Memory Subsystem: Bandwidth

As we have reported before, measuring the full bandwidth potential with John McCalpin's Stream bandwidth benchmark has become a matter of extreme tuning, requiring a very deep understanding of the platform. 

If we used our previous binaries, both the first and second generation EPYC could not get past 200-210 GB/s. It gave the impression of running into a "bandwidth wall", despite the fact that we now had 8-channel DDR4-3200. So we used the results that Intel and AMD best binaries produce using AVX-512 (Intel) and AVX-2 (AMD). 

The results are expressed in gigabytes per second.

Stream Triad

AMD can reach even higher numbers with the setting "number of nodes per socket" (NPS) set to 4. With 4 nodes per socket, AMD reports up to 353 GB/s. NPS4 will cause the CCX to only access the memory controllers with the lowest latency at the central IO Hub chip.

Those numbers only matter to a small niche of carefully AVX(-256/512) optimized HPC applications. AMD claims a 45% advantage compared to the best (28-core) Intel SKUs. We have every reason to believe them but it is only relevant to a niche. 

For the rest of the enterprise world (probably 95+%), memory latency has much larger impact than peak bandwidth. 

Benchmark Configuration and Methodology Memory Subsystem: Latency
Comments Locked

180 Comments

View All Comments

  • steepedrostee - Thursday, August 8, 2019 - link

    if i had to guess who is more full of crapola, i would think you
  • Smell This - Thursday, August 8, 2019 - link


    Pooper Lake?

    Cascade is obviously, The Mistake By The Lake

    Chipzillah will certainly strike back but it reminds me of the old 'virgin' joke. "My first wife was an OB/GYN, and all she wanted to do was look at it. My second wife was a psychiatrist, and all she wanted to do was talk about it, and ...

    My third wife was an Intel Fan Girl, and all she could say was, "Wait until next year!"

    HA!
  • RSAUser - Thursday, August 8, 2019 - link

    Do we finally have a contender to run Crysis on max?
  • Tunnah - Thursday, August 8, 2019 - link

    I'd normally just ignore this but this really needs proofreading, There's multiple mistakes on every page, becomes a bit difficult to follow.
  • cerealspiller - Thursday, August 8, 2019 - link

    Well, 50% of the sentences in your post have a grammatical error, so I will just try to ignore it.
  • GreenReaper - Friday, August 9, 2019 - link

    He's a reader, not a writer. ;-)
  • Oliseo - Thursday, August 8, 2019 - link

    Pot. Meet Kettle.
  • steepedrostee - Thursday, August 8, 2019 - link

    wow amd !
  • umano - Thursday, August 8, 2019 - link

    Intel people will welcome "Rome" like gladiators "Ave, Caesar, morituri te salutant"
    I am really happy for Amd and I really hope their sells will be a lot more than they could ever dream.
    Because they become more than competitive with the most rightful strategy, just deliver an awesome product. The fact they think that they will just double their shares shows how sick the market(s) are. The epyc is faster, greener and way way cheaper.
  • 29a - Thursday, August 8, 2019 - link

    "AMD does not blow fuses on cheaper SKUs to create artificial 'value' for buying more expensive SKUs"

    I like this guy, more reviews from him. He's not afraid to bite the hand that feeds him.

Log in

Don't have an account? Sign up now