Sun's Rock canceled?
- Thread starter rpg.314
- Start date
Rock is just a beefed up Niagara with more FP performance and double the number of cores though fewer threads per core. Niagara and Rock were many core multithreading SPARC chips while the chips from Fujitsu were the standard single and dual core SPARC chips.
Last edited by a moderator:
This is actually quite surprising if true. Niagara performs quite well for its size, and Rock is supposed to fix some of its major shortcomings. Furthermore, Sun has put quite a few new innovations into Rock, including the so-called hardware scout and support for transactional memory. When it's announced that Oracle will buy Sun, many thought it would be for Rock.
If Rock is actually cancelled, then the current best SPARC processor would be from Fujitsu, the SPARC64V line. However, it's said that Fujitsu is also considering leaving this market... if so, then this could be the end for the entire SPARC line.
Of course, it's also possible that Sun is going to make a more traditional improve on Niagara rather than the more radical Rock.
If Rock is actually cancelled, then the current best SPARC processor would be from Fujitsu, the SPARC64V line. However, it's said that Fujitsu is also considering leaving this market... if so, then this could be the end for the entire SPARC line.
Of course, it's also possible that Sun is going to make a more traditional improve on Niagara rather than the more radical Rock.
No, Rock is not just a beefed up Niagara. In fact, it included two totally unique features: Out of Order Commits and Scout Threads. Moreover, the OoOC system's hardware checkpointing was going to be used to implement transactional memory for the first time. There were a lot of innovations in Rock, and even if it wasn't going to be a world beater, it's sad to see a chip that taped out not even make it to live hardware. This is like 3Dfx Rampage in a way.
I true, this might have to do with Oracle's interests not being served by staying in hardware, as it might have been argued that Sun's intrests were not being served particularly well either.
Rock had already been delayed, and the OOC and transactional memory support would have been two very new things to implement in a performant chip.
OOC should theoretically be simpler than full-blown OOE, although with no experience in what makes a good OOC chip, it might be hard to get completely right from a perfomance standpoint.
Transactional memory sounds like a bear to implement and verify. It was imposing enough that far more successful CPU companies with more resources and better track records in engineering have taken their sweet time to get around to it.
Rock had already been delayed, and the OOC and transactional memory support would have been two very new things to implement in a performant chip.
OOC should theoretically be simpler than full-blown OOE, although with no experience in what makes a good OOC chip, it might be hard to get completely right from a perfomance standpoint.
Transactional memory sounds like a bear to implement and verify. It was imposing enough that far more successful CPU companies with more resources and better track records in engineering have taken their sweet time to get around to it.
I agree 3d, but it's still sad, because we'll never get the experimental data or experience on OOC chips now. There's very few companies in the world now that the resources to build high end CPUs and the field has continued to diminish over the years, paralleling what has happened with GPU vendors.
Maybe Oracle will sell off the IP to Intel or AMD.
Maybe Oracle will sell off the IP to Intel or AMD.
Similar threads
- Locked
