AMD Zen 4 to have 5nm by 2021

[thedangerine]

AMD looks to be keeping the pressure on Intel with many reports saying we could see 5nm chips based on their Zen 4 Architecture by 2021. This going in hand with their constant increase in core count will likely continue their current domination in the workstation and server space. All of this is just more bad news for Intel who is expected to have 7nm by 2022 maybe? Regardless of which side you are rooting for this is good news as a consumer as Intel still has a lot of cards they can play to still compete, the future should be very interesting.

https://www.techradar.com/news/amds-5nm-zen-4-cpus-could-arrive-in-2021-really-turning-the-heat-up-on-intel 

[mr moose]

Quote

 

according to a new rumor.

...

This is the case according to a report from China Times (spotted by Wccftech),

 

 

In other news, extinct dodo bird could be resurrected as early as 2024 according to....

[yian88]

7 minutes ago, AdvancedMicroDisapointment said:

.... Regardless of which side you are rooting for this is good news as a consumer as Intel still has a lot of cards they can play to still compete, the future should be very interesting....

hold up, how many more ++++ do you think intel has? 14nm from intel is done they have no more tweaks to it, maybe they can slightly improve yields but thats about it.

 

And this is not NEWS, the article says its a rumour. I think we need a different website section for rumours and this one for real news.

[thedangerine]

5 minutes ago, yian88 said:

hold up, how many more ++++ do you think intel has? 14nm from intel is done they have no more tweaks to it, maybe they can slightly improve yields but thats about it.

 

And this is not NEWS, the article says its a rumour. I think we need a different website section for rumours and this one for real news.

I don't think Intel is going to do any more +++ but they can adjust things like pricing, core count, cache, PCIe lanes, etc. plus it would not surprise me if they end up releasing some desktop 10nm despite talk of them skipping straight to 7nm. I'm hoping we see some interesting moves by each company. The server space has me the most intrigued, especially with the chipset getting moved to the cpu on Epyc chips, that makes a lot of sense considering the north bridge move all those years ago

[Unixsystem]

And Intel 10nm will definitely be out by 2016.

 

Nodes are getting harder and harder to achieve and delays are getting more and more common. Regardless of what AMD and TSMC are planning, I'll wait and see. 

[thechinchinsong]

Its interesting cause if you take this rumor along with the interview by Mark Papermaster about how they don't see a saturation point for cores (with the caveat that one needs to be careful to add cores before software can catch up in a meaningful way), it builds a case that there is possibility for even more cores than we current have in Zen 2. 5nm will certainly allow AMD to physically put more cores in the same space, but it'll be up to the software ecosystem and other architecture/market factors to determine if AMD will actually add them in the near future.

[amdorintel]

1 hour ago, thechinchinsong said:

allow AMD to physically put more cores in the same space

how many cores does one need for triple a games anyway

used to be 4 cores 8 threads was plenty and good enough, now it seems 6 cores 12 threads is plenty to have for gaming atleast for the triple a games that are programmed 6/12

[Sauron]

2 hours ago, yian88 said:

hold up, how many more ++++ do you think intel has? 14nm from intel is done they have no more tweaks to it, maybe they can slightly improve yields but thats about it.

 

And this is not NEWS, the article says its a rumour. I think we need a different website section for rumours and this one for real news.

Intel supposedly has a 7nm node coming at some point, the failure of their 10nm node shouldn't have impacted the development of their next one - these things are drafted years in advance and both have been in development at the same time.

[amdorintel]

5 minutes ago, Sauron said:

these things are drafted years in advance and both have been in development at the same time.

yea i was guessing intel would have something up their sleeves or have something to fast track to get to the market

 

the next couple years will be interesting

[thechinchinsong]

27 minutes ago, amdorintel said:

how many cores does one need for triple a games anyway

used to be 4 cores 8 threads was plenty and good enough, now it seems 6 cores 12 threads is plenty to have for gaming atleast for the triple a games that are programmed 6/12

That's a good question and I'm not a game dev so I honestly have no clue.

[Unixsystem]

1 hour ago, thechinchinsong said:

That's a good question and I'm not a game dev so I honestly have no clue.

Today I'd say 6/12 is enough for pretty much any current AAA game, assuming you aren't doing anything in the background. 

 

That said, I personally wouldn't recommend anything below an 8 core going forward unless you are on a very tight budget. The cost bump at this point is relatively low and it gives you a whole lot more flexibility in terms of multitasking while also potentially giving a bit more longevity as devs slowly start to acclimate to the new market of affordable high-core-count CPUs. 

 

Spoiler

And no, I'm not saying an 8 core will "future proof" you, just that it will give you more room to grow, especially once new consoles come out with much more modern hardware. It might add an extra year or two before your system goes from "outdated but fully functional" to "struggling to keep up". 

 

[thorhammerz]

6 hours ago, yian88 said:

hold up, how many more ++++ do you think intel has? 14nm from intel is done they have no more tweaks to it, maybe they can slightly improve yields but thats about it.

Rocket-Lake S (11th Gen) is also supposedly coming on 14nm as well. Although it is also supposedly going to be a 14nm backport of Willow Cove, so IPC should be at least in the ballpark of Zen 2.

[Taf the Ghost]

43 minutes ago, thorhammerz said:

Rocket-Lake S (11th Gen) is also supposedly coming on 14nm as well. Although it is also supposedly going to be a 14nm backport of Willow Cove, so IPC should be at least in the ballpark of Zen 2.

Rocketlake might be getting eDRAM and/or some sort of EMIB-attached iGPU. There's some funky stuff going on with that architecture that we won't be seeing until 2021.

 

In theory, desktop will go to 10nm on Alder Lake in 2022 with 7nm (Intel) on Meteor Lake in 2023. As a note. AMD will be the similar node to Intel's 7nm (TSMC 5nm) in 2021. Just to clarify where everything is right now.

[Hugh54321]

2 hours ago, Waffles13 said:

as devs slowly start to acclimate to the new market of affordable high-core-count CPUs

You might be interested in this from 2013. 2 minutes in John talks about functional programming.

 

 

[redbread123]

I'd assume AMD's TSMC 5nm is equivalent to intel's 7 nm? anyways you guys fall right into marketing crap, the nm is just a marketing term, if this amd gen 7 nm is so superior, than why does it fall behind intel (slightly, don't wanna trigger amd kids) in gaming. anyways EUV has a lot of potential and reduces cost due to less layers needed for the smaller size.

[leadeater]

2 hours ago, redbread123 said:

if this amd gen 7 nm is so superior, than why does it fall behind intel (slightly, don't wanna trigger amd kids) in gaming. anyways EUV has a lot of potential and reduces cost due to less layers needed for the smaller size.

Intel has the luxury of actually being a fab company and can design things exactly as they want for the product they intend to build, something AMD had which was why it was so shocking they sold that business and gave that ability up. TSMC 7nm is still at it's heart a low power node because that's really what most customers want, Intel 14nm is high power.

 

EUV is still really slow so will only be used on layers where triple/quad patterning would have been used is it's wafer output ~equivalent. Immersion and UEV will be used together, EUV where it actually matters.

 

I wonder how fast EUV technology will advance, with fab production already being so constrained by huge demand we really need to be able to make more wafers per day not just more chips per wafer.

[yian88]

4 hours ago, redbread123 said:

I'd assume AMD's TSMC 5nm is equivalent to intel's 7 nm? anyways you guys fall right into marketing crap, the nm is just a marketing term, if this amd gen 7 nm is so superior, than why does it fall behind intel (slightly, don't wanna trigger amd kids) in gaming. anyways EUV has a lot of potential and reduces cost due to less layers needed for the smaller size.

And you are falling for the same old subject that is wrong, it doesnt matter tsmc 7nm vs intel 7nm, maybe for you personally, but for the rest is doesnt no one buys chips based on 7nm process alone, always look at performance. What it really matter is how much better is 7nm tsmc vs 14nm tsmc? how much better is 7nm+ vs 7nm from tsmc? or 5nm vs 7nm+? thats all that really matters because we need to compare AMD chips from previous generation to next not with intel.

Do you really think it matters who has less "real nm" distance between transistors? there are so many variables in making transistor that even if intel had 3nm real right now ready for production if they had 20% yields and frequency max 3ghz it would be crappy.

[Valentyn]

I'm still kicking on Haswell-E here. 

If Zen4 ends up on the same socket as Zen3, I'll upgrade next year. Otherwise I'll try holding off that smidge more to see what happens.

[CephDigital]

20 minutes ago, Valentyn said:

I'm still kicking on Haswell-E here. 

If Zen4 ends up on the same socket as Zen3, I'll upgrade next year. Otherwise I'll try holding off that smidge more to see what happens.

My bet is that they'll change sockets once DDR5 becomes mainstream. Doesn't make sense to lock an entire generation of CPUs to a new socket and then later release an entirely new socket for DDR5 *cough* Intel *cough*

[RejZoR]

26 minutes ago, TheUberMedic said:

My bet is that they'll change sockets once DDR5 becomes mainstream. Doesn't make sense to lock an entire generation of CPUs to a new socket and then later release an entirely new socket for DDR5 *cough* Intel *cough*

Which is what I'm waiting for. My 5820K is good enough for now and I see no point in upgrading. Upgrading to AM4 at its end of life is also rather pointless. So, ideal will be AM5 and DDR5 release to upgrade. Partially early adopter of new tech but AMD has enough experience now and support from vendors it'll go without much issues imo.

[CarlBar]

54 minutes ago, Valentyn said:

I'm still kicking on Haswell-E here. 

If Zen4 ends up on the same socket as Zen3, I'll upgrade next year. Otherwise I'll try holding off that smidge more to see what happens.

 

 

AMD only said they'd stay on AM4 until 2020. So i expect Zen 4 to be a new socket.

[Doobeedoo]

It's looking very good if yields are as good so early on though. But yeah, big node jump with probably new socket, memory, PCIe, USB4 well it's looking very exciting. 

[Donut417]

Very exciting time to be alive. Finally we got competition back in the CPU market. Hopefully AMD can keep the pressure up. Get Intel to get off its ass and see both of these companines inovate. 

[thorhammerz]

14 hours ago, leadeater said:

something AMD had which was why it was so shocking they sold that business and gave that ability up

Well, there's the financial quagmire Global Foundries found itself in trying to develop new nodes (they even ended up licensing 14nm from Samsung at one point....

 

It works out OK for Intel in large part because of their scale (most of the time.... 10nm being the most recent road bump), a luxury AMD never had.

[descendency]

13 hours ago, redbread123 said:

I'd assume AMD's TSMC 5nm is equivalent to intel's 7 nm? anyways you guys fall right into marketing crap, the nm is just a marketing term, if this amd gen 7 nm is so superior, than why does it fall behind intel (slightly, don't wanna trigger amd kids) in gaming. anyways EUV has a lot of potential and reduces cost due to less layers needed for the smaller size.

There seems to be pros and cons to rapidly shrinking the process node that is used to make the die. Shrinking allows you to jam more transistors into a smaller space without having electrical interference from nearby ones. The end result is either more logic per core or more cores. 

 

Pushing everything closer together, though, has the negative of shrinking the surface area to cool these parts, in spite of the fact that it should maintain roughly the same TDP. This means getting the clock speed as high as previous nodes requires more work to make it happen. This is why process refinements occur (14nm+, 14nm++). Each process refinement gets harder and harder to achieve, ultimately making shrinking the node better long term. 

 

Until recently, no one would have argued that a process refinement could be better than shrinking the process node, but Intel's 10nm mobile CPUs have regressed in some metrics.