Alder Lake-S leaks

[Grim5]

Forget Rocket lake, that's old news; let's talk about Intel's next gen cpu coming at the end of this year.

Alder Lake-S has been spotted with photos too and specs

this sample has 16 cores and 24 threads with 1.8ghz base clock and 4ghz boost clock, 30mb L3 cache and was paired with DDR5 4800mhz memory for benchmarking

8 of the cores are "high performance" and have 8 hyper threads. The other 8 cores are "low performance" and have no hyper threads - that's why windows reports just 24 total threads


https://videocardz.com/newz/intel-a...6-cores-at-4-ghz-and-ddr5-4800-memory-spotted

 

[Grim5]

Idling doesn't need 8 efficient cores. Maybe one
But even then regular cores are already very efficient at their lowest p-state. It is the power delivery wasting energy, at least in desktops.

We've had a long time where on pc cpus had one core or when they had multiple cores they were cores of equal value

now when alder lake arrives you have a cpu with some cores that are way faster than other, we're talking like double the performance or more. There is going to be a lot of teething problems and people having issues with single threaded applications and games getting put on one of the little cores etc even with multi thread it could be interesting since some cores will complete their task again well before other cores.


There is another alternative Intel and MS could use - like mobile phones, they could keep the little cores just for web browsing, sitting idle etc and the big cores only for data crunching, games and power hungry tasks - if you can switch cores on/off depending on load them you can avoid some of the potential scheduler issues. I believe this is what mobile phones do, only the Big cores are used during gaming, the Little cores get turned off to avoid confusing the OS.


But if they turn cores off based on load it creates another issue - what's the point of the product, desktop users don't care about 8 little cores for web browsing to save a few watts when those cores don't do crap while gaming. This big.little approach looks like it's made from the ground up for laptops, not desktops but desktop users will just get stuck with the same processors

And so from a gamers point of view, Alder Lake is in fact another generation of the 8 core CPu despite the move to 10nm. In fact I would not at all be surprised if you can permanently shut down the 8 little cores from the z690 BIOS to ensure maximum performance - just useless wasted die space

 

[Grim5]

More leaks: Alder Lake-S and the z690 boards have a target launch date for September. This means Rocket Lake-S has just 4 to 5 months of lifespan before its rapidly replaced

 

[Grim5]

Heterogenous CPU cores then. It makes a certain amount of sense. I wouldn’t expect all threads to be the same so you could economise with smaller cores; darkbahamut’s point about maximising performance for a die area with some giant, some little (or efficiently) makes sense.

This could just be the first step of many to come, as Moore’s law slows and we seek other ways to optimise: many different core sizes, specialised cores and so on.

Its not about optimizing performance in the way that most people care about, they are optimizing performance per watt for a task, not maximum performance - desktop users usually care about maximum performance, Big.Little is not for maximum performance.

Big.Little is not new, it's been used in processors since 2013 and ever since it's existence has been for one reason and one reason only - to maximize battery life.

Intel is moving to Big.Little because it believes the Desktop market is dying and mobile is the future - it's as simple as that, people can feel how they want to feel about it, it's a simple fact.

 

[Grim5]

I got one look forward to the big little design tbh. I am not sure if their top range sku should sport this design. 16core part should be full fat not some trimmed down skinny latte version.

but the lower SKU especially 6c 8c 10c ones definitely have some low powered ones. I am happy if the CPU can save me a few pennies while I am browsing and watching YouTube or typing up reports. No one can be gaming all the time on their PC while it is on right.

Current top of the range models from amd and intel sit at around 40-50w while you web browse, YouTube or even just idle. I assume big.little will bring that down to like 10-15w, but do you care?

 

[Grim5]

If your processors using 50w while web browsing there's something seriously wrong with your setup. Moderm processors put unused cores to sleep and they use virtually no power.

im watching a youtube video right now and my 5950x is sitting at 55w

 

[Grim5]

Countered somewhat by modern browsers being so well threaded. Every core has something to do and can't go to deep sleep.
40-50W idle is whole system, right?
My 5800X idles at 20-23W PPT. So including IO die and VRM inefficiencies. Cores themselves are at 1-2W all together.

So far all idle problems people had (mostly high 45-55C temps) could be traced to some background app, usually RGB.

No not whole system lol

Gpu idles at 45w
Cpu idles at 50-60w
Water pump probably another 20w
Not sure how much ram and rgb uses and NVMe and fans

Temps are fine, CPU at 35c, GPU at 30c

I say idle, but as you mention there is background stuff running all the time (rgb, sound control panel, mouse control panel etc)

 

[Grim5]

Alder Lake-S pops up again on Geekbench.

Again its the 16 core (8 Big 8 Little) configuration.

It scored slightly less than the 10900k in single core but higher than 9900k in single core.
It scored slightly more than the 10900k in multi core but still 46% behind the 5950x.

https://videocardz.com/newz/intel-1...gineering-sample-is-faster-than-core-i9-9900k

 

[Grim5]

^^^ Well, if we divide the reported fake 27.2 GHz by 8 if it's the number of cores that it recognises, then we get 3.4 GHz all core boost.

Alternative theory, it's reading the clock speed of the little cores = 3.4ghz on the little cores.
We know from earlier leaks months ago that they were getting over 4ghz on the big cores, so there is no way the score has increase and dropped 1ghz to 2ghz clocks

The most likely answer here is that the little cores run like 1 to 2ghz slower than the big cores and software like Geekbench gets very confused

 

[Grim5]

It's an engineering sample - your statement is wrong. The frequency can be in any range.

you don't gain performance by dropping clock speed, not rocket science

 

[Grim5]

Massive leak from Intel from someone taking photos at an internal presentation

Alder Lake will enter mass production in 5 months from now. Built on 10nm superfin, +20% IPC over Rocket lake, up to 38mb cache, up to 16 cores(8 big 8 little), 125w tdp on k models, 65w on non K models and 35w on T models, supports both ddr4 and DDR5,supports pcie4, up to 2x multithreaded performance over Rocket lake, hardware scheduler onboard processing, uses z690 chipset


In addition to all of this, Alder Lake will also be used for HEDT using a W680 chipset. HEDT models are up to 38 cores with 4.0ghz frequency


https://wccftech.com/intel-alder-la...ion-cpus-leaked-up-to-16-cores-w680-platform/

 

[Grim5]

In what way is Rocket Lake not a dud? The only area it wins in is AVX512, which nothing uses... Everything else it's slower and hotter than even the previous gen Intel.

It'll be interesting to see how well the 'gaming crown' thing works going forwards, fundamentally it's actually irrelevant as to see the differences you need to be running 1080p, or even lower, with a 3090... Most people don't have a 3090 and most people don't game at <1080p. For marketing sure, for sales, who knows.

Just because a company spent millions/billions in R&D doesn't make something good, Bulldozer and Rocket Lake are good arguments for that.

AMD aren't exactly struggling with the process, just capacity, and that's largely because of consoles. With Apple dropping to 3nm opening up 5nm for next gen processors/GPUs I don't see it being a problem for Zen3+/Zen4.

If Alder Lake comes out in 'the fall', actually hits the 20% IPC in more than just one specific case *and* can still hit the clock speeds then they'll take the lead... That's a few ifs. I doubt Alder Lake will see any real availability in 2021, I also doubt the 20% IPC claim (it will likely hit that in one or two very specific benchmarks so it can still be claimed), and given Ice Lake results I'm not sure they'll hit the clock speeds on 10nm they can on 14nm (with all the +'s)

Who knows what Zen3+ will be and when it will come out, if it's 'just' Zen3 tweaked on 6nm or even 5nm that could well be enough to be in the same ballpark as Alder Lake, and quite possibly out earlier as well.

In Intel's recent slides they talked about the HEDT Alder Lake models having an all core clock of 4ghz, so that will give us some indication of where the clocks will land

 

[Grim5]

Fresh Leaks for the i9-12900k

Base-clocks big core: 1.8ghz core 1-8
Base-clocks small core: 800mhz core 1-8

Turbo Clocks Big core: 4.6ghz for core 1-2, 4.4ghz for core 3-4, 4.2ghz for core 5-6 and 4.0ghz for core 7-8

Turbo clock small core: 3.4ghz core 1-4 and 3.0ghz core 4-8

TDP: 125w PL1, 228w PL2

TjMax: 100c


https://wccftech.com/intel-alder-la...sktop-cpu-leak-up-to-4-6-ghz-clocks-228w-tdp/

 

[Grim5]

Wonder how long it'll take Microsoft to sort out the Windows scheduler for these "big&small" cores

doesn't matter what Microsoft do because alder lake doesn't give a crap, it's uses its own hardware scheduler and doesn't use the windows scheduler

 

[Grim5]

If it is an early engineering sample, then that generally is not reflective of later clocks.
Seems such a complex mess of things going on with the chip.
They're going to have to spend ages trying to configure then reconfig to get the best out of it for different use cases.
Interesting.
Although wtftech so might all be complete BS.

It's not from wccrtech they just repost it, it's internal leaked specs provided by igors lab

 

[Grim5]

Interesting, how on each will that work, sees a speciifc app and use big cores, sees another and uses small cores, or completely app independant?

Don't know, Intel hasn't given specifics of how it works - if I had to guess it will distribute work based on the instruction set of incoming data and data load, like a web browser generated light load and could be placed on a single small core, while a old game like csgo would be placed on the 1st two big cores. Maybe it's Intel has its own software layer that is premprogrammed for the optimal work distribution based on the application exe name, kinda like a driver? Who knows

What we do know is that Intel internal leaked slides claims to have up to 200% performance gain in multi thread loads from 11900k to 12900k. Part of that 200% is higher ipc, part is the extra small cores that 11900k doesn't have but most of that will be the hardware scheduler

 

[Grim5]

hmm and yet they have twice the core count (big and little) between 11900 and 12900, so probably they're still behind the 5950x.... Whoopsie.

Yea 8 cores over the 11900k, but the small cores are very weak and run very slow speed and don't do HT - good enough to run background tasks like drivers, Windows processes, web browsers etc and keeping the high performance cores exclusively for the game threads with no core load swapping like the 5950x does

that is where the benefit comes, the 5950x is unable to keep its best cores exclusively for the game, Windows will keep pushing work around which means all cores get used for remedial stuff like background processes, alder lake resolves that by forcing background tasks to the mobile chip performance small cores and leaves high performance desktop cores just for the game/compute work loads

 

[Grim5]

How will the hardware scheduler know that function A / instruction B is part of an extremely demanding benchmark / production software / strategy game or just Windows bloat ?
How will the Gracemount cores operate? Will they be seen as normal "cores" 8-15 or will there be new ACPI level stuff that Windows will need to implement or will they not be visible at OS level at all?

No idea how the scheduler "just knows what to do" as such. However, I would not be surprised if Intel launches software for it that delivers this information to the scheduler. Sort if how AMD and Nvidia release drivers for new games once a month that allow the GPU to have better performance, maybe Intel will do the same for its CPU's starting with Alder Lake

 

[Grim5]

lol https://videocardz.com/press-releas...erclocked-ddr4-5300-memory-will-cost-1245-usd

 

[Grim5]

"leaks" from Moore's Law is Dead

Interesting points:

* Intel plans to stick with big.little design for a long time, so it's not going anywhere - anyone who believes big.little will fail on desktop therefore believes Intel will go out of business.
* The upcoming due to be announced new version of Windows includes an upgraded scheduler that brings compatibility for big.little CPU's (this new Windows is due to be launched a couple months before Alder Lake does)
* Intel is telling it's partners that Alder Lake delivers 100% improvement in multithreaded performance vs Rocket Lake and 20% improvement in single core performance.
* LGA1700 socket will be used for several generations starting with Alder Lake
* Alder Lake supports both DDR5 and DDR4 and PCIE5, but PCIE5 is only provided to the PCIE Slot, not M.2
* The updated iGPU is up to 32EU units (12900k)
* The maximum size core configuration is as follows: 8+8 for i9, 8+4 for i7 and 6+4 for i5. There are further variations depending on specific CPU model, thats the maximum (so the top i5 will have 6 big core and 4 little core)