Original link: Intel may have been right about killing Hyper-Threading after all / digitaltrends.
I both think that HT made sense when it was introduced and that it makes sense to drop HT now.
The cost of adding HT is obviously a fixed percentage - adding X% more silicium to a core adds Y% to cost.
On 2/4/6/8 core CPU's then HT did (usually) improve performance. All my measurements indicate 0-30% improvement.
But I can easily imagine that it is much rarer to see improvements with 20/28/32/48 core CPU's. There are simply enough full cores that HT does not improve much. And with highly parallelizable workloads then memory becomes the bottleneck not the CPU.
So with constant cost and decreasing benefits of HT then it is only a matter of time before HT does not make sense anymore.