隔离的 42GB 实例运行。
满载功耗可达 600W。在如此功率下，该卡很可能采用英伟达备受争议的 12VHPWR 供电接口——这一接口在高负载条件下曾多次出现故障，而以 AI 为主要负载的显卡恰好长期处于这种工况。
RTX Pro 5500 主要面向机架式工作站部署，而非传统桌面系统。标准版尺寸为 4.4 × 11.1 英寸，占据两个插槽并采用主动风冷；英伟达表示后续还将推出液冷版本。
截至发稿，英伟达尚未公布价格。无论官方定价如何，实际售价几乎肯定会进入五位数美元区间，RTX Pro 6000 Blackwell 已经在该价位停留了相当一段时间。
（本文由 NotebookCheck 报道，中文由编辑整理翻译）
本文来源：https://www.notebookcheck.net/Nvidia-launches-new-Blackwell-workstation-GPU-with-84-GB-VRAM.1398858.0.html
若有版权问题，请来信告知：13951431913#139.com 或在右方提交评论。
NVIDIA has quietly expanded its Blackwell architecture workstation GPU lineup with the introduction of the RTX Pro 5500. This new entry is positioned between the RTX Pro 5000 and the flagship RTX Pro 6000, but its specifications—particularly in terms of memory capacity and power consumption—align more closely with the latter, primarily due to its massive memory pool. The official status is currently listed as "Coming soon," without a specified release date.
PNY has revealed some preliminary specifications that provide further insight. The RTX Pro 5500 features a 750 mm² Blackwell GPU (likely a cut-down version of the GB202) housing 92.2 billion transistors and 21,760 CUDA cores. This core count is identical to that of the GeForce RTX 5090. In comparison, the RTX Pro 6000 boasts 24,064 CUDA cores, while the RTX Pro 5000 has 14,080 cores.
The most notable feature is the memory configuration. PNY reports that the RTX Pro 5500 is equipped with 84 GB of ECC-capable GDDR7 memory, offering a bandwidth of 1398 GB/s. However, this is paired with a stated 416-bit memory bus width, which does not mathematically correlate with the reported 84 GB capacity. A 416-bit bus typically corresponds to 13 standard 32-bit memory channels, which cannot accommodate the precise 84 GB total.
Conversely, 14 memory chips (each 6 GB) would perfectly total 84 GB and utilize a 448-bit bus. If the effective memory rate is approximately 25 Gbps, the resulting bandwidth would be nearly 1400 GB/s, which matches the NVIDIA specification almost exactly. This suggests the 416-bit label may be a typographical error or placeholder, while the actual hardware utilizes a 448-bit bus.
Other specifications include support for PCIe 5.0 x16, up to four DisplayPort 2.1b outputs, three ninth-generation NVENC encoders, three sixth-generation NVDEC decoders, and the ability to run as a single 84 GB instance or two isolated 42 GB instances via MIG.
The maximum power consumption is estimated at 600 W. Given this high power draw, the card is expected to use the controversial 12VHPWR power connector, which has a history of failing under high load conditions.
The RTX Pro 5500 is primarily intended for rack-mounted workstation deployments rather than tower desktops. It has a physical size of 4.4 by 11.1 inches and requires two expansion slots. For cooling, it features a blow-through cooler with dual fans, with a liquid-cooled version planned for later release.
Pricing for the RTX Pro 5500 has not yet been disclosed. Based on the high specifications and the positioning of the RTX Pro 6000 (which already commands a five-figure price), it is highly likely that the new model will also fall into the five-digit dollar range.
(Reported by NotebookCheck, translated by our editors.)