CJ Newburn

Distinguished Engineer at NVIDIA driving tech strategy for Data Ctr arch/security, storage/NW IO, programming models for scale. 100+ patents. Community builder.

[CJ Newburn @CrisasCJ]

@storagereview @Supermicro_SMCI @nvidia Most people don't lose sleep over modest perf gains. But many lose sleep over compromises to data integrity in storage. I'm more focused on DPUs for manageability, storage security and network isolation. Once there, they can also boost perf! #supermicrostorage2024

[CJ Newburn @CrisasCJ]

@storagereview @James_DDN @DDNStorage @Supermicro_SMCI Absolutely: new app usage models may induce new software APIs. An example is SCADA=scaled accelerated data access for per-thread fined-grained GPU-initiated storage. Also user-level storage drivers on GPUs like Infinia. #supermicrostorage2024

[CJ Newburn @CrisasCJ]

@storagereview @James_DDN @DDNStorage @Supermicro_SMCI Continue the conversation at Supercomputing 24 in the Emerging Tech BoF in Atlanta in November. #SC24 #supermicrostorage2024 #HPC

[CJ Newburn @CrisasCJ]

@storagereview @James_DDN @DDNStorage @Supermicro_SMCI We added file support to CUDA a few years ago with cuFile. We're now exploring an object interface with leading partners. Stay tuned, and see how that could be used in your infrastructure/middleware. #supermicrostorage2024

[CJ Newburn @CrisasCJ]

@WendellW It's kinda fun to help reinvent GPUs for the new generation of apps. Isn't it crazy that every couple of years we now care about something completely different for computing? Join us and other partners in definiting new generations of storage! #supermicrostorage2024

[CJ Newburn @CrisasCJ]

@storagereview @Supermicro_SMCI @nvidia Storage clients operating on the DPU are within the more-trusted infra control plane making the storage stack harder to hack than if it were on the untrusted compute node where user code can break out of containers/VMs to root the node. DPU Secure Storage! #supermicrostorage2024

[CJ Newburn @CrisasCJ]

@storagereview @Supermicro_SMCI @nvidia NVIDIA takes on the interests of customers in advocating RDMA everywhere. RDMA enables a remote agent, thru the local NIC, to directly transfer data between the GPU and storage, potentially skipping the CPU. RDMA is available on IB & Ethernet (RoCE). #supermicrostorage2024

[CJ Newburn @CrisasCJ]

@storagereview @Supermicro_SMCI @nvidia Unfortunately, CPU vendors tend to lack full support for peer-to-peer PCI, sometimes dropping bandwidths by half. Supermicro offers compute nodes with a PCIe switch that help enable that direct path from NIC to GPU. #supermicrostorage2024

[CJ Newburn @CrisasCJ]

We're seeking to build and strengthen collaboration in the emerging storage tech space to respond to demands from a new generation of applications. Continue the conversations from #SupermicroStorage2024 at the #SC24 BoF149 on Emerging Storage Tech in Atlanta in November.

[CJ Newburn @CrisasCJ]

Looking forward to joining colleagues and partners at #SupermicroStorage2024 to talk about collaboration around emerging storage tech.

[CJ Newburn @CrisasCJ]

@Raptors lack of clarity and courage to pull down the statement on Women's Day