Telum II
Mainframe processor for IBM z17 with AI acceleration
Telum II is a mainframe central processing unit designed by IBM to power the z17 system. It features eight high-frequency cores, on-chip AI accelerators and a unique virtual cache hierarchy to support real-time analytics in financial services.
- Manufactured on Samsung 5 nm process node
- Features eight cores running up to 5.5 GHz
- Includes 360 MB of on-chip Level-2 cache
- Provides 4x AI compute power over predecessor
- Uses virtual L3 and L4 cache hierarchy
What it is
Telum II is a specialised central processing unit built for IBM mainframe systems. It serves as the engine for the IBM z17 platform, targeting industries that require high reliability and low latency, such as banking and insurance. The chip integrates artificial intelligence capabilities directly onto the silicon, allowing it to perform real-time data analysis without needing separate hardware. This design helps organisations process large volumes of transactions while simultaneously running security checks and fraud detection algorithms.
How it came about
The processor follows the original Telum chip, which powered the previous generation z16 mainframe. IBM developed Telum II to build upon that foundation by increasing memory capacity and boosting artificial intelligence performance. The company presented details about the chip at the Hot Chips 2024 conference. It was designed to address the growing need for immediate business insights within mission-critical applications, continuing IBM’s long history of creating high-performance servers for enterprise use.
Inside the chip
The chip contains eight cores organised into four interconnected clusters. These cores run at speeds up to 5.5 GHz. A key feature is its caching strategy, which uses ten 36 MB Level-2 caches. Eight of these serve the cores, one supports a data processing unit for input and output, and one remains unattached. The system creates a virtual Level-3 cache by moving unused data between these Level-2 slices based on saturation metrics. It also forms a 2.8 GB virtual Level-4 cache across multiple chips in a drawer, reducing data duplication and improving access times.
How it was made
Samsung manufactures the Telum II using its 5 nanometre process node. This advanced manufacturing technique allows IBM to pack significant computing power and cache memory into a single die. The design includes a dedicated data processing unit to handle input and output tasks efficiently. By centralising IO and using this specific fabrication process, IBM claims the new design reduces core power consumption by up to fifteen per cent compared to earlier models, improving overall energy efficiency for large server installations.
Where it went
This processor powers the IBM z17 mainframe, which is used primarily in data centres for critical business operations. It supports workloads in banking, finance, trading and insurance where fast transaction processing is vital. The on-chip AI accelerators enable real-time analytics for customer interactions and fraud prevention. The system can scale by connecting up to thirty-two Telum II processors to form a large shared memory complex, handling massive enterprise workloads and generative AI tasks without external accelerators.
Versions and family
Telum II is the successor to the original Telum processor found in the IBM z16. While the predecessor had eight cores on a 7 nm process, Telum II moves to a 5 nm node and increases cache sizes significantly. The Level-2, Level-3 and Level-4 caches are all forty per cent larger than in the previous generation. It also offers four times the AI compute power. IBM pairs this chip with the Spyre Accelerator for even larger AI models, but Telum II itself handles near real-time inference directly on the mainframe.
Why it matters
The chip represents a shift in how mainframes handle artificial intelligence, moving inference from separate cards to the processor itself. Its unique virtual cache hierarchy addresses the latency issues common in large memory systems. By maintaining high clock speeds and large on-die caches, it prioritises single-threaded performance for fast transaction completion. This approach supports the continued relevance of mainframes in modern cloud and hybrid environments, particularly for sectors requiring strict data security and quantum-safe encryption methods.
Questions people ask about Telum II
What is the main purpose of the Telum II processor?
It powers the IBM z17 mainframe, focusing on high-speed transaction processing and real-time AI analytics for industries like banking and insurance. It integrates AI acceleration directly on the chip to reduce latency.
How does the cache system in Telum II work?
It uses ten 36 MB Level-2 caches. Unused space in these caches forms a virtual Level-3 cache, while data evicted from there creates a 2.8 GB virtual Level-4 cache across multiple chips, reducing data duplication.
Who manufactures the Telum II chip?
Samsung manufactures the Telum II processor using its 5 nanometre process technology. IBM designs the chip architecture and integrates it into their z17 mainframe systems.
How many cores does Telum II have?
The processor features eight high-performance cores arranged in four interconnected clusters. These cores operate at frequencies up to 5.5 GHz to ensure fast response times for critical workloads.
Good reading and sources
There is no English Wikipedia article on Telum II yet, so the facts on this page come from the sources listed above: ibm.com/z/telum, chipsandcheese.com/p/telum-ii-at-hot-chips-2024-mainframe-with-a-uniqu, servethehome.com/the-ibm-z17-mainframe-brings-ai-with-telum-ii-and-spy. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryChip wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.