Samsung announces 3nm MBCFET

[dalekphalm]

5 minutes ago, pas008 said:

point below

Faster CPU, more cores, faster single threaded performance, less power consumption or more performance for the same power consumption.

 

Seriously? The consumer might not care how those improvements are made, but they are made in part due to node shrinks.

[Guest]

AMD and Intel aren’t really competing with the same space as Samsung, but it would be interesting to see the next Apple processor.

[pas008]

1 minute ago, dalekphalm said:

Faster CPU, more cores, faster single threaded performance, less power consumption or more performance for the same power consumption.

 

Seriously? The consumer might not care how those improvements are made, but they are made in part due to node shrinks.

 

broadwell to coffeelake refresh has shown that more mghz , more cores,  and single threaded can be had pretty much on same node lol

 

[dalekphalm]

2 minutes ago, pas008 said:

 

broadwell to coffeelake refresh has shown that more mghz , more cores,  and single threaded can be had pretty much on same node lol

 

Yes, some can be squeezed out - often at the expense of OC headroom. But node shrinks allow more headroom for more cores, and higher frequencies, due to lower power consumption and better heat dissipation (comes with lower power consumption).

 

Node shrinks, just like arch refreshes, work in tandem with one another. One, by itself, cannot solve all the problems. Nor can the other. Each one can slowly chip away at the problems, making for a better CPU.

 

Do you seriously want us to be stuck at 32nm?

 

We literally ran into a GPU wall because of GPU's being stuck on 28 nm for 3 generations worth of GPU's, with incredibly minor improvements between each generation (due to TSMC cancelling their 22nm node - so NVIDIA and AMD had to wait for 16/14nm).

[dalekphalm]

10 minutes ago, floofer said:

AMD and Intel aren’t really competing with the same space as Samsung, but it would be interesting to see the next Apple processor.

AMD doesn't compete with Samsung at all, since AMD doesn't have any fabs.

 

Intel doesn't do a lot of custom fab work for third parties, so there's less overlap. But there's nothing stopping one or the other from leveraging their fabs to compete in the same market segment.

 

If Samsung's 3nm fab kills it, you can be damn sure AMD and NVIDIA will be looking their way for CPU's and GPU's. And of course, any ARM production too.

[Trixanity]

26 minutes ago, pas008 said:

point below

 

 

What has nodes ever done for consumers then? Why is 32nm the cut off point? There's some logic missing here.

 

In any case, nodes are decidedly a B2B product but the products they spawn are what consumers benefit from.

[Sauron]

1 hour ago, pas008 said:

where are you getting a full ghz?

i7 2700k boost: 3.9GHz

i9 9900k boost: 5.0GHz

1 hour ago, pas008 said:

amd is responsible for more cores
 

they couldn't have done that without a smaller production node

seriously, this isn't a complicated concept. Stop insisting, you're just wrong.

1 hour ago, pas008 said:

like i said they get more we get less

look at it in silicon size we are getting less of the wafer by far they are getting more per wafer

What are you even talking about? Transistor count is what matters here, and you're getting at least twice as many transistors in the same space.

1 hour ago, pas008 said:

32nm we had bigger chunk of the wafer 5ghz was achievable for enthusiasts and barely still is lol amd isnt hitting 5ghz

False, and even if it were true a stock 2700x still smokes a 5GHz 2600k.

1 hour ago, pas008 said:

and intel is doing more cores because of AMD

and they can do that because they're on a smaller die size than they were when sandy bridge came out.

 

In other news, 1+1 = 2.

[pas008]

46 minutes ago, Sauron said:

i7 2700k boost: 3.9GHz

i9 9900k boost: 5.0GHz

they couldn't have done that without a smaller production node

seriously, this isn't a complicated concept. Stop insisting, you're just wrong.

What are you even talking about? Transistor count is what matters here, and you're getting at least twice as many transistors in the same space.

False, and even if it were true a stock 2700x still smokes a 5GHz 2600k.

and they can do that because they're on a smaller die size than they were when sandy bridge came out.

 

In other news, 1+1 = 2.

2700k 3.9 boost with stock cooling under 50 degrees (enthusiast headroom 5 ghz)

7700k 4.5 boost with stock cooling under 50 degrees lol good luck (enthusiast headroom 5.1)

Oh wait they are pushing them farther to their limit with smaller headroom for overclocks

9900k  5.0 boost with stock cooling oh wait you cant use stock cooling unless you want constant throttling

 

you know we had 10 cores on 32nm and 18 on 22nm right

wasnt offered to consumers because it didnt evolve to needing them yet and intel milking of course til amd came in with ryzen with chiplet design that is more efficient

 

does a chip designer/maker get more chips per wafer with a smaller node?

so we consumers are getting less of the silicon pie at the cost of a few hundred mhz and thx to amd more cores now

 

2700x vs 2600k point? evolving forward so it should happen better isa/ipc

we had 32nm doing 5ghz with sandy bridge  overclocked and stock pile driver too

again you know we had 10 cores on 32nm and 18 on 22nm right

 

 

 

 

 

[dalekphalm]

6 minutes ago, pas008 said:

2700k 3.9 boost with stock cooling under 50 degrees (enthusiast headroom 5 ghz)

7700k 4.5 boost with stock cooling under 50 degrees lol good luck (enthusiast headroom 5.1)

Oh wait they are pushing them farther to their limit with smaller headroom for overclocks

9900k  5.0 boost with stock cooling oh wait you cant use stock cooling unless you want constant throttling

 

you know we had 10 cores on 32nm and 18 on 22nm right

wasnt offered to consumers because it didnt evolve to needing them yet and intel milking of course til amd came in with ryzen with chiplet design that is more efficient

 

does a chip designer/maker get more chips per wafer with a smaller node?

so we consumers are getting less of the silicon pie at the cost of a few hundred mhz and thx to amd more cores now

 

2700x vs 2600k point? evolving forward so it should happen

we had 32nm doing 5ghz with sandy bridge  overclocked and stock pile driver too

again you know we had 10 cores on 32nm and 18 on 22nm right

 

 

 

 

 

The best Sandy Bridge Xeon, Sandy Bridge-EP E5-4650 was an 8-core CPU w/ a 130W TDP and a base clock of 2.7 GHz. Want to know the boost clock? Yeah that's a mega-fast 3.3 GHz.

https://ark.intel.com/content/www/us/en/ark/products/64622/intel-xeon-processor-e5-4650-20m-cache-2-70-ghz-8-00-gt-s-intel-qpi.html

 

AMD had some 16-core single threaded CPU's on 32nm. The fastest of which was the Opteron 6386 SE, which at a 140W TDP and a boost clock of 3.5 GHz.

 

So yeah. Now we CPU's with the same core count and much faster frequency, plus your typical architectural single thread performance increases, at nearly half the power consumption. So, again, you're telling me this doesn't benefit the consumer? Really?

[Results45]

10 hours ago, Stroal said:

.

Spoiler

 

https://www.anandtech.com/show/14333/samsung-announces-3nm-gaa-mbcfet-pdk-version-01

.

 

 

 

 

My thoughts:

While I think 3nm MBCFET is a ways off for consumer end, this announcement is a game changer, especially considering Intel's current position, their inability to hit 10nm, and considering their rival, AMD, is just starting on 7nm. Hopefully we'll see some collaboration between the 2 major CPU players in regards to this technology, which I think could bring it to the market even faster.  

 

Indeed it is:

 

ASML designs and makes EUV Lithography machines

 

Both Samsung and TSMC begin production of 5nm this year so phones and tablets that transform into laptops seem more likely than ever. I'm thinking Android phones that unfold into Chromebooks, folding iPad Pros with OS X Mojave, and Qualcomm "always-on" ultrabooks competitive with Intel's offerings.

 

I guess we'll see what comes along for 2nm in 2024/2025 

 

 

P.S. If you're into what it takes to develop 7, 5, and 3nm transistors, this a pretty great read: https://semiengineering.com/transistor-options-beyond-3nm/ 

[mr moose]

This is good news,  I wouldn't call it a game changer, in fact it's basically par for the course (they already are using 7nm for their phone processors and 5nm is due next year),  but the good thing is that manufacturing that small isn't without new limitations so the announcement would indicate they are confident of work around's.

 

 

[pas008]

52 minutes ago, dalekphalm said:

The best Sandy Bridge Xeon, Sandy Bridge-EP E5-4650 was an 8-core CPU w/ a 130W TDP and a base clock of 2.7 GHz. Want to know the boost clock? Yeah that's a mega-fast 3.3 GHz.

https://ark.intel.com/content/www/us/en/ark/products/64622/intel-xeon-processor-e5-4650-20m-cache-2-70-ghz-8-00-gt-s-intel-qpi.html

 

AMD had some 16-core single threaded CPU's on 32nm. The fastest of which was the Opteron 6386 SE, which at a 140W TDP and a boost clock of 3.5 GHz.

 

So yeah. Now we CPU's with the same core count and much faster frequency, plus your typical architectural single thread performance increases, at nearly half the power consumption. So, again, you're telling me this doesn't benefit the consumer? Really?

westmere had 10 core

https://ark.intel.com/content/www/us/en/ark/products/53580/intel-xeon-processor-e7-8870-30m-cache-2-40-ghz-6-40-gt-s-intel-qpi.html

 

 

it also has been shown by intel a mature node can do those same things from broadwell to coffelake refresh

single thread, cores, and frequency barely

 

such a hurry to keep up with moores law that the consumers are trading few hundred mhz for these chip maker/designers to bank on

[dalekphalm]

1 minute ago, pas008 said:

westmere had 10 core

https://ark.intel.com/content/www/us/en/ark/products/53580/intel-xeon-processor-e7-8870-30m-cache-2-40-ghz-6-40-gt-s-intel-qpi.html

Missed that one. Okay so we've got a 10 core CPU that has a boost clock of 2.8 GHz, and still runs at 130W TDP.

1 minute ago, pas008 said:

it also has been shown by intel a mature node can do those same things from broadwell to coffelake refresh

single thread, cores, and frequency barely

You haven't shown that at all. Your 10-core CPU has frequencies that are embarrassing by today's standards.

1 minute ago, pas008 said:

such a hurry to keep up with moores law that the consumers are trading few hundred mhz for these chip maker/designers to bank on

There's no doubt that Intel has been holding back, before Zen came out. But your insistence that node shrinks have done nothing for the consumer is literally nonsense. It's not based on any logic or facts.

[pas008]

16 minutes ago, dalekphalm said:

Missed that one. Okay so we've got a 10 core CPU that has a boost clock of 2.8 GHz, and still runs at 130W TDP.

You haven't shown that at all. Your 10-core CPU has frequencies that are embarrassing by today's standards.

There's no doubt that Intel has been holding back, before Zen came out. But your insistence that node shrinks have done nothing for the consumer is literally nonsense. It's not based on any logic or facts.

broadwell to coffeelake clocks on 4 core 8 threads and even all of coffeelake refresh lineup with more cores not to mention isa/ipc

 

logic? facts?

of course it helps us but at what cost? are they getting more chips per wafer on node shrink? is that ever been passed on to us on any node shrink

 

[dalekphalm]

1 minute ago, pas008 said:

broadwell to coffeelake clocks on 4 core 8 threads and even all of coffeelake refresh lineup with more cores not to mention isa/ipc

 

logic? facts?

of course it helps us but at what cost? are they getting more chips per wafer on node shrink? is that ever been passed on to us on any node shrink

I'm really not sure what you fail to understand. ISA/architectural improvements can increase IPC. No one doubts this. There are numerous examples. These increases are limited, however. And come increasingly at a higher fixed base cost, with lower yields, while maintaining the same node.

 

So, what's your point? We've done really well on 14nm. Are you saying we should just stop here?

[pas008]

21 minutes ago, dalekphalm said:

I'm really not sure what you fail to understand. ISA/architectural improvements can increase IPC. No one doubts this. There are numerous examples. These increases are limited, however. And come increasingly at a higher fixed base cost, with lower yields, while maintaining the same node.

 

So, what's your point? We've done really well on 14nm. Are you saying we should just stop here?

nevermind i'm just dumb

[leadeater]

1 hour ago, dalekphalm said:

Missed that one. Okay so we've got a 10 core CPU that has a boost clock of 2.8 GHz, and still runs at 130W TDP.

Today we can do 24 cores with a boost of 3.6Ghz at 135W TDP. No chance of that happening on that old node, not only will the die be likely bigger than those fabs could actually print it would be some stupid 600W+ TDP.

[Dylanc1500]

32 minutes ago, leadeater said:

Today we can do 24 cores with a boost of 3.6Ghz at 135W TDP. No chance of that happening on that old node, not only will the die be likely bigger than those fabs could actually print it would be some stupid 600W+ TDP.

Hey, if we aren't using the whole wafer for the die we are doing something wrong.

[Stroal]

2 hours ago, mr moose said:

This is good news,  I wouldn't call it a game changer, in fact it's basically par for the course (they already are using 7nm for their phone processors and 5nm is due next year),  but the good thing is that manufacturing that small isn't without new limitations so the announcement would indicate they are confident of work around's.

 

 

They are using FinFet though currently, and FinFet for 5nm as well. This is a new way of doing GAA at 3nm. We are talking a 40% reduction in power consumption , 40% increase in performance, compared to 7nm FinFet. We are talking sub .75v power consumption. That's nutty. 

[Results45]

1 hour ago, leadeater said:

Today we can do 24 cores with a boost of 3.6Ghz at 135W TDP. No chance of that happening on that old node, not only will the die be likely bigger than those fabs could actually print it would be some stupid 600W+ TDP.

 

.......or 255/400W TDP with 28/56 cores at 3.8Ghz:

• https://ark.intel.com/content/www/us/en/ark/products/189452/intel-xeon-w-3175x-processor-38-5m-cache-3-10-ghz.html
• https://ark.intel.com/content/www/us/en/ark/products/194146/intel-xeon-platinum-9282-processor-77m-cache-2-60-ghz.html

[leadeater]

1 minute ago, Results45 said:

 

.......or 255/400W TDP with 28/56 cores at 3.8Ghz:

• https://ark.intel.com/content/www/us/en/ark/products/189452/intel-xeon-w-3175x-processor-38-5m-cache-3-10-ghz.html
• https://ark.intel.com/content/www/us/en/ark/products/194146/intel-xeon-platinum-9282-processor-77m-cache-2-60-ghz.html

Yea I went for the same TDP so the comparison was easier to look at 

[Results45]

1 minute ago, leadeater said:

Yea I went for the same TDP so the comparison was easier to look at 

 

K. Cuz I was looking at Ghz

[mr moose]

6 hours ago, Stroal said:

They are using FinFet though currently, and FinFet for 5nm as well. This is a new way of doing GAA at 3nm. We are talking a 40% reduction in power consumption , 40% increase in performance, compared to 7nm FinFet. We are talking sub .75v power consumption. That's nutty. 

It's still what I would call within the expected realms of advancement.    Every node change comes with new tech to overcome issues that the older tech can't.   To be Honest, I now fully expect 3nm to be thing, I just don't expect it to be ground breaking.

[pas008]

13 hours ago, Dylanc1500 said:

Hey, if we aren't using the whole wafer for the die we are doing something wrong.

 

14 hours ago, leadeater said:

Today we can do 24 cores with a boost of 3.6Ghz at 135W TDP. No chance of that happening on that old node, not only will the die be likely bigger than those fabs could actually print it would be some stupid 600W+ TDP.

 

but thats my point, are they getting more chips per wafer now? we use to get a bigger piece of the silicon pie for the price their profits per wafer has increased significantly considering their consistent markup

but we consumers/mainstream in return really only got few hundred mhz if that and some power savings

intel has shown its 14nm broadwell to coffeelake refresh was able to give frequency, ipc/isa improvements, cores(thx amd but wait its still a chiplet design), other features, etc

not saying we should say lets stay but hopefully amd ends this nonsense with 7nm

 

edited multitasking at work

[leadeater]

28 minutes ago, pas008 said:

but thats my point, are they getting more chips per wafer now? we use to get a bigger piece of the silicon pie for the price their profits per wafer has increased significantly considering their consistent markup

That doesn't matter, older nodes were far cheaper. Die area means extremely little as most of the cost is in R&D recovery and all the other overheads like operating costs and capital expansion projects. The actual dies are not materially worth much at all and is actually mostly inconsequential to the final product cost.