Rumor: Intel to Introduce Big.Little Architecture for Desktop With Alder Lake-S, New LGA 1700 Socket

[SL2]

The thing is, the 8 core processor with SMT will be less expensive to manufacturer and actually use less power than an 8+8 processor. Big.Little makes no sense whatsoever on a desktop.

If you're right then I suppose Intel wouldn't even think about it, but they do. Maybe they know something we don't?

Using only HT will obviously cost less, but that doesn't mean anything if we don't know the pro's of going with this solution.

 

[ppn]

Big core design could mean that one such core can be outputting the same IPC without the split Into logical threads since the software threads do the same thing by executing on the same core, so SMT can be avoided.
Cpus can idle at 5 watts since sandy, big.little makes no sense in the way that it is implemented in smartphones, will it idle at 0.1 watt, or little core can just be used to offload some less demanding threads that would otherwise fragment and bottleneck the big core pipeline with lost clock cycles. I think Intel plans to move all the pcie lanes to the cpu substrate, including onchip thunderbolt usb to avoid the pcie4x bottleneck with PCH therefore the need for 500 more contacts.

 

[R0H1T]

The rumour is PCIe 4.0. Sometimes pins are reserved for future use as well.

Yet we get new sockets virtually every gen with Intel, when they aren't actually needed cause $$$

This is exciting so long as they're priced appropriately but knowing Intel it's safe to say you'll probably get the exact opposite

 

[DeathtoGnomes]

Socket Hockey!

Uhh, will this fit? How about this one?

 

[ppn]

Better than GPUs than throw away a perfectly good cooler, pcb and GDDR.
if intel move the PCH part to the CPU, motherboard will become really cheap. so this is good.

 

[TheinsanegamerN]

Meanwhile AMD is going into, what, its 4th year on the same socket with a 4th generation of CPU, each of which shows substantial improvements in performance.

Pathetic showing here. Intel shouldnt even bother until they have something to actually show performance wise.

 

[ppn]

And each of which you have to pay again and again for this slighlty better version of 8 core $299 r1700, sell for 99, buy $299 2700, $299 r3700, r4700, buy and sell end you end up paying $1000 for the r4700 8 core that will be dwarfed by the big intel core, it will be demolished.

 

[ironcerealbox]

If you're right then I suppose Intel wouldn't even think about it, but they do. Maybe they know something we don't?

Using only HT will obviously cost less, but that doesn't mean anything if we don't know the pro's of going with this solution.

Unless it's a temporary implementation until Intel gets HT security issues fixed (assuming there are security issues with HT)?

 

[Gmr_Chick]

And each of which you have to pay again and again for this slighlty better version of 8 core $299 r1700, sell for 99, buy $299 2700, $299 r3700, r4700, buy and sell end you end up paying $1000 for the r4700 8 core that will be dwarfed by the big intel core, it will be demolished.

Kind of a pointless argument on behalf of Intel there, bud. Ryzen's only gotten better with each generation. I just upgraded from a 1600 to a 3600 and the difference is like night and day for me. And, "end up paying $1000 for the r4700 8 core"? Uh...you DO know that not even the Ryzen 9 CPUs even cost $1000 even at this very moment, right? The simple fact of the matter is whenn it comes to the mainstream desktop platform, Intell's got absolutely nothing in their current product stack that competes with Ryzen 9. Hell, their even seeing their dominance over AMD in recent years rapidly shrinking on account of what AMD has been able to do with each phase of Ryzen, so much so that AMD's Ryzen and Threadripper are just overall better buys at this point.

 

[ratirt]

Kind of a pointless argument on behalf of Intel there, bud. Ryzen's only gotten better with each generation. I just upgraded from a 1600 to a 3600 and the difference is like night and day for me. And, "end up paying $1000 for the r4700 8 core"? Uh...you DO know that not even the Ryzen 9 CPUs even cost $1000 even at this very moment, right? The simple fact of the matter is whenn it comes to the mainstream desktop platform, Intell's got absolutely nothing in their current product stack that competes with Ryzen 9. Hell, their even seeing their dominance over AMD in recent years rapidly shrinking on account of what AMD has been able to do with each phase of Ryzen, so much so that AMD's Ryzen and Threadripper are just overall better buys at this point.

I know something about that. I got 2x threadrippers 3970x. I have considered Intel's products but it just didn't add up to me. For the price of Intel's offering, I got easily two threadrippers and actually got spare cash.

With the little, big cores. Hmm. More pins. Hmm. two different cores two different power deliveries maybe more pins are needed for separate power delivery. Somehow this tells me it is not about performance increase it is about power savings due to 10nm insufficient manufacturing process. This kinda sounds to me, instead of increasing performance, giving more, faster cores, Intel goes smaller cores packed with regular calling (probably, since I haven't seen this as of yet) a better solution? For me it sound and looks like crippling product for obvious reasons. That also means Intel doesn't have a lot to offer against AMD.

 

[ZoneDymo]

So that Means 1700 pins right?
i Mean AM4 has 1331 so i wonder what those extra 369 pins will bring to the table...
More Power? Faster/Wider Bandwich to the CPU? PCI-E 4-5? Better RAM performance?

see that is what I dont understand about Intel, they have socket 2011...they have had that for a while now, why not...stick with that? pins for days AND no need to produce/develop yet another new socket.
Keep sticking to 2011 untill there are not enough pins anymore to push the data.

 

[londiste]

Small cores are OoO but not usually HT , probably used for system level tasks and battery life help, that wattage!,

This, undoubtedly. Atom SoCs are an under 5W affair even on 14nm. Operating system, web browsing etc do not really need a big core. There are things in web like video decoding that might take advantage of performance but iGPU will have hardware decoding for that. Such layout would leave big cores either idle or to be used only when more performance is required. LGA1700 seems to be a desktop-class socket and 8+8 cores can do well in this area.

 

[dj-electric]

see that is what I dont understand about Intel, they have socket 2011...they have had that for a while now, why not...stick with that? pins for days AND no need to produce/develop yet another new socket.
Keep sticking to 2011 untill there are not enough pins anymore to push the data.

The amount of pins is never the engineering crunch on a socket. They could have 6000 pins for all they care. Its the designation and layout planning that needs improving and modifications on those. You can't just deploy the same socket and expect it to do new things and work differently.

 

[ARF]

AMD's Bulldozer had those modules with two half-baked cores per module and got sued for misleading marketing.

The idea for little cores on the desktop is somehow contradictory, though.
In gaming, for example, 8 normal cores can be used by the game, and 8 small cores by the OS.
But these small cores have much lower IPC and performance, and is it even meaningful to use such instead of normal cores that just switch between the tasks rapidly (in case the cores number is not enough) ?

 

[Vayra86]

Please be Golden Cove. It’ll easily be the fastest CPU core in the world.

Don't spoil it for me... is it because they will make it out of gold?

As far as I'm concerned, they can shove this lake right up there with the rest of the watery graveyard. I'm approaching Intel like I do AMD GPU: Product pls!

 

[ratirt]

Don't spoil it for me... is it because they will make it out of gold?

As far as I'm concerned, they can shove this lake right up there with the rest of the watery graveyard. I'm approaching Intel like I do AMD GPU: Product pls!

At least someone had the guts to say it.
When I see it I believe it. I don't take in on faith.

Anyway it is weird what Intel is offering and what rumors say. Hard to see any advantages with these big little cores.

 

[Vayra86]

At least someone had the guts to say it.
When I see it I believe it. I don't take in on faith.

Anyway it is weird what Intel is offering and what rumors say. Hard to see any advantages with these big little cores.

 

[londiste]

AMD's Bulldozer had those modules with two half-baked cores per module and got sued for misleading marketing.

Bulldozer is a bad analogy. What is describes in this story is very different from that.

The idea for little cores on the desktop is somehow contradictory, though.
In gaming, for example, 8 normal cores can be used by the game, and 8 small cores by the OS.
But these small cores have much lower IPC and performance, and is it even meaningful to use such instead of normal cores that just switch between the tasks rapidly (in case the cores number is not enough) ?

These small cores also need much more power and are... well, much smaller.

Edit:
4-core 14nm Apollo Lake die is roughly 9 x 10.2 mm - 91.4 mm^@. From that die, 4 Goldmont cores are 11mm^2.
For comparison:
- One 7nm Zen2 core (without L3) is 3.64mm^2 - 7nm scaling over 14nm is about 60%.
- One 14/12nm Zen/Zen+ core is about 7mm^2.
- One 14nm Skylake core is about 8.73 mm^2 - Kaby Lake, Coffee Lake/Refresh and supposedly Comet Lake are all about the same.

 

[wurschti]

Nope... knowing Intel, those little cores will be Atom ones and the others are going to be able to handle real work.
AMD already has a big 16-core CPU, why bother with this one?
Also, the implementation will take a lot of time to develop, leading to more issues with the OS.

Interesting concept, but I think it will flop really soon.

 

[Vayra86]

There is a place for this concept but it is mobile/laptop/SFF. And probably only serves a purpose untill 7nm is mainstream for Intel. They're gonna have to shrink radically at some point...

I can totally see the Little core serve a purpose for idle usage with the big ones being shut off completely. You only need a few hundred mhz to run some sensors, gyro, gps, etc, perhaps even a simple browser or background process, push services from the OS. This will give Intel an edge in battery life, probably greater than anything AMD can bring to Zen chiplets atm.

10nm ain't happening, that is for sure, and 14nm won't cut it here for battery life. Those laptop CPUs are fast becoming Atoms with a ridiculous turbo the way they're going with baseclocks... heck even the desktop CPUs are starting to lean that way. Not touching these with a 10nm pole tho.

 

[ARF]

There is a place for this concept but it is mobile/laptop/SFF. And probably only serves a purpose untill 7nm is mainstream for Intel. They're gonna have to shrink radically at some point...

I can totally see the Little core serve a purpose for idle usage with the big ones being shut off completely. You only need a few hundred mhz to run some sensors, gyro, gps, etc, perhaps even a simple browser or background process. This will give Intel an edge in battery life, probably greater than anything AMD can bring to Zen chiplets atm.

AMD for sure can control the clocks of its chiplets and that's the root for the battery duration problem.

 

[londiste]

AMD for sure can control the clocks of its chiplets and that's the root for the battery duration problem.

AMD does not use chiplets in mobile CPUs, primarily because power is an issue. Ryzen 3000 will use 10-15W at idle due to SoC or more likely IF between dies (considering this was not an issue for Zen/Zen+). Ryzen 4000 APUs are monolithic and sidestep this problem.

 

[Vayra86]

AMD for sure can control the clocks of its chiplets and that's the root for the battery duration problem.

Just controlling clocks isn't enough anymore, that is what Intel was doing five years ago. Yes, Zen is very dynamic but its still a server/enterprise arch first and foremost, with limitations.

 

[ARF]

Just controlling clocks isn't enough anymore

There are no limitations. They can clock at 100 MHz and 0.4 V and it will run just as fine.

 

[Vayra86]

There are no limitations. They can clock at 100 MHz and 0.4 V and it will run just as fine.

And here is the result of that: at least 20% more idle usage, in this example.