Firstly, awesome project! I am so impressed at how far this project has come in just a week. I want to thank @ir88ed
, good work guys!
I did some testing with my 14 core chip today. I tested a few different variables to see what effects the total run time. The results are surprising to say the least.
Here is my test setup:
Core/Thread Testing i9 7940X14 Cores / 28 Threads = 93.93 seconds14 Cores / 14 Threads = 93.26 seconds2 Cores / 2 Threads = 93.29 seconds
- New install of W8.1 64
- Office Pro 2010
In the pictures below you can see, there is basically no scaling potential with core/thread count. If you watch the cores while the bench is running you will see that only 50% of 1 core is being utilized. The score difference is negligible.Core Test Screenshots (Click to show)
Amount of Memory31.8 Gb usable by OS = 93.93 seconds2.4 Gb usable by OS = 93.68 seconds
I am using quad channel (8x4) for both tests. To test less memory I used a console command to limit the amount of memory the OS can use. I reused the 32GB score from my core testing since its the same 14/28 for the baseline. As you can see, there is basically no scaling potential with the physical amount of memory. The score difference is negligible.Amount of Memory Screenshots (Click to show)
Memory TimingsCL16-16-16-36 2T = 93.68 secondsCL12-11-11-24 1T = 91.4 seconds
31.8Gb System Memory
2.41Gb System Memory
For this test I will compare the timings only. The DRAM frequency will be 3600 for both tests. The test only compares super tight timings vs XMP timings. In order to run super tight memory with high frequency in 64bit OS's, the amount of memory used by the OS needs to be less than 4gb. For this test I will reuse my result from the memory quantity testing as the baseline.
Finally we see a gain!! This is logical, we expect the performance to be increased with tight timings.Memory Timings Screenshots (Click to show)
3600 CL16-16-16-36 2T
3600 CL12-11-11-24 1T
Thats all for now, but I have more tests planned for the future. Next time I run some LN2 on 7740X, I will try and run this bench at high frequencies and see what happens