# Linux 7.5 expands DRM mmap range to 8TB for high-VRAM multi-GPU systems

> **Key takeaway:** The Linux kernel is increasing the fake mmap offset range for GEM/TTM buffer objects from 1TB to 8TB to prevent memory allocation failures in multi-GPU setups.

**Published:** 2026-11-20T19:41:29+00:00 
**Source:** Phoronix Benchmarks &amp; Drivers 
**Category:** kernel-os 
**Canonical URL:** https://fosswire.org/news/linux-75-expands-drm-mmap-range-to-8tb-for-high-vram-multi-g.html 

## Summary
Intel is introducing a significant change to the Direct Rendering Manager (DRM) subsystem for the upcoming Linux 7.5 kernel cycle. The update expands the maximum size of the mmap range for GEM and TTM buffer objects, including imported dma-bufs, from the current 1TB limit to 8TB.

This modification specifically targets multi-GPU configurations that utilize massive amounts of video memory. By increasing the DRM_FILE_PAGE_OFFSET_SIZE on 64-bit kernels, the patch prevents buffer object creation and PRIME import failures that occur when the 64-bit address space for fake offsets is exhausted.

## Technical details of the DRM offset expansion
- **Subsystem Modification:** The change increases the `DRM_FILE_PAGE_OFFSET_SIZE` constant on 64-bit architectures to support a fake mmap offset range of up to 8 TiB (SZ_8T).
- **Resource Management:** Every GEM/TTM buffer object, including those imported via dma-bufs, relies on a fake mmap offset allocated from the per-device `drm_vma_offset_manager`.
- **Error Prevention:** The expansion prevents `-ENOSPC` errors during buffer object (BO) creation or PRIME imports, which previously occurred when peer-to-peer dma-buf imports across multiple high-VRAM devices exhausted the 1 TiB limit.
- **Compatibility Constraints:** The new 8TB limit is designed to remain strictly below the `DRM_PANTHOR_USER_MMIO_OFFSET_32BIT` threshold to avoid breaking existing userspace requirements.

## Why it matters
This update is critical for high-performance computing and professional workstation users employing multi-GPU setups with massive VRAM capacities, such as the upcoming Intel GPU D Series. It ensures that Linux-based drivers and compute runtimes can scale to handle peer-to-peer memory transfers without hitting artificial address space bottlenecks.