Thanks for the thorough review! I wanted to give it a close read, which I only now got around to.
other native Chinese companies have joined the fray, but with non-x86 architectures.
Indeed, RISC-V is going to be a big story. Probably the leading non-x86 ISA for China, in the foreseeable future.
ARM has the dual problems of being Japanese-owned and seemingly susceptible to US embargoes. Neither MIPS nor POWER appear to be gaining a lot of momentum, so I gather RISC V is where things are headed. I'd be surprised if MS didn't even have Windows support for it, before long.
These types of scaling losses can come from cache and fabric contention
Also memory. If bandwidth-limited, faster or wider memory would help. As for memory latency, about the only thing you can really do is tweak around with the caches.
However, what the results show is that their per-core scaling is quite good.
@PaulAlcorn , were the IPC tests single- or multi- threaded?
In the Rendering benchmarks, the
LuxMark label is confusing.
CPU OpenCL and
CPU C++ should be two
different benchmarks, each with different results. I wonder which these are. Or, is it an average of both?
Using the disable AVX option yielded some extra performance, to the tune of an extra 5.5% in the multi-threaded test and 5.8% in the single-threaded test,
What I read into the Cinebench AVX vs. non-AVX benchmarks is probably that their vector pipelines are only 128-bits wide, causing AVX instructions to be executed in two pieces. And, unlike how Zen1 managed this trick, they seem to incur some overhead in doing so.
there aren't any established SM3/SM4 tests available yet from the roster of ISVs that provide reputable test software. We're sending out requests for tests that reflect the advantages/disadvantages of those algorithms.
I wouldn't bother benchmarking them unless/until they start seeing use outside of China. Otherwise, it's irrelevant to your readership. Furthermore, comparing a feature they implement in hardware with one that other CPUs don't, basically tells us nothing about the level of sophistication and refinement in their CPUs - it's a benchmark they'll win because they're the only ones even in that particular race.
That begs the question: ...
Heh, you just
had to pull that one on us.
😉