Hoping to build momentum in two burgeoning markets, IBM unveiled its latest mainframe system containing its new Telum processor with a built-in AI accelerator. It is also the company's first system that can fend off cyber attacks launched from quantum computers.
The built-in AI accelerator significantly increases the system's AI inferencing capabilities, allowing banks and payment networks to run deep learning models at scale to greatly reduce latency issues. This means users in financial markets can carry out analysis of fraud at the precise time transactions take place, according to IBM.
The new system can process up to 300 billion inference requests per day with only one millisecond of latency. This level of speed and performance also expands the opportunities for corporate and third-party developers to create new AI-powered applications and services, IBM said.
"Having a massive increase in AI capability at an application programmer's fingertips now could prove to be a game-changer," said Ross Mauri, general manager of IBM's Z and LinuxOne division, in a press briefing.
One of the benefits of the new AI capabilities is the reduction in time and energy typically required to handle fraudulent credit card transactions for both cardholders and merchants. This in turn could lessen frustrations associated with false credit card declines and prevent consumers from turning to competing credit cards for future transactions, Mauri said.
New IBM z16 takes on AI-based systems
While many enterprise users see the benefits of integrating AI into their existing workloads, two key challenges have kept them from committing to the technology or maximizing their existing implementations, according to Elpida Tzortzatos, CTO for AI at IBM Systems.
"First was the difficulty of just implementing AI into existing applications," Tzortzatos said. "But the even bigger challenge was, once they integrated it in, the ability to scale AI to meet their service-level agreements. It proved simply unattainable to them," she said.
By loading up the z16 with a number of different technologies, IBM is trying to differentiate its mainframes from a host of AI-based high-end computing systems that Nvidia has now and that Intel and AMD plan to deliver over the next couple of years.
"Some AI supercomputer makers are stacking up impressive numbers of GPUs and CPUs believing they can deliver the performance users will need down the road, but it doesn't work that way over the long haul," said Frank Dzubeck, president of Communications Network Architects, Inc. "IBM is doing something different with the Telum chip/AI accelerator combined with improved security that might prove better over the long term."
He was impressed with the z16's overall speed, but added that given the level of sophistication the next generation of AI applications will require, even the upper reaches of the z16's performance could be challenged.
"A key factor is that AI processing, especially with machine learning, takes a really long time," Dzubeck said. "Even Nvidia's biggest machines today still can't do some of the things in real time that IBM is talking about doing."
Another analyst agreed a z16 loaded up with the latest AI and cryptographic software is IBM's attempt to head off high-end supercomputer makers hoping to grab a major share of the cloud-based market.
"IBM is not a top-three leader in the cloud, so they are doubling down on the high-performance computing market in order to differentiate themselves and give the mainframe a new identity," said Judith Hurwitz, an independent analyst focused on emerging technologies.
But one disadvantage IBM may have in going up against Nvidia is that Nvidia has established a network of partners who add technical and marketing value to its systems. The z16 is very much a self-contained system that IBM will largely sell to enterprises directly or through resellers.
Judith HurwitzIndependent analyst
"Nvidia has all these hardware and software partners to work with, which is obviously not the cornerstone of IBM's strategy with this system," Hurwitz said. "Plus, they still have a difficult time getting users to stay in love with the mainframe when competitors tell them to decommission their systems in favor of less expensive cloud services."
Using IBM's Pervasive Encryption and Confidential Computing software as a foundation, the new quantum-safe security scheme uses lattice-based cryptography. This approach allows for the construction of security primitives that protect data and applications against current and future system threats, including those from quantum computers, IBM said.
New cryptography security module
IBM also debuted a hardware security module that can be used with both classical and quantum-safe cryptographic technology for uses involving information confidentiality, integrity and non-repudiation.
The Crypto Express 8S in the z16 system offers a quantum-safe secure boot and cryptography that can address quantum-related threats including 'harvest now, decrypt later' attacks that could lead to extortion and loss of intellectual property, according to IBM.
"You must think about a couple of key things from attacks originating from a quantum computer," said Anne Dames, an IBM distinguished engineer and a specialist in cryptographic technology, in a press briefing. "Number one is algorithms we have been using can be completely broken. Second, they can break things we use for key exchange which affects secure communications."
Real-time payments and new payment methods such as cryptocurrencies are testing the limits of existing fraud detection software, IBM's Mauri said. With the improved security software, IBM looks to solve existing user problems but will take aim at a range of new use cases it couldn't pursue before.
The z16 will be generally available on May 31, IBM said. Pricing for the system is not available.
As editor at large with TechTarget's news group, Ed Scannell is responsible for writing and reporting breaking news, news analysis and features focused on technology issues and trends affecting corporate IT professionals.