You take your 13900kf bump the core clock 200mhz and voila you have a 14900k. Why do these companies even do these pointless refreshes. Intel should have just sit on its ass until arrow lake is ready.

Undying:

You take your 13900kf bump the core clock 200mhz and voila you have a 14900k. Why do these companies even do these pointless refreshes. Intel should have just sit on its ass until arrow lake is ready.

But really though - they don't have to release a new product. The tick-tock strategy is fine but they're more like doing tick-tick-tick-tock, and all that does is make the tocks not as appealing.

The market expects them to do annual product refreshes. It's a thing now. You can thank Apple for that.

I wonder about power/heat. If you look at that chat, it shows the 14900KF at a 125w tdp so it might use less power at 6ghz then a 13900KS. I also wonder if that is a all core 6ghz or just a few cores.

mackintosh:

The market expects them to do annual product refreshes. It's a thing now. You can thank Apple for that.

This, unfortunately. Companies are expected to present something new every year, so they have to resort to "new" parts like this or just rebadge older products...

Looks like a waste of sand.

Prod:

I wonder about power/heat. If you look at that chat, it shows the 14900KF at a 125w tdp so it might use less power at 6ghz then a 13900KS. I also wonder if that is a all core 6ghz or just a few cores.

The 6GHz boost is not allcore, atleast it has not been like that the last 10 years. The fact that the multicore passmark for 13900k and 14900k is practically identical, indicates that there is a power or temperature limit on both chips, the allcore is probably in the 5.4GHz range with a good cooler, just like the 13900k.

Tbh yearly launches are a relatively new thing. In the old days of CPUs, they would launch a new one with a few Mhz extra every few months, or even quicker. For example, Pentium II started with 233 Mhz, then later 266, 300, 350, 400, 450 were launched. Then it continued with Pentium III, which wasn't that much different, with basically exact same instruction set and similar IPC, and the clock continued to increase... 500, 533, 550, 600, 667, 700, 733, 750, 800, 850, 900, 933, 1000, 1100, 1133, 1200, 1266, 1333, 1400 OMGGGGG!! And don't get me started with the list of clock speeds for Pentium 4s ... it's... BIG. This trend kept going all the way into "Core" processors, where they finally got model numbers and an entire stack launched at once with different performance levels.

wavetrex:

Tbh yearly launches are a relatively new thing. In the old days of CPUs, they would launch a new one with a few Mhz extra every few months, or even quicker. For example, Pentium II started with 233 Mhz, then later 266, 300, 350, 400, 450 were launched. Then it continued with Pentium III, which wasn't that much different, with basically exact same instruction set and similar IPC, and the clock continued to increase... 500, 533, 550, 600, 667, 700, 733, 750, 800, 850, 900, 933, 1000, 1100, 1133, 1200, 1266, 1333, 1400 OMGGGGG!! And don't get me started with the list of clock speeds for Pentium 4s ... it's... BIG. This trend kept going all the way into "Core" processors, where they finally got model numbers and an entire stack launched at once with different performance levels.

And those few MHz jumps actually made a difference.

True, but most (if not all) of these releases made sense. They were mostly die shrinks and speed bumps that genuinely improved efficiency and performance. Very few were just straight up refreshes for the sake of releasing something. My memory is extremely hazy after 20+ years, but I seem to recall that back in those days Intel's roadmap was fairly accurate, so most of these CPUs and their release windows didn't catch us by surprise. That said, perhaps I'm looking at this through rose-tinted nostalgia glasses.

Undying:

You take your 13900kf bump the core clock 200mhz and voila you have a 14900k. Why do these companies even do these pointless refreshes. Intel should have just sit on its ass until arrow lake is ready.

They simply got slightly better bins over the year. That +200mhz is due to that. Rather than continue to sell 13900k at current price, may as well milk that 200mhz for all its worth. Whatever 13900k can OC to, the 14900k will likely do it +200mhz better.

too bad for e-cores, such high IPC CPUs would be sweet cheap alternatives in servers needing high single thread performance (knock knock, gaming hosting servers). So far all high single thread performance servers not using epyc or old xeon are all AMD ryzen , Threadripper or old intel pre-e cores (with very small exceptions using the very few recent intel models without e-cores)

Alessio1989:

too bad for e-cores, such high IPC CPUs would be sweet cheap alternatives in servers needing high single thread performance (knock knock, gaming hosting servers). So far all high single thread performance servers not using epyc or old xeon are all AMD ryzen , Threadripper or old intel pre-e cores (with very small exceptions using the very few recent intel models without e-cores)

? The E cores is the only reason Intel is able to keep up with AMD in the desktop segment. The entire Intel server stack without E cores is getting smashed by AMD on almost all points.

TLD LARS:

? The E cores is the only reason Intel is able to keep up with AMD in the desktop segment. The entire Intel server stack without E cores is getting smashed by AMD on almost all points.

"Server" is a huge galaxy of different needs. On performance servers, where single thread performance matters as well low latency matters, e-cores are a cancer. Gaming host servers are usually an example of such servers. Almost nobody wants e-cores in a high performance gaming host server. Most game servers still likes single thread performance and low latency, e-cores are all but this: no gaming server is meant to take advantage of such architecture, plus context switch that will be caused by OS scheduler between e-cores and p-cores creates performance issues. The fact e-cores are in higher number than p-cores is another factor that nobody likes high performance gaming servers on such architectures.

Alessio1989:

"Server" is a huge galaxy of different needs. On performance servers, where single thread performance matters as well low latency matters, e-cores are a cancer. Gaming host servers are usually an example of such servers. Almost nobody wants e-cores in a high performance gaming host server. Most game servers still likes single thread performance and low latency, e-cores are all but this: no gaming server is meant to take advantage of such architecture, plus context switch that will be caused by OS scheduler between e-cores and p-cores creates performance issues. The fact e-cores are in higher number than p-cores is another factor that nobody likes high performance gaming servers on such architectures.

Maybe at a local Ethernet connected CSGO tournament a high speed 8 core would make sense, but when internet connected a 64 core Epyc would be able to service 6 times as many users as a fast 8 core would. The handover of users and data between for example 20x 64 core Epyc servers is going to be a lot easier to handle then 120x 8 core servers. 8 high speed cores at 5400-5600MHz or so, like the 14900k probably will be (with E-cores off), are ridiculously bad on power efficiency, it is possible to get 64 Epyc cores at 3500MHz at the same power usage with a lot more capacity for users. Users on a fast low core count server would get faster game data as long as the user count is low enough to not bog down the server, but the cost of the service would need to be much higher to pay for the server running cost. E-cores are not cancer if the software actually works correctly. If a E-core is able to solve a task before the task can be slottet into a timeslot on a busy P-core, the latency is lower, but those numbers are ridiculously low compared to internet latency anyway, so it does not really matter.

I have been doing some read up on the next generation Intel ( Desktop Performance cpu's ). As i understand from what Intel is saying we will be stuck with Raptor lake refresh for ~ 1 year on the performance Intel cpu side , as they might release laptop and low end/low wattage entry level / mainstream Meteor Lake-S Up To 6 Performance cores and X ( unknown ) E-cores energy efficient cores , launch date around Q4 2023 / Q2 2024. New arch Arrowlake-S wish ( Core i9 / Core i7 ) will replace refresh Raptor Lake Refresh performance cpu's is far , far away late 2024/2025 from what i have read all over the net and Intel. " Intel is on track to release its codenamed Meteor Lake-S processors for desktops in the coming quarters, but there's a catch. The company will only address entry-level and mainstream systems with these CPUs, whereas performance-demanding PCs will have to use Raptor Lake Refresh or Arrow Lake-S processors. " " [Meteor Lake] desktop will come in 2024," said Michelle Johnston Holthaus, Executive Vice President & General Manager of Client Computing Group at Intel, in an interview with PCWorld's Mark Hachman. " Sources - https://www.tomshardware.com/news/intel-confirms-meteor-lake-comes-to-desktops-next-year https://www.igorslab.de/en/intels-internal-performance-projection-for-raptor-lake-s-refresh-and-arrow-lake-s/ https://www.techpowerup.com/314018/intel-confirms-meteor-lake-desktop-cpus-coming-in-2024 So hold on to your LGA 1700 Motherboards and Alder / Raptor / Raptor Lake Refresh cpu's as it will be a long time to wait for a new upgrade for us enthusiasts and the LGA1851 and Arrow lake performance / enthusiast new cpu's.

Gamin

TLD LARS:

Maybe at a local Ethernet connected CSGO tournament a high speed 8 core would make sense, but when internet connected a 64 core Epyc would be able to service 6 times as many users as a fast 8 core would. The handover of users and data between for example 20x 64 core Epyc servers is going to be a lot easier to handle then 120x 8 core servers. 8 high speed cores at 5400-5600MHz or so, like the 14900k probably will be (with E-cores off), are ridiculously bad on power efficiency, it is possible to get 64 Epyc cores at 3500MHz at the same power usage with a lot more capacity for users. Users on a fast low core count server would get faster game data as long as the user count is low enough to not bog down the server, but the cost of the service would need to be much higher to pay for the server running cost. E-cores are not cancer if the software actually works correctly. If a E-core is able to solve a task before the task can be slottet into a timeslot on a busy P-core, the latency is lower, but those numbers are ridiculously low compared to internet latency anyway, so it does not really matter.

I think you didn't look a lot on what is the gaming servers market now. It's not only big companies like nitrado, it's also VPS, root server and bare metal dedicated servers. On such market you will find xeon (both old and recent models), epyc, as well workstation and desktop CPUs. On such market who rents a server usually needs single thread performance to run multiple server instances, sometimes in a cheap way. E-cores are a cancer: nobody wants e-cores on performance gamin server, even big companies: they create too many issues, not only software side but also just because they are: context switch, latency, cache flush on having a server having threads executed on p-core and e-cores simultaneously. You don't just use a single machine to run a single server, that's a huge waste amount of money, electric power and computational power. The problem is, e-cores are too many on the single packet for such applications and for an optimal usage you not only would need to map each server on a specific thread, knowing it if it is a p-core or an e-core but also running different gaming servers of different games, like 1 hardware demanding game and one less hardware demanding game on the same machine. This is unlikely to happen outside server farms. Even on server farms, e-cores requires such careful thread mapping, plus most gaming servers have "smart" idle functions that will lower to almost stop the tick rate until a client get connected. This is far more power-saving that using e-cores on an environment that will suffer of context switch between p-cores and e-cores of threads of the same process.

Alessio1989:

Gamin I think you didn't look a lot on what is the gaming servers market now. It's not only big companies like nitrado, it's also VPS, root server and bare metal dedicated servers. On such market you will find xeon (both old and recent models), epyc, as well workstation and desktop CPUs. On such market who rents a server usually needs single thread performance to run multiple server instances, sometimes in a cheap way. E-cores are a cancer: nobody wants e-cores on performance gamin server, even big companies: they create too many issues, not only software side but also just because they are: context switch, latency, cache flush on having a server having threads executed on p-core and e-cores simultaneously. You don't just use a single machine to run a single server, that's a huge waste amount of money, electric power and computational power. The problem is, e-cores are too many on the single packet for such applications and for an optimal usage you not only would need to map each server on a specific thread, knowing it if it is a p-core or an e-core but also running different gaming servers of different games, like 1 hardware demanding game and one less hardware demanding game on the same machine. This is unlikely to happen outside server farms. Even on server farms, e-cores requires such careful thread mapping, plus most gaming servers have "smart" idle functions that will lower to almost stop the tick rate until a client get connected. This is far more power-saving that using e-cores on an environment that will suffer of context switch between p-cores and e-cores of threads of the same process.

The ping on a service like this is often over 100ms. A "slow" 3500 MHz Epyc is going to be something like 120ms instead of a 5500 MHz Intel with 100ms ping. Nobody is going to be able to tell them apart in a blind test and the cost of running the faster Intel is going to be 5 times as high, because it supports a lot less paying customers. E-cores are plenty good enough to backup player and server data, they are professional they know how to setup a server. They can always turn the E-cores off if they want to ruin the economy of hosting a fast 8 core server even more.

100ms is an 8 bit '60s era calculator man. if you turn e-core off you are paying for using 40% of the CPU core amount. The day intel will fix different ISA from p-core and e-core and ALSO the context switch issues, we can talk again about e-cores as better alternative than SMT.. And this requires a total redesign of cache and memory controllers too.