#GPU-Initiated
rocm-xio was published a couple of days ago, which is billed as GPU-initiated IO (for context, I think this theoretically could open up the GPU streaming weights/KV cache directly via NVME from an SSD, bypassing the host CPU/memory).

github.com/ROCm/rocm-xio
GitHub - ROCm/rocm-xio: A ROCm library for GPU-Initiated IO. This provides support for initiating IO from a ROCm-capable GPU against a range of targets including RDMA NICs, NVMe SSDs and AMD GPU DMA e...
A ROCm library for GPU-Initiated IO. This provides support for initiating IO from a ROCm-capable GPU against a range of targets including RDMA NICs, NVMe SSDs and AMD GPU DMA engines (SDMAs). - ROC...
github.com
April 27, 2026 at 6:54 AM
mKernel: a set of fast multi-node, multi-GPU fused kernels.

mKernel fuses compute + communication into one persistent GPU kernel, covering both intra/inter-node with GPU-initiated communication.

Code: github.com/uccl-project...
Blog: uccl-project.github.io/posts/mkernel/
mKernel: Fast Multi-GPU, Multi-Node Fused Kernels
mKernel is a collection of multi-GPU, multi-node fused kernels that fuse intra-node NVLink communication, inter-node RDMA, and compute into a single persistent kernel.
uccl-project.github.io
May 27, 2026 at 2:52 PM
Rothschild Redburn initiated Sell ratings on Nebius Group and CoreWeave, citing falling GPU prices, hyperscaler captive capacity, and costly financing that may compress neocloud returns.
Save What Matters
Curate Feeds | Make Collections | Customize Email Briefs
briefly.co
September 21, 2026 at 7:17 PM
Rothschild Redburn initiated Sell ratings on Nebius Group and CoreWeave, citing falling GPU prices, hyperscaler captive capacity, and costly financing that may compress neocloud returns.
Save What Matters
Curate Feeds | Make Collections | Customize Email Briefs
briefly.co
September 21, 2026 at 7:17 PM
Rothschild Redburn initiated Sell ratings on Nebius Group and CoreWeave, citing falling GPU prices, hyperscaler captive capacity, and costly financing that may compress neocloud returns.
Save What Matters
Curate Feeds | Make Collections | Customize Email Briefs
briefly.co
September 21, 2026 at 7:16 PM
In this paper is proposed SwarmIO, an IOPS-scalable NVMe SSD emulation framework that supports end-to-end modeling of GPU-centric storage systems with GPU-initiated I/O.

arxiv.org/pdf/2604.06668
April 9, 2026 at 11:40 PM
#$NBIS #$CRWV both initiated at Sell by Rothschild Redburn.

- Bear thesis: AI compute unit economics are starting to crack
- Lower GPU pricing + in-house buildouts threaten third-party demand
- Expensive financing compresses returns as the cycle matures
September 21, 2026 at 11:08 AM
Breaking: CoreWeave is up 5.4% today

$CRWV added ~$2,600,000,000 in market cap today

Why it's moving:

• Brought up its first Nvidia Vera Rubin NVL72 cluster. Non-GPU revenue already topped $400M annualized

• Rothschild & Co Redburn initiated Sell with a $54 p...

🔗 x.com/joinautopilot/status/...
September 21, 2026 at 7:48 PM
SwarmIO faithfully models IOPS-optimized SSDs at target performance levels of up to 40 MIOPS, achieving a 303.9x speedup over the state-of-the-art baseline SSD emulator under GPU-initiated I/O.
April 9, 2026 at 11:40 PM
[2025-11-20] 📚 Updates in #GNN

(1) Importance Ranking in Complex Networks via Influence-aware Causal Node Embedding
(2) <a href="https://researchtrend.ai/papers/2511.15076" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">GPU-Initiated Networking for NCCL
(3) GPU-Initiated Networking for NCCL

🔍 More at researchtrend.ai/communities/GNN
November 20, 2025 at 3:09 AM
The connection can be trivially initiated via a simple script from other parts of the environment where the AI chip is deployed, but just talk to the GPU driver for handshake initiation/client key exchange with the EXPORT_CERT. This minimizes the technical reqs for AI chips
July 16, 2025 at 1:02 PM
🎥 Now on YouTube!

Watch a preview of our July #CephTechTalk, then catch the full session:
For Whom the Doorbell Tolls: GPU-Initiated Key-Value #Storage on #Ceph

📺 Full talk: t.ly/ceph-radosnk...
#CephCommunity #GPU #OpenSource
July 10, 2026 at 3:37 PM
It’ll be either a text-to-speech engine narcissistic vanity model in a couple of days, or a GPU initiated housefire. Exciting times!

#piper-tts
September 4, 2025 at 12:00 PM
(1) Karl B. Torp is presenting his @itu.dk MSc project work on GPU-Initiated I/O at the DaMoN workshop on Monday. This work is a collaboration with Simon Lund from Samsung Denmark. We are exploring the next steps in this line of work. If you are interested, come talk to us during SIGMOD. 👇
June 21, 2025 at 9:26 AM
Intel Begins Open-Source Xe3P GPU Driver Enablement

Intel has quietly initiated open-source efforts to lay the groundwork for its next-generation Xe3P graphics within the Mesa OpenGL “Iris” and...

🔗 https://www.madshrimps.be/news/intel-begins-open-source-xe3p-gpu-driver-enablement/
March 6, 2026 at 10:32 AM
Join our July 8 #CephTechTalk to see how rados-nkv explores #GPU-initiated key-value storage on #Ceph.

🕙 10:00 AM PDT / 17:00 UTC

Register: t.ly/JulyCephTech...
#CephCommunity #Storage #OpenSource
July 1, 2026 at 2:53 PM
Join us July 8 at 10:00 AM PDT for the next #CephTechTalk!

"For Whom the Doorbell Tolls: #GPU-Initiated Key-Value Storage on #Ceph"

Learn how rados-nkv enables #AMD GPUs to retrieve key-value data directly from Ceph.

Register: t.ly/TheDoorbellT...
#OpenSource #NVMeKV
June 18, 2026 at 9:59 PM
Goldman Sachs initiated coverage on cloud GPU provider #Nebius with a Buy and a $68 PT.

💻 Backed by Nvidia and boosted by a $72M round led by Jeff Bezos’s firm, the company is poised to grow.

Shares soared 17.3% Monday, up 80% YTD, but Goldman still sees it as undervalued vs. peers like CoreWeave.
July 15, 2025 at 12:06 PM
Accelerate multi-node shared memory and RDMA for distributed #HPC and #AI compute. Intel #SHMEM: GPU-initiated RDMA, following #OpenSHMEM extended with open multi-vendor multiarchitecture #SYCL APIs.
Now with Intel #oneAPI HPC Toolkit 2025.1:
www.intel.com/content/www/...
Intel® SHMEM: Multi-Node Data Sharing with SYCL*
Intel® SHMEM: Accelerate Multi-Node Data Sharing for GPUs and SYCL* Devices
www.intel.com
March 31, 2025 at 11:45 PM
Goldman Sachs starts Nebius at “buy,” cites AI infrastructure strength
Investing.com -- Goldman Sachs has initiated coverage on Nebius Group (NASDAQ:NBIS) with a “buy” rating, pointing to the company’s position as a leading provider of AI infrastructure and forecasting over 50% upside from current levels. With a market capitalization of $10.5 billion, Nebius operates in the AI Neocloud market, offering outsourced access to GPU computing power via the cloud. Analyst Alex Duval views the company as a direct beneficiary of rising demand for generative AI and the growing need for specialized GPU infrastructure. According to the initiation note, Nebius’ key advantages include a full-stack software offering, cost efficiencies, and the ability to operate at scale. These features, Duval writes, position the company to serve the increasing number of enterprise and developer clients seeking access to AI compute without investing in their own hardware. Duval sees a favorable risk/reward skew for investors, estimating a 4:1 upside-to-downside ratio between bull and bear scenarios. He also flags that Nebius is currently trading at 3x EV/Sales, well below its closest peer CoreWeave, which trades at 5x. Goldman’s report identifies a clear path to growth catalysts, including potential capital raises, expanded infrastructure capacity, and global market entry. These factors, Duval suggests, could further support revenue growth and valuation upside. Despite recent share price gains, the brokerage maintains that Nebius remains attractively priced relative to peers, particularly given its role as a pure-play in AI compute infrastructure. With valuations skyrocketing in 2024, many investors are uneasy putting more money into stocks. Sure, there are always opportunities in the stock market – but finding them feels more difficult now than a year ago. Unsure where to invest next? One of the best ways to discover new high-potential opportunities is to look at the top performing portfolios this year. ProPicks AI offers 6 model portfolios from Investing.com which identify the best stocks for investors to buy right now. For example, ProPicks AI found 9 overlooked stocks that jumped over 25% this year alone. The new stocks that made the monthly cut could yield enormous returns in the coming years. Is NBIS one of them?
www.investing.com
July 14, 2025 at 10:20 AM
NVIDIA is expanding beyond GPU sales into AI infrastructure investments and partnerships. BMO Capital initiated coverage with a $340 price […]
Nvidia’s AI Ecosystem Support Key to Upside, BMO Says
NVIDIA is expanding beyond GPU sales into AI infrastructure investments and partnerships. BMO Capital initiated coverage with a $340 price […]
bitnewsbot.com
August 21, 2026 at 5:34 PM
Evolving Comms Libraries in ROCm for Future AI Workloads
This talk covers advances in AMD ROCm communication libraries for large-scale AI. RCCL innovations include copy-engine offloading, symmetric memory, GPU-initiated collectives, congestion-aware load balancing, and port failover. rocSHMEM extends an OpenSHMEM-like model with Python and Triton support, while NIXL contributions enable efficient inference. Together, they support diverse NICs like AMD Pensando AINICs and Broadcom Thor. Discover more: https://www.amd.com/en/corporate/events/advancing-ai/sessions-catalog/evolving-comms-libraries-in-rocm-for-future-ai-workloads.html *** Subscribe: https://bit.ly/Subscribe_to_AMD Join the AMD Gaming Discord Server: https://discord.gg/amd-gaming Visit the AMD Gaming Community Website: https://www.amdgaming.com/ Like us on Facebook: https://bit.ly/AMD_on_Facebook Follow us on Twitter: https://bit.ly/AMD_On_Twitter Follow us on Twitch: https://Twitch.tv/AMD Follow us on LinkedIn: https://bit.ly/AMD_on_Linkedin Follow us on Instagram: https://bit.ly/AMD_on_Instagram ©2026 Advanced Micro Devices, Inc. AMD, the AMD Arrow Logo, and combinations thereof are trademarks of Advanced Micro Devices, Inc. in the United States and other jurisdictions. Other names are for informational purposes only and may be trademarks of their respective owners.
www.youtube.com
August 16, 2026 at 7:20 AM
DDN and Nvidia Collaborate on GPU-Initiated Data Access for AI $NVDA
August 4, 2026 at 3:02 PM
Graid Technology Achieves Industry-First 100 Million IOPS on a Protected RAID 5 Volume for GPU-Initiated I/O

https://www.byteseu.com/2250818/

Graid Technology announced an industry-first result: 100 million random 512-byte IOPS on a protected storage volume built from 32 Kioxia XD8 NVMe …
Graid Technology Achieves Industry-First 100 Million IOPS on a Protected RAID 5 Volume for GPU-Initiated I/O - Bytes Europe
Graid Technology announced an industry-first result: 100 million random 512-byte IOPS on a protected storage volume built from 32 Kioxia XD8 NVMe SSDs.The
www.byteseu.com
August 4, 2026 at 12:52 PM