On 9/15/2026 5:32 PM, Dennis Clarke wrote:
On 9/15/26 16:36, Aaron B. wrote:On Tue, 15 Sep 2026 15:13:38 -0400 Dennis Clarke <dclarke%blastwave.org@localhost> wrote:with the exception of a single "new" SPARC S7-2 server. It is a 128-core wonder beast with 256G of memory. Sadly it can only run Solaris 11.4 andmy attempts with NetBSD have all failed.OpenBSD supports this model: https://www.openbsd.org/sparc64.html
<snip>
Which would be a good thing to have happen - lots of older sun4v hardware is out there for quite cheap, with plenty of life left. My Netra T2000 is still going strong; I haven't had problems finding parts for this 20 years old machine.Some of the sun4v hardware can sport a lot of active cores. Sadly the floating point unit in the early units was a single FPU core surroundedby a matrix of eight SPARCv9 type cores. A single FPU instruction would and could wait for seven or more clock cycles just to get access to the FPU. Great for integer stuff however.
While I believe your description of the v9 FPU is correct, I wonder how frequently that was an issue for the sorts of workloads most people would put on a v9 machine.