The read performance hit a little bit of a bump at 2.67GHz to 2.93GHz, but improves again at 3.2GHz - Everest was consistent in reporting this and we can allude it's possibly due to core-uncore ratio changes, perhaps: 1:1.23 works better than 1:1.1. Overall though there are significant increases all round, although even though memory controller frequency elevation does give it a lower latency, as we saw previously, optimising the memory timings works out better in that equation.
L3 Cache Performance
IMC: CPU L3 Cache Read Test
Everest 4.60 beta
CPU Uncore Clock: 2133MHz
CPU Uncore Clock: 2667MHz
CPU Uncore Clock: 2933MHz
CPU Uncore Clock: 3200MHz
24778.0
24932.0
25343.0
24912.0
0
5000
10000
15000
20000
25000
MB/s (higher is better)
IMC: CPU L3 Cache Write Test
Everest 4.60 beta
CPU Uncore Clock: 2133MHz
CPU Uncore Clock: 2667MHz
CPU Uncore Clock: 2933MHz
CPU Uncore Clock: 3200MHz
15313.0
18032.0
19028.0
20236.0
0
5000
10000
15000
20000
MB/s (higher is better)
IMC: CPU L3 Cache Copy Test
Everest 4.60 beta
CPU Uncore Clock: 2133MHz
CPU Uncore Clock: 2667MHz
CPU Uncore Clock: 2933MHz
CPU Uncore Clock: 3200MHz
23667.0
24958.0
25428.0
25753.0
0
5000
10000
15000
20000
25000
MB/s (higher is better)
IMC: CPU L3 Cache Latency Test
Everest 4.60 beta
CPU Uncore Clock: 2133MHz
CPU Uncore Clock: 2667MHz
CPU Uncore Clock: 2933MHz
CPU Uncore Clock: 3200MHz
4.1
4.0
4.3
3.9
0
1
2
3
4
MB/s (higher is better)
Just to confirm our theory above - the latency of a 2.93GHz CPU Uncore gave us a consistently 0.3ns higher latency than 2.67GHz, indicating the other three are slightly more optimised for Core-Uncore data transfer. Increasing the Uncore frequency helps L3 cache writes hugely - by over 32 percent - but it does little for copy performance and even less for read.
Prev
