+1 to Michael's question.
Also, to highlight the importance of testing in your own environment, here are a series of questions that your original email doesn't answer that are highly relevant:
1. TCP or UDP? There's additional overhead associated with the reliability of TCP. That said, UDP can also cause problems in high traffic environments due to a lack of flow control.
2. Calls per second? Call setup can be expensive in a lot of ways. If you're thrashing through calls, and using TCP, you may run out of ephemeral ports to open new connections. Freeswitch may spend more time setting up resources to handle new calls than it does processing the actual call.
3. Is SIP signaling traffic encrypted with TLS? Is RTP encrypted? Both of these, of course, require additional CPU proportional to the strength of the encryption chosen.
4. What codecs are you intending to use? Is transcoding required? Transcoding = more CPU. Better audio quality codecs (generally) = more bandwidth.
5. Are your calls spending a lot of time in the dialplan? Or actively talking? Conferencing? Parked/waiting? Each of these requires different amounts of CPU.
6. What's your conference configuration? Lower baseline energy level means mixing more conference participants together, which means more CPU.
7. Are calls all being recorded to the same disk? Are they going to compete for IO? You may not just have CPU to worry about. Your ability to handle 1k concurrent calls is dependent on all parts of the system holding up to that load, not just the CPU.
Of course there are other (indirect) factors as well. What else is running on the machine (It's easy to say nothing - but it's rarely true. Common applications might include log rotation, backup software, monitoring tools, etc)? What else is running on the network?
The list goes on and on, but each one of these items has the ability to significantly affect the amount of resources your environment will need. I doubt anyone will be able to give you an answer besides: test it for yourself and see what your setup can do. Then, as you test, you'll probably land on tuning/performance questions that others could help with.
Final note: at 1k calls per second it sounds like the system gets a good amount of use, and is likely important to your business/operation. I'd suggest that if you haven't started thinking about HA/fault tolerance now, you should - even if it's just an active/passive setup. The same ideas that get you to HA can also land you closer to a horizontally scalable solution in many cases. Both of which seem like they'd be beneficial for you.
Best,
Colin