Gabriele Svelto (gabriele.svelto.delete@this.gmail.com) on August 16, 2014 11:33 pm wrote:
> Michael S (already5chosen.delete@this.yahoo.com) on August 16, 2014 12:43 pm wrote:
> > That's already the same or worse SPECint_2006/W then many Intel's year old Xeon E3L processors:
> > E3-1285L v3: SPECint_2006=210, 65 W
> > E3-1265L v3: SPECint_2006=194, 45 W
> > E3-1230L v3: SPECint_2006=168, 25 W
> >
> > So, Cavium has to execute perfectly just to come close to *old* Xeon E3s!
>
> Not only that but with all these rate scores been thrown around it seems that nobody is paying attention
> to the single-threaded performance these cores will offer. One of the huge upsides of the low-power E3s
> is that they can pretty much all turbo to within 70-80% of the top frequency of the highest, most expensive
> bins. That means you get up to 70-80% of the single-threaded performance of the state-of-the-art as far
> as single-threaded performance goes. So how fast are these ARM cores going to be when running a single
> thread? I think that's a very important question because while it's true that low-power Xeons are bought
> with throughput in mind it's also true that they can still guarantee very fast response times in the
> face of serial code sections and that's very important in the server market IMHO.

What cores? ThunderX? Gene? Vulcan? K12? Cavium product is optimized for throughput workloads, its goal is not to compete with top Xeons in single thread performance, however offers (according to Microprocessor report) between 50% and 100% more throughput performance per watt than Xeon E5.

As I mentioned before, Applied Micro and Broadcom are promising 80--90% of Xeon single thread performance.