Question Raptor Lake - Official Thread

[Hulk]

Since we already have the first Raptor Lake leak I'm thinking it should have it's own thread.
What do we know so far?
From Anandtech's Intel Process Roadmap articles from July:

Built on Intel 7 with upgraded FinFET
10-15% PPW (performance-per-watt)
Last non-tiled consumer CPU as Meteor Lake will be tiled

I'm guessing this will be a minor update to ADL with just a few microarchitecture changes to the cores. The larger change will be the new process refinement allowing 8+16 at the top of the stack.

Will it work with current z690 motherboards? If yes then that could be a major selling point for people to move to ADL rather than wait.

 

[DrMrLordX]

Benchmarks favor the 13600k over the 12700K.

Pretty sure he was referring to the non-k i5-13600.

 

[Heartbreaker]

Pretty sure he was referring to the non-k i5-13600.

I think he was just speaking generically given the second sentence:

Is it just me or am I the only one to prefer the 8x P-core + 4x e-core 12700 over the 6x P-core +8x e-core 13600.

For example I would highly recommend the Core i7 12700KF vs Core i5 13600KF.

 

[Wolverine2349]

Is there a workload where that works better?

Benchmarks favor the 13600k over the 12700K. You are also getting tweaked performance cores with more L2, in 13th Gen K parts.

For any games that potentially scale beyond 8 cores if there are nay, for sure with 2 extra P cores. Though I do not know of any games that do. Though for running background tasks and streaming 2 extra P cores way way way better than more e-cores. Nevermind the no need to deal with the WIN11 and hybrid arch crap and Intel Thread director gimmicks.

Spamming more e-cores only is a bit better for productivity tasks that scale to infinite threads.

 

[DrMrLordX]

I think he was just speaking generically given the second sentence:

Maybe. But it's a bit confusing since the -k parts have updated ring speed, cache, etc. The non-k 13600 is an inferior product (not speaking wrt price; per Anand, there are no bad products, just bad prices). It's hard to lump both the 13600 and 13600k into the same example.

 

[Heartbreaker]

For any games that potentially scale beyond 8 cores if there are nay, for sure with 2 extra P cores. Though I do not know of any games that do. Though for running background tasks and streaming 2 extra P cores way way way better than more e-cores. Nevermind the no need to deal with the WIN11 and hybrid arch crap and Intel Thread director gimmicks.

Spamming more e-cores only is a bit better for productivity tasks that scale to infinite threads.

Anything beyond empty theory crafting? Because the gaming results clearly favored the 13600K over the 12700K.

Also you still have to deal with thread allocation issues between P and E cores because both chips still have them.

 

[Hulk]

Intel hasn't offically said anything about HEDT.
They are dumping the entire HEDT brand i think and going full workstation, which is a total bummer.

Workstation CPU's are expensive, always have been.

I know i am sounding like a broken record playing a tiny violin, but i really hope HEDT does not die out, as i really need those PCI-E lanes which neither 13000 series or 7000 series can provide enough of for my usage.

I'm not totally sure I really understand the main things differentiating desktop, HEDT, and workstation other than more cores/compute available? Can you give me the really quick points that differentiate them? Also I'm curious as to what you are doing that saturates your PCI-E lanes. To be fully clear, I am not in any way doubting you it's just something I'd like to be brought up to speed on by someone who knows.

 

[Wolverine2349]

Anything beyond empty theory crafting? Because the gaming results clearly favored the 13600K over the 12700K.

Also you still have to deal with thread allocation issues between P and E cores because both chips still have them.

Thats because very few games benefit from more than 6 cores. And the 13600K has equal L3 cache to 12700K and its 6 P cores are faster and have more L2 cache than 12700K. So of course it is better just like the 5600X is better than the 10700K and 9900K despite having 2 less cores.

Your right games do not benefit from more than 6 cores much less 8, but some rare games may benefit form 8.

The e-cores do nothing for gaming though, Not a thing and are e-waste for any gaming loads.

Best would be then for Intel to create an 8 P core only Raptor Lake with 72MB or 108MB L3 cache and sell it as 3D cache gaming chip like AMD has 5800X3D as games love L3 cache and clock speed and IPC with 6 cores or sometimes 8.

And yeah thread allocation between P and e cores. Well disable the e-cores and problem solved and you have yourself the most powerful 7 and 8 core gaming chips in existence and will be even better not being held back by the e-waste cores. Even with e-waste cores on, they still are the best gaming chips as long as no game threads do not get stuck on one of those peasant cores which in the benchmarks I imagine they do not. But still they hold back P core clocks and ring a little bit and use more power so they still should be disabled for gaming.

 

[ondma]

Basically, make it a 7950X.

Hah, hah. Will be interesting to see what the e core haters have to say if, as rumored, AMD goes to hybrid architecture with Zen 5.

For any games that potentially scale beyond 8 cores if there are nay, for sure with 2 extra P cores. Though I do not know of any games that do. Though for running background tasks and streaming 2 extra P cores way way way better than more e-cores. Nevermind the no need to deal with the WIN11 and hybrid arch crap and Intel Thread director gimmicks.

Spamming more e-cores only is a bit better for productivity tasks that scale to infinite threads.

Do you have any data for streaming results? Admittedly, one would expect big cores to be better. However, without carefully controlled tests (I haven't seen any), it is just speculation. As for productivity, e cores are not a "bit better" as 13900k (8 + 16) is essentially equal to 7950x overall in productivity tasks. So essentially 16 e cores is equal to 8 AMD big cores.

That said, I am not a fan of big.little on the desktop either, but one cant argue with the data that at least with the current software and gaming landscape it seems to work. The real problem with big.little for Intel is that it should allow higher efficiency, but due to core design and a worse process, AMD is still more efficient.

 

[Wolverine2349]

Do you have any data for streaming results? Admittedly, one would expect big cores to be better. However, without carefully controlled tests (I haven't seen any), it is just speculation. As for productivity, e cores are not a "bit better" as 13900k (8 + 16) is essentially equal to 7950x overall in productivity tasks. So essentially 16 e cores is equal to 8 AMD big cores.

That said, I am not a fan of big.little on the desktop either, but one cant argue with the data that at least with the current software and gaming landscape it seems to work. The real problem with big.little for Intel is that it should allow higher efficiency, but due to core design and a worse process, AMD is still more efficient.

I agree the hybrid arch sucks and is a gimmick and scheduling issue nightmare when we have been in an SMP world for more than 2 decades.

And not saying e-cores are better for productivity than the P cores. P cores on equal count are tremendously better!! Only that Intel is unable to add more P cores in that the e-cores can be good for productivity apps that scale to infinite threads because Intel can just add so many of them to compensate for lack of more than 8 P cores. They need 16 extra e-cores to compete with an 8 additional P cores from AMD because those e-cores are so inferior. And thats factoring in Intel already has 8 Raptor Cove P cores that have better IPC and more consistent clocks all load than AMD Zen 4 cores. Yet they still need 16 e-cores plus their better 8 P cores to trade blows with AMD in productivity apps. Thus no matter how inferior e-cores are, you add enough of them and apps that can scale to infinite threads/CPU cores, will eventually equal performance top much less far superior P cores in raw compute workloads. Unfortunately though most software does not scale to infinite CPU cores, so they are a waste in that regard.[/quote]

 

[ondma]

I agree the hybrid arch sucks and is a gimmick and scheduling issue nightmare when we have been in an SMP world for more than 2 decades.

And not saying e-cores are better for productivity than the P cores. P cores on equal count are tremendously better!! Only that Intel is unable to add more P cores in that the e-cores can be good for productivity apps that scale to infinite threads because Intel can just add so many of them to compensate for lack of more than 8 P cores. They need 16 extra e-cores to compete with an 8 additional P cores from AMD because those e-cores are so inferior. And thats factoring in Intel already has 8 Raptor Cove P cores that have better IPC and more consistent clocks all load than AMD Zen 4 cores. Yet they still need 16 e-cores plus their better 8 P cores to trade blows with AMD in productivity apps. Thus no matter how inferior e-cores are, you add enough of them and apps that can scale to infinite threads/CPU cores, will eventually equal performance top much less far superior P cores in raw compute workloads. Unfortunately though most software does not scale to infinite CPU cores, so they are a waste in that regard.

[/QUOTE]
Dont let the data invalidate your preformed conclusions about the e cores. Bottom line is they work, and those 16 e cores use up a lot less die space than another 8 p cores would. Also, RC and Zen 4 big cores are very close in per core performance, so Intel being competitive in multithreaded benchmarks is no longer due to superior big core performance.

Again, I am not a big fan of big.little, but in the current software environment, despite your claims of "scheduling nightmares" it works. My concern about big.little is that eventually more than 8P cores will be needed, and Intel just seems intent on adding more e cores.

(BTW, you dont have to yell and bold your entire post. I only bolded part of your post to highlight the statement that I was responding to.)

 

[Wolverine2349]

Dont let the data invalidate your preformed conclusions about the e cores. Bottom line is they work, and those 16 e cores use up a lot less die space than another 8 p cores would. Also, RC and Zen 4 big cores are very close in per core performance, so Intel being competitive in multithreaded benchmarks is no longer due to superior big core performance.

Again, I am not a big fan of big.little, but in the current software environment, despite your claims of "scheduling nightmares" it works. My concern about big.little is that eventually more than 8P cores will be needed, and Intel just seems intent on adding more e cores.

(BTW, you dont have to yell and bold your entire post. I only bolded part of your post to highlight the statement that I was responding to.)
[/QUOTE]


Yes the e-cores are much more die space efficient and they can get more performance out of them than e-cores on their node for apps that can scale to as many cores as you throw at them. Cause they can fit 4 -cores on the die space of 1 P core. And of course the e-cores while much weaker, they are not near weak enough that 4 e-cores provides more performance than 1 P core in apps that can fully parallelize and 100% use as many CPU cores as you can throw at them.

And the scheduling issues are very real especially if you are not on WIN11 which is hot garbage. Though with Process Lasso you can work around the issues in WIN10. Yes it works mostly, but pure SMP scheduling with no big.little is so much easier and better overall.

And yeah there may be need for more P cores than 8 could be very real, but all Intel is doing is adding more and more e-cores which I do not like.

Do you foresee games actually meaningfully benefiting form more than 8 P cores in the next few years assuming no streaming or running heavy background tasks along with the game??

 

[Heartbreaker]

I agree the hybrid arch sucks and is a gimmick and scheduling issue nightmare when we have been in an SMP world for more than 2 decades.

It really isn't. The E cores don't appear to harm gaming, and enable a big boost in productivity. If you look at the reviews it doesn't really falter in any type of workload.

With Intel, and nearly all ARM SoCs designed like this, it's probably inevitable that AMD will eventually also have multiple classes of cores.

 

[Wolverine2349]

It really isn't. The E cores don't appear to harm gaming, and enable a big boost in productivity. If you look at the reviews it doesn't really falter in any type of workload.

With Intel, and nearly all ARM SoCs designed like this, it's probably inevitable that AMD will eventually also have multiple classes of cores.

I still do not care for it. For a desktop power user, I want only P cores and fast ones at that. For mobile devices, it makes sense. But not in the desktop space. Plus you need WIN11 to be ensured it works correctly

 

[biostud]

A bit off topic, but why isn't the OS made aware if the cores are E or P cores, or which ccx they reside on?

 

[Kocicak]

I observed how assigning work to the cores works in 13900K and as the thread count increases, it works like this:

1) P cores are getting 1 thread each
2) after P core has 1 thread each, E cores are starting to get work
3) P cores are starting to get work on their second threads only after E cores are highly utilised.

It seems it knows what to do with those cores... I would not worry too much about that.

I wrote this on this topic a while ago.

 

[sze5003]

Picked up a 13900k from microcenter and some 6600mhz G skill DDR5. I'll be running the 3080Ti in this build probably until I can find a msi liquid 4090 or I'll probably wait until the Ti editions are out.

Would love to get it put together and run some benchmarks but I need to wait for Nzxt to ship me an lga 1700 bracket for the x73 AIO.

 

[nicalandia]

First Linux GB5 Results for 13900K

N/A - Geekbench

Benchmark results for a N/A with an Intel Core i9-13900K processor.

browser.geekbench.com

 

[Wolverine2349]

First Linux GB5 Results for 13900K

N/A - Geekbench

Benchmark results for a N/A with an Intel Core i9-13900K processor.

browser.geekbench.com

View attachment 69737

That's impressive. My 5GHz 12700K got like 2052. SO 20% performance uplift. A monster chip disable e-cores and have 8 extremely fast cores on a 13900K with lots more L3 cache.

 

[nicalandia]

I don't know, but I always thought that the 13900K and the 7950X were going to Surpass The 5965WX ThreadRipper PRO on MT Workloads, but most of the MT Benchmarks I have seen put The 5965WX at par or slightly ahead.

 

[Markfw]

I don't know, but I always thought that the 13900K and the 7950X were going to Surpass The 5965WX ThreadRipper PRO on MT Workloads, but most of the MT Benchmarks I have seen put The 5965WX at par or slightly ahead.

8 channel memory and fast stock speeds are hard to beat.

 

[nicalandia]

8 channel memory and fast stock speeds are hard to beat.

Geekbench Seems to love Memory Bandwidth more than memory speed, DDR4 13900K results are really underwhelming..

13900K DDR5 vs 5965WX

N/A vs N/A - Geekbench

browser.geekbench.com

 

[aigomorla]

I'm not totally sure I really understand the main things differentiating desktop, HEDT, and workstation other than more cores/compute available? Can you give me the really quick points that differentiate them? Also I'm curious as to what you are doing that saturates your PCI-E lanes. To be fully clear, I am not in any way doubting you it's just something I'd like to be brought up to speed on by someone who knows.

Workstation.... basically rebranded Enterprise Class CPU.
These Chips were not overclockable... Intel had the Designation W- in front. 3rd Gen ThreadRipper Pro, would also qualify as a workstation and not HEDT. Has typically double the PCI-E Lanes of regular consumer CPU's.

HEDT - fully overclockable, Enterprise CPU.
Intel had 1 or 2 of them in the lineup. They started as dual socket overclocking server boards, skulltrail 2, eVGA SR-2.
ThreadRipper (non pro) for 1 2 and 3rd gen. 4th Gen Thread Ripper Pro also qualifies as HEDT as they allowed overclocking on 4th gen.
Also has double the PCI-E Lanes of CPU's.

I saturate my lanes because i run 2 video cards, which drives 3 monitors + 1VR Headset. (Rift-S which uses DP port and not HDMI or USB-C.)
3090 is my dedicated gaming card, along with a alienware 38 Ultra wide with the Rift S.
3070ti to drive the others, which incudes 1 32inch 4k, however I will probably replace the other non 4k into a OLED 42inch C2 LG.
I also have 3 x nVME's.

32+12 = 44 which is where i am at max'd out....

I would want more lanes so i can add a 10gbe card onto this system, however current platform doesn't allow for it unless i run 1 of my GPU's in 8x.
Which my OCD will not allow, as i did not get those expensive cards to run them @ 8x.

 

[nicalandia]

Workstation.... basically rebranded Enterprise Class CPU.
These Chips were not overclockable... Intel had the Designation W- in front. 3rd Gen ThreadRipper Pro, would also qualify as a workstation and not HEDT. Has typically double the PCI-E Lanes of regular consumer CPU's.

Have you seen the numbers posted by The 5965WX? Those can be Overclocked to 5 Ghz All Cores too on Water and just trample over the 13900K/7950X.. But they are $2,500 a pop and 1K Mother Board.

 

[poke01]

But they are $2,500 a pop and 1K Mother Board.

Well for that price they should no?

They have to be the best everything not providing more lanes and cores.

 

[aigomorla]

Have you seen the numbers posted by The 5965WX? Those can be Overclocked to 5 Ghz All Cores too on Water and just trample over the 13900K/7950X.. But they are $2,500 a pop and 1K Mother Board.

I was.
That was the platform i almost hit the trigger on, until i realized, that will probably be the price of the 5th gen Thread Ripper, or a RaptorLake-S.
And with 5th Gen TRPro and RL-S, i get PCI-E 5.0
So i am playing that dog chases tail game right now, in the infinite loop of lets wait and see.

But originally i was going to get that setup, and roll down my current setup to my server box, but i ended up getting surplus EYPC with 256GB ECC RAM + 7601 + Board for less then HALF what that cpu costs alone, and went that route instead to replace my aging server.