Lets start out with a little history before we get to the painful stuff, the performance for the current Intel Ice Lake and AMD Milan CPUs. Both hold the high water mark for their respective companies, and both have multiple published benchmark scores from multiple vendors. You can see the results we are using at SPEC but we will concentrate on the best published SpecIntRate2017 Peak for both sides to compare.
The best we can find as of this writing for Intel’s Ice Lake is a submission by Nettrix of 625 for a dual Xeon 8380 platform. For completeness the next best is a Gigabyte submission of 613. AMD’s best for Milan is a score of 928 by ASUSTeK with an Epyc 7773X, followed by a 923 from Gigabyte. For the record, the best Epyc 7763 score we can find is another ASUSTeK submission of 913.
Since this is about numbers, we will go with the highest available scores but feel free to use any ones you feel are better, they are all public. In theory you can buy these systems as submitted and with a little tuning, you can get the same numbers. If you are a skeptic, you might say that both sides massage their submissions equally. In any case SemiAccurate thinks these scores are pretty good reflections of what each chip is capable of. If you want to use different results, base scores, or whatever, feel free.
If you divide the best Milan score, 928, by the best Ice Lake score, 625, you get 1.4848. For the non-math inclined, that means AMD currently has a 48% advantage over Intel in server performance on the integer side. AMD has 60% more cores thank Intel, 64 vs 40, but both have the same 8 DDR4 channels so core performance is pretty equal. 1.6/1.4848 = 1.0776 or roughly an 8% advantage for Intel on paper. Take memory bandwidth into account and that number drops a bit so Intel seems to have a bit better cores but not in a different league as it was in the past.
Last up we have TDP. AMD’s Milan 7773X and 7763 draw a peak of 280W while Intel’s Ice Lake 8380 only pulls 250W. We won’t go into the performance per Watt numbers but that slightly more than 10% higher power draw per socket doesn’t carry over to the system as a whole so the PPW numbers will be less than the raw CPU PPW delta. In short PPW is a little better for Intel than raw performance but not enough to move the needle.
So where are we now? AMD’s Milan is about 48% faster than Intel’s Ice Lake for Integer workloads as shown by SPECIntRate 2017. What people really want to know is where we are going with AMD’s Genoa later this week and Intel’s Sapphire Rapids in a few more weeks. Lets answer the Sapphire Rapids half of that right now.
Note: The following is for professional and student level subscribers.
Disclosures: Charlie Demerjian and Stone Arch Networking Services, Inc. have no consulting relationships, investment relationships, or hold any investment positions with any of the companies mentioned in this report.
Latest posts by Charlie Demerjian (see all)
- What’s going on with Qualcomm’s Oryon SoC? - Sep 26, 2023
- What is the code name for the next Qualcomm laptop SoC? - Sep 19, 2023
- How fast is Qualcomm’s Oryon SoC - Sep 19, 2023
- How is Qualcomm’s Oryon SoC doing? - Sep 12, 2023
- A new player enters the ARM laptop SoC space - Aug 16, 2023