Wilco (wilco.dijkstra.delete@this.ntlworld.com) on December 11, 2019 4:23 am wrote:
> anon (spam.delete.delete@this.this.spam.com) on December 11, 2019 4:05 am wrote:
> > Wilco (wilco.dijkstra.delete@this.ntlworld.com) on December 11, 2019 3:33 am wrote:
> > > juanrga (nomail.delete@this.juanrga.com) on December 11, 2019 12:39 am wrote:
> > > > Wilco (wilco.dijkstra.delete@this.ntlworld.com) on December 10, 2019 5:13 pm wrote:
> > >
> > > > > The A1100 was a good implementation and went into several server products - it is still being
> > > > > sold into storage servers. But what's ironic is that AMD then cancelled the Arm variant of
> > > > > Zen, and now their flagship EYPC2 servers are being outcompeted by Neoverse N1...
> > > > >
> > > > > Wilco
> > > > >
> > > >
> > > > And soon by TX3.
> > >
> > > And Ampere with 80 cores. With the highest throughput per socket
> > > Arm servers should gain a lot of market share in 2020.
> > >
> > > Wilco
> >
> > So you're expecting Ampere to beat Zen3 by more than 60% per thread? Seems a tad over-optimistic.
>
> How do you get 60% per thread? Zen3 increases IPC by 15% and EPYC3 (Milan) is claimed to have
> 64 cores. So with 25% more cores I'd expect Ampere will stay ahead of the increased IPC.
>

I'd imagine anyone going for throughput would keep SMT on. To beat 128 threads with 80 threads they need to be 60% faster.
Unless Ampere is going for SMT as well.