Intel “Kaby Lake” Xeon E3 Sets the Server Cadence
nextplatform.com
nextplatform.com
> Intel kind of quietly slipped out the Kaby Lake Xeon E3 processors right smack in the middle of its Technology and Manufacturing Day last week, and it did not make a big deal about it.
This is relevant to customers buying Xeon E3 hardware. Over the last few weeks, OEMs quietly added Kaby Lake Xeons to some existing devices. Because the chips are compatible with existing motherboards, there were no new devices to be announced. With little coverage of this announcement, it is easy to end up buying an i7 Kaby Lake or a E3 Skylake on a device, if you did not know that E3 Kaby Lakes are now shipping.
Are people really saying that? There's a lot of discussion about languages and software that have trouble taking advantage of many cores, but I've never heard anyone saying that the benefits of parallelisation are going to reach a limit soon?
Perhaps that's what the author meant in shorthand.
Nope, especially not on servers. Of course Intel will have to scale up the core interconnects at some point and might also need to scale I/O and memory busses up in a similar vein as AMD does in Naples.
Note that this article is about Xeon E3 though, which is not the mainstream server platform, but is usually used for lowish-end workstations and entry-level servers.
By definition, if there are no relationships between components, then Amdahl's law does not apply (or rather there's 0% serialization), and you'll always benefit from more parallelization.
For example, my typesetter doesn't care that my compiler is running which doesn't care that my web server is running. There's no relationship between the inputs or outputs of those processes. And there's no synchronization.
Therefore, all else equal, as long as the cost per core continues to decrease, it's beneficial to add more cores: because you can run more things.
Similarly a lot of stuff which workstations are used for are (comparatively) easy to parallelize, e.g. rendering and many kinds of simulations (fluidics, FEM, ...).
> The neat bit is that activating the GPU on one of these Xeon E3 processors only adds a single watt to the thermal profile of the chip and only a few bucks to the cost.
Well I very much doubt that in practice. Perhaps Intel's TDP scenario here is "no outputs connected and no framebuffer active" or something like that.
The GPU can steal power from the CPU cores and force them to throttle, which may explain the TDP.