AMD's CPU roadmap reveals Zen 3, Zen 4, and 5nm

[midian182]

Forward-looking: After years of living in its rival’s shadow, AMD has been challenging Intel in the CPU space since the arrival of its Zen architecture, and it looks like that trend is going to continue. Team Red has just revealed its latest processor roadmaps, showing the release of Zen 3 and the 5nm Zen 4 EPYC Genoa chips.

The consumer and enterprise CPU roadmaps were unveiled at AMD’s Financial Analyst Day. In the former category, the graph only goes as far as Zen 3. The fourth-generation Ryzen chips will arrive at the end of this year, but AMD says that the full range of consumer products, which includes desktop CPUs, should all be available by the end of 2021.

The enterprise product roadmap extends further into the future than the consumer version, a result of a longer product cycle and the financial investment long-term maps bring from customers. AMD also hinted that it didn’t want to reveal too much of its future plans for the competitive consumer CPU market.

AMD confirmed that EPYC Rome’s successor, EPYC Milan, will arrive late in 2020. The Zen 3-based CPUs will be built on the 7nm process, but it will be a more advanced version of the 7nm process AMD uses in its current products.

Looking further ahead, AMD revealed it would have its Zen 4 EPYC Genoa processors available by the end of 2022. These use the 5nm process node, which will likely also be used for the Zen 4-based consumer CPUs when they arrive.

With Intel’s upcoming Comet Lake desktop CPUs based on a refined version of the 14-nanometer process it’s been using since Skylake, Chipzilla is struggling to catch up with its rival, which is eroding Team Blue’s market share all the time.

Permalink to story.

https://www.techspot.com/news/84295-amd-cpu-roadmap-reveals-zen-3-zen-4.html

 

[rrwards]

Do we know if the current Zen 2 is the end of the AM4 socket?

 

[Impudicus]

Do we know if the current Zen 2 is the end of the AM4 socket?

Last I've heard is that ZEN 3 will likely be the last of the AM4's.

 

[rrwards]

Last I've heard is that ZEN 3 will likely be the last of the AM4's.

Well that's good to know. Heck, I'm already happy with Zen2 still being supported on AM4. Longer than any intel socket I've used.

 

[StupidPeopleSuk]

So why don't they just get on with it, and move to 0.0000001nm zen 20 or whatever.....

oh, I forgot.....can somebody say "milkin it", which is the exact same thing intel has been doing with Skylake 14nm+++++++ for like 5 years now...... go figure

 

[Ravalo]

Well that's good to know. Heck, I'm already happy with Zen2 still being supported on AM4. Longer than any intel socket I've used.

The real question is.... will zen 3 support b450 chipsets?

 

[wiyosaya]

So why don't they just get on with it, and move to 0.0000001nm zen 20 or whatever.....

oh, I forgot.....can somebody say "milkin it", which is the exact same thing intel has been doing with Skylake 14nm+++++++ for like 5 years now...... go figure

If it were possible to break the laws of physics, maybe they would. Unfortunately, it is not possible to break those laws, and even at the dimensions that they are at now, it is not an easy problem to solve. If it were easily solvable, it would have been solved by now.

IMO, AMD is far from milking it. Intel is the expert at milking.

 

[eafshar]

So why don't they just get on with it, and move to 0.0000001nm zen 20 or whatever.....

oh, I forgot.....can somebody say "milkin it", which is the exact same thing intel has been doing with Skylake 14nm+++++++ for like 5 years now...... go figure

there is a limit to Moore's law. its expected to end by 2025.

 

[Cod3nflame]

there is a limit to Moore's law. its expected to end by 2025.

Just so you know.. TSMC doesn't base their node progress off Moore's Law. They work in smaller, more reliable increments.

 

[Cod3nflame]

So why don't they just get on with it, and move to 0.0000001nm zen 20 or whatever.....

oh, I forgot.....can somebody say "milkin it", which is the exact same thing intel has been doing with Skylake 14nm+++++++ for like 5 years now...... go figure

AMD can only work with whatever TSMC has at the moment. They don't make their own chips TSMC does. AMD sold their fabs off long ago.

 

[Cod3nflame]

If it were possible to break the laws of physics, maybe they would. Unfortunately, it is not possible to break those laws, and even at the dimensions that they are at now, it is not an easy problem to solve. If it were easily solvable, it would have been solved by now.

IMO, AMD is far from milking it. Intel is the expert at milking.

If it were possible to break the laws of physics, maybe they would. Unfortunately, it is not possible to break those laws, and even at the dimensions that they are at now, it is not an easy problem to solve. If it were easily solvable, it would have been solved by now.

IMO, AMD is far from milking it. Intel is the expert at milking.

If it were possible to break the laws of physics, maybe they would. Unfortunately, it is not possible to break those laws, and even at the dimensions that they are at now, it is not an easy problem to solve. If it were easily solvable, it would have been solved by now.

IMO, AMD is far from milking it. Intel is the expert at milking.

You're right about the first part, very mistaken about the second.. Intel tried a 2.25x scale with 10nm. Much higher than what TSMC targets with their node progressions. Had Intel accomplished what they set out to do with 10nm, Ryzen would literally be Bulldozer 2.0 right now. Intel's performance targets were way too aggressive. Intel is far from lazy. They tried to give us the performance jump we all wanted and failed. They aren't milking anything. They took a massive gamble and it didn't work. They aren't milking 14nm, they just literally have no other alternative until they figure how where to go next. That's why they're going to start outsourcing some stuff to Global Foundry and Samsung. At this point 10nm is a lost cause and they've already begun working on 7nm so that's what we'll probably be seeing from Intel next.

 

[Evernessince]

The real question is.... will zen 3 support b450 chipsets?

That's going to depend mostly on the manufacturer.

 

[AS692313]

So why don't they just get on with it, and move to 0.0000001nm zen 20 or whatever.....

oh, I forgot.....can somebody say "milkin it", which is the exact same thing intel has been doing with Skylake 14nm+++++++ for like 5 years now...... go figure

Someone clearly doesn't understand that atoms are a half a nanometer thick, pretty hard to get down as small as we already have, I'd like to see you fabricate your subatomic particle utilizing CPU

 

[AS692313]

You're right about the first part, very mistaken about the second.. Intel tried a 2.25x scale with 10nm. Much higher than what TSMC targets with their node progressions. Had Intel accomplished what they set out to do with 10nm, Ryzen would literally be Bulldozer 2.0 right now. Intel's performance targets were way too aggressive. Intel is far from lazy. They tried to give us the performance jump we all wanted and failed. They aren't milking anything. They took a massive gamble and it didn't work. They aren't milking 14nm, they just literally have no other alternative until they figure how where to go next. That's why they're going to start outsourcing some stuff to Global Foundry and Samsung. At this point 10nm is a lost cause and they've already begun working on 7nm so that's what we'll probably be seeing from Intel next.

but they really had to spend that long figuring out that 10nm sucked?

 

[JimboJoneson]

but they really had to spend that long figuring out that 10nm sucked?

I'm sure they figured that out a long time ago. But with Intel, their shareholders are the "customers" that they cater to, people who buy their stuff are just the milk cow.

So all communication from Intel lately has been for the shareholder benefits only - even if that means selecting and bending truths carefully to keep them happy at the chance of getting outed.

In order to keep your shareholders happy (and if that's goal #1), you have to be extremely careful with what you say publicly - even if you have to tell them about an upcoming 5.0ghz 28 core processor in a somewhat less than honest way, or craft your marketing slide data to show laptop and tablet data and pretend to the shareholders that its desktop and HEDT data, or tell them 10nm is "on track!", setback after setback, when its clearly not on track, etc.

When the guy running the company's expertise is accounting and finance, this is what happens.

 

[amstech]

First of all, "AMD 7nm" isnt exactly the correct term because AMD does not manufacture their own 7nm dies. In fact, the company that was spun-off from AMD due to mismanagement issues, now called GlobalFoundries, have halted all the R&D and plans for 7nm, and stick with the older 14nm tech instead for god knows how long.

"AMD 7nm" is akin to saying "Samsung Android". Samsung just produces hardware that happens to run Android, while Android itself is made by Google. 7nm AMD chips will be manufactured by TSMC, the same one responsible for producing Apple, Qualcomm, Nvidia, and various other chips.

Now that we got the basics out of the way, when it comes to Intel 10nm vs. TSMC 7nm, the TSMC has slightly denser die. However, the difference is so narrow, that in the end its a matter of which architecture (IPC) is better, and which can clock higher. Those two would have a much larger impact than the density itself.

As others have said, the 10nm or 7nm name are pretty much marketing numbers these days.

Don't get too caught up with the 7NM and 10NM terms, as the size of the distance between each transistor is going down (from about 75 to 54 or so) is far more than the transistor and far more important to the density calculation.

 

[wiyosaya]

You're right about the first part, very mistaken about the second..

I was specifically referring to the tiny increases in performance starting at maybe Sandy Bridge and on and then trying to extract many times that in hard earned cash from their customer's wallets. My Ivy Bridge proc was competitive with a few of the subsequent generations. IMO, it is hard to not call that kind of behavior milking.

Perhaps it took AMD to light a fire under that pile of trash; it is great they tried to really push the limits; but, that they did and failed spectacularly may just be another sign of the same kind of complacency / corporate hubris that kept them charging substantially higher prices for mediocre performance gains through those years specifically.

 

[wiyosaya]

I'm sure they figured that out a long time ago. But with Intel, their shareholders are the "customers" that they cater to, people who buy their stuff are just the milk cow.

So all communication from Intel lately has been for the shareholder benefits only - even if that means selecting and bending truths carefully to keep them happy at the chance of getting outed.

In order to keep your shareholders happy (and if that's goal #1), you have to be extremely careful with what you say publicly - even if you have to tell them about an upcoming 5.0ghz 28 core processor in a somewhat less than honest way, or craft your marketing slide data to show laptop and tablet data and pretend to the shareholders that its desktop and HEDT data, or tell them 10nm is "on track!", setback after setback, when its clearly not on track, etc.

When the guy running the company's expertise is accounting and finance, this is what happens.

Yup. Just like Rory Reed and AMD and Bulldozer - then that ridiculous announcement that AMD was going to shift their product line to game consoles.

I see that as very similar to Intel's new accounting and finance oriented CEO saying "we make chips for other, non-cpu markets, too" since AMD is clearly highly competitive with them now. IMO, those kind of statements are like saying something like, "Oh, we fell down once, so that must mean that we can never get up and run again."

IMO, Intel's current spin-doctoring is just lame and unacceptable.

 

[Randall Markus]

Anyone else see a marketing opportunity? Zen 4 will be 5000 series chips. On 5nm. Almost definitely coming with ddr5 support. Plausibly coming with pcie 5.0 (especially if they give themselves some time).

Rename Zen 4 to Zen 5, and everything will have a 5 in it. Who cares there will never be a zen4. Zen 3 will make Intel's 14++++++ CPUs look silly anyways, likely 10nm desktop as well, maybe push "zen5" to 2022, may 5th (which would give 7nm+ Zen3 chips a year and a half of likely domination, and give AMD the time to actually test their hardware so it doesn't need 4 months of weekly bios updates). Would also give them some time to get the cpu/gpu infinity fabric buttoned up too.

 

[JimboJoneson]

Yup. Just like Rory Reed and AMD and Bulldozer - then that ridiculous announcement that AMD was going to shift their product line to game consoles.

I see that as very similar to Intel's new accounting and finance oriented CEO saying "we make chips for other, non-cpu markets, too" since AMD is clearly highly competitive with them now. IMO, those kind of statements are like saying something like, "Oh, we fell down once, so that must mean that we can never get up and run again."

IMO, Intel's current spin-doctoring is just lame and unacceptable.

Indeed. Well Bob did also make the statement indicating that "we not interested in chasing AMD ... look at all the money over there! That's were we are going .. buy stocks!"

I really appreciate Lisa Su's much more blunt and honest approach, even if she does sound like a roboscript at times (she's also very careful about what she says, but just keeps a distinction in her words between fluff and tech (and knows how to speak tech pretty well) - so it seems less like she is hiding something when she makes statements). I was going to mention the comparison of Bob to Rory, in that post.

Intel is carving a weird place for itself while really trying to keep investors lined up; I fully think they do plan on coming back and trying o deliver another whoopin' on AMD like they did with Core, but since it will take a few years probably, they need to distract investors with what the left hand is pointing at, while the right hand feverishly tries to work on stuff.

There's quite a bit of psychology involved actually. (because it usually works)

 

[wiyosaya]

Anyone else see a marketing opportunity? Zen 4 will be 5000 series chips. On 5nm. Almost definitely coming with ddr5 support. Plausibly coming with pcie 5.0 (especially if they give themselves some time).

Rename Zen 4 to Zen 5, and everything will have a 5 in it. Who cares there will never be a zen4. Zen 3 will make Intel's 14++++++ CPUs look silly anyways, likely 10nm desktop as well, maybe push "zen5" to 2022, may 5th (which would give 7nm+ Zen3 chips a year and a half of likely domination, and give AMD the time to actually test their hardware so it doesn't need 4 months of weekly bios updates). Would also give them some time to get the cpu/gpu infinity fabric buttoned up too.

If AMD does go down such a path, I hope it is because they have a truly great product and and not to coverup a rebrand of a previous generation's part.

 

[wiyosaya]

Indeed. Well Bob did also make the statement indicating that "we not interested in chasing AMD ... look at all the money over there! That's were we are going .. buy stocks!"

Yes, that is the very definition of lame, IMO!

I really appreciate Lisa Su's much more blunt and honest approach, even if she does sound like a roboscript at times (she's also very careful about what she says, but just keeps a distinction in her words between fluff and tech (and knows how to speak tech pretty well) - so it seems less like she is hiding something when she makes statements). I was going to mention the comparison of Bob to Rory, in that post.

IMO, Su is a genius.

Sticking to an honest tack serves more than one purpose - protecting them from the ever present threat of investor lawsuits, and not making promises about how awesome the next gen of products will be and then suffering an unexpected extreme hiccup like they did with Bulldozer. The fact that she does not reveal everything is not necessarily nefarious in nature. Hopefully, as a corporate culture, AMD has learned its lesson WRT over-promising something that is, essentially, an unknown. She will have far fewer people to answer to if "secret project X" turns out to be a disappointment.

Intel is carving a weird place for itself while really trying to keep investors lined up; I fully think they do plan on coming back and trying o deliver another whoopin' on AMD like they did with Core, but since it will take a few years probably, they need to distract investors with what the left hand is pointing at, while the right hand feverishly tries to work on stuff.

There's quite a bit of psychology involved actually. (because it usually works)

That old adage comes to mind - The proof is in the pudding.
Not to mention -

You can fool some of the people all of the time, and all of the people some of the time, but you can not fool all of the people all of the time.

- Abraham Lincoln

 

[AS692313]

I'm sure they figured that out a long time ago. But with Intel, their shareholders are the "customers" that they cater to, people who buy their stuff are just the milk cow.

So all communication from Intel lately has been for the shareholder benefits only - even if that means selecting and bending truths carefully to keep them happy at the chance of getting outed.

In order to keep your shareholders happy (and if that's goal #1), you have to be extremely careful with what you say publicly - even if you have to tell them about an upcoming 5.0ghz 28 core processor in a somewhat less than honest way, or craft your marketing slide data to show laptop and tablet data and pretend to the shareholders that its desktop and HEDT data, or tell them 10nm is "on track!", setback after setback, when its clearly not on track, etc.

When the guy running the company's expertise is accounting and finance, this is what happens.

I'm not saying that they have to appeal to the shareholders, I'm just saying that upon realizing that 10nm wasn't going to work, you think they'd have gone a different path much faster, rather than simply delaying a disappointing release. I think the shareholders would've been happy to hear "Hey, guess what, we're ditching 10nm because we've got an even BETTER idea, why not shrink even further and get far more performance and efficiency?" Then, they could just delay that release, not have to deal with lying about 10nm anymore and put themselves in a position to openly develop 7nm..

 

[JimboJoneson]

I'm not saying that they have to appeal to the shareholders, I'm just saying that upon realizing that 10nm wasn't going to work, you think they'd have gone a different path much faster, rather than simply delaying a disappointing release. I think the shareholders would've been happy to hear "Hey, guess what, we're ditching 10nm because we've got an even BETTER idea, why not shrink even further and get far more performance and efficiency?" Then, they could just delay that release, not have to deal with lying about 10nm anymore and put themselves in a position to openly develop 7nm..

I think AMD was the issue here ... I imagine initially they didn't take the Ryzen launch seriously at all, in fact I think I read a quote somewhere of someone inside saying they expected Ryzen to be another "bulldozer" - so when it launched perfroming where it did, then zen+ came with 10% improvement and then zen2 with an additional 15% ... it probably very quickly absorbed all the time they thought they had to keep using 14nm while working on 10nm. They got lazy and complacent. Had Ryzen actually been another bulldozer, no one would care about 10nm being late, because people would still be buying Intel whether 14nm or 10nm - Intyel had no reason to care (from its erroneous perspective) ... that complacency just carried over from during the bulldozer days.

By the time they actually admitted to themselves that AMD was going to over take them in CPU tech, well, it wasn't that long ago, they were in plain denial before that. Had they seen Ryzen for the threat it was, I am sure their roadmap would have changed a lot faster than it eventually did.

 

[JimboJoneson]

...
IMO, Su is a genius.
Sticking to an honest tack serves more than one purpose ...

Honesty is always the best policy in the long run ... unfortunately, after thousands of years, humans are still too stupid to have learned this for some bizarre reason ... its amusing.