Tuesday, April 16th 2019

AMD Zen3 to Leverage 7nm+ EUV For 20% Transistor Density Increase

AMD "Zen 3" microarchitecture could be designed for the enhanced 7 nm+ EUV (extreme ultraviolet) silicon fabrication node at TSMC, which promises a significant 20 percent increase in transistor densities compared to the 7 nm DUV (deep ultraviolet) node on which its "Zen 2" processors are being built. In addition, the node will also reduce power consumption by up to 10 percent at the same operational load. In a late-2018 interview, CTO Mark Papermaster stated AMD's design goal with "Zen 3" would be to prioritize energy-efficiency, and that it would present "modest" performance improvements (read: IPC improvements) over "Zen 2." AMD made it clear that it won't drag 7 nm DUV over more than one microarchitecture (Zen 2), and that "Zen 3" will debut in 2020 on 7 nm+ EUV.
Source: PCGamesN
Add your own comment

90 Comments on AMD Zen3 to Leverage 7nm+ EUV For 20% Transistor Density Increase

#51
Vario
notb
Intel replaces sockets often. It's not a problem for most and the rest accept it. After all, motherboard features age faster than CPU performance.

Right from the beginning we were told that AM4 will last for few years and will also work with a future revamped architecture on improved semiconductor process (i.e. more cores, higher clocks etc)
It was meant to be a big jump in performance.

Until now we got 2 generations: Zen and Zen+, the latter being just slightly faster and a bit more polished. That's hardly different from what Intel is doing.
Zen 2 is rumored to be problematic. Zen 3 seems very unlikely.

So in a fairly probable scenario AM4 will be compatible with... 2.5 generations...

Intel, MSI, Asus etc...
Keep going and soon you'll be butthurt about so many companies that you'll have problems assembling a whole PC...
In addition, X470 probably won't be compatible with Zen 3 due to DDR5. Unless DDR5 support is postponed to Zen4.
Posted on Reply
#52
londiste
Xaled
No. Even at 3 generations (Zen, Zen+ and Zen) it would be much greater than what Intel has been doing for almost 8 years, changing chipset every year even when there were no "technical" need for that at all.
Intel changes sockets every 2 generations, a new chipset every generation.
AMD has so far done 2 chipsets and 2 generations and most likely outcome is 3 generations and 3 chipsets.
Posted on Reply
#53
Xaled
notb
Intel replaces sockets often. It's not a problem for most and the rest accept it. After all, motherboard features age faster than CPU performance.

Right from the beginning we were told that AM4 will last for few years and will also work with a future revamped architecture on improved semiconductor process (i.e. more cores, higher clocks etc)
It was meant to be a big jump in performance.

Until now we got 2 generations: Zen and Zen+, the latter being just slightly faster and a bit more polished. That's hardly different from what Intel is doing.
Zen 2 is rumored to be problematic. Zen 3 seems very unlikely.

So in a fairly probable scenario AM4 will be compatible with... 2.5 generations...

Intel, MSI, Asus etc...
Keep going and soon you'll be butthurt about so many companies that you'll have problems assembling a whole PC...
You don't really want AMD to give the best performance possible from a new architecture AND to keep AM4 socket forever.
You are biased and just trying to find flaws. If they kept the AM4 you'll say AHA AMD isnt giving us the promised performance increqse. While if went the opposite way and made zen3 at maximum possible performance, but with a new socket you'll say: AMD didn't keep its promise.
Posted on Reply
#54
efikkan
CTO Mark Papermaster stated AMD's design goal with "Zen 3" would be to prioritize energy-efficiency, and that it would present "modest" performance improvements (read: IPC improvements) over "Zen 2.
Perhaps it's just me, but that sort of sounds like IPC gains are less exciting or something. "IPC" gains is one of the best types of performance gains.

Jism
If they could only get those clocks up, modest 4.5 to 5GHz or so.
Single core boost?
I think people overestimate the clock gains from the new nodes. Even if TSMC 7nm beats the current nodes, it's only going to be a marginal gain. Both Zen+ and Coffee Lake are already pushing their respective nodes into throttling territory.

bug
I'm guessing the biggest news here is there's not that much room left to improve IPC after Zen2 :(
I have really wondered about Zen 3-5, will it be just "small" incremental changes?
*ahem*GCN…
We need AMD to push harder, not pull back.
Posted on Reply
#55
Vario
efikkan
Perhaps it's just me, but that sort of sounds like IPC gains are less exciting or something. "IPC" gains is one of the best types of performance gains.


Single core boost?
I think people overestimate the clock gains from the new nodes. Even if TSMC 7nm beats the current nodes, it's only going to be a marginal gain. Both Zen+ and Coffee Lake are already pushing their respective nodes into throttling territory.


I have really wondered about Zen 3-5, will it be just "small" incremental changes?
*ahem*GCN…
We need AMD to push harder, not pull back.
Higher clock with modest IPC gain might be enough to close the gap.
Posted on Reply
#56
notb
Vario
In addition, X470 probably won't be compatible with Zen 3 due to DDR5. Unless DDR5 support is postponed to Zen4.
Zen 3 will have to come with all the latest goodies. Intel is sure to launch a DDR5 platform as soon as it becomes available. That's actually the main advantage of launching new sockets all the time. ;-)

Zen 3 is coming in 2020. AM4 is supported until 2020. IMO this means Zen 3 will use a new socket already. I don't see other reason why they would phrase it like that.
Posted on Reply
#57
Xaled
bug
I'm guessing the biggest news here is there's not that much room left to improve IPC after Zen2 :(
Cause 7nm was pretty much a given for Zen3 anyway.
Die Intel really improved IPC since Sandy'Bridge?
Posted on Reply
#60
R0H1T
notb
Zen 3 will have to come with all the latest goodies. Intel is sure to launch a DDR5 platform as soon as it becomes available. That's actually the main advantage of launching new sockets all the time. ;-)

Zen 3 is coming in 2020. AM4 is supported until 2020. IMO this means Zen 3 will use a new socket already. I don't see other reason why they would phrase it like that.
You keep saying that but how early/quickly do you think Intel can support DDR5 within their consumer lineup? I predict not anytime before 2021 - ICL will have to be respun, before it can support DDR5 so chances are TGL or whatever comes after ICL will get DDR5. ICL isn't coming anytime soon, so that's anywhere between 2~3 years before we see DDR5 Intel chips with mass availability. Intel could surprise us of course, but they'll probably bring DDR5 to servers first.
Posted on Reply
#61
Tomgang
But i love intel:cry:. Joke a side. I really hope amd as well as intel comes out with a interesting cpu line up, as i next year planning a new pc. Yes i am now moving on from X58 next year.

If the rumers are true, amd ryzen 3000 series shut be pretty darn interesting.
Posted on Reply
#62
notb
R0H1T
You keep saying that but how early/quickly do you think Intel can support DDR5 within their consumer lineup?
From day 1 consumer DDR5 modules arrive in shops. Just like Intel server CPUs will arrive just in time for server-grade DDR5.
RAM is made for CPUs. There's no reason to launch DDR5 if there are no CPUs to use it.

It's obvious that announcements of DDR5 and new Intel CPUs will be synchronized.
I predict not anytime before 2021 - ICL will have to be respun, before it can support DDR5 so chances are TGL or whatever comes after ICL will get DDR5. ICL isn't coming anytime soon, so that's anywhere between 2~3 years before we see DDR5 Intel chips with mass availability. Intel could surprise us of course, but they'll probably bring DDR5 to servers first.
If DDR5 comes 2-3 years from now, it would be awful news for AMD. What will they do with their CPUs? Keep AM4 for at least 1 year longer? Change to some "AM5" and give us "AM5+" year later?

Intel, being able to quickly replace their lineup (from both technological and PR point of view), will not be harmed by this in any way.
Posted on Reply
#63
R0H1T
Why do you think Intel will be the first with DDR5, it could well be IBM, AMD or even ARM? I also speculated in another thread that Apple will likely be the first to transition towards LPDDR5, since they're moving Macbooks to Axx & low idle power is nearly an obsession for them. Don't assume Intel is the leader in everything everywhere, when history's shown quite the opposite!

https://www.tomshardware.com/news/apple-mac-arm-cpus-2020-intel,38668.html
Posted on Reply
#64
qcmadness
notb
From day 1 consumer DDR5 modules arrive in shops. Just like Intel server CPUs will arrive just in time for server-grade DDR5.
RAM is made for CPUs. There's no reason to launch DDR5 if there are no CPUs to use it.

It's obvious that announcements of DDR5 and new Intel CPUs will be synchronized.

If DDR5 comes 2-3 years from now, it would be awful news for AMD. What will they do with their CPUs? Keep AM4 for at least 1 year longer? Change to some "AM5" and give us "AM5+" year later?

Intel, being able to quickly replace their lineup (from both technological and PR point of view), will not be harmed by this in any way.
With AMD now utilizing chiplet systems, AMD could just re-spin the DDR-5 I/O chip. While for Intel, they either need to build a chip with DDR-4 and DDR-5 MC or re-spin a new chip for DDR-5.

I don't agree how you think that Intel is in a superior position to support DDR-5 than AMD.

Vario
In addition, X470 probably won't be compatible with Zen 3 due to DDR5. Unless DDR5 support is postponed to Zen4.
In fact AMD could let X470 to support Zen 3 if Zen 3 can still pair up with the Zen 2 I/O chip.
Posted on Reply
#65
notb
R0H1T
Why do you think Intel will be the first with DDR5, it could well be IBM, AMD or even ARM? I also speculated in another thread that Apple will likely be the first to transition towards LPDDR5, since they're moving Macbooks to Axx & low idle power is nearly an obsession for them. Don't assume Intel is the leader in everything everywhere, when history's shown quite the opposite!

https://www.tomshardware.com/news/apple-mac-arm-cpus-2020-intel,38668.html
Because Intel is the dominant supplier of CPUs. AMD may align with their launch or not. It's of lesser importance.
Next get Macs will keep using Intel CPUs. Not enough time to switch.

Intel is the leader of computing at the moment. It has nothing to do with what I think or "assume".
I know you don't like this, but your aversion towards Intel won't change anything either.
Posted on Reply
#67
qcmadness
notb
Because Intel is the dominant supplier of CPUs. AMD may align with their launch or not. It's of lesser importance.
Next get Macs will keep using Intel CPUs. Not enough time to switch.

Intel is the leader of computing at the moment. It has nothing to do with what I think or "assume".
I know you don't like this, but your aversion towards Intel won't change anything either.
Intel is so dominant that people would choose RAMBUS over DDR. And IA-64 over x86-64.
Posted on Reply
#68
notb
qcmadness
With AMD now utilizing chiplet systems, AMD could just re-spin the DDR-5 I/O chip. While for Intel, they either need to build a chip with DDR-4 and DDR-5 MC or re-spin a new chip for DDR-5.
No. AMD has to design their chiplets for RAM. Having an I/O die doesn't change much in this regard.

I don't really understand what you mean by "re-spin". DDR5 won't come as surprise. CPU makers are taking part in memory development. DDR5 and appropriate CPUs have been developed together and can be launched together.
Posted on Reply
#69
R0H1T
notb
Because Intel is the dominant supplier of CPUs. AMD may align with their launch or not. It's of lesser importance.
Next get Macs will keep using Intel CPUs. Not enough time to switch.

Intel is the leader of computing at the moment. It has nothing to do with what I think or "assume".
I know you don't like this, but your aversion towards Intel won't change anything either.
This (year) next gen of Apple Macs will still use 8th gen, pardon me - the vastly improved 9th gen Intel core. The move however totally depends on Apple not Intel, their first ARM Macbook will probably feature Axx SoC on 5nm - https://www.tomshardware.com/news/tsmc-5nm-euv-process-node,38995.html

So when that arrives, Apple will likely use them with LPDDR5. These Axx SoC (less cores?) will also feature in iPhone & iPads - sorry fogot to mention that, but I guess $200~250 billion of sales is peanuts compared to Intel?



That depends on where you look at, for 5 billion people ARM multiverse matters so much more & if/when ARM servers catch on Intel will be in a world of pain that even Zen couldn't inflict upon them.
Posted on Reply
#70
qcmadness
notb
No. AMD has to design their chiplets for RAM. Having an I/O die doesn't change much in this regard.

I don't really understand what you mean by "re-spin". DDR5 won't come as surprise. CPU makers are taking part in memory development. DDR5 and appropriate CPUs have been developed together and can be launched together.
I think you missed the question.

Designing / re-spinning an I/O die is surely easier than designing / re-spinning the whole chip with CPU and memory controller. This is not rocket science, right?
Posted on Reply
#71
notb
R0H1T
This (year) next gen of Apple Macs will still use 8th gen, pardon me - the vastly improved 9th gen Intel core. The move however totally depends on Apple not Intel, their first ARM Macbook will probably feature Axx SoC on 5nm - https://www.tomshardware.com/news/tsmc-5nm-euv-process-node,38995.html

So when that arrives, Apple will likely use them with LPDDR5.
Macs depend on x86 software ported from Windows. This is the reason Apple switched to Intel in the first place. They'll have to make sure every important piece of software is available for ARM. Most isn't.

Apple could offer an ARM powered Macbook next to x86 one (just like Microsoft does with Surface).
ARM exclusive lineup? Think 2025+. But don't bet your house on it.
That depends on where you look at, for 5 billion people ARM multiverse matters so much more & if/when ARM servers catch on Intel will be in a world of pain that even Zen couldn't inflict upon them.
ARM servers are not compatible with x86 and - assuming ARM will become a more attractive option at some point (likely!) - it'll take a decade before a significant part of market migrates. And x86 will stay with us for many years.
Both Intel and AMD are thinking about the ARM threat. They'll join if necessary. Don't worry too much. :)
Posted on Reply
#72
R0H1T
Yes & we have windows on ARM, how cool is that ~ Huawei Readies Windows 10 Laptop Based on Qualcomm Snapdragon 850

Besides the switch from PPC to x86 was much more painful yet they did it in one fell swoop, I also said Macbooks first because I'm not sure how they'll replace high end Xeon or desktop chips in iMacs/Mac Pro but Intel does make so much more from Macbooks.

ARM servers is a long shot atm, for ARM & the likes of QC, Huawei etc. However it's possible that we'll see more custom (in house) ARM cores than the top end x86 chips being replaced by them - Amazon Launches ARM-Based Custom Graviton CPU
Posted on Reply
#73
Lindatje
las
Huh what? It's a fact. Ryzen performance is hit and miss depending on workload and especially in games (high fps gaming that is).

In tons of applications Ryzen sucks too. Handbrake to mention one.
Lol, no.
Sorry but you just talk nonsense now, I can't really take this seriously.
Posted on Reply
#74
notb
R0H1T
Yes & we have windows on ARM, how cool is that ~ Huawei Readies Windows 10 Laptop Based on Qualcomm Snapdragon 850

We have Windows for ARM, we have Linux for ARM and we may have OSX for ARM. That's not a problem.
We don't have software for ARM. Not much anyway.
Besides the switch from PPC to x86 was much more painful yet they did it in one fell swoop, I also said Macbooks first because I'm not sure how they'll replace high end Xeon or desktop chips in iMacs.
No it wasn't. They had a separate platform with incompatible software.
And suddenly it became pretty easy to port all the good stuff from Windows. This is what made MacBooks a sensible alternative for most people.
Earlier OS X had 3-4% market share - mostly in US. Now they're at 13%.
Macs can be used by financial companies, by software developers, by scientists, by artists. Most mainstream professional software is available.
More importantly, Macs and Windows machines can be used in the same organization. It used to be almost impossible.
ARM servers is a long shot atm, for ARM & the likes of QC, Huawei etc. However it's possible that we'll see more custom (in house) ARM cores than the top end x86 chips being replaced by them - Amazon Launches ARM-Based Custom Graviton CPU
2 separate aspects: ARM servers existance and ARM servers becoming a viable alternative to x86.
ARM servers are pretty close to going mainstream (i.e. offered by big OEMs).
They could be used in particular tasks and newly developed systems.

Viable alternative they will be not. For a very long time. :-)
Posted on Reply
#75
mtcn77
Intel shut the EUV door on itself so deliberately that they sort of masterminded their own disadvantage. Lasertec inventing light inspection machines that got booked for years on end had them drop even further in the node race. Now, the rest of the manufacturers have a fast path to both DUV which spare a portion of mask verification runtime and EUV which is now even more readily deployed as 5nm. GJ Intel, you have successfully let yourself become a fast follower.
Posted on Reply
Add your own comment