jimmysmitty :
I only have two requests.
1. Same OS for both to eliminate any optimization variables or OS overhead variables, again Windows adds more overhead than IOS by far.
2. Do it with non synthetic benchmarks.
Well, if they switch Mac to this platform, you'll get #1 and people will undoubtedly run plenty of #2.
jimmysmitty :
bit_user :
jimmysmitty :
HT adds at best 20% performance when it is programmed to take advantage.
Wow, have you looked at any CPU benchmarks, this decade? Do you see only ~20% difference between i5's and i7's, or between Ryzen 3 and Ryzen 5 models (after accounting for clock speed)?
https://www.tomshardware.com/reviews/intel-kaby-lake-core-i7-7700k-i7-7700-i5-7600k-i5-7600,4870-4.html
https://www.tomshardware.com/reviews/amd-ryzen-3-2200g-raven-ridge-cpu,5472-7.html
Of course, the amount of benefit from HT depends on the workload, but it's routinely
much better than 20%.
Its not routinely much better than 20%. In most games its not much better at all and in some programs it can bet better than 20%. How about I say its more average 20%? Would that be a better term? Because SMT is not powerful enough to beat out a real core.
BTW the Kaby Lake review you posted is not easy to use. Not only does the i7 have a clock speed advantage (3.8GHz base vs 4.2GHz base, the same as the i5s turbo speed) it also has a cache advantage (8MB vs 6MB, 33% more) on top of the SMT.
https://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/15
https://www.gamersnexus.net/game-bench/3227-ffxv-hyperthreading-smt-on-vs-off-benchmarks-cpu
Some benchmarks do show that it can hit up to 30%. Some show little to no gains. Still doesn't replace 4 full cores, low power or not.
To argue that one CPU is not faster because it doesn't speed up a certain non-CPU bound game relative to another is absurd on the face of it. I feel embarrassed for you, over that one.
When we're debating CPU performance and the merits of uArch features, you need to focus on CPU-bound tasks. And though I initially resisted, you've successfully trolled me into whipping out my spread sheets and computing the per-clock speed up on the CPUs I mentioned. Kaby Lake and Ryzen, because they're both 4c/4t vs 4c/8t comparisons.
First, Kaby Lake (i7 7700K vs i5 7600K):
■ Blender: 46.5% faster per clock
■ Cinebench: 29.1% faster per clock
As you say, the i7's L3 cache is 33% larger. There's no way to intuitively say how much that benefits the i7 - it will vary on a case-by-case basis, potentially not even helping at all.
Now, the Ryzen 2000G series presents a better opportunity, both since they have the same L3 cache size and because the review included many more compute-bound benchies. So, comparing Ryzen 3 2200G with Ryzen 5 2400G, the clock-speed normalized improvements are:
■ Cinebench: 38.4%
■ Blender: 36.6%
■ LuxMark (CPU/C++): 35.3%
■ LuxMark (CPU/OpenCL): 46.9%
■ POV-Ray: 38.6%
■ 7-Zip decompress: 65.3%
■ 7-Zip compress: 23.5%
Finally, there's my own experience of compiling C/C++ code on 4c/8t and 8c/16t machines, where I typically see about a 60% benefit from HT.
So, going back to the question of how HT compares with low-power cores, let's see how the Tempest (little) cores stack up with the Vortex (big) cores.
■ Clock speed: 1.6 vs. 2.5 Ghz
■ Instruction issue: 3 vs. 13
■ ALU pipelines: 2 vs. 6
Source:
https://www.anandtech.com/show/13392/the-iphone-xs-xs-max-review-unveiling-the-silicon-secrets
So, it's really not hard to view their contributions in a similar vein as HT. And while reading through that article for those specs, I came across this juicy morsel:
Monsoon (A11) and Vortex (A12) are extremely wide machines – with 6 integer execution pipelines among which two are complex units, two load/store units, two branch ports, and three FP/vector pipelines this gives an estimated 13 execution ports, far wider than Arm’s upcoming Cortex A76 and also wider than Samsung’s M3. In fact, assuming we're not looking at an atypical shared port situation, Apple’s microarchitecture seems to far surpass anything else in terms of width, including desktop CPUs.
By all accounts, what Apple has done is simply amazing. Many have predicted ARM
eventually overtaking x86, but we now finally find ourselves standing at the edge, peering over the precipice.