Intel Picks LPDDR5X Over HBM for Inference GPU
Intel's GPU D-Series uses lower-cost LPDDR memory, reflecting a broader industry push to cut AI inference infrastructure costs.
Updated on Nov. 12, 2026 in Semiconductors
Live Poll
Do you expect the cost of your everyday tech to rise over the next few years?
Intel unveiled its inference-focused GPU D-Series last month, adopting LPDDR5X memory instead of high-bandwidth memory (HBM). The choice reflects LPDDR's larger capacity at a relatively lower cost, despite its slower data transfer speeds.
Why it matters
As demand for AI inference surges, reducing costs for equipment, power, and cooling has become critical for companies building infrastructure. Qualcomm has similarly chosen LPDDR for its AI200 inference accelerator.
AMD claims its PACE software lifts CPU inference speeds by an average of 1.6 times over conventional methods, though the claim is not independently verified. TrendForce projects AI servers could overtake smartphones as the largest LPDDR demand source between 2028 and 2030.
The players
Intel
Intel is a major U.S. semiconductor company that designs CPUs and, with the GPU D-Series, is entering the inference accelerator market.
AMD
Advanced Micro Devices is a U.S. chipmaker known for its EPYC server processors and is developing software to run AI tasks on CPUs.
Qualcomm
Qualcomm is a U.S. wireless and mobile chip designer that has chosen LPDDR memory for its AI200 inference accelerator.
TrendForce
TrendForce is a Taiwan-based market research firm that tracks memory and semiconductor industry trends and issues demand forecasts.
The details
Intel's GPU D-Series pairs LPDDR, the memory type used in smartphones, with its inference accelerator instead of HBM. Separately, AMD developed technology to run certain AI tasks, such as document classification, on its EPYC server CPUs using its PACE software, which optimizes AI operations and memory usage for CPUs.
Timeline
Last month: Intel unveiled the GPU D-Series.
2028 to 2030: TrendForce forecasts AI servers could surpass smartphones as the largest LPDDR demand source.
The Tech Race
Intel's LPDDR choice follows TrendForce's forecast that AI servers could overtake smartphones as the largest LPDDR demand source by 2028 to 2030. It places Intel alongside Qualcomm's AI200 and U.S. startups Possiton and D-Matrix in a race to cut inference hardware costs.
For businesses buying AI infrastructure, lower-cost LPDDR-based accelerators could reduce equipment, power, and cooling expenses compared with HBM-based systems. Users of AI services may eventually see those savings reflected in pricing, though the source does not specify timing.
The takeaway
Memory choices now weigh cost and capacity as heavily as raw speed when designing AI inference hardware. Watch how LPDDR adoption in servers develops over the next several years.
Further reading
For broader context on chip design and memory choices, explore the Semiconductors section.
Source note: This article includes information reported by 조선일보.
Live Poll
Do you expect the cost of your everyday tech to rise over the next few years?
