20201114, 22:53  #45 
If I May
"Chris Halsall"
Sep 2002
Barbados
24BA_{16} Posts 

20201115, 00:30  #46 
"Mike"
Aug 2002
17406_{8} Posts 
Our study is a catfree environment. The rest of the house?

20201115, 00:52  #47  
"Mike"
Aug 2002
2×11×19^{2} Posts 
Quote:
Here is an explanation from https://www.reddit.com/r/overclockin...ram_overclock/ Quote:
Quote:
Old: https://www.gskill.com/product/165/1...35V16GB(2x8GB) New: https://www.gskill.com/product/165/1...5V32GB(2x16GB) We "erased" the motherboard's memory timings and had it go through the training process again. It ended up with the same numbers as before even though the signal/clock/whatever load is significantly increased. As an experiment, we then forced geardown mode off and the command rate to 1. It passed a severe memory check with that setting, but any gain we measured was lost in the runtorun variation of our benchmarks. IOW, we think the difference was negligible. So we enabled geardown mode to have a safety net for stability. We like fast things but only if they are utterly reliable. 

20201115, 00:54  #48 
"Mike"
Aug 2002
2·11·19^{2} Posts 
Attached are benchmark timings for our usual 560K and 6144K FFT lengths.
You may notice that the 560K FFT single rank and dual rank timings are very similar. We figure this is because the data is cached. The 6144K FFT data shows a surprising (to us) increase in throughput, up to 25% higher with six cores running. 
20201116, 05:41  #49 
"/X\(‘‘)/X\"
Jan 2013
5561_{8} Posts 
25% higher! Wow!
That does say something about how memory starved the 6 core chip is though, if rank interleaving can provide that much more bandwidth. Could you test that dual rank memory configuration at different CPU clock speeds? I'm curious where "knee" in performance is for the 6144k FFT. That could save a lot of power. 
20201116, 20:32  #50  
"Mike"
Aug 2002
2×11×19^{2} Posts 
Quote:
ECO = 40W STK = 57W PBO = 105W Code:
ECO Timings for 6144K FFT length (1 core, 1 worker): 18.20 ms. Throughput: 54.95 iter/sec. Timings for 6144K FFT length (2 cores, 1 worker): 9.67 ms. Throughput: 103.40 iter/sec. Timings for 6144K FFT length (3 cores, 1 worker): 6.94 ms. Throughput: 144.17 iter/sec. Timings for 6144K FFT length (4 cores, 1 worker): 5.55 ms. Throughput: 180.20 iter/sec. Timings for 6144K FFT length (5 cores, 1 worker): 4.85 ms. Throughput: 206.26 iter/sec. Timings for 6144K FFT length (6 cores, 1 worker): 4.30 ms. Throughput: 232.78 iter/sec. STK Timings for 6144K FFT length (1 core, 1 worker): 18.11 ms. Throughput: 55.21 iter/sec. Timings for 6144K FFT length (2 cores, 1 worker): 9.49 ms. Throughput: 105.32 iter/sec. Timings for 6144K FFT length (3 cores, 1 worker): 6.67 ms. Throughput: 149.83 iter/sec. Timings for 6144K FFT length (4 cores, 1 worker): 5.23 ms. Throughput: 191.16 iter/sec. Timings for 6144K FFT length (5 cores, 1 worker): 4.41 ms. Throughput: 226.98 iter/sec. Timings for 6144K FFT length (6 cores, 1 worker): 4.00 ms. Throughput: 249.94 iter/sec. PBO Timings for 6144K FFT length (1 core, 1 worker): 18.33 ms. Throughput: 54.55 iter/sec. Timings for 6144K FFT length (2 cores, 1 worker): 9.58 ms. Throughput: 104.38 iter/sec. Timings for 6144K FFT length (3 cores, 1 worker): 6.85 ms. Throughput: 146.04 iter/sec. Timings for 6144K FFT length (4 cores, 1 worker): 5.24 ms. Throughput: 190.98 iter/sec. Timings for 6144K FFT length (5 cores, 1 worker): 4.48 ms. Throughput: 223.26 iter/sec. Timings for 6144K FFT length (6 cores, 1 worker): 3.84 ms. Throughput: 260.43 iter/sec. 

20201116, 20:51  #51 
"/X\(‘‘)/X\"
Jan 2013
29·101 Posts 
That eco mode is super efficient! What clock speeds do you see running in eco?

20201116, 23:07  #52  
"Mike"
Aug 2002
1F06_{16} Posts 
Quote:


20201116, 23:18  #53 
"Composite as Heck"
Oct 2017
3×5×7^{2} Posts 
Ah this is the good stuff. If only there were a way to pipe benchmarks directly into a vein.

20201116, 23:46  #54 
Feb 2005
Colorado
577 Posts 

20201117, 00:03  #55 
"Composite as Heck"
Oct 2017
1011011111_{2} Posts 
I can quit whenever I want!
But there'll be more right... 