What do you guys use for reliable benchmarks for modernish CPU's?

[Unknown_K]

For example, I have an E5 2667 V3 XEON (8C/16T) and was comparing it to a desktop I7-5960X (8C/16T) same socket LGA 2011-3

https://cpu.userbenchmark.com/Compare/Intel-Xeon-E5-2667-v3-vs-Intel-Core-i7-5960X/m505323vs2580

https://www.cpubenchmark.net/compar...old-G5420-vs-Intel-i7-5930K-vs-Intel-i7-5960X

Passmark says that they are pretty much the same in single and multi-core, Userbench says the I7 is 29% faster on average (which would be very noticeable).

 

[GiGaBiTe]

Userbench should be taken with a grain of salt, their benchmark suite is very basic and has a ton of variability since they don't enforce any testing consistency.

For all we know, the specific test they used for the compare could have been run on a heavily loaded machine and not been able to perform at its best. Or the test could have been run on a motherboard that was abusing the turbo expiry and letting the CPU run in turbo indefinitely.

Both CPUs are from the Haswell generation and should perform virtually identically on a clock for clock basis. The E5-2667v3 should be the faster part because it has both a higher base clock (3.2 vs 3.0 GHz) and a higher boost clock (3.6 vs 3.5 GHz.)

Passmark has a more believable benchmark score, 29% faster from Userbenchmark is nonsense. That sounds like Intel cutting off the bottom of a bar chart and zooming in 1000x to make a 1% difference look like something orders of magnitude different.

 

[Chuck(G)]

Back in the day, we used to misquote Mark Twain as "There are lies, damned lies and benchmarks".
If you want to test raw math compute power, there's a large database of LINPACK resutls, going back years: https://www.top500.org/project/linpack/

 

[lowen]

Since I run Linux I use the old Byte benchmarks as a baseline. To get meaningful increases that can actually be felt at the UI the speed needs to be at least double. That's why I've used a rule of three in upgrading for several years; the machine has to be at least three times faster for me to pay the price to upgrade. So I went from a Core 2 Duo (Precision M4300) to an i7-740M (Precision M6500), then to an i7-3740M (Precision M6700), and now an i7-9750H (Precision 7740). The biggest jumps have been going from the fast HGST 7200 RPM spinning drives in the M4300 to mSATA SSD and then to NVMe in the 7740.

I was given a System76 LemurPro (10), which has an i7-1165G7. This 11th-gen i7 is slower on many tasks than the 9th-gen (like compiling SDRangel and then running SDRangel; on the 7740 this takes 15 minutes, on the LemurPro it takes 23), but that's mostly due to the larger number of cores in the i7-9750H. A cheap i3 or lower end i5 of the latest generation will not be any faster than the i7-9750H; in fact, it will very likely be significantly slower. I wouldn't have paid for the LemurPro based solely on my rule of three, but I will definitely accept the gift of it.

The Precision 7740 is a beast; it's going to be quite a while before systems that are three times faster come down in price enough to make it worthwhile. EDIT: all these systems were bought used; the 7740 is a 2019 model and I purchased in 2021 for less than one-fifth MSRP; it had been a corporate lease unit, rotated out after two years.