🚀 Toward GPU-first computing, one doorbell at a time.
What does that mean in practice? Moving control of communication closer to the computation, so the GPU can trigger work itself.
👇 Read Part 6: “A GPU Thread Rings Its Own Doorbell”
medium.com/ai-infrastructure…
I just published Getting Started with SPDK Inside VPP medium.com/fd-io-vpp/getting…
If you guys really want to get the best of you GPU and want to understand what CUDA is doing under the hood, go have a look at this article: medium.com/ai-infrastructure…
#AI #Cisco
What does a GPU command stream look like before the GPU sees it?
In Part 4 of our GPU internals series, we intercept @nvidia doorbells and recover command streams before execution. Work from our AI infrastructure team at @Cisco .
medium.com/ai-infrastructure…
What really happens during a cudaMemcpy?
At @Cisco , our team instrumented @nvidia's open-source kernel driver to watch the closed-source libcuda library in action.
Part 3 of our deep dive into GPU internals is now available: medium.com/ai-infrastructure…
Jérôme Tollet retweeted
Replying to @jtollet
thnx! just finished writing it up
toonk.io/wireblast-building-…
As part of our work on AI infrastructure at @Cisco, we've launched a series of 7 technical articles exploring the GPU ecosystem . I'm happy to share Part 2, focused on the five software layers that enable modern GPU computing: medium.com/ai-infrastructure…
As part of our work on AI infrastructure at @Cisco, we've launched a series of 7 deep-dive technical articles exploring the GPU ecosystem from hardware to software.
Today, I'm happy to share Part 2: medium.com/ai-infrastructure…
At @Cisco , we are working on AI infrastructure technologies to optimize all parts of inference and training clusters. Here is the first article of a series of 7 technical articles we are publishing on recent work we did on advanced GPU mechanisms.
medium.com/ai-infrastructure…
Here is an article I just published: medium.com/fd-io-vpp/spdk-in…
SPDK is inserted as a @FDioProject / VPP plugin for NVMe/TCP acceleration under @nvidia BlueField 3. Initial performance results are extremely good !
This is the kind of really exciting things we do develop at @Cisco
Here is a brief introduction to VPP stateful layer (SFDP): medium.com/fd-io-vpp/a-state…
Almost 2y ago I did a VPP crypto benchmark comparing VPP native engine vs Openssl engine perfomance. I did it again on a Mac M4. You can compare performance speedup...
Jérôme Tollet retweeted
How does #SONiC’s modular control plane meet #VPP’s high-speed data plane?
Discover the architecture behind a fully software-defined router powered by open source.
Read Part One on @linuxfoundation’s hubs.la/Q03Qk7jv0:
🔗 hubs.la/Q03Qk7FQ0
#FDio #LFNetworking
Jérôme Tollet retweeted
Vector Packet Processor (VPP) release 25.10 extends the sFlow implementation to include support for dropped packet notifications. blog.sflow.com/2025/10/vecto…
Jérôme Tollet retweeted
When SONiC’s control plane meets VPP’s user-space data plane — the result is a high-performance, open, software-defined router.
Explore the architecture in Part One on @linuxfoundation’s hubs.la/Q03Qkbcv0:
🔗 hubs.la/Q03Qk7lm0
#SONiC #VPP #LFNetworking
People using @FDioProject VPP in the background may be interested in applying for a FOSDEM talk
lists.fd.io/g/vpp-dev/messag…
Here is the kind of exciting things we are doing @CiscoSecure to protect our cloud security services
blogs.cisco.com/security/imp…
Jérôme Tollet retweeted
From home setups to data centre monsters, this talk breaks the limits on what's possible with DIY internet infrastructure. Big tech energy, built by the community. Watch the full talk here: youtu.be/xdSMqhQVjDw











