• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

14th Gen "Raptor Lake Refresh"

Maybe it's more cost effective just to wait until their improved 'Intel 3' process is ready, most likely in the first half of 2024?

Couldn't Intel release another generation on 'Intel 7' (10nm), but shift desktop CPUs to a new socket (LGA 1851)? And skip 'Intel 4' for desktops in 2023.

It's notable that there's no 'Intel 6', and 'Intel 5' process technologies also...

Or, even use TSMC's 4/5nm EUV technologies as a stop gap in 2023?
 
Last edited:
Disappointing if true. Shows that they are limited by their 'Intel 4' process technology, and need to adopt the improved 'Intel 3' as soon as possible (presumably the 1st half of 2024).

I hope they at least try to explain why reducing the P-Cores is necessary, or considered to be a good idea.

I also think they've got it wrong about Arrow Lake being the generation after Meteor Lake, as that does not explain where the 'Intel 3' process would fit in.
 
Last edited:
There's some assumptions there about P-core count too:

Intel-Meteor-Lake-Arrow-Lake-Lunar-Lake_-Hot-Chips-34_16-1030x579.png


There's 6 P cores in the image on the top left, for a single compute tile. Isn't it possible that there could be 2 compute tiles though?
 
Based on the other diagram from Intel, desktop meteor lake contains one CPU tile, one I/O tile, one GPU tile and one SOC tile.

They have previously said its very scalable, so if they want they can use two GPU tiles instead of 1, two CPU tiles instead of one etc - its all about playing Tetris with the package.
So in that regard, yes a second CPU tile is definitely possible if they cut something else out, like removing the GPU tile.

I don't know how much flexibility will be used for desktop parts, but Intel is using it for their server parts where they include multiple CPU tiles to get higher cores counts
 
Last edited:
6P = D.O.A. for gaming. It will get absolutely walloped by Zen 4 3D and maybe even lose to 13900KS in key titles that have that 8 core scaling (see below). This is also why it's so hard to choose RPL over Zen 4, because there's so much to look forward to on AM5 even though RPL wins hard in some games over it, but Meteor Lake is just so pathetic looking for gaming.

BaBt1TX.jpg

6 P-cores do not always bring the full performance
The Core i5-12600K is missing an additional 2 P-Cores and that limits the gaming performance significantly more. The flagship Core i9-12900K delivers 10 and 11 percent more FPS than the smallest Alder Lake offshoot, with the Core i7-12700K it is 8 and 9 percent. But the Core i5-12600K is on average still as fast as the Ryzen 9 5950X and thus the fastest AMD CPU, the Core i9-11900K is just beaten.

In individual cases, however, the difference can also be significantly larger. In Age of Empires 4, the Core i9-12900K is 15 percent faster than the Core i5-12600K and then the Ryzen 9 5950X is also 4 percent faster. The Core i9-12900K is up to 17 percent ahead of the Core i5-12600K in Far Cry 6, up to 20 percent in Resident Evil Village and up to 23 percent in The Ascent as a worst-case scenario.
 
6P = D.O.A. for gaming. It will get absolutely walloped by Zen 4 3D and maybe even lose to 13900KS in key titles that have that 8 core scaling (see below). This is also why it's so hard to choose RPL over Zen 4, because there's so much to look forward to on AM5 even though RPL wins hard in some games over it, but Meteor Lake is just so pathetic looking for gaming.

BaBt1TX.jpg


Maybe Intel will have a tonne of cache on it, it does say scalable cache in the slides + we don't know if we'll have more than one compute tile or tiles with more than 6 P-cores.
 
6P = D.O.A. for gaming. It will get absolutely walloped by Zen 4 3D and maybe even lose to 13900KS in key titles that have that 8 core scaling (see below). This is also why it's so hard to choose RPL over Zen 4, because there's so much to look forward to on AM5 even though RPL wins hard in some games over it, but Meteor Lake is just so pathetic looking for gaming.


Meh, It's a minimum of another 12 months away, A lot can change in that time.
 
6P = D.O.A. for gaming. It will get absolutely walloped by Zen 4 3D and maybe even lose to 13900KS in key titles that have that 8 core scaling (see below). This is also why it's so hard to choose RPL over Zen 4, because there's so much to look forward to on AM5 even though RPL wins hard in some games over it, but Meteor Lake is just so pathetic looking for gaming.

BaBt1TX.jpg



Too early to count Meteor Lake out - though Intel would have to innovate something special to make up for 2 less P cores.
 
Assuming there is a large improvement in IPC, like 20% vs Golden Cove, I think 6 P cores will be OK. I doubt they would be much faster than a Zen 4 /Golden Cove CPU with 8 cores though, at least in games that benefit from higher core counts.
 
Last edited:
Games don't really care about cores though if the raw IPC and clockspeed is high enough.

No one has really figured out how to make dynamic multithreaded applications efficient. So lots of cores work best when the application is not dynamic, and is performing the same task just multiplied across the threads (like cinebench r23).

Developers do to some extent allocate certain game tasks to specific core threads, but it's not dynamic so there is a lot of inefficiency and most game tasks are still just running one by one on the same thread like waiting in a queue to get on the bus.
 
Last edited:
Games don't really care about cores though if the raw IPC and clockspeed is high enough.
The fact that you can compensate with IPC/Clk Spd doesn't mean they don't care about cores. Secondly, the obvious thing to point out is there's no magic 'just add more IPC' button to press, so compensating with that for the missing cores will be difficult, plus MTL is almost certain to have clock speed regressions compared to RPL so another hurdle to jump over. Really their only chance is a lot more cache, but we don't know how that looks for them. AMD we know for sure, and they've already demonstrated it.

But really the biggest thing is, why stop at 6P? We know for a fact there are multiple games which scale very nicely with 8 cores so it's a baffling decision. To me if they really only release a 6P MTL it signals an admission of defeat and that they are still confused as to how to catch up to AMD who remains two steps ahead. Leaving the companies aside tho, for a consumer this is yet another vote in favour of buying AM5.
 
Not sure that's true about clockspeed on the 'Intel 4' process technology. Here is a graph comparing Intel 7 and Intel 4:

Intel-PPW-Curve.png


The base clock of the Intel 4 chip is significantly improved with this new process. It looks like Meteor Lake will have a base clock of 3.6-3.7ghz, vs a max base clock of 3.3ghz on Intel 7 at 1.1v, and using less power also.

It suggests an improvement of about 10% in base clockspeeds, it explains why Intel is focusing on Mobile CPUs so much for this generation.

Base clock speeds have been declining since the 9th/10th gen Intel CPUs, so it's a nice improvement.

It looks like similar clock speed improvements (slightly less) are planned for the 'Intel 3' process:

From here:
 
Last edited:
Why do you say that :)

I think it's a bit disingenuous (or lazy) to say that it doesn't matter. It clearly matters a lot in terms of power usage. Compare the base clocks and rated TDPs of Zen 4 vs Golden Cove CPUs, there is a huge difference in power consumption, which also scales at the turbo boost frequencies of these CPUs.
 
Last edited:
Why do you say that :)

I think it's a bit disingenuous (or lazy) to say that it doesn't matter. It clearly matters a lot in terms of power usage. Compare the base clocks and rated TDPs of Zen 4 vs Golden Cove CPUs, there is a huge difference in power consumption, which also scales at the turbo boost frequencies of these CPUs.

It only matters if it was the other way round, that's how this works, get with the program dude. :D
 
Why do you say that :)

I think it's a bit disingenuous (or lazy) to say that it doesn't matter. It clearly matters a lot in terms of power usage. Compare the base clocks and rated TDPs of Zen 4 vs Golden Cove CPUs, there is a huge difference in power consumption, which also scales at the turbo boost frequencies of these CPUs.
Because the context is gaming (desktop) which means you will never not run boost clocks, so if you want to tell me MTL(-S) won't have clock regressions then that's the relevant scenario. In fact I can't even think of a scenario where you won't run past base clocks on desktop.
 
It's true that Intel has had yield and defect problems on new process technologies, 10nm was particularly bad at first, it took many years to release viable 10nm desktop chips.

Intel's 7nm technology has been in the works for years already also. It apparently does not suffer from the complex patterning /design problems that their 10nm process had, and so Intel prioritized the transition to this technology years ago. There's also a high probability of much greater transistor densities, due to the use of EUV lithography, which unfortunately, Intel has adopted quite late in the game, for both Intel 4, and Intel 3.

So, if they officially announce an Intel 4 based desktop product at some point in 2023, I think things should go well. If not, it's very likely that there are still serious production issues with Intel 4.
 
Last edited:
Intel Saphire Rapids looks set to launch o on 10 January 2023, Intel just announced. It uses the same chiplet technology that meteor lake would use and has 4 CPU core chiplets each containing 14 performance cores for a total of 56 Golden Cove performance cores (no E cores) and a 350w TDP
 
Last edited:
Back
Top Bottom