If you want to see the power of monopoly in action, look at the enterprise market and how Intel is able to hold market shares despite their inferior products. Some corporates still blindly buy Intel.
Caveat: 500W 6980P vs 360W 9754/9654.
Also, 6980 has 128 cores over 96 cores in the old Epyc 9754.
Next month AMD will launch their current generation, which will also have 128 cores in their model competing with 6980 (and 192 cores in the model competing with Sierra Forest).
Intel has chosen well to do the Granite Rapids launch a few weeks earlier than expected, so that they can be compared only with the old AMD CPUs.
Granite Rapids will have the advantage over AMD Turin to use some non-standard memory modules "MCR DIMM" (Multiplexer Combined Ranks DIMM), originally developed by Hynix, but now also made by Micron, which have higher throughput than the standard DDR5 memories.
There is a pending JEDEC standard for a similar kind of memories (MRDIMMs, multiplexed rank buffered DIMMs), which is slightly different from what Granite Rapids uses now. AMD will support MRDIMMs eventually, but presumably only in future CPUs, because the JEDEC standard is not finalized.
For additional confusion, Intel and Micron have renamed "MCR DIMM" into MRDIMM, to match the name of the standard JEDEC memory, even if their specifications are different.
Therefore the MRDIMMs for Intel will not be usable in different computers, e.g. in those with AMD CPUs, even when AMD will begin to support MRDIMMs.
While Granite Rapids will have the advantage of faster non-standard memory (presumably also more expensive), it will have the disadvantage of using obsolete Redwood Cove cores, like those of Meteor Lake but with the server features added to them, e.g. AVX10.1/512 and AMX, while AMD Turin will have up-to-date Zen 5 cores, which have a significantly higher IPC (perhaps by 15%) than the Redwood Cove cores.
For AI/ML applications, Granite Rapids will be superior, due to the AMX instruction set extension, which offers features similar to the "tensor cores" of the NVIDIA GPUs. AMX is the most important part of the Intel ISA for server CPUs that is not available yet in the AMD CPUs.
So Intel needs 40% more power for 10-20% more performance? Cool story, bro.
And AMD is about to release something even faster.
Is TDP directly comparable in this case? e.g. on desktops or even laptops AMD/Intel tend to define it pretty differently.
Hopefully this leads to price competition between Zen 5 128 Core and Intel 6900P.
The new SuperMicro X14 can fit Dual 6900P, or 256 Core / 512 Thread in a single 2U along with 6TB of Memory ( if you can afford it ).
For comparison, Stack Overflow is running on 8 Server with a total of 64 CPU Core. Not to mention the CPU are from 2015 era.