By Intel's standards, 64 cores is a lot. But AMD had 64 cores per socket since 2019. Granted, Emerald Rapids will mop the floor with a 7002 Epyc, but, not only does AMD more than double Intel's offering today, Zen 4 cores are very competitive, especially in terms of performance-per-watt. Not to mention, I'm sure these pairs of gargantuan dies are going to be absurdly expensive. Where these chips are likely to handily beat AMD is in latency-bound tests.

schmidtbag:

By Intel's standards, 64 cores is a lot. But AMD had 64 cores per socket since 2019. Granted, Emerald Rapids will mop the floor with a 7002 Epyc, but, not only does AMD more than double Intel's offering today, Zen 4 cores are very competitive, especially in terms of performance-per-watt. Not to mention, I'm sure these pairs of gargantuan dies are going to be absurdly expensive. Where these chips are likely to handily beat AMD is in latency-bound tests.

I am not sure about the latency this one seems to have 2 identical chips glued .... So 2 x32 core chips there is for sure some interconnection between em . We will see 😉

Venix:

I am not sure about the latency this one seems to have 2 identical chips glued .... So 2 x32 core chips there is for sure some interconnection between em . We will see 😉

Yes, but compare that to AMD's 13 dies (one being the I/O die) and Intel's interconnect is effectively negligible.

schmidtbag:

Yes, but compare that to AMD's 13 dies (one being the I/O die) and Intel's interconnect is effectively negligible.

Most likely although threadrypper 1950x had only 2 dies as well and had more latency than the 3995 if I am not mistaken .... On the other hand it was working likenuma so most likely not even apples to apples here. We will see !

Venix:

Most likely although threadrypper 1950x had only 2 dies as well and had more latency than the 3995 if I am not mistaken .... On the other hand it was working likenuma so most likely not even apples to apples here. We will see !

1st gen Ryzen had a lot of issues to work out, with OS schedulers being one problem. By the time the 3995 came out, the schedulers were in better shape, although still bad, particularly in Windows.

schmidtbag:

1st gen Ryzen had a lot of issues to work out, with OS schedulers being one problem. By the time the 3995 came out, the schedulers were in better shape, although still bad, particularly in Windows.

Yeah they where terrible on latency sensitive load ...and bargains for tons of PCI express lanes or workloads that just wanted MOAR cores :P