Intel Enables Arc iGPUs to Use Up to 93% Memory

Wccftech reports Intel released a HotFix workstation driver for Arc Pro Graphics that increases the maximum share of host system memory available to built-in Arc Pro iGPUs and select Arc Pro cards to 93%, per the driver notes cited by Wccftech. The coverage names specific hardware families as supported, including built-in Arc Pro GPUs in select Intel Core Ultra processors and discrete Arc Pro models such as Arc Pro B390 and Arc Pro B370, and it notes compatibility with the Alchemist and Battlemage Arc families. Wccftech reports the driver supports multiple Windows hosts, and gives the example calculation that a 64 GB system can allocate 59.5 GB (64 x 0.93) to the GPU. The article also compares the figure to AMD Ryzen AI, which it reports allows up to 87% memory allocation.

Increasing the host-memory ceiling for integrated GPUs affects how much working set a GPU can address without relying on device-local memory. For practitioners, allocating a larger fraction of DRAM to an iGPU can let larger language models or larger batch sizes fit entirely within the GPU address space on machines without discrete GPUs. Industry-pattern observations: comparable vendor updates that raise host-allocation limits typically trade off potential system-memory pressure for greater GPU-accelerated workload capability on single-socket workstations.

For ML engineers who prototype or run inference on local hardware, the change is a practical enabler rather than a paradigm shift. Industry context: larger host-memory allocations make integrated GPUs more useful for lightweight or midsized LLM inference workflows on laptops and compact workstations, but they do not change the raw compute or on-die memory bandwidth. Observed patterns in similar shifts show that ISV certification and driver maturity determine real-world usability for production ML workloads.

• •Adoption in vendor toolchains and frameworks, including whether frameworks can reliably place model tensors into the expanded iGPU addressable region.
• •Any OS-level behaviors under high host-memory reservation, such as paging, thermal, or performance regressions reported by users.
• •Progress on ISV certifications for Arc Pro GPUs cited by the driver notes, and comparative benchmarks versus discrete GPUs and AMD Ryzen AI systems.