New Platform Lets GPUs and FPGAs Use Intel Optane Memory Modules

Intel
(Image credit: Lenovo)

Liqid, MemVerge, and Intel have developed a platform solution that allows to pool and orchestrate system memory and storage-class memory (SCM) like Intel Optane Persistent Memory (PMem) modules together and then use these ultra-large memory pools with CPUs, GPUs, FPGAs, and other accelerators. The solution supports not only existing PMems, but also future memory hardware from Intel, and upcoming CXL accelerators. 

DRAM is fast, but it has inherent physical capacity limitations and high per-bit costs. While the DDR5 standard was designed with extreme memory capacities in mid, the problem of costs is still there. Intel's Optane and other upcoming SCM are considerably cheaper than DRAM on per-bit basis, which is why PMem modules are often installed alongside traditional DRAM modules in machines that are used to run large databases or in-memory applications.  

But many modern workloads use  accelerators for processing, including GPUs and FPGAs. While accelerators usually have their DRAM memory onboard, in many cases they need more memory for higher performance. The looming CXL technology will enable upcoming systems to pool-in memory resources easily on the hardware level, but these systems still need software that understands how to use these multi-tier memory pools.  

The platform solution that Intel, Liqid, and MemVerge developed uses the Liqid Matrix composable disaggregated infrastructure (CDI) software, MemVerge Memory Machine software for in-memory computing and Intel Xeon Scalable platforms supporting Optane PMem to build high-capacity multi-tier memory pools that can be used both by CPUs and accelerators (GPUs, FPGAs, etc.).  

The platform allows to granularly configure and compose all resources (memory, network, GPU, FPGA to support unique workload requirements. With the new platform, it is now possible to accelerate in-memory computing applications with GPUs or FPGAs, something that was rather hard to do previously. 

"Our composable big memory solutions are an important part of an overall CDI architecture," said Bernie Wu, vice president of business development at MemVerge in a press release. "The solutions are also a solid platform that customers can leverage for deployment of future memory hardware from Intel, in-memory data management services from MemVerge, and composable disaggregated software from Liqid." 

The solution supports not only Optane Persistent Memory modules, but also future memory hardware from Intel, and CXL accelerators. At least initially, the platform will require Intel hardware to run, but expect Liqid and MemVerge to tailor their software for other platforms eventually when CXL-based memory accelerators become widely available.  

"Collaboration between organizations like MemVerge and Liqid, whose respective expertise in big memory and PCIe-composability are well-recognized, deliver solutions that provide functionality now that CXL will bring in the future," said Kristie Mann, vice president of product for Intel Optane Group at Intel in the same release. "Their solution creates a layer of composable Intel Optane-based memory for true tiered memory architectures. Solutions such as these have the potential to address today's cost and efficiency gaps in big-memory computing, while providing the perfect platform for the seamless integration of future CXL-based technologies." 

The new platform solution will be available from Liqid. To learn more about the product, interested parties need to set up an appointment with Liqid.

 

Anton Shilov
Contributing Writer

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.