ScaleMP and AMD Announce Collaboration for Scale-Up EPYC Servers

[cageymaru]

ScaleMP is a leading provider of virtualization solutions such as vSMP Foundation 9.0 for high-end computing. ScaleMP has announced that it is working together with AMD to enable AMD server OEM manufacturers to create systems with 4, 8, and up to 128 processor sockets, up to 8,192 CPUs and 256 terabytes of shared memory. With EPYC processor-based servers available from many leading server vendors, ScaleMP and AMD have partnered to enable cloud, hyperscale, enterprise, and HPC customers to benefit from the superior value provided by AMD EPYC processors, with ScaleMP offering both fabric-based as well as storage-class-memory solutions.

"Supporting AMD-based servers has always been a goal of ours, as AMD has a long history of offering powerful, yet cost-effective solutions. We see a natural fit for ScaleMP technologies and AMD processors, especially with the strong demand for scale-up systems based on AMD EPYC processors," said Shai Fultheim, founder and CEO of ScaleMP. "Through this collaboration, ScaleMP enables AMD customers to create scale-up servers that will deliver breakthrough performance and help lower the TCO for organizations -- enabling customers to use more of their budget for their applications and tackle the most demanding workloads while enjoying a broader choice of multi-socket servers."

 

[velusip]

I get that sometimes you need to satisfy a growing pain NOW -- and that in most cases the need for an SDS stems from urgent demand, sudden influx of money, or just more money than sense -- but holy smokes, the larger these systems get the funnier this whole solution is.

I mean, how much budget needs to get burned before someone realizes, "wait, we have this flow chart up-side-down."

 

[Cactusj]

My company has a 9 node ScaleMP system that is now sitting there unused after acquiring a true HPC system. The biggest problem with ScaleMP, is that if one of the blades crashed, the whole system would die. This lack of stability was one of the top reasons to pursue a traditional HPC. That said, the underlying technology is kinda cool.

 

[velusip]

My company has a 9 node ScaleMP system that is now sitting there unused after acquiring a true HPC system. The biggest problem with ScaleMP, is that if one of the blades crashed, the whole system would die. This lack of stability was one of the top reasons to pursue a traditional HPC. That said, the underlying technology is kinda cool.

Exactly. SDSs can be feeble. I haven't personally witnessed any multimodal systems in action yet, but the idea of smoothly transitioning from the brute-force, aggregate model to a clustered model once the software problem has been sorted out is ideal. It also means users are more likely to stick with the brand through all forms of growth.