Geekbench Blog
www.geekbench.com
Weird choice of baseline CPU and even weird is that the baseline score is 2500.
i7-12700 does hardly 2000 in GB5 with the fastest DDR5.
How many of these? https://www.inven.ai/company-lists/top-28-microprocessor-production-companiesAll CPU design companies I know are using GB because they know how to exploit its results (and because their customers want to know).
I don't know most of them. And did you notice AMD is missing and Arm is named Armagh
Attending a benchmark committee just once would allow you to answer thatI suppose there are more than a few good developers with deep knowledge of microprocessors. What is keeping such developers from creating a GB competitor with popular opensource software as tests? Could it be due to legal reasons? (a CPU performs badly in their benchmark and they get contacted by the CPU manufacturer's lawyers or something).
What science? There are lies, damn lies, and benchmarksAh... the science of benchmarks..
What shall we name it? Benchmarkology ?
I think that's why GB has a single developer? How is SPEC possible then? Must've been torturous meetingsAttending a benchmark committee just once would allow you to answer that
All what you say is applicable to all benchmarks, not only GB. Do you think it makes more sense to compare SPEC FP results of a CPU without AVX against a CPU with AVX?
It's not only funny, but again it's definitely useful for people developing CPUs. Properly used any benchmark shows you something. It's just not for the typical end-users (where your comment below applies), but, outside of CPU design, it is still useful if you know what you're looking at.So yes, comparing Geekbench scores may be pretty funny, but it's a useless metric to compare the CPU performance on different platforms. To a certain degree, you can compare those results within one platform, but the test suite should be objective. Geekbench 6 is not an objective test.
On that I definitely agree.Ultimately, the only viable metric is the actual performance of the apps you're using on the fixed power limit.
It's not only funny, but again it's definitely useful for people developing CPUs. Properly used any benchmark shows you something. It's just not for the typical end-users (where your comment below applies), but, outside of CPU design, it is still useful if you know what you're looking at.
OK that becomes more specific What do you consider a correct benchmark? Don't list properties, give me a name.Yes, benchmarks, in general, may be pretty useful, especially when they are correct.
That's another topic that doesn't apply to GB only.At the same time, many companies tend to manipulate that data for marketing purposes. Geekbench is the best example here.
Another field of manipulation is power efficiency.
What do you consider a correct benchmark?
I never use R23, becuase it's not good for comparing x86 vs ARM CPUs.I prefer using Cinebench R23 mostly because of its consistency, relatively long benchmark time, and great scalability.
AMD also used CB 2024 when comparing X Elite to Strix Point. It’s a much better benchmark to compare x86 and ARM CPUs.I never use R23, becuase it's not good for comparing x86 vs ARM CPUs.
I never use R23, becuase it's not good for comparing x86 vs ARM CPUs.
Having native support doesn’t mean it’s properly makes use of NEON, it just meaning it’s an arm64 application. Anyway, CB23 does not fully use SIMD pipelines on Apple CPUs.Cinebench R23 has native support for Apple Silicon
I prefer using Cinebench R23 mostly because of its consistency, relatively long benchmark time, and great scalability.
Cinebench 2024 fixes the issues that CB R23 had.
By the way, I find it really difficult to understand how some can prefer Cibenench to Geekbench as a general-purpose benchmark.
Cinebench R23 used an automatic translation layer to support NEON. It is a joke. Much worse than GB6.Cinebench R23 has native support for Apple Silicon, and it does not use specific extensions missing on some platforms (SME, AVX512). So, unlike Geekbench v6, you can compare those results between platforms.
As for Cinebench R24, it's less representative from my perspective but also pretty consistent.
I have been saying for months that the only GB6 test that seems to scale is Ray Tracer. Here we see that 7950x scales by about 19, while 9754 scales by 108.According to Geekbench, 16-core AMD Ryzen 9 7950X is 32% faster in the multi-core test than AMD EPYC 9754 with 128 cores:
View attachment 100597
Again. 16 cores are faster than 128 cores using the same architecture. Does it match the reality?
AMD's Larrabee killer!9754 scales by 108.
I have been saying for months that the only GB6 test that seems to scale is Ray Tracer.
I know you're kidding, but claiming GB6 is a poor benchmark for MT just proves the point I've made today multiple times: if you don't properly use benchmark results, you get wrong answers. Don't blame the benchmark, blame the people that are cluelessGeekbench 6 is a great multi-core benchmark!
I agree but there's no harm in telling everyone that comparing GB composite MT score is very limited.I know you're kidding, but claiming GB6 is a poor benchmark for MT just proves the point I've made today multiple times: if you don't properly use benchmark results, you get wrong answers. Don't blame the benchmark, blame the people that are clueless