Section by Andrei Frumusanu

CPU MT Performance: SPEC 2006, SPEC 2017

We’ve noted the earlier discussions of Intel’s TDP handling and how Tiger Lake has 15W and 28W operating modes, and where this comes into play the most is in multi-threaded scenarios where the platform is generally power envelope limited, having to otherwise clock down.

We’re showcasing the MT performance in SPEC for both the Tiger Lake modes, comparing it to both the 15W Ice Lake and AMD Renoir chips. As a note, the 15W Ice Lake platform had a sustained power draw of 18W which makes things not quite as apples-to-apples. Also as a reminder, the Intel systems have 4 cores and are running 8 thread instances, while the AMD system has 8 cores and is running 16 threads.

SPECint2017 Rate-N Estimated Scores

At first glance, the Tiger Lake system performs quite well versus its predecessor, but that’s mostly only in the 28W mode. At 15W, the generational boost, while it is there, isn’t that significant. This might point out that efficiency isn’t all that much better this generation.

AMD’s platform scales incredibly well in execution-bound workloads as it fully takes advantage of double the core count. In more memory-heavy workloads, the Zen2 cores here seem to be lacking sufficient resources and scale below the performance of Intel’s 4-core designs in some workloads.

SPECfp2017 Rate-N Estimated Scores

In the floating-point results, it’s again a matter of TDP headroom as well as memory performance scalability. In the 15W results, the Tiger Lake chip posts rather small improvements over its Ice Lake counterpart, whilst in the 28W mode the gains are more considerable and even manages to outperform the AMD system more often than not.

SPEC2017 Rate-N Estimated Total

In the overall scores, the verdict on Tiger Lake is dependent on how you evaluate Intel’s performance gains. At an (semi)equal-TDP level between Tiger Lake and Ice Lake, the improvements in performance are 17%. Intel does reach a larger 51% generational performance boost in its 28W configuration, but at that point we’re talking about quite different cooling solutions inside of a laptop, no longer making this a valid apples-to-apples comparison.

We haven’t had opportunity to test out higher TDP -HS model of Renoir yet, but with the 15W 4800U already mostly tied with the 28W i7-1185G7, we would expect it to notably outperform the Tiger Lake chip.

Overall, Tiger Lake seems to be offering roughly 20% better performance per watt over its predecessor, with increased performance beyond that coming at a cost of higher power consumption.

CPU ST Performance: SPEC 2006, SPEC 2017 CPU Performance: Office and Web
Comments Locked

253 Comments

View All Comments

  • Spunjji - Friday, September 18, 2020 - link

    Yet actual performance isn't as far apart as IPC alone would indicate, because the designs differ in some fundamental ways. It'll be interesting to see how this shakes out in practice.
  • RedOnlyFan - Friday, September 18, 2020 - link

    Lol you have got your info wrong. Sunny cove and willow cove IPC is more or less the same. Sunny cove is 18% higher IPC over skylake. Willow cove performance is higher because of higher clock ~20% more. It's impossible to define what's "sufficient".
  • Rudde - Saturday, September 19, 2020 - link

    From Spec2017: Willow Cove has 15% higher integer IPC and 12% higher floating point IPC compared to Zen 2. Zen 3 should be on par with Willow Cove on IPC. Golden Cove will of course keep a gap to Zen 3.
  • zepi - Friday, September 18, 2020 - link

    You should also publish the frequency vs. time graph for some of the tests. This would make it much easier to estimate how the chips scale with TDP.
  • Sychonut - Friday, September 18, 2020 - link

    For me personally, the real star of the show are those ARM processors (especially Apple's) performing so admirably at a smaller power envelope.
  • abufrejoval - Friday, September 18, 2020 - link

    For me the biggest motivator for getting one of these in a NUC would be to play with the shadow stack, control flow integrity (CFI) and memory encryption, because the ability to run secured corporate VMs on personal home-office hardware has a lot of appeal, even if it's originally a cloud issue.

    I'd obviously want those same features from AMD and wonder where they stand: Are their VM encryption mechanisms sufficiently similar to what Intel is pushing (would such things actually be covered under their intellectual property agreements)?

    Any word on CFI from AMD? Or actual implementation of similar extensions on the ARM side?
  • vinay001 - Friday, September 18, 2020 - link

    @Anandtech, What happened to 3080 review??
  • Rudde - Saturday, September 19, 2020 - link

    CA wildfires
  • andracass - Saturday, September 19, 2020 - link

    The side ports on that reference model and the way the screen props the laptop up when you open it makes the device VERY reminiscent of the MSI Prestige.
    Like, identical.
  • Ian Cutress - Sunday, September 20, 2020 - link

    It is. Intel initially asked the press not to put too much emphasis on the OEM they partnered with, as retail units will be different and more optimized. But a lot of press straight up mentioned it in their reviews, so I guess the cat is out of the bag.

Log in

Don't have an account? Sign up now