Because it would need to have around 2.23x fp throughput per core and per clock Vs Deneb to do so.Which won't happen. In numbers ,Deneb QC @ 3Ghz scores 10.4K ,so it would take 2.23x the throughput of Deneb to score 23K with QC Zambezi @ 3Ghz and another ~16% clock speed to get to 27K.Originally Posted by Mechanical Man
Although Zambezi's fmacs are really powerful ,they are not more than 2x faster than one family 10h core in SSE(comparing one 128bit fmac with one Deneb core). Looking at AMD's own projections for spec fp rate and then looking at leaked donanimhaber slide with C11.5 score,we see a curious matching of performance increase in fp workload,of around 1.88x .This is the combined effect of 33% more cores,more fp execution resources and more clockspeed.Translated into IPC terms,around 1.3x or 30% improvement fmac vs core at similar clock. Note that this may vary a lot since we have no idea in which specific cases fmacs can excel and in which they may have lower throughput.
edit:
@undone
Nope it hasn't been posted and thanks! Looks nice. All the ISA extensions crammed into Orochi design need a wider CPUz window in order to be displayed properly
.
Can you tell us from where did you get this image?
Reply With Quote
Bookmarks