What do you care about in a Benchmark?

[Mussels]

What benchie program?

IBT (intel burn test) - was mostly using it to test out the cooling, but also helped educate the owner.

 

[Aquinus]

the key is to keep in mind that benchmarks are often not comparing apples to apples.

Of course not but, the point of that is to gauge relative performance but, people seem to forget that these cases tend to be highly workload specific and not very generalized to reflect the actual kind of performance given more than one characteristic about say, a CPU or a GPU or both. For example:

i just used a benchmark to demonstrate to somehow that their PC was upgraded, and worth it.

40 Gflops in an 8 core, vs 103 Gflops on a 4 core i5.

Pointed out that meant each core was 5x faster, and his games are single threaded. He got the hint.

You got that with IntelBurnTest, right? That uses 256-bit AVX which requires both FMA FPUs on a Bulldozer module acting as one which is a very specific part of the CPU carrying out a very specific function. Considering most workloads don't use AVX, it's why many other benchmarks don't show such a discrepancy (but rather to a much lesser extent.)

On the other hand, if you're going to be doing things that heavily rely on AVX, it's a great benchmark and an awesome way to determine if your FPUs are stable. It does very little to ensure that integer cores are stable though and almost completely cuts out integer performance which is arguably the most unrealistic way to measure a CPU's performance since FPU math is not done nearly as often as integer math, either on the ALU (numerical math,) or the AGU (memory location math.)

This is kind of the reason why I started this thread up because, if people are only concerned about a high score, you're missing the point of a benchmark in the first place. A benchmark means nothing if the load doesn't reflect what you'll be doing with it... which is why I asked the very question, "What do you care about in a benchmark?"

So, if you want to educate the person, tell them what the benchmark is testing and how it compares to others. Most people don't understand that AVX isn't commonly used and might not realize that IBT is testing one very particular feature on a CPU. It's kind of disingenuous to say AVX performance is a good gauge of general usage of a CPU though, in my opinion.

 

[Mr.Scott]

I'd love to see some benchmark that includes all:

- CPU single core
- CPU multi core
- Memory throughput (read, write)
- Storage performance (system drive: read, write, latency)
- Graphics performance
- Combined score so we can compare our systems

PCMark 05 had all of this.

 

[Mussels]

Of course not but, the point of that is to gauge relative performance but, people seem to forget that these cases tend to be highly workload specific and not very generalized to reflect the actual kind of performance given more than one characteristic about say, a CPU or a GPU or both. For example:

You got that with IntelBurnTest, right? That uses 256-bit AVX which requires both FMA FPUs on a Bulldozer module acting as one which is a very specific part of the CPU carrying out a very specific function. Considering most workloads don't use AVX, it's why many other benchmarks don't show such a discrepancy (but rather to a much lesser extent.)

On the other hand, if you're going to be doing things that heavily rely on AVX, it's a great benchmark and an awesome way to determine if your FPUs are stable. It does very little to ensure that integer cores are stable though and almost completely cuts out integer performance which is arguably the most unrealistic way to measure a CPU's performance since FPU math is not done nearly as often as integer math, either on the ALU (numerical math,) or the AGU (memory location math.)

This is kind of the reason why I started this thread up because, if people are only concerned about a high score, you're missing the point of a benchmark in the first place. A benchmark means nothing if the load doesn't reflect what you'll be doing with it... which is why I asked the very question, "What do you care about in a benchmark?"

So, if you want to educate the person, tell them what the benchmark is testing and how it compares to others. Most people don't understand that AVX isn't commonly used and might not realize that IBT is testing one very particular feature on a CPU. It's kind of disingenuous to say AVX performance is a good gauge of general usage of a CPU though, in my opinion.

i did comment that i primarily did it to test the cooling.

 

[Aquinus]

i did comment that i primarily did it to test the cooling.

You also commented that:

Pointed out that meant each core was 5x faster

Which is a straight up fabrication unless you only consider IBT. How many ways can you say "cherry picked."

 

[Recon-UK]

Single thread and GPU.

All else can suck my mechanical ball sack.

 

[Jetster]

I like well rounded benchmarks. Like OCCT I just want to stress the system in multiple ways to make sure it solid

 

[Mussels]

You also commented that:

Which is a straight up fabrication unless you only consider IBT. How many ways can you say "cherry picked."

i told him 5x faster per core in that benchmark, not that it'd be 5x faster in minesweeper or making his e-penis 5 feet longer. This same guy was running 1080i on a 1024x768 plasma HDTV, so he's not exactly going to understand the fine details.