F5F Stay Refreshed Hardware Desktop Looking for tools to evaluate IPC on certain Intel processors? Software recommendations are available.

Looking for tools to evaluate IPC on certain Intel processors? Software recommendations are available.

Looking for tools to evaluate IPC on certain Intel processors? Software recommendations are available.

Pages (2): Previous 1 2
S
196
06-28-2018, 12:11 AM
#11
The reference was briefly mentioned earlier, but the exact detail is unclear to me
S
StampyKittenNZ
06-28-2018, 12:11 AM #11

The reference was briefly mentioned earlier, but the exact detail is unclear to me

A
achoo_ca
Member
69
06-28-2018, 12:11 AM
#12
Conducted preliminary checks on the 12100F. For stable timing I disabled turbo, keeping it at the standard 3.3 GHz base. I also experimented with running it with and without hardware acceleration, while monitoring power usage. This aims to understand the impact of performance enhancements versus energy costs. Early reports suggest hardware acceleration could boost performance more than the power increase would suggest. The system uses a single-channel dual-rank 3200 RAM module; I haven’t verified if these tests have any restrictions. Cinebench R15 showed a 40% score improvement with a 19% rise in power consumption. Y-cruncher recorded a 13% gain at both 25m and 1b, while 14% was seen in the 1b run. Power usage jumped to 14% higher on the 25m test. POV-Ray noted an 18% performance lift with a 17% power increase. Linpack Xtreme reported only a 1.5% throughput boost without any power change. For compression tools, 7-Zip improved speed by 17% and power use by 47%, though results varied. Cinebench R15 and decompression saw strong gains from HT, but POV-Ray, Y-cruncher, and 7-Zip compression didn’t see major perf improvements. Linpack Xtreme and Prime95 showed no change in performance, with Prime95’s P95 dropping significantly. Actual power draw varied: Cinebench R15 at 35W, Y-cruncher at 40W, POV-Ray at 35W, Linpack at 37W, 7-Zip at 29W, Prime95 at 45W (41W without HT). This explains why Cinebench R15 is labeled a "light" task and Prime95 a "heavy" one. If you set a fixed power cap instead of a fixed clock, heavier jobs may run slower than lighter ones. Other notes: I skipped Cinebench 2024 due perceived slowness; I might add it later. R23 isn’t readily available as a standalone app, so I avoided installing BenchMate. Blender benchmark launcher wouldn’t download benchmarks, so they’re missing. I planned to test HWBOT x265 for video encoding, but it won’t run without HPET and I didn’t enable it.
A
achoo_ca
06-28-2018, 12:11 AM #12

Conducted preliminary checks on the 12100F. For stable timing I disabled turbo, keeping it at the standard 3.3 GHz base. I also experimented with running it with and without hardware acceleration, while monitoring power usage. This aims to understand the impact of performance enhancements versus energy costs. Early reports suggest hardware acceleration could boost performance more than the power increase would suggest. The system uses a single-channel dual-rank 3200 RAM module; I haven’t verified if these tests have any restrictions. Cinebench R15 showed a 40% score improvement with a 19% rise in power consumption. Y-cruncher recorded a 13% gain at both 25m and 1b, while 14% was seen in the 1b run. Power usage jumped to 14% higher on the 25m test. POV-Ray noted an 18% performance lift with a 17% power increase. Linpack Xtreme reported only a 1.5% throughput boost without any power change. For compression tools, 7-Zip improved speed by 17% and power use by 47%, though results varied. Cinebench R15 and decompression saw strong gains from HT, but POV-Ray, Y-cruncher, and 7-Zip compression didn’t see major perf improvements. Linpack Xtreme and Prime95 showed no change in performance, with Prime95’s P95 dropping significantly. Actual power draw varied: Cinebench R15 at 35W, Y-cruncher at 40W, POV-Ray at 35W, Linpack at 37W, 7-Zip at 29W, Prime95 at 45W (41W without HT). This explains why Cinebench R15 is labeled a "light" task and Prime95 a "heavy" one. If you set a fixed power cap instead of a fixed clock, heavier jobs may run slower than lighter ones. Other notes: I skipped Cinebench 2024 due perceived slowness; I might add it later. R23 isn’t readily available as a standalone app, so I avoided installing BenchMate. Blender benchmark launcher wouldn’t download benchmarks, so they’re missing. I planned to test HWBOT x265 for video encoding, but it won’t run without HPET and I didn’t enable it.

Pages (2): Previous 1 2