On Mon, 25 Aug 2014 05:43:04 -0700 (PDT) RFat <raan...@gmail.com> wrote:
> Hi all, > > I am trying to overclock the A31 higher than 1008MHz Are you fine with the deterioration of reliability and/or overheating under high load? > and I am getting deterioration in performance. > > The way I work is execute a small program from u-boot which redefines the N > multiplayer of PLL1 and then runs some CPU-intentive routine (to measure > performance). > > Here are the results I get (the N multiplier is given in the small brackets > starting from 18): > > CPU Freq = 864Mhz (18) > run in 10Hz, 94ms > PLL1 is ENABLED P:2 M:1 K:3 N:18 Sigma:0 Lock:0 Freq:432Mhz coreclk:864Mhz > CPU Freq = 912Mhz (19) > run in 11Hz, 89ms > PLL1 is ENABLED P:2 M:1 K:3 N:19 Sigma:0 Lock:0 Freq:456Mhz coreclk:912Mhz > CPU Freq = 960Mhz (20) > run in 11Hz, 85ms > PLL1 is ENABLED P:2 M:1 K:3 N:20 Sigma:0 Lock:0 Freq:480Mhz coreclk:960Mhz > CPU Freq = 1008Mhz (21) > run in 12Hz, 81ms > PLL1 is ENABLED P:2 M:1 K:3 N:21 Sigma:0 Lock:0 Freq:504Mhz coreclk:1008Mhz > CPU Freq = 1056Mhz (22) > run in 12Hz, 77ms > PLL1 is ENABLED P:2 M:1 K:3 N:22 Sigma:0 Lock:0 Freq:528Mhz coreclk:1056Mhz > CPU Freq = 1104Mhz (23) > run in 9Hz, 101ms > PLL1 is ENABLED P:2 M:1 K:3 N:23 Sigma:0 Lock:0 Freq:552Mhz coreclk:1104Mhz > CPU Freq = 1152Mhz (24) > run in 9Hz, 101ms > PLL1 is ENABLED P:2 M:1 K:3 N:24 Sigma:0 Lock:0 Freq:576Mhz coreclk:1152Mhz > CPU Freq = 1200Mhz (25) > run in 9Hz, 101ms > PLL1 is ENABLED P:2 M:1 K:3 N:25 Sigma:0 Lock:0 Freq:600Mhz coreclk:1200Mhz > CPU Freq = 1248Mhz (26) > run in 9Hz, 101ms > PLL1 is ENABLED P:2 M:1 K:3 N:26 Sigma:0 Lock:0 Freq:624Mhz coreclk:1248Mhz > CPU Freq = 1296Mhz (27) > run in 9Hz, 101ms > PLL1 is ENABLED P:2 M:1 K:3 N:27 Sigma:0 Lock:0 Freq:648Mhz coreclk:1296Mhz > CPU Freq = 1344Mhz (28) > run in 9Hz, 101ms > PLL1 is ENABLED P:2 M:1 K:3 N:28 Sigma:0 Lock:0 Freq:672Mhz coreclk:1344Mhz > CPU Freq = 1392Mhz (29) > run in 9Hz, 101ms > PLL1 is ENABLED P:2 M:1 K:3 N:29 Sigma:0 Lock:0 Freq:696Mhz coreclk:1392Mhz > > As you can see, after 1008 (or 1056) there's a _decrease_ in performance. I can see that the performance flattens out and remains at the same 101ms level for your CPU-intensive routine. If you are running the code in u-boot from SRAM and the L1 cache is not enabled, then the instructions fetch may become a bottleneck. > Does anyone knows how higher performance can be achieved? Maybe buy a different higher performance SoC or better optimize the software? -- Best regards, Siarhei Siamashka -- You received this message because you are subscribed to the Google Groups "linux-sunxi" group. To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
