On the Intel side, we’ll mention the end of the Ponte Vecchio GPUs. These are GPUs designed for Intel supercomputers. This “technical marvel” (as Wccftech puts it) is bowing out in favor of the development of other solutions!
Ponte Vecchio is finished!
With Ponte Vecchio, Intel offered high-performance chips for supercomputers. In fact, these have been incorporated into Aurora, an exascale super machine. At the same time, it has to be said that it has what it takes, as we find a huge 1280 mm² chip. Intel does cheat a little, though… To come up with such a chip, 47 chiplets had to be assembled in a single package, no less!
Internally, the configuration is huge, with 16,384 shaders, 1024 TMUs, but 0 ROP. The chip also boasts 128 GB of HBM2E memory, divided between 8 stacks. All this, coupled with an 8192-bit bus, contributes to a memory bandwidth of around 3 TB/s!
But all good things must come to an end, and the blues are moving on. In order to rationalize costs and focus resources on growth sectors (AI), the brand is now banking on Falcon Shores. We’re talking about a GPU for the promising AI sector, combining the best of Gaudi and the future Xe GPU architecture. Obviously, the launch is scheduled for 2025… It remains to be seen whether the blueprints will face production delays or not.
So if you’re interested, make sure you let us know soon, as Blue won’t be rolling out any new clusters. The available offer will simply be based on existing ones.
