I am aware that the differences between high-end i7 and Xeon are:

  1. compatibility with dual socket motherboards,
  2. total L2/L3 cache size.
  3. overclocking capability.
  4. extra cores - Haswell i7-4960X vs E7.
  5. 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.

  • 6
    Real bottleneck would be connection between computer port and detector. – huseyin tugrul buyukisik Jun 11 '13 at 21:11
  • 3
    FPGA may be a better (and cheaper) solution for such application. – gronostaj Jun 11 '13 at 23:22
  • 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 Jun 12 '13 at 5:34
  • That is totally doable. But don't run windows. Xeon wouldn't be needed. Overclocking probably won't help as much as you think. – hookenz Sep 10 '13 at 0:09

Anandtech did a detailed review of the Core i7 4960x recently.

Memory Latency vs. Access Range

But honestly, these times don't mean all that much without knowing what kind of algorithm you're using. Assuming there isn't that much processing to be done, then the ethernet connection is going to be the real bottleneck, like @huseyin said


Xeon Processors only advantage over i7 is the Memory Handling capacity and its resilience. i7 has certainly more clock rate and it cannot handle large amount of memory. Extra cache just makes the Physical Memory able to handle large data and i/o processing. Also Hyper-threading does that for Intel. It will show a difference only if you are going to directly control processor by your algorithm.

If you are hesitant to use ASIC/RISC+RTOS strategy then you can go forward with any Linux Based 64-bit OS on Intel® Core™ i7-990X Processor Extreme Edition (12M Cache, 3.46 GHz, 6.40 GT/s Intel® QPI) the best offered by Intel.

Also if you are thinking to use the processor for large amount of data in near future my pick will be definitely Intel® Xeon® Processor E7-8870 (30M Cache, 2.40 GHz, 6.40 GT/s Intel® QPI) . this is having a lesser clock rate and hence more latency.


I'd recommend a Nation Instruments cRIO. It has both an Intel i7 CPU and an FPGA. It can run either Windows, Linux or an RTOS. You can program it in a high level language and run code either on the i7 or on the FPGA. The low level hardware connectivity to your sensors will give you far lower latency than any PC can.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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