Usual Suspects 🙂
Geekbench 3/4, Luxmark, Cinebench r15, 3D Mark FS Standard. And if you have some game with Benchmark built-in.
I sorted them out based on expected ability to show difference.
All done. Run/tested multiple time to eliminate the variations as much possible.
Also checked with a few other program/benchmarks but difference was either none or negligible.
So, some of the results below I would consider them a "best case" for pointing out differences.
But, in others, yeah, they are clearly there.
P.S: And excuse my horrendous table/format results. 😛
All done. Run/tested multiple time to eliminate the variations as much possible.
Also checked with a few other program/benchmarks but difference was either none or negligible.
So, some of the results below I would consider them a "best case" for pointing out differences.
But, in others, yeah, they are clearly there.
Great Work. Geekbench is most interesting. Some of workloads are practically unaffected, and some show even 25% scaling up from 2400 to 3200MHz.
Those which did not scale up are workloads which did not hit memory bandwidth limitation.
Those which did scale up require good bandwidth. Now I wonder for one last test:
May you please disable 2cores in BIOS? And run 2400/3200MHz memory + 4C/8T as controlling sample. (Geekbench only)
Looking at ST SGEMM which is in either case 133GFlops, and then MT scales with bandwidth to:
2400MHz - 504
3200MHz - 630
4266MHz - 710
I wonder if just 4C/8T drop it to theoretical 133*4 = 532. (Maybe geekbench 4 started to support SMT, maybe not.)
Great Work. Geekbench is most interesting. Some of workloads are practically unaffected, and some show even 25% scaling up from 2400 to 3200MHz.
Those which did not scale up are workloads which did not hit memory bandwidth limitation.
Those which did scale up require good bandwidth. Now I wonder for one last test:
May you please disable 2cores in BIOS? And run 2400/3200MHz memory + 4C/8T as controlling sample. (Geekbench only)
Looking at ST SGEMM which is in either case 133GFlops, and then MT scales with bandwidth to:
2400MHz - 504
3200MHz - 630
4266MHz - 710
I wonder if just 4C/8T drop it to theoretical 133*4 = 532. (Maybe geekbench 4 started to support SMT, maybe not.)
Results table in the previous post updated. A GB links also at the end for the 4C/8T.
At 2400MHz the results varied a bit, but I would say you were spot on. 🙂
jaggerwild:
Stop highjacking the thread, with your AMD blow fish lips! Post your numbers when they get a CPU, highjack that thread no mods in here?
If they did have an 8 core, maybe they did the sensible thing and got rid of the integrated GPU. A very considerable number of people with Intel processors use them for gaming, so they get the higher spec CPU's. They then use a discrete graphics card and the GPU goes to waste. For businesses, they tend to buy the 'lesser' processors, those that gamers won't buy, so integrated GPU's for them makes sense.
I like my Intel HD
Total power-in of my PC, while idling on desktop, browsing Guru3D.
75W on Intel HD (via HDMI input/output cable)
125W on R9 290 (via display port 1.2) (mem goes between 300-1250MHz, but this doesn't seem to be culprit)
------------------
That's cool 50W difference just one click away:
https://abload.de/img/screenshot2018-03-281bzsfx.png
Results table in the previous post updated. A GB links also at the end for the 4C/8T.
At 2400MHz the results varied a bit, but I would say you were spot on. 🙂
https://s7.postimg.org/teqon47iz/05_Home_Alone.jpg
Thanks. Tiny Variation does not matter at all. Differences between each configurations are incomparably bigger.
https://browser.geekbench.com/v4/cpu/compare/7655359?baseline=7637463
Left: 4C/8T + 3200MHz memory ; Right: 6C/12T + 2400MHz memory.
Shows how starved even 6C/12T can be with 2400MHz. I did series of calculations and following are results:
Approximated based on performance scaling from 4C/8T+3200MHz to 6C/12T+2400MHz; 6C/12T+3200MHz; 6C/12T+4266MHz. While considering average 35% scaling as target (instead of 50%).
Memory clock [MHz] to extract close to maximum in given Workload from all threads.
6C/12T 8C/16T
AES 4898 6531
LZMA 2863 3818
JPEG 2643 3524
Canny 4422 5896
Lua 2221 2961
Dijkstra 4385 5847
SQLite 2579 3439
HTML5 Parse 2862 3816
HTML5 DOM 4562 6083
Histogram Equalization 3826 5101
PDF Rendering 4891 6521
LLVM 2390 3186
Camera 2852 3803
SGEMM 4161 5548
SFFT 2173 2898
N-Body Physics 2165 2887
Ray Tracing 2411 3215
Rigid Body Physics 2206 2941
HDR 4209 5613
Gaussian Blur 2863 3818
Speech Recognition 4552 6070
Face Detection 2184 2912
Memory clock [MHz] to extract close to maximum in given Workload from all threads. 6C/12T 8C/16T AES 4898 6531 LZMA 2863 3818 JPEG 2643 3524 Canny 4422 5896 Lua 2221 2961 Dijkstra 4385 5847 SQLite 2579 3439 HTML5 Parse 2862 3816 HTML5 DOM 4562 6083 Histogram Equalization 3826 5101 PDF Rendering 4891 6521 LLVM 2390 3186 Camera 2852 3803 SGEMM 4161 5548 SFFT 2173 2898 N-Body Physics 2165 2887 Ray Tracing 2411 3215 Rigid Body Physics 2206 2941 HDR 4209 5613 Gaussian Blur 2863 3818 Speech Recognition 4552 6070 Face Detection 2184 2912