First pictutres of Intels Hybrid Design Alder Lake CPUs (LGA1700) emerge with a rectangular package

[Ankh Tech]

It does since that means more transistors in the same size cpus as the node size makes them smaller 

[Master Disaster]

55 minutes ago, Ankh tech tips said:

It does since that means more transistors in the same size cpus as the node size makes them smaller 

Not at all. Transistor efficiency matters just as much as density, if you have twice as many transistors that only have half the efficiency of the old ones then the net result is the same performance.

 

In layman's terms node size simply means the thickness of the lithography used to print the PCB onto the wafer and the smaller you print the more problems arise, most notably is something called quantum tunnelling. Essentially the traces on the wafer are so small they cannot hold onto the electricity and the ions tunnel between the traces causing shorts (again this is an ultra simplistic explanation). This can be avoided by using different lithography techniques and by using silicon compounds but the reality is, eventually we will reach a point where physics (or at least our current understanding of it) won't allow us to go any smaller. This is why people are investing in quantum computing and investigating different CPU technology such as light based.

[Ankh Tech]

4 minutes ago, Master Disaster said:

Not at all. Transistor efficiency matters just as much as density, if you have twice as many transistors that only have half the efficiency of the old ones then the net result is the same performance.

 

In layman's terms node size simply means the thickness of the lithography used to print the PCB onto the wafer and the smaller you print the more problems arise, most notably is something called quantum tunnelling. Essentially the traces on the wafer are so small they cannot hold onto the electricity and the ions tunnel between the traces causing shorts (again this is an ultra simplistic explanation).

Well amd is more efficient than intel and has a smaller node so it should be better overall right?

[Master Disaster]

1 minute ago, Ankh tech tips said:

Well amd is more efficient than intel and has a smaller node so it should be better overall right?

Total CPU efficiency is not the same thing as transistor efficiency.

[Ankh Tech]

well they are related right?

[Master Disaster]

12 minutes ago, Ankh tech tips said:

well they are related right?

In a very removed way, yes. Remember the CPU is more than just transistors, it also has interconnects, integrated memory controllers, cache, SOCs for micro control of the package and if the CPU has one, a GPU as well.

 

Transistor efficiency is more related to IPC than anything else, total CPU efficiency takes into account the full package.

[Arika]

8 minutes ago, Ankh tech tips said:

well they are related right?

node size can make a processor more efficient but it's not the deciding factor.

[Ankh Tech]

Wait just a question each transistor connect to the 3 or 4 adjacent transistors right, why don't they make a cpu where transistors connect to each transistor,like your brains nervous system,  that would reduce load on each transistor and make it faster right?

[porina]

9 minutes ago, Ankh tech tips said:

Wait just a question each transistor connect to the 3 or 4 adjacent transistors right, why don't they make a cpu where transistors connect to each transistor,like your brains nervous system,  that would reduce load on each transistor and make it faster right?

Connections are only present where actually needed, which depends on the function. Going vertical does help, but not without its downsides. Mostly cost and power density. ICs are built up in layers. More layers is more complexity and more risk. For something as simple as flash, they seem to have that under control. Let's say you double stack in one area, you're doing double the work, double the power. That's going to be a challenge. And this is ignoring the risk that the layers will interfere with each other, making design more complicated.

[CommanderAlex]

3 hours ago, Deli said:

Does it mean CPU coolers require a new mount?

I doubt it as Skylake-X (Intel's 9000 series HEDT) is 52.5mm x 45.0mm so Alder Lake is 37.5mm x 45.0 mm and my Noctua NH-D15 works great on Skylake-X.

[Vector0102]

I mean why hasn't Intel moved the igpu off die and make it a chiplet design? Having that Igpu as a seperate die could help yeilds and costs, in laptops they could probably have it monolithic, just like AMD does it. Removing it may not be an option, because there are people who need a powerful CPU but not much GPU power. So why dont they just do that?

[Master Disaster]

8 minutes ago, Vector0102 said:

I mean why hasn't Intel moved the igpu off die and make it a chiplet design? Having that Igpu as a seperate die could help yeilds and costs, in laptops they could probably have it monolithic, just like AMD does it. Removing it may not be an option, because there are people who need a powerful CPU but not much GPU power. So why dont they just do that?

I have to imagine there is a reason why they haven't gone down the chiplet route, perhaps there's an architectural issue that prevents it? The thing with CPU design is that its just as complicated as you would expect it to be and the majority of us on here (myself included) probably wouldn't understand the reasons anyway

[George.]

I think they are planning on sort of battling Apple silicon by not making an ARM cpu but rather trying to make x86 shine a bit more as far as energy efficiency is concerned.

DDR5 maaybe, PCI-e 5 I doubt it .... they've been optimizing the 14nm platform for ages now, idk if any of these optimizations will transfer over to newer lithographies. Let's hope they bring sth good to the table, competition is always good for us in the end of the day.

[Vector0102]

afaik just shifting the same architecture to a smaller node can do wonders for efficiency, but maybe it dont work like that

[sazrocks]

6 hours ago, LAwLz said:

Intel is already producing chips on their 10nm node, and (if it doesn't get delayed) they will ship desktop chips on their 10nm node in Q1 2021.

That seems a little early.
From the OP:

Quote

Intel’s first desktop processors based on 10nm SuperFin architecture will officially launch in the second half of 2021. Our sources believe that the launch of Alder Lake will be closer to the fourth quarter, 

Rocket lake (14nm) is going to launch ~Q1 2021. Alder Lake will be closer to the end of the year, when AMD will be significantly closer to having Zen 4 ready on 5nm.

[straight_stewie]

8 hours ago, Ankh tech tips said:

By the time Intel reaches 10nm AMD would be 5nm

I can say with 100% certainty that this is false.

Intel has already reached 10nm, just not on socketed processors. However, if you'd read the op, you'd know that they plan to release 10nm socketed processors in late 2021.

[Ankh Tech]

Then amd would reach 5nm 5 months after that at most, that makes amd in the lead

[straight_stewie]

20 minutes ago, Ankh tech tips said:

Then amd would reach 5nm 5 months after that at most, that makes amd in the lead

In the lead how? AMD's current 10nm offerings still lose to Intels 14nm offerings in basically every real world performance benchmark. This has been the case for 2 generations now.

I don't see any reason why the move to 5nm itself will change this. 

I'll be very blunt here: From my perspective, it's a very tough sell to make any other metric more important than raw performance in desktop processors. I couldn't possibly care any less about what node size a manufacturer is using. The sole care should be benchmark results, and retail cost if that's your jam.

Also, welcome to the forum! Please remember to quote users when you reply to them so that they get a notification that you've replied.

[Ankh Tech]

Just now, straight_stewie said:

In the lead how? AMD's current 10nm offerings still lose to Intels 14nm offerings in basically every real world performance benchmark. This has been the case for 2 generations now.

I don't see any reason why the move to 5nm itself will change this. 

I'll be very blunt here: From my perspective, it's a very tough sell to make any other metric more important than raw performance in desktop processors. I couldn't possibly care any less about what node size a manufacturer is using. The sole care should be benchmark results, and retail cost if that's your jam.

Also, welcome to the forum! Please remember to quote users when you reply to them so that they get a notification that you've replied.

Ok but how to qoute a specific sentence?

[Sauron]

9 hours ago, Master Disaster said:

Rumours suggest Intel intends to keep LGA1700 for 3 generations though which could make this a compelling upgrade.

Just because the pinout is the same doesn't mean it's going to be compatible across releases.

8 hours ago, Ankh tech tips said:

Some people do car about efficiency more than performance

Sure, but for enthusiasts efficiency improvements are extremely boring. To an individual customer this makes pretty much no difference.

8 hours ago, Moonzy said:

and node size matters because...????

Typically a smaller node allows for higher density and therefore... more CPU on the same package. It also often comes with a higher frequency overhead due to lower temperatures (though that's becoming less and less true as taking the heat out of ultradense chips is proving to be a bit of a challenge). None of this is guaranteed of course.

8 hours ago, TetraSky said:

All jokes aside, really makes you wonder though, where will it stop?

Does it really matter how large the CPU package is? It's not even that much larger than LGA115x...

[Moonzy]

caution: derailing

3 minutes ago, Sauron said:

Typically a smaller node allows for higher density and therefore... more CPU on the same package.

only if we're comparing same architecture and such

smaller node is just one factor in many to dictate a CPU's performance, which in the end is what matters.

so using node size in an argument is pointless, bring out the benchmarks.

 

not to mention the node size names have long been decoupled from actual node sizes.

 

[spartaman64]

9 hours ago, LAwLz said:

Do we even know if that's the case?

It might be that the current 6 core will be replaced with a 6+2 (6 big, 2 little) core config. I seriously doubt Intel will just replace big cores with small ones and call it a day.

 

 

Those numbers are fairly meaningless.

What Intel calls "10nm", TSMC calls "7nm".

 

Intel is already producing chips on their 10nm node, and (if it doesn't get delayed) they will ship desktop chips on their 10nm node in Q1 2021.

AMD is ahead of Intel a lot right now, but that's only because AMD just launched their new products and Intel will launch theirs in 3-5 months. I don't think AMD will release the 6000 series of CPUs in ~3-5 months.

its not meaningless its just not the only factor to transistor density. 

intel's 10nm has a density of 100.8 million transistors per mm^2 and tsmc's 7nm n7ff and n7p has a density of 96.5 MT/mm^2 while n7ff+ is 113.9MT/mm^2

and tsmc 5nm is 173 MT/mm^2

 

the problem was never whether or not they can make 10nm chips but if they can make them with sufficient yields

 

22 minutes ago, straight_stewie said:

In the lead how? AMD's current 10nm offerings still lose to Intels 14nm offerings in basically every real world performance benchmark. This has been the case for 2 generations now.

I don't see any reason why the move to 5nm itself will change this. 

I'll be very blunt here: From my perspective, it's a very tough sell to make any other metric more important than raw performance in desktop processors. I couldn't possibly care any less about what node size a manufacturer is using. The sole care should be benchmark results, and retail cost if that's your jam.

Also, welcome to the forum! Please remember to quote users when you reply to them so that they get a notification that you've replied.

if you only count gaming benchmarks as real world benchmarks sure but in pretty much everything else amd's 7nm cpus beat intel's cpus.

this is because intel's ring bus architecture has lower latency than amd's mesh architecture and games which are reliant on latency benefit a lot from ring bus. but ring bus doesnt scale to more cores as well as mesh so intel will be forced eventually to switch

50 minutes ago, straight_stewie said:

I can say with 100% certainty that this is false.

Intel has already reached 10nm, just not on socketed processors. However, if you'd read the op, you'd know that they plan to release 10nm socketed processors in late 2021.

intel has been able to produce 10nm chips since 2018 the problem is they dont have sufficient yields for large scale production 

[Sauron]

1 minute ago, Moonzy said:

only if we're comparing same architecture and such

Yes, I did say "typically".

2 minutes ago, Moonzy said:

smaller node is just one factor in many to dictate a CPU's performance, which in the end is what matters.

so using node size in an argument is pointless, bring out the benchmarks.

We're speculating at this point so we have to make do with what information we have. Yes, obviously, benchmarks will have the final word but we can make a rough guess in the meantime. It's kind of like saying that a given car has a V12 engine... that doesn't give you all the information to determine its performance but you can assume it's probably some sort of supercar.

5 minutes ago, Moonzy said:

not to mention the node size names have long been decoupled from actual node sizes.

Well... there isn't really a set definition so anything goes. In this case we know that this is smaller than Intel's previous nodes though which is all that matters.

[sazrocks]

2 hours ago, straight_stewie said:

AMD's current 10nm offerings still lose to Intels 14nm offerings in basically every real world performance benchmark.

I'm a bit confused by this statement. Ryzen 3000 CPUs win quite a few benchmarks against their Intel equivalents, the main loss being specifically gaming. Additionally, the 5000 series is poised to become much more competitive in this regard, and is still on 7nm (AMD has no 10nm parts).

[CommanderAlex]

1 hour ago, Moonzy said:

caution: derailing

only if we're comparing same architecture and such

smaller node is just one factor in many to dictate a CPU's performance, which in the end is what matters.

so using node size in an argument is pointless, bring out the benchmarks.

 

not to mention the node size names have long been decoupled from actual node sizes.

 

I totally agree with this as you can't compare Intel's nm to AMD's nm. They are totally different and are not representative of the actual node the processors are made from as each silicon maker has their own standards. 

 

You just can't compare TSMC's 5nm to Intel's 5nm, or any number you'd like to plug in (7nm and 7nm respectively) it doesn't work like that.