Intel Core i9-14900KS vs. Intel Core Ultra 9 285K

[Bobby5109]

No, it is much worse.

Much worse? Why so? Why would Intel do something which would damage their reputation even more?

 

[Dr. Dro]

Much worse? Why so? Why would Intel do something which would damage their reputation even more?

It's a regression from going to a fully monolithic, to a first generation tile based design. It doesn't matter for most use cases, yours is just a VERY sensitive niche case where that does matter.

 

[ir_cow]

Changed from monolithic to tile die. But it makes up for being much more efficient and hyper threading is gone too.

Best Single thread performance? Wouldn't that be the AMD Ryzen 9800X3D?

 

[mb194dc]

Buy both current generation options from AMD and Intel and TEST THEM to see, which will work better for you.

This, because op has as ultra niche use case that general reviews won't show the performance for.

 

[Visible Noise]

I am building a PC that I am going to use for serious programming, compiling large source codes, running a lightweight web server, and, most importantly, running a stock trading bot (requires extremely good single-thread performance). For my use case, I think the Intel Core i9-14900KS or the Intel Core Ultra 9 285K would be the right fit because of their extremely good single-thread performance and higher clock speeds. But I am concerned about the manufacturing defects, instability, and oxidation issues that Intel had lately with their early 13th/14th generation chips. I am not sure if it also impacted the 14900KS. Intel claims to have released the microcode fixes which is supposed to work like a silver bullet, but I don't trust Intel at this point. It is also worth mentioning that I don't use Windows and I am going to use it along with Arch Linux or any other GNU/Linux distribution.

Which processor should I choose and why? Also, feel free to recommend me any processor from the team red (AMD) too.

honestly, this is not something you should be wanting local. There are many remote systems available for rent and if you’re worried about latency, they actually run their servers in the basement of the NYSE. Any possible performance gains you could get from a CPU is going to be lost in communications latency.

At one time, the NYSE was actually measuring fiber links, and if you were in a closer rack to their mainframe, they would make you make more loops of your fiber so that you couldn’t get a trading advantage on somebody else.

edit: I used to do Forex trading

 

[Dr. Dro]

Changed from monolithic to tile die. But it makes up for being much more efficient and hyper threading is gone too.

Best Single thread performance? Wouldn't that be the AMD Ryzen 9800X3D?

No, speaking purely from a compute perspective, they are actually quite ordinary. The standard Ryzens and the Intel chips tend to be faster (often significantly faster) if the workload doesn't benefit from the extra cache memory. The X3D cache slice, from a latency perspective, also introduces a penalty of an extra few cycles, which very slightly reduces the processor's theoretical instructions per clock rate, although that is offset by the cache capacity in most workloads.

honestly, this is not something you should be wanting local. There are many remote systems available for rent and if you’re worried about latency, they actually run their servers in the basement of the NYSE. Any possible performance gains you could get from a CPU is going to be lost in communications latency.

At one time, the NYSE was actually measuring fiber links, and if you were in a closer rack to their mainframe, they would make you make more loops of your fiber so that you couldn’t get a trading advantage on somebody else.

edit: I used to do Forex trading

That's awesome. I think the OP should go with the 285K, if anything, the option to upgrade to Panther Lake might come in handy.

 

[Bobby5109]

honestly, this is not something you should be wanting local. There are many remote systems available for rent and if you’re worried about latency, they actually run their servers in the basement of the NYSE. Any possible performance gains you could get from a CPU is going to be lost in communications latency.

At one time, the NYSE was actually measuring fiber links, and if you were in a closer rack to their mainframe, they would make you make more loops of your fiber so that you couldn’t get a trading advantage on somebody else.

edit: I used to do Forex trading

I agree with you; Co-location services offered by the exchanges are a good fit for the HFTs. But our proprietary trading firm doesn't engage in market making business. Rather, we use a proprietary algorithm where security and latency are utmost important but it should also be balanced. So we decided to run it locally.