9
1
I am aware that the differences between high-end i7 and Xeon are:
- compatibility with dual socket motherboards,
- total L2/L3 cache size.
- overclocking capability.
- extra cores - Haswell i7-4960X vs E7.
- power consumption.
(See: What is the difference between an Intel i7 and a Xeon quad core processor?)
What I'd like to ask is, what are the implications of the cache size on achieving the lowest latency?
It seems that I'd much rather have a heavily-overclocked i7 Extreme than a Xeon processor and I cannot envision how an extra 5~15 MB cache will make a significant difference except for a few edge cases where the problem set is too large for the L2 cache and small enough that it doesn't have to rely on main memory. Am I missing some advantage of Xeon processors?
P.S.: We are deploying hardware for neutrino detectors for high-energy physics and need a very fast response time but not large matrix computations. Some will point out that I should be using some ASIC/RISC+RTOS strategy to achieve my goals, but the way the experiments are designed, I just need a 10 microsecond internal latency (hitting socket to egress), which seems possible with regular processors.
5Real bottleneck would be connection between computer port and detector. – huseyin tugrul buyukisik – 2013-06-11T21:11:02.210
3FPGA may be a better (and cheaper) solution for such application. – gronostaj – 2013-06-11T23:22:33.690
Can you be more precise about when the 10 microseconds starts and when it ends? As huseyin pointed out, minimizing hardware latency is going to be critical. General-purpose hardware and software is lousy at this. – David Schwartz – 2013-06-12T05:34:32.487
That is totally doable. But don't run windows. Xeon wouldn't be needed. Overclocking probably won't help as much as you think. – Matt H – 2013-09-10T00:09:35.947