Same CPU family and ghz/mhz but less cache.

[mgv]

Hello I find the single threaded benchmark page of great interest since that is mostly what I focus in on when developing my software for third parties that don't use multiple threads/cpus/etc:

http://www.cpubenchmark.net/singleThread.html

HOWEVER! The page seems to run some type of benchmark in which larger cache sizes skew the results quite heavily. For example the e5-1660v2 at 2030 handily beats the e5-1620v2 at 1892. (nearly a 7% difference). However in theory since both are the same cpu family and the same mhz/ghz (3.7ghz) then they would have the same results for low memory footprint single-threaded processes.

This is where my weakness in process benchmarks falls apart. Could someone please confirm that the benchmark is skewed towards higher CPU caches and that a low memory footprint (I.E. 1024k process using 1024k in ram) would not benefit whatsoever running on the e5-1660v2 vs the e5-1620v2 for example.

Obviously this makes a material difference in cost for my clients! ($310 vs $1100 according to the benchmark website)

Thanks much for the expertise!

[mgv]

Upon closer examination I'm actually starting to think that I may have answered my own question and that this is more turbo mode related. For example how could the e3-1290v2 smoke the e5-1660v2 ? (both at 3.7ghz, but e3-1290v2 has smaller cache). I think the answer to that question and my question above is that the turbo mode is superior (3.9-4.1 e3-1290 v2,3.8-4.0 e5-1660 v2, 3.7-3.9 e5-1620 v2), not so much that one has a bigger cache. I suppose then, that the "drill down" cpu benchmark page has some stats that are a bit misleading and should be corrected. For example the e5-1660 v2 says "Turbo Speed: 4.0 GHz" and the e3-1290 v2 says "Turbo Speed: 3.8 GHz" when it is actually the e3-1290 that has the 3.9ghz (@4 cores) and the e5-1660 that has the 3.8ghz (@4 cores). (even at 1 core the e3-1290 is at 4.1ghz, and the e5-1660 is at 4.0ghz)

That is, unless someone would like to offer a different explanation as to why the e3-1290 v2 is beating out all the other non-haswell CPUs in the benchmark. (perhaps a pipeline or architecture difference)

Thanks again.

[David (PassMark)]

Yes the main differences between the E5-1660V2 and E5-1620V2 CPUs is cache size, turbo clock speed & number of cores. Number of cores isn't important for the single threaded test, but the turbo clock speed is. As when only one thread is running one would expect the max turbo speed to the reached. So with 50% extra cache and 2.6% higher clock speed I think it is reasonable to expect that the E5-1660V2 is faster.


The massive price difference is way out of proportion to the small performance increase however.


I have fixed up the incorrect Turbo speed that was listed for the E3-1290 v2. If you notice any others that are wrong, please let me know.