Take the 2-minute tour ×
Super User is a question and answer site for computer enthusiasts and power users. It's 100% free, no registration required.

I was browsing the system76 page for servers https://www.system76.com/servers/model/jacp8 and i realized that when u check the option for a second processor you must choose memory for that processor. In dual processor systems, does the memory get shared? In my old powermac g4 with 2 processors i remember they are shared.

share|improve this question
    
Considering that a XEON processor only is able to access 4 channels of memory: ark.intel.com/products/64596/… its not clear how they are even offering ( 12 x 16GB ) options. I suggest you just contact SYstem76 to answer this question. –  Ramhound Feb 27 '13 at 13:00
    
4 channels != 4 DIMMs. The motherboard image shows 12 DIMMs per socket. –  Paul A. Clayton Feb 27 '13 at 14:53

1 Answer 1

up vote 1 down vote accepted

Multiple socket systems that do not share memory (along with cache coherence) are relatively exotic (Calxeda's EnergyCard is an example) and System76 is more of a whitebox vendor than a specialized system designer. The use of 2xxx versions indicates System76 are paying Intel for support of cache coherence and memory sharing, so it is very unlikely that they would be using them in some kind of mini-cluster-on-a-motherboard design.

share|improve this answer
    
So would a multithreade program only be able to run on one processor? –  agz Feb 27 '13 at 20:34
    
For a traditional system like the System76 example, threads can be distributed across multiple sockets. (The performance impact of thread distribution is a complex subject, but for only two sockets with a modern OS such considerations are usually not significant.) –  Paul A. Clayton Feb 27 '13 at 23:39
    
Where would the programs memory be stored? Will the contents of the program in ram be transferred over? –  agz Mar 11 '13 at 21:47
    
@agovizer For a 2 socket system the difference in latency and bandwidth for remote (alternative socket) memory is not that great, but modern NUMA-aware OSes will tend to map memory according to locality of use so less than half of the accesses should be remote. Other than the access being a little slower, it does not really matter much where the memory is mapped; the content will be transfered in cache line sized (64B) chunks from (local or remote) memory to the local cache as it is accessed. –  Paul A. Clayton Mar 11 '13 at 23:20

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.