News report

Perplexity Portable Computer Comes to Windows, but It Needs a 24GB RTX GPU

Perplexity Portable Computer now runs local agent workflows on compatible Windows PCs, with NVIDIA specifying GeForce RTX or RTX PRO GPUs with at least 24GB of VRAM.

On this page
  1. Portable Computer moves Perplexity’s agent workflow onto Windows RTX PCs
  2. The 24GB VRAM floor excludes most gaming GPUs
  3. Local does not mean the agent is cut off from cloud services

At a glance

Key facts

Windows availability
Available now; announced September 14, 2026
Supported GPU families
Compatible NVIDIA GeForce RTX and RTX PRO
VRAM requirement
24GB or more
Example local model
Qwen 3.8 27B
Local usage
Locally completed work does not consume Perplexity Computer credits
Cloud escalation
Optional; Portable Computer asks permission before sending information off-device

Portable Computer moves Perplexity’s agent workflow onto Windows RTX PCs

Portable Computer is the local counterpart to Perplexity Computer, designed to plan and carry out multistep tasks while running the model and agent workflow on the user’s own hardware. NVIDIA says the Windows release builds on existing support for DGX Spark systems and RTX PCs running Linux.

The practical difference is where the work happens. NVIDIA says sensitive information can remain on the PC for locally completed tasks, and those tasks do not consume Perplexity Computer credits. When a task needs more advanced cloud reasoning, Portable Computer can request permission before sending information off-device rather than silently treating every workflow as cloud-only.

What the Windows release requires and provides
AreaWindows Portable Computer status
GPUCompatible NVIDIA GeForce RTX or RTX PRO GPU
VRAMAt least 24GB
Local model exampleQwen 3.8 27B
Local filesCan be processed on-device
Cloud modelsCan be used with user permission when additional reasoning is needed
Connectors named by NVIDIAOutlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub
DGX StationSupport expected later

The 24GB VRAM floor excludes most gaming GPUs

The Windows launch is broader than requiring a dedicated DGX system, but it is not a general “any RTX PC” feature. NVIDIA explicitly sets the floor at 24GB of VRAM. That requirement matters more than the RTX branding by itself because many current and recent GeForce cards carry less memory.

Tom’s Hardware notes that consumer cards meeting the stated capacity include the 24GB GeForce RTX 3090, RTX 3090 Ti and RTX 4090, plus the 32GB RTX 5090. That list is useful as an illustration of the VRAM gate, not as an exhaustive compatibility guarantee: NVIDIA’s requirement is stated in terms of compatible RTX hardware with 24GB or more, and individual software support can still matter.

Local does not mean the agent is cut off from cloud services

Perplexity is positioning Portable Computer as a hybrid workflow rather than a permanently offline assistant. NVIDIA says local models can analyze data and work across files on the PC, while the agent can identify tasks that would benefit from stronger cloud reasoning and ask the user before escalating them.

The Windows version also connects to services including Outlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub. That makes “local” an execution and data-handling option, not a claim that every connected workflow is isolated from external services. Users should still consider the permissions and data paths of whichever connectors they enable.

Practical context

What this means for PC users

The release makes a capable local-agent workflow available on ordinary Windows workstations and some high-end gaming desktops without requiring a separate DGX Spark system. At the same time, the 24GB VRAM requirement keeps the addressable consumer hardware pool narrow.

  • Owners of qualifying high-VRAM RTX cards can use existing PC hardware for local agent tasks instead of buying a dedicated AI box solely for Portable Computer.
  • Local completion can reduce reliance on cloud credits and keep sensitive files on the machine for workflows that stay local.
  • Most mainstream and upper-midrange gaming GPUs remain outside the stated requirement because VRAM capacity, not just GPU compute performance, is the immediate gate.
  • The launch is another example of VRAM becoming a capability constraint for local AI workloads, separate from its traditional role in games and professional graphics.

Evidence status

Confirmed versus not established

Confirmed

  • NVIDIA confirms Portable Computer is available in the Perplexity app for Windows on compatible GeForce RTX PCs and RTX PRO workstations.
  • NVIDIA states a 24GB-or-more VRAM requirement for the Windows release.
  • NVIDIA says locally completed work does not consume Perplexity Computer credits and cloud escalation requires user permission.
  • NVIDIA names Qwen 3.8 27B as an example local model and lists Outlook, OneDrive, Word, Google Drive, Gmail, Slack and GitHub connectors.

Unconfirmed

  • The announcement does not establish performance parity across every qualifying 24GB-or-higher GPU, so this article does not infer equal speed or practical capacity from VRAM eligibility alone.
  • A qualifying VRAM capacity should not be treated as a guarantee that every RTX model, laptop configuration or unusual system setup is supported unless Perplexity or NVIDIA explicitly lists it.

Sources

Primary and technical sources

These sources support the reporting and analysis above. Current stories are updated when later evidence materially changes the facts.

  1. 01 NVIDIA

    Perplexity Portable Computer Is Now Available on Windows, Powered by NVIDIA RTX
  2. 02 Tom's Hardware

    Perplexity’s local AI agent comes to Windows, but only for RTX GPUs with at least 24GB of VRAM

Related