Skip to main content
  1. Home
  2. Computing
  3. News

Nvidia’s new 84GB GPU can split itself in two for AI-heavy workplaces

Nvidia built an 84GB GPU to give entire teams workstation power from one rack

Add as a preferred source on Google
Nvidia RTX Pro 5500 Blackwell
Nvidia

Nvidia has another enormous Blackwell graphics card, and this one isn’t built to run Cyberpunk 2077 a little faster. The new RTX Pro 5500 Blackwell Workstation Edition comes with a whopping 84GB of GDDR7 ECC memory and is designed primarily for rack-mounted professional workstations.

The GPU was built for companies that need to centralize GPU horsepower and share it among teams working on AI models, robotics simulations, rendering, and other workloads that can quickly overwhelm a conventional PC. The new RTX Pro 5500 sits right between Nvidia’s 72GB RTX Pro 5000 configuration and the 96GB RTX Pro 6000 in terms of memory capacity.

84GB isn’t there just to look impressive

The company has packed the large memory pool for a very specific reason. It allows the RTX Pro 5500 to hold larger AI models, longer context windows, or multiple models simultaneously without having to constantly offload data elsewhere. So the GPU can be very useful for local LLM inference, AI agents, computer vision, and multimodal generative AI. The same capacity can also accommodate enormous 3D scenes and datasets used for engineering or scientific work.

Recommended Videos

Under the hood is Nvidia’s Blackwell architecture with 21,760 CUDA cores, fifth-generation Tensor Cores, and fourth-generation RT Cores. Memory bandwidth reaches 1,398GB/s, while maximum power consumption can climb all the way to 600 watts. The card uses PCIe 5.0 x16 and supports up to four DisplayPort 2.1b outputs. Nvidia also offers air-cooled and liquid-cooled configurations, which also adds to the fact that this GPU is intended for managed workstation racks rather than the PC sitting beside your feet.

One GPU can become two

One of the coolest features is Multi-Instance GPU, or MIG. The RTX Pro 5500 can operate as one 84GB GPU or divide itself into two isolated 42GB instances. Each gets its own memory, cache, and compute resources, allowing two users or workloads to share the same physical accelerator while maintaining predictable performance.

The RTX Pro 5500 gives enough GPU resources to an organization to distribute. Nvidia hasn’t announced pricing yet and currently lists the RTX Pro 5500 as “coming soon.” Considering the specifications and enterprise focus, I wouldn’t expect GeForce pricing when it eventually arrives.

Vikhyaat Vivek
Vikhyaat Vivek is a tech journalist and reviewer with seven years of experience covering consumer hardware, with a focus on…
Google’s next Googlebook could make your Android tablets feel outdated
Googlebooks are moving beyond traditional laptops, and the next one might look more like a tablet.
Computer, Electronics, Laptop

Googlebooks have only been around for a little over a month, but Google is already looking beyond traditional laptops. The company has confirmed that detachable Googlebooks are on the way, meaning we could soon see devices that work as both a tablet and a laptop without carrying two separate gadgets.

The announcement came during the Android Dev Summit 2026, where Google outlined its plans to expand the Googlebook lineup beyond the clamshell and convertible designs currently available. While the company hasn't revealed an official launch date or shown any hardware, an earlier leak suggests Lenovo could be the first manufacturer to bring one to market.

Read more
Meta is already working on its second-gen humanoid robot, but don’t expect it to land in homes anytime soon
Meta wants a robot to do your chores, but it still has to learn how to climb stairs
Fourier’s GR-3 humanoid robot, which the company unveiled in August 2025.

Meta appears to be getting serious about bringing artificial intelligence out of the screen and into the physical world. The company has built a humanoid robot prototype with Chinese robotics firm Fourier, and it is already developing a second-generation version. However, before anyone starts imagining a robot folding laundry or handling household chores, there is a fairly important caveat: this is still research and development, with no launch date in sight.

According to The Information, which cited people familiar with the project, Meta and Fourier built the prototype, internally called Max, in around three months. The roughly 48kg robot is part of Meta's broader ambition to develop humanoids that could eventually carry out everyday tasks in people's homes.

Read more
Anthropic cracks down on video game cloning with new Claude restrictions
Claude can still help develop original games, but copying existing titles is now off the table.
Claude website open on laptop

AI has been making its way into video game development for a while now, but the recent developments have started raising some serious concerns. Popular games are being reverse-engineered with AI assistance, recreated for web browsers, and made playable on platforms they were never released for. Between all that, indie developers have started pulling down their demos over fears of getting copied. Anthropic has now stepped in with new restrictions for Claude Code.

The change was first spotted on X and subsequently confirmed by Anthropic's Addy Osmani. The new instructions prevent Claude from recreating specific existing games, even if they are renamed or presented as original projects. Users can still make original games, work on games within a genre, and use their own designs and assets.

Read more