Intel's Benchmarking Shenanigans

 24 November 2020 Opinion
If benchmarks do not show the result you want, create your own.

Intel is struggling. Though it is still a behemoth among chip companies, but it feels increasingly like Nokia or Blackberry after the iPhone launch. AMD has been taking market share almost every quarter with superior products and now Apple not only replaced Intel's chip in their laptop lineup with their own but did so while improving performance.

Meanwhile Intel's 10nm products are delayed.

Since Intel cannot compete head on, it looks like they went back to some of their old tricks, in this case misleading benchmarks.

AMD laptop performance on battery

Apple's M1 chip just dealt a huge blow to Intel. Intel did not just lose a customer, but they also got some serious competition in the laptop space at the same time.

But instead of targeting Apple, Intel attacked AMD, claiming that AMD sacrifices performance for better battery life.

This attack on AMD, rather than Apple, may seem counterintuitive at first, but it makes sense. OEMs like HP, Dell and Lenovo are worried about Apple’s new M1 chip and they need something competitive. Fortunately, AMD will release a new laptop chip at the beginning of next year that is most likely going to be better in multi-threaded performance than the M1 chip.

This is a nightmare scenario for Intel. Not only will it allow AMD to take even more market share, but it will also force OEMs to optimize their models to work better with AMD chips.

To prevent this from happening, Intel is trying to show that AMD chips are not as good as everybody says. Their claim: AMD laptop chips have better battery life and power efficiency than Intel's Tiger Lake, because AMD limits performance while the laptop is running on battery.

Intel’s conclusion: "AMD performance on battery negates value of its product stack".

Intel is partially right, that some AMD models have lower performance on battery life. Unfortunately (for Intel), this has nothing to do with AMD as Extreme Tech discovered.

Corona Render encoding time on Ryzen 4800U actually improves slightly to 2.62 hours. Meanwhile on Intel's i7 1065G7 it takes 12:32 hours instead of 5:17 hours on battery.

The AMD 4800U ON BATTERY is twice as fast at rendering than Intel’s previous Core i7 PLUGGED IN. Source: Extreme Tech

It is up to OEMs to set power limits on battery and it can be further tuned by the user, e.g. in Windows by changing the power options. This has been the case for a long time and as Extreme Tech noticed, is also the case for Intel chips.
Something that appears to be new is a 12 second delay before the CPU speeds up, that some OEM’s added to some of their AMD laptops. Again, this is a decision by the OEM to keep laptops from running out of battery too quickly.

The other option is to ignore battery life and fan noise to achieve maximum performance in benchmarks, as some OEM’s seem to have done with Tiger Lake laptops:
"…the MSI Prestige 14 Evo [i7 1185G7]—which reached nearly gaming-laptop levels of fan noise almost immediately—and the HP EliteBook [4700U], which took more than a minute to ramp its fans up to max and remained much quieter than the MSI throughout the run" – Source: Ars Technica

Intel Ice Lake Server vs AMD EPYC Rome

Intel released information about their upcoming 32 core Ice Lake server a couple of days ago. They also showed some benchmarks where the 32 core Ice Lake server beats the 64 core AMD EPYC 7742.

This looks impressive at first glance, but the fine print shows how Intel achieved it. Intel literally created the Monte Carlo benchmark themselves, using Intel’s Math Kernel Library (MKL) and Intel’s compiler. If you need to create your own benchmark, so that your upcoming product can beat your competitors last year’s product by 30%, you are in trouble. Especially since the MKL alone can reduce performance of AMD chips by a factor of 3 by using slower code paths and by disabling AVX2 acceleration.

Intel claims 20-30% better performance compared with the 64 core EPYC Rome server.

Look at this, our CPU is faster in benchmarks we have developed. Source: Intel News Release

Conclusion: Apple may be next

Intel is at it again. It worked last time, so why not do it again? Unfortunately for Intel, times have changed: There is no Core microarchitecture to save the day. 7nm, initially considered to be the solution, is already delayed by 12 months until late 2022. Their current 10nm processors are even worse than the old Pentiums, because even if they are competitive, Intel still cannot manufacture them in sufficient quantities.

Additionally, AMD is very competitive in Notebooks for the first time and to top it off, Apple with their chips is putting a lot of pressure on Windows laptop manufacturers: OEMs cannot simply postpone the release of AMD laptops just because Intel pays them something extra, because then they will lose (even more) market share to Apple that they are unlikely to get back.

So, we can expect more Intel benchmarks soon. Perhaps even attacks on Apple's M1 chip.