Intel Core Ultra 9 285K “Arrow Lake” CPU Blazes Past Core i9-14900KS & Ryzen 9 9950X

[AnotherReader]

Since geekbench is stupid and puts waayy too much AI stuff into its score weighting, lets show it broken down:

There doesn't really seem to be all that much difference in most benches:

View attachment 360301
Though it depends on where you look, I've seen some with worse performing 14900KS.

Broken down multi-core (reminder: ASUS one is 285K):

View attachment 360302
View attachment 360303
There's definitely been an improvement in some areas, but there seems to be some regressions. Background Blur is one example where the 14900KS seems to be a *lot* better at?
Clang is probably the best representation of synthetic performance, and that seems to be ~15% better.

Most improvements seem to be all on the E-Core front, as single-core is almost identical, though Clang on single-core is also a fair bit better.


Honestly I'd rather wait for Cinebench scores to make a conclusion.

If we take the scores at face value the overall improvement in Geekbench could skew towards zero.

Clang is a good analog for real world branchy code. On the other hand, the HTML5 browser score is significantly lower than Raptor Lake and that's reflective of the most common workload for most people.

 

[Bobaganoosh]

Since geekbench is stupid and puts waayy too much AI stuff into its score weighting, lets show it broken down:
...

I don't use GB that much, so thanks for showing that you can look at breakdowns. I took a look at my OC'd 13900k vs that 285k. This is a 13900k running at 6GHz on 1-4 core-loads, 5.8, 5.7, 5.6, 5.5GHz as you increase from 5-8 cores active. e-cores at 4.6GHz. 6000MT/s cl36 memory (it's a z690 that can't really do faster memory).

Looking at the comparison, it's not too far off your 14900ks, but I'm not sure if there's a big difference between GB 6.2.1 vs 6.3.0. One thing that seems confusing to me is that the 14900ks has a lower SC score, but it should be boosting to 6.2GHz, right? e-cores at 4.5GHz and P-core all-core at 5.7GHz stock config should end up outperforming mine, but it is almost the same.

Anyway, compared to the 285k I'm curious if any of the tests specifically benefit (or are hindered) from hyperthreading. HTML5 Browser for example, the 285k is ~12% faster in the SC test, but the 13900k is ~12% faster at multi-core. Object Detection is another interesting one where the results are within 2% on single-core, but ~13% better score on 285K.

Overall, I wouldn't put too much weight on this result either. I'd want to see if there's any actual difference in gaming or gaming-specific benchmarks. So far, I don't see anything out or out soon that makes me want to deal with a new MB, Memory, water block and CPU lol.

 

[Melvis]

it's 24 core

Really? Gosh ok! more and worse then I thought! Thats Threadripper Territory! HDT Chip?

 

[Zach_01]

Really? Gosh ok! more and worse then I thought! Thats Threadripper Territory! HDT Chip?

Yes but core count is not the only qualification to call a CPU HEDT.
PCIE lanes, memory channels… to name a few.
Remember that 16 out of those 24 are E-cores. Overall it’s like 12-14P-cores.
3-4 E-cores = 1 P-core maybe?

 

[Melvis]

Yes but core count is not the only qualification to call a CPU HEDT.
PCIE lanes, memory channels… to name a few.
Remember that 16 out of those 24 are E-cores. Overall it’s like 12-14P-cores.
3-4 E-cores = 1 P-core maybe?

Yeah I know, I just mean with it having that many cores, regardless how many "E" cores there are its ALOT for just a basic Desktop PC. These things need to go head to head against threadripper if there getting that high in core counts.

 

[Kodehawa]

On the other hand, the HTML5 browser score is significantly lower than Raptor Lake and that's reflective of the most common workload for most people.

The lack of hyperthreading might be hurting it on this benchmark as the pipeline could be stalling. HT was a great workaround to keep the pipeline busy even in scenarios where it could stall without it.