Slashdot Mirror


Intel Launches Xeon Scalable CPUs: Dual Xeon Platinum 8176, 112 Threads Tested (hothardware.com)

MojoKid writes: Intel announced its new Xeon Scalable processor family based on the 14nm Skylake-SP microarchitecture a few weeks back, though today marks the official launch of the platform. Not only do these processors feature a new microarchitecture, but Intel has also revamped the naming convention and arrangement of the Xeon product stack, branding them with Platinum, Gold, Silver, and Bronze model families. Intel Xeon Scalable series processors feature core counts ranging from 4 to 28, with varied frequencies and cache configurations. Workstation processors and lower-core count server chips top out in the 3.2GHz -- 3.6GHz range, while the higher-core count products typically fall in the 2GHz -- 2.7GHz range. Six memory channels are supported and the chips have 48 lanes of integrated PCIe 3.0 connectivity. Power envelopes range all the way from 70W on up to 205W. The Xeon Scalable series also introduces new security, virtualization, and storage-related features, more memory bandwidth, support for AVX-512 extensions, a mesh interconnect, and enhanced hardware controlled power management, among a host of other architectural improvements. Testing of a 2P Xeon Platinum 8176 system, sporting 56 physical cores / 112 threads shows significantly increased performance and bandwidth, with only moderately higher power consumption versus a previous-gen 2P Xeon E5-2679 v4-based system.

6 of 54 comments (clear)

  1. Imagine a Beowulf cluster... by Red+Herring · · Score: 5, Funny

    Too old a joke? Anyone? Anyone? This thing on?

    --
    #include "standard_disclaimer.h"
  2. Hilarious by Anonymous Coward · · Score: 5, Informative

    Very expensive (the top one is $11722) and very fragmented portfolio. There actually is a review comparing one of the top Xeon Platinums 8176 ($8719) to the AMD EPYC 7601 (4200$) - http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade

    The results might be surprising and speak of desperation in pushing the Core-derived architectures too far. Another indicator is the recent HEDT platform's problems on Tom's Hardware - http://www.tomshardware.com/reviews/-intel-skylake-x-overclocking-thermal-issues,5117.html

    Not to mention another security nightmare on the level of Intel ME/AMT:

    "The chipset will also include a new feature called Intel’s Innovation Engine, giving a small embedded core into the PCH which mirrors Intel’s Management Engine but is designed for system-builders and integrators. This allows specialist firmware to manage some of the capabilities of the system on top of Intel’s ME, and is essentially an Intel Quark x86 core with 1.4MB SRAM."

    Because we all know the motherboard manufacturers are known for stable and secure code, so let's let them botch yet another thing.

    It's high time AMD kicked Intel's butt into actually trying and not milking their customers constantly.

    1. Re:Hilarious by gweihir · · Score: 3, Interesting

      I think this is a sign of pretty extreme desperation: They do not have anything on par with what AMD offers and they will not have anything for years to come, as developing new architectures takes a lot of time, regardless of how much money you throw at it. AMD, meanwhile can optimize their new design for the next 5 years or more before they are even remotely threatened by Intel. Will take all the mindless sheep a while to understand, but eventually it might even dawn on them how thoroughly Intel has fucked them over the last few years..

      --
      Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
  3. Re: Obsolete computer architecture by Rockoon · · Score: 2

    Its powered by Wishful Thinking. The fastest data lines ever devised.

    --
    "His name was James Damore."
  4. Databases by nojayuk · · Score: 3, Insightful

    The overall takeaway is that the AMD system is faster in a lot of workloads (databases being the one big notable area where it loses badly) than the Intel system and at a much lower cost.

    Databases are what the Big Iron servers live to support so AMD losing badly against Skylake on that front means they've lost the sales war. Web servers are databases, order processing systems are databases, pretty much everything that's computationally intensive has a database or six on the backend.

    High Performance Computing (HPC) is shiny and prominent but the sales are limited and a lot of new HPC kit is based around non-CPU computation elements derived from GPUs rather than general-purpose CPUs so even good performance in that area won't save AMD in the datacentre markets.

    1. Re:Databases by epine · · Score: 2

      Databases are what the Big Iron servers live to support so AMD losing badly against Skylake on that front means they've lost the sales war.

      Big language. Been watching too much Bruno Ganz lately?

      I appears Ars tested MySQL Percona Server 5.7.0 as their chosen representative for the entire category. I wouldn't recall Rommel's tanks just yet.

      Typically when high response times were reported, this indicated low single threaded performance. However for EPYC this is not the case. We tested with a database that is quite a bit larger than the 8 MB L3-cache, and the high response time is probably a result of the L3-cache latency.

      I have about 30 different database products listed in my notes (many oriented at graphs or machine learning, along the entire sharing spectrum). Would they all suffer this much?

      What does this mean to the end user? The 64 MB L3 on the spec sheet does not really exist. In fact even the 16 MB L3 on a single Zeppelin die consists of two 8 MB L3-caches. There is no cache that truly functions as single, unified L3-cache on the MCM; instead there are eight separate 8 MB L3-caches.

      Well, that does make the present EPYC implementation suck for a popular worker-thread model used to concurrently access a single, large datastore.

      I suspect, however, that a database server server hundreds of small databases as part of a WordPress server farm would hardly suffer at all (so long as CPU locality is stable at the OS level).

      Web servers are databases, order processing systems are databases, pretty much everything that's computationally intensive has a database or six on the backend.

      Or six. You even said it yourself.