Intel Announces Scalable Platform Family, Skylake Xeons Imminent

Status
Not open for further replies.

bit_user

Titan
Ambassador
This whole precious metals naming scheme is going to backfire on them. I can just see it: IT department files a purchase order for Platinum Xeon processors and it gets shot down because some MBA decides they should be able to make do with Silver.

Back when processor families were just E3/E5/E7, the same people would just silently nod and approve the order, for fear of speaking up and exposing their ignorance. But everybody thinks they know something about precious metals. Especially the finance folks.

This was a huge marketing fail. Intel should've used the Platinum/Gold/Silver/Bronze naming for consumer products, where some people are ignorant enough to fall for it. In the enterprise IT sector, where products are chosen based purely on specs and price, this isn't going to upsell anyone.
 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
876
394
19,360



Fixed, thanks! :D

 


Based on what? There is very little information out there but Skylake-EP, which is that this socket is designed for, is going to have up to 32 cores and 64 threads as well. No solid info but I am seeing 6 memory channels which will be a slight disadvantage to AMDs 8 but TBH AMD hasn't had a competitive HPC processor for a very long time. People think Intels market share on the consumer end is high, in the server market it is almost every last server.
 


Based on what? There is very little information out there but Skylake-EP, which is that this socket is designed for, is going to have up to 32 cores and 64 threads as well. No solid info but I am seeing 6 memory channels which will be a slight disadvantage to AMDs 8 but TBH AMD hasn't had a competitive HPC processor for a very long time. People think Intels market share on the consumer end is high, in the server market it is almost every last server.
 


Those are Skylake. This platform is Skylake-EP. Think of LGA1150/1151 vs LGA2011. While there are some based on the same uArch, they are geared towards higher end use. For example, the top Skylake EP CPU that will be used for this socket will have 32cores and 64threads. The best Skylake does is 4/8 With Skylake-X (equivalent to what used to be the -X paltfor) on LGA2066 does 10/20.

Skylake EP and the Purley platform are geared towards HPC, especially since the same socket can plug in a KL chip. I imagine you can have a dual socket board and one is a Skylake-EP for OS and the other is a KL chip for high performance computing. Even if AMD has more memory bandwidth the ability to throw a KL chip in there might give Intel an advantage so long as the proper support for it exists.
 

bit_user

Titan
Ambassador
I'm with you on most points, but not this one. Xeon Phi x200 (KNL) doesn't support multi-CPU. You can find an interview with one of the architects on nextplatform, I think, where he said it just needed too much bandwidth to be worthwhile (i.e. more than they could provide).

Also, KNL can run Windows and Linux, so you don't need another CPU for the OS.
 

bit_user

Titan
Ambassador
That's great, but none of the CPUs mentioned in this article will work with budget boards. These will replace the E5/E7 Xeons, and will occupy enormous LGA3647 sockets with 6-channel memory & (I think) up to 48-lanes of PCIe.

In other words, these are not the Xeons you're looking for.
 


I will wait and see. I think that would be an edge up to be able to do that. I doubt it is that hard.

Not sure what kind of bandwidth KNL will need since no matter what it will be running on Purley which has the OmniPath at 100Gbps with or withour a SK-EP CPU.



Current specs say at 48 lanes per CPU. So a dual socket board will be 96 total PCIe 3.0 lanes.
 

jn77

Distinguished
Feb 14, 2007
587
0
18,990
I have been seeing Xeon mobile processors show up in high end laptops now, I guess everything is a "Xeon" now, so basically a Xeon means nothing anymore. What is after "Xeon"?
 

bit_user

Titan
Ambassador

https://www.servethehome.com/intel-xeon-phi-x200-knights-landing-boots-windows/

There's a difference. When you're talking about a cache-coherent multi-CPU system, it needs to support jobs running on one CPU that are accessing memory attached to another. The bigger these CPUs each get, the fatter that pipe needs to be. To the point where you're talking about Xeon Phi, and it's no longer practical (or at least cost-effective) to build so much inter-processor communication bandwidth. There's also the overheads associated with maintaining cache coherency that really start to overtake the amount of useful communication you're doing.

When you're talking about scaling up over OmniPath, it's no longer cache-coherent. The communication over the 100 (or 200) Gbps link really takes the form of inter-process communication, rather than process <-> memory communication. I will go out on a limb and speculate that processes typically need at least an order of magnitude higher bandwidth to local memory than they do inter-process communication bandwidth. Of course, that figure will be highly application-dependent.
 


Xeon is a name for their workstation and server class chips. Those high end laptops are normally workstation class laptops and normally will be paired off with a Quadro GPU.
 

bit_user

Titan
Ambassador
Yeah, Xeon is basically their enterprise branding. Xeon CPUs differ from their mainstream counterparts mostly by offering features like support for ECC memory and enterprise-level manageability. I think some virtualization features might also be disabled in the non-Xeon versions.

Some Xeons might also be higher-binned parts, in order to provide better reliability.
 


Xeons are always higher binned. You don't put a more error prone CPU into a server.

It is also why server chips tend to demand a higher price than their desktop equivalents.
 
Status
Not open for further replies.