News TensorWave just deployed the largest AMD GPU training cluster in North America — features 8,192 MI325X AI accelerators tamed by direct liquid-cooling

[Admin]

TensorWave has built North America's largest AMD AI cluster with 8,192 liquid-cooled MI325X GPUs, delivering 21 exaFLOPS of FP8 throughput. It’s a bold move against NVIDIA’s dominance and marks ROCm’s first major public cloud-scale deployment for general-purpose AI training.

TensorWave just deployed the largest AMD GPU training cluster in North America — features 8,192 MI325X AI accelerators tamed by direct liquid-cooling : Read more

 

[jp7189]

This is a great first step, but there are still missing pieces. We need legions of young people just graduating from school that have been tinkering with rocm on their gaming rigs and are willing to put in the extra effort to save a buck. Maybe if amd enables the full rocm stack on everything they make with a GPU and can keep it going for a few years.. then maybe they'll have something. They are working towards it, but they have a ways to go.

The problem is, I dont think they have the bank roll to keep it going that long. As it stands, AMD ventures is paying tensor wave to buy gpus from AMD to show design wins and revenue on the AMD datacenter side. That's a very expensive shell game. The snake can only eat its tail for so long.

 

[thisisaname]

Just leaves the big question how does it compare with Nvidia's offering?

 

[DS426]

This is a great first step, but there are still missing pieces...

AMD knows and pretty much figures hardware and software won't be [mostly] directly comparable to nVidia's until MI450X and ROCm of that time in 2026. That said, yeah, kids and adults alike need to be tinkering with this stuff now. The whole ecosystem has changed rapidly enough that anecdote of professors

AMD Ventures doesn't have to keep the shell games going for long -- it's just getting the ground work laid and confidence built in startups like TensorWave.