Same memory speed and latencies? The E5-2650 is newer, it might be simply better at multithreading, there are small improvements with every generation. The IPC increased also significantly between those generations, you see that on the lower CPU time, the difference much larger than the 6.25% difference between the turbo clocks (if they are running at base clocks, that difference is even larger). So the E5-2650 is doing more work in the same amount of time and that means, even if it has one more channel, the slow down caused by a cache miss might be more significant and that makes more threads per WU faster. Well, just a guess. The results for 2-4 threads are also very close, 1% for the E5-2650 and 2.5% for the X5660, would be interesting to see how they change with a different, longer running WU, maybe that will give a more clear result.

---

Same memory speed and latencies?

1866 MT/s in the E5-2650 v2, 1333 MT/s in the X5660. Not sure about latencies.

Here are some numbers from my dual Xeon E5-2670 (2 sockets, 16 cores, 32 threads, 40 MB L3 cache):

WU: de_nbody_orbit_fitting_10_01_2025_v193_OCS_lmc_pm__data__02_1758550502_1411512
Application version: 1.93

        threads/WU     run time     CPU time     threads in use        WU/day
ST             1      7980         7969           2× 16× CPU           346.5
MT             2      3719         7184           2× 16× CPU           371.7
MT             4      1868         6848           2× 16× CPU           369.9
MT             8       994         6654           2× 16× CPU           347.8
MT            16       593         6628           2× 16× CPU           291.3
MT            32       404         6685           2× 16× CPU           214.0

This one surprised me by being so much slower than the E5-2650 v2, which I had thought was basically the same performance with better power efficiency.

According to this and this, the Intel Xeon E5-2650 v2 has 870 millions transistors less than the Intel Xeon E5-2670, perhaps they removed all those, which were slowing down everything. ;-)

Just kidding, but that indicates significant change, not just a die shrink from 32 to 22 nm.

---