Core Ultra 3 205 flaunts 30% higher multi-core performance over Core i3-14100 in early benchmarks — extra E-cores seemingly contribute to uplift

I have a 13100T works great maxed out the 6700xt no problems... some games need more coreeeeeees powa.
but for 99% will be great for a budget systems.

And intel made a new cpu for a dead platform :S holly S.

So doubling the number of cores yields better multicore score ...

adamXpeter said:

So doubling the number of cores yields better multicore score ...

Except Geekbench's multicore score is garbage. It doesn't mean what you think it does and has little to do with the number of cores.

People simply shouldn't use it. Its name suggests it's a synthetic multi-core score, like Cinebench. But, it's really designed to be more of a "typical application benchmark", like PC Mark or whatever.

Geekbench MT is sometimes more sensitive to differences in memory configuration than even core count.

bit_user said:

Except Geekbench's multicore score is garbage. It doesn't mean what you think it does and has little to do with the number of cores.

People simply shouldn't use it. Its name suggests it's a synthetic multi-core score, like Cinebench. But, it's really designed to be more of a "typical application benchmark", like PC Mark or whatever.

Geekbench MT is sometimes more sensitive to differences in memory configuration than even core count.

Yes people, for heavens sake don't use a benchmark that actually shows you how much performance you as a home owner will get out of your system, only look at server benchmarks.

Geekbench launches one task that uses all cores instead of running something that can break up its work into individual parts that don't need anything from any other part.
It will not show you how good of an server CPU you have so why in the world would you ever even look at it?!

(Emphasis by me)
https://www.geekbench.com/doc/geekbench6-benchmark-internals.pdf

Multi-ThreadingGeekbench 6 uses a “shared task” model for multi-threading, rather than the “separate task” model used in earlier versions of Geekbench. The “shared task” approach better models how most applications use multiple cores. The "separate task" approach used in Geekbench 5 parallelizes workloads by treating each thread as separate. Each thread processes a separate independent task.
This approach scales well as there is very little thread-to-thread communication, and the available work scales with the number of threads.
For example, a four-core system will have four copies, while a 64-coresystem will have 64 copies. The "shared task" approach parallelizes workloads by having each thread processes part of a larger shared task. Given the increased inter-thread communication required to coordinate the work between threads, this approach may not scale as well as the "separate task" approach.

TerryLaze said:

Yes people, for heavens sake don't use a benchmark that actually shows you how much performance you as a home owner will get out of your system, only look at server benchmarks.

I have no problem with benchmarks that are clearly advertised to measure typical application performance.

The issue I have is with a synthetic benchmark that claims to measure "Multithreaded performance". People expect such benchmarks to scale will with the number of cores. Geekbench MT doesn't, because they explicitly designed it not to!

TerryLaze said:

Geekbench launches one task that uses all cores instead of running something that can break up its work into individual parts

Taking that statement out of context is misleading. The intent of that excerpt is to contrast itself to "shared-nothing" benchmarks, like SPEC_rate=n, which is fair. They're not saying all parts of GB6 MT use all cores. If it did, then the result shown for the Intel Core Ultra 3 205 would mean it's a trash CPU, and I'm sure you don't think that!

If you actually read that doc, they go on to describe how the MT benchmark is a mix of heavy and lightly-threaded tasks, because their intent was not to measure how well a system scales to all-core workloads, but rather to be more of a "typical MT application" benchmark.

TerryLaze said:

It will not show you how good of an server CPU you have so why in the world would you ever even look at it?!

People look to MT benchmarks as a guide of all-core performance, because there are lots of things that actually people do (like compiling code) on client CPUs that use those cores. Otherwise, why would client CPUs have so many cores and why would anyone buy them?

bit_user said:

I have no problem with benchmarks that are clearly advertised to measure typical application performance.

The issue I have is with a synthetic benchmark that claims to measure "Multithreaded performance". People expect such benchmarks to scale will with the number of cores. Geekbench MT doesn't, because they explicitly designed it not to!

That's confirmation bias, people that are normal people and don't have a 30+ year history of working with PCs expect a benchmark to tell them how much performance they might get out of one CPU compared to another.

bit_user said:

People look to MT benchmarks as a guide of all-core performance, because there are lots of things that actually people do (like compiling code) on client CPUs that use those cores. Otherwise, why would client CPUs have so many cores and why would anyone buy them?

What year do you live in?!?
Have you ever seen an average "client CPUs" desktop in action?
They need that many cores just for all the web adds and bloatware alone.