>At the core of Vera is Olympus, NVIDIA’s custom CPU core, which delivers 50% higher instructions per cycle than NVIDIA Grace
It sounds exciting until you do the maths.
NVIDIA Grace is based on Neoverse V2, also used by Graviton 4, it is derived from the ARM Cortex-X3, used in Snapdragon 8 Gen 2 and Mediatek Dimensity 9200.
With the context above, 50% IPC improvement in 2028 doesn't sounds that impressive. We since have Cortex-X4, Cortex-X925 and currently C1-Ultra used in Dimensity 9500. And we are expecting C2 this year. Considering every generation has had higher than 10% IPC improvement. That Rosa would be about the same as C2, or roughly M5.
What's exciting is the line of investment, not the first product.
Nvidia isn't just releasing this and washing their hands. They probably have 2-3 generations in development at the same time. And they wouldn't make this investment if they didn't have strong evidence they can compete with the Apple and ARM products in the same timeline.
Agree and I think Server CPU segment desperately needs more competition. Pushing ahead with PCIe 6.0. They now have full stack from CPU, GPU and Networking. Working with partners that pushes Rack Improvements and Water cooling development forward.
You're right. I bet they're going to buy companies that do those things soon, if they haven't already! They're just starting the ball rolling on the corpdev side of things, I think their processes aren't that mature but they'll get it.
>At the core of Vera is Olympus, NVIDIA’s custom CPU core, which delivers 50% higher instructions per cycle than NVIDIA Grace
It sounds exciting until you do the maths.
NVIDIA Grace is based on Neoverse V2, also used by Graviton 4, it is derived from the ARM Cortex-X3, used in Snapdragon 8 Gen 2 and Mediatek Dimensity 9200.
With the context above, 50% IPC improvement in 2028 doesn't sounds that impressive. We since have Cortex-X4, Cortex-X925 and currently C1-Ultra used in Dimensity 9500. And we are expecting C2 this year. Considering every generation has had higher than 10% IPC improvement. That Rosa would be about the same as C2, or roughly M5.
Neither ARM or Apple is sitting still.
What's exciting is the line of investment, not the first product.
Nvidia isn't just releasing this and washing their hands. They probably have 2-3 generations in development at the same time. And they wouldn't make this investment if they didn't have strong evidence they can compete with the Apple and ARM products in the same timeline.
Agree and I think Server CPU segment desperately needs more competition. Pushing ahead with PCIe 6.0. They now have full stack from CPU, GPU and Networking. Working with partners that pushes Rack Improvements and Water cooling development forward.
You're right. I bet they're going to buy companies that do those things soon, if they haven't already! They're just starting the ball rolling on the corpdev side of things, I think their processes aren't that mature but they'll get it.