anon (anon.delete@this.anon.com) on September 19, 2020 11:08 am wrote:
> Andrei F (andrei.delete@this.anandtech.com) on September 19, 2020 10:43 am wrote:
> > Maynard Handley (name99.delete@this.name99.org) on September 19, 2020 10:34 am wrote:
> > > Andrei F (andrei.delete@this.anandtech.com) on September 19, 2020 8:29 am wrote:
> > > > Maynard Handley (name99.delete@this.name99.org) on September 18, 2020 5:00 pm wrote:
> > > > > itsmydamnation (no.delete@this.way.com) on September 18, 2020 3:19 pm wrote:
> > > > > > Andrei F (andrei.delete@this.anandtech.com) on September 18, 2020 3:26 am wrote:
> > > > > > > Jose (1.delete@this.2.com) on September 18, 2020 1:24 am wrote:
> > > > > > > > Doug S (foo.delete@this.bar.bar) on September 17, 2020 10:36 pm wrote:
> > > > > > > > > anon (anon.delete@this.anon.com) on September 17, 2020 7:10 pm wrote:
> > > > > > > > > > AnandTech's (SPEC ST performance) review is here: anandtech.com/show/16084/intel-tiger-lake-review-deep-dive-core-11th-gen/8
> > > > > > > > > >
> > > > > > > > > > All of TigerLake's extra performance over IceLake is due to higher clocks. And it significantly
> > > > > > > > > > outperforms the Zen 2 based AMD Renoir chip (Ryzen 7 4800U). However not all is good: TigerLake
> > > > > > > > > > experiences a noticeable IPC regression compared to IceLake. The memory subsystem is unable
> > > > > > > > > > to keep up with the higher clocks, and the reworked cache is not enough.
> > > > > > > > > >
> > > > > > > > > > And it just barely beats Apple's A13 despite using significantly more power.
> > > > > > > > > > Makes you think the A14 based Mac coming later this year will blow past Intel's
> > > > > > > > > > best CPU microarchitecture in single threaded integer performance.
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > Going back what Apple stated about A14 being 40% faster than A12, A14 should be about 16% faster than A13.
> > > > > > > > > Add 16% to the A13 scores and it easily beats those Tiger Lake numbers in SPEC2006, for both int and fp.
> > > > > > > > >
> > > > > > > > > That doesn't even include the bigger caches and wider memory that the A14X will get, nevermind
> > > > > > > > > if they have something even better than A14X in store for that Macs arriving later this year.
> > > > > > > > I always have my doubts with benchmark numbers under different operating systems.
> > > > > > > > How can we compare SPEC/Geekbench etc numbers in iOS vs Windows?
> > > > > > > >
> > > > > > >
> > > > > > > Because the OS is irrelevant. There's no system calls in either
> > > > > > > SPEC or GeekBench, you're just running binary code on the CPU.
> > > > > > >
> > > > > > > The only valid concern is compiler differences between Apple's Xcode LLVM toolchain vs the
> > > > > > > public Clang toolchains, and I've been told by industry people the differences are minor.
> > > > > >
> > > > > > Why are there still large differences been Linux and Windows in GB5?
> > > > > >
> > > > >
> > > > > For a start the GB5 binaries are built with different compilers.
> > > > >
> > > > > Then, especially on x86, the OSs provide different policies regarding DVFS.
> > > > >
> > > > > Finally "large difference" is a problematic phrase. I'd consider anything less than 10% to
> > > > > be noise, but there's a certain class of hysterical internet user who thinks it makes sense
> > > > > to build their theory of conspiracy and persecution on 2% differences in benchmarks.
> > > > >
> > > >
> > > > GB5 uses the same compilers.
> > >
> > > I saw an article claiming that for GB4 they were different. You know they are the same for GB5?
> > >
> > > OK, checking now. You are correct!
> > > They were different for 4, but not for 5:
> > > https://www.geekbench.com/doc/geekbench4-cpu-workloads.pdf
> > >
> > > https://www.geekbench.com/doc/geekbench5-cpu-workloads.pdf
> > >
> > > Do you think my point for DVFS is correct, or does the board firmware have much more
> > > control than the OS over that? On x86 Win/Linux I have no idea where the knobs live.
> > >
> >
> > I don't see the claim being that GB5 is performing differently in Win/Linux being valid in the first place.
> >
> > DVFS might affect a few milliseconds but nothing significant.
>
>
> Andrei regarding the SPEC2006 ST test, you say TGL single core power draw
> is 20-21W (at up to 4.8G) but this is for the 28W TDP part right?
>
> What is the TGL power draw in this test for the 15W TDP part?

Well, 15W, since it throttles. It no longer sustains 4.8G, and clock down depending on workload.