Fungible sells scale-out storage appliances based on a technology called Data Processing Units, or DPUs. DPUs are specialised processing devices, that offload processing from the CPU in much the same way that GPUs do floating-point maths better than CPUs.
This is the first stage on what Fungible hopes will be a broader portfolio of disaggregated, composable infrastructure. Fungible is betting that the need for big, fast data will require scale-out systems that offload work to specialised components to keep up with the amount of volume and performance required to quickly access massive datasets. This is a fair enough bet, given what we’re seeing in the industry more broadly, but I’m not sure that Fungible’s approach is a winner.
Fungible also sells storage initiator cards for NVMe/TCP, a bit like the TCP Offload Engine (TOE) cards you used to get. TOE cards mostly went away as the functionality became available in the CPUs themselves at roughly the same efficiency, so there was no need for a specialised card any more. I wonder if the same thing will happen with these storage initiator cards.
As I write this, and as foreshadowed at #SFD23, Fungible has also started selling the Fungible GPU-Connect, which seems to be a box that you can put GPUs in and then share them between multiple other servers using the DPUs to handle the interconnect over Ethernet.
Too Clever By Half
One of my frustrations with Fungible’s presentation was due to something I see fairly often from tech companies: trying to be all things to all people.
There is an art to finding the right level of detail to talk about. If you’re too high level—too abstract—everything is too vague and you can’t really get a clear idea of what’s going on. The concepts are too nebulous and you can’t get a firm grasp of them. There’s a lack of concrete detail you need to understand the high-level concepts.
But if there’s too much focus on detail, you get overwhelmed and struggle to tell where each piece fits in the overall picture. You can’t relate the individual, concrete details to why you should care about them. They become isolated factoids, adrift from the world and untethered from reality.
It’s a challenge to tell a coherent story that ties everything together that also has enough concrete details that the story feels grounded and real. There is a science to storytelling, but it’s also an art. Art takes both skill that comes from learning and practice, but also a wisdom and aesthetic sense that required good judgement about what not to do. Technical companies tend to struggle with telling good stories about themselves through a combination of lack of practice and a lack of wisdom.
Of course, that’s what we sell at PivotNine, so I would say that.
Fungible seems to be trying to be multiple things at once, which is really tricky. It’s a storage company, but it also sells DPUs, and it also sells composable infrastructure. Some of this is painting a vision of the future, and thus a roadmap for product development, but that’s hard to buy today to solve a real problem you currently have.
Fungible needs to succeed both now and in the future, but I worry that it is trying to design a
ComposableInfrastructureBrokerFactory when all I want is a storage array or a network card. It’s the risk inherent in straddling multiple markets: you aren’t going to be as good as something custom designed for that niche, so you need a market that values the flexibility more than the niche capability.
Hopefully Fungible can show some really good traction in one of its many markets and that will sustain it as it moves into all the other markets it wants to be in. Or until someone buys it for the DPU tech that drives everything. AMD just bought Pensando, so it’s a possibility.