Intel Core Ultra 5 245K Linux Performance

I intend to send an email to Intel's top executives, as a share holder, and tell them they need to change the way they are thinking.

Intel has the superior technology, no one can deny it, but they are not using it.

If I were running Intel I would simplify the lineup with one Core i3, one Core i5, one Core i7 and one Core i9 and that's it.

They would not have any P cores, only E-cores and lots of them.

Core i3 - 30 E-cores
Core i5 - 50 E-cores
Core i7 - 70 E-cores
Core i9 - 90 E-cores

I would only use on-die ram, no discrete ram modules, in fact the motherboards would not even have any ram slots:

Core i3 - 20 gb HBM
Core i5 - 40 gb HBM
Core i7 - 60 gb HBM
Core i9 - 80 gb HBM​

They would still have NPU:

Core i3 - 1 NPU
Core i5 - 2 NPU
Core i7 - 3 NPU
Core i9 - 4 NPU​

And they still have QSV and Arc graphics.

That would be the lineup.

I would focus on multitasking benchmarks were these processors would excel and AMD would be caught flatfooted, they don't have anything to counter such a lineup.

I would also have the Pentium lineup as a catchall for processors that failed to meet minimum quality standards to qualify as one of the above SKU.

I think reviewers and the general public would for crazy for this lineup and Intel would sell them like hotcakes.

I have to find their emails, if anyone knows where I can find them let me know.

In case I wasn't clear, I am not that enthused by Arrow Lake.

Every other media outlet with gaming testing: AAA games

Phoronix: Imma do Unvanquished and Xonotic

Not sure what kind of thought process goes on here but those are not the titles majority of people are curious.

If counter agument is: Other outlets already does them, why should we include them too? Then i would suggest simply refrain yourself from adding a "gaming" section because of how useless it is currently.

I'm about to upgrade my E3-1275 sandy bridge to a 9950X, so I'm obviously happy with this test results. I decided a while ago to ditch intel, but they've been making some inroads that I was close to considering them. But at the end, AMD having a stable motherboard platform and all chipsets supporting ECC is just so sweet.

I also am worried about this hybrid/big.little stuff. It reminds me of the early days where linux would perform worse with HT on than it did off. I don't even want X3D because I doubt the kernel can schedule for it correctly, since x3d is only on a subset of cores. It would be interesting to see how a benchmark compares when you manually pin tasks to the fast/x3d cache cpus, vs just letting the kernel do it's thing.

Also, Michael you aren't listing the AEGSA version for AMD, isn't that important w.r.t. performance these days?

Tbh, based on the leaks I was expecting mostly a side-grade performance-wise. Intel managed to have done slightly better. Sure, they don't that well when it comes to gaming. But on the other hand, power draw seems to be tamed a lot. I can see this being an option for someone like me, coming from a 12600k. But it will depend on the price of RAM, motherboard (most important) and even CPU (I would like to see it a tad cheaper, but even the 12600k launched at $300, it will be a while).

Gamers don't care about cores or multi-threaded performance, all they care is about high single-core performance so they would want p-cores. A gamer would rather have 8 P-cores and 0 E-cores than 90 E-cores and 0 P-cores.

As an inhabitant of Earth and aspiring dictator, I have sent my thoughts and prayers to God that poor people receive $1 million each without any consequence on inflation.

No. Intel needs to:

1. Focus back on high performance x86 IP (instead AI learning and dGPU) and finish Royal Core project, which by the way was precisely opposite: very big core with ability to change IPC/hw thread combo in the run-time. Nobody really needs a lot more cores in gaming or general consumer space.
2. Work on high performance ARM IP in parallel to 1;
3. Understand the fact they will not catch NVIDIA in AI learning solutions and focus on inference solutions instead. Inference market will be growing a lot in the near future.

In case of coremark the really interesting metrics is score ( aka iterations per second) per Mhz. need to lock cpu to some freq and use taskset to tie coremark to P-core. Otherwise the error margin is too big.