You can’t beat the price of the Skylake-H Xeon E3 chips, as the table below shows: But for workloads that do not require low latency interconnect and that can fit their data and applications in the relatively small memory footprint of the Xeon E3 chip, this is clearly an option. The Knights Landing chip is expected to deliver more than 3 teraflops at double precision and twice that at single precision, so it packs a punch to be sure. Yes, that is about 20 percent of the expected single precision performance of the “Knights Landing” Xeon Phi parallel processor, which has 72 active “Silvermont” Atom cores with a pair of 512-bit AVX 3.0 vector engines on each core.
#INTEL SKYLAKE IRIS PRO PRO#
We asked Intel for the floating point ratings for the GPUs inside the Skylake-H chips, and it turns out the Iris Pro P580 delivers only 331.2 gigaflops at double precision running at its Turbo Boost speed of 1.15 GHz, but the single precision performance is a very respectable 1.32 teraflops. Was is intriguing is the fact that the new Iris Pro P580 graphics engine on the chip has 72 graphics cores, up from 48 graphics cores with the P555 engine used in the Broadwell Xeon E3 v4 hybrid chips. While the performance of the CPU cores on the Xeon E3-1500 v5 processors is interesting for single-threaded applications, these four cores with eight threads do not provide all that much oomph. (We would expect something more on the order of 8 percent to 10 percent, given past history for microarchitecture improvements.) From early tests of the desktop variants of Skylake cores, it looks like the IPC has gone up around 3 percent, but that may not hold for the server implementations and the number could be a bit higher. With Skylake being a “tock” in the tick-tock scheme Intel uses to advance its chips, the process stays the same but the microarchitecture gets improvements. We do not have the instruction per clock (IPC) ratings relative to the Broadwell cores at a constant clock, but we will expect to see that when Intel puts together the presentations for the Skylake Xeons next year. They feature the Skylake Xeon core, and give us a preview of sorts about what we can expect from next year’s Skylake Xeon E5 and E7 rollouts. The Skylake-H Xeon processors are implemented in the same 14 nanometer processes that Intel uses to etch its Broadwell Xeon E5 v4 and Xeon E7 v4 processors, which have both been recently announced. The same hyperscale, service provider, and cloud builder organizations that might be looking at the Skylake-H processors to speed up video streaming or encoding workloads can also think about deploying the same infrastructure to do parallel floating point operations on the relative cheap. This, as far as The Next Platform is concerned, is where the Xeon E3-1500 v5 processors get interesting. The sales pitch from Intel about the Xeon E3-1500 v5 processors did not mention that the 72 GPU cores on the Skylake-H processor could be used for floating point math acceleration per se, but of course this GPU can be turned into an offload engine like any other GPU linked to a CPU.
![intel skylake iris pro intel skylake iris pro](https://www.userbenchmark.com/resources/img/generic/gpu/intel.jpg)
![intel skylake iris pro intel skylake iris pro](https://www.gizlogic.com/wp-content/uploads/2016/05/Gizlogic-Intel-Iris-Pro-580-Core-i7-6785R-Core-i5-6685R-y-Core-i5-6585R.jpg)
![intel skylake iris pro intel skylake iris pro](https://m.media-amazon.com/images/I/61ymNkteSnL._AC_SX450_.jpg)
But the relatively inexpensive and powerful GPUs are nonetheless compelling for those who are looking to cram a lot of nodes into a rack of servers that can do a mix of single-threaded and floating point work and do so with very good bang for the buck.
#INTEL SKYLAKE IRIS PRO SERIES#
We would be the first to admit that the Xeon E3 would be an unconventional choice for hybrid CPU-GPU workloads, much as was the case with AMD’s “Kyoto” X1150 and “Berlin” X2150 lines of Opteron X series APUs.
![intel skylake iris pro intel skylake iris pro](https://askgeek.io/images/gpu/Intel.png)
Enough in fact that for certain kinds of floating point math on hybrid workloads that system architects should probably give them consideration as they are building out clusters to do various kinds of processing. The “Skylake-H” Xeon E3-1500 v5 chips that Intel recently announced with its new Iris Pro Graphics P580 GPUs pack quite a wallop. AMD gets a lot of credit for creating Accelerated Processing Units that merge CPUs and GPUs on a single package or on a single die, but Intel also has a line of chips Core and Xeon processors that do the same thing for workstation and server workloads.