Tuesday, March 7th 2023

Intel's Ponte Vecchio HPC GPU Successor Rialto Bridge Gets the Axe

Late on Friday in a newsroom posting by Intel's Interim GM Jeff McVeigh a roadmap uplift was quietly revealed. Rialto Bridge, the process improved version of Ponte Vecchio currently shipping under the Max Series GPU branding, has been pulled from the roadmap in favor of doubling down on the future design code-named Falcon Shores. Rialto Bridge was first announced last May at SC22 as the direct successor to Ponte Vecchio, and was set to begin sampling later this year. In the same post Intel also cancelled Lancaster Sound, their Visual Cloud GPU meant to replace the Arctic Sound Flex series of GPUs based on similar Xe cores to Arc Alchemist. In its stead the follow-up architecture Melville Sound will receive focused development efforts.

Falcon Shores is described as a new foundational chiplet architecture that will integrate more diverse compute tiles, creating what Intel originally dubbed the XPU. This next architectural step would combine what Intel is already doing with products such as Sapphire Rapids and Ponte Vecchio into one CPU+GPU package, and would offer even further flexibility to add other kinds of accelerators. With this roadmap update there is some uncertainty as to whether the XPU designation will make the transition as it is notably absent in the letter. It is clear though that Falcon Shores will directly replace Ponte Vecchio as the next HPC GPU, with or without CPU tiles included.
With these changes to the roadmap Intel intends to restructure their data center GPU business to a two-year cadence. Stating that:
"This matches customer expectations on new product introductions and allows time to develop their ecosystems."
Source: Intel
Add your own comment

6 Comments on Intel's Ponte Vecchio HPC GPU Successor Rialto Bridge Gets the Axe

#1
Tomorrow
Big surprise. At this point any Intel product is just a placeholder that may or may not be released at some point but rarely on time.
Posted on Reply
#2
lemonadesoda
The is a beautiful example of Product Managers running ahead of themselves. In their attempt to manage the Product Lifecycle of their product portfolio, they issue roadmaps to test the market interest even before the engineers and project managers have working prototypes. And on this occasion, the upper management said no, or the R&D couldnt get anything working that would have performance exciting enough to create a market niche.
Posted on Reply
#3
Daven
Makes sense. PV was so late that it bunches up too many planned releases. Cancel and spread out is probably the best way to go.
Posted on Reply
#4
_JP_
I swear that "Rialto Bridge" and Intel in the same sentence confused me a bit...
Posted on Reply
#5
AusWolf
So basically, they figured it's better to release something good whenever it's ready instead of incremental upgrades on a regular basis. I wonder why the rest of the industry hasn't come to this realisation yet.
Posted on Reply
#6
Minus Infinity
Next we'll hear Meteor Lake cancelled as they decide to go straight from Raptor lake+ to Arrow Lake
Posted on Reply
Nov 18th, 2024 17:20 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts