Zen 2 rumors

AMD never stated that Zen wasn't as good as they hoped. Quite the opposite. Their target was a 40% IPC uplift vs. Bulldozer, and they ended up with well over 50% (52% or 54% IIRC). Zen exceeded expectations. Zen+ was just some process tweaks and latency tweaks, combined with a better boosting algorithm. Zen 2 will almost certainly incorporate design tweaks that will grant a little more IPC, but 15% is very doubtful. I'd be happy with 5%. The difference between Zen+ and Skylake in IPC is roughly 9%, excluding heavy AVX workloads. If AMD gets that difference down to a couple percentage points and makes up a little more of the clockspeed deficit, they are right where they should be, provided they are bringing extra cores to the table (which is likely).

https://www.techspot.com/news/70887-amd-ryzen-worst-case-scenario-threadripper-built-engineers.html

While it turned out with good performance they admitted they left stuff to improve in Zen 2. Intel may be left with only 5% gains, but to apply that same restriction to AMD may be a bit foolish since its a brand new architecture.
 
I want it to be something that can hit 4.8 on a great chip, 4.6 on a good chip, and 4.4 on the average chip.
Oh, and to out perform Intel.

Frankly frequency is not something we can expect from AMD, it's not happening, unless they make some design compromises, aka Bulldozer. I would rather they continue on the path they are on now, chasing frequency when you don't own a fab, and can't dictate process design is a loosing proposition.
 
Frankly frequency is not something we can expect from AMD, it's not happening, unless they make some design compromises, aka Bulldozer. I would rather they continue on the path they are on now, chasing frequency when you don't own a fab, and can't dictate process design is a loosing proposition.

Curious on why you would think that? Zen & Zen+ clock speeds seem to be held back more by the process than the architecture itself. I would be surprised if we don't see 4.5+ ghz on Zen 2.
 
Curious on why you would think that? Zen & Zen+ clock speeds seem to be held back more by the process than the architecture itself. I would be surprised if we don't see 4.5+ ghz on Zen 2.

What makes you think they are held back by the process? I don't think so. What makes you think, that the next available process at TSMC or GF is going to target high clocks? They won't, they never have, at least not when you compare them with Intel fabs across all of history. The only CPU made by AMD that came close to the same clockspeed as Intel was Buldozer, and that because of the long pipeline, which of course came at the cost of reduced IPC, and wasn't worth it at all.

I'm sure it's possible, that next gen Ryzens will be able to achieve higher clockspeeds, but the opposite is also possible. Clockspeed is nice to have, but overall design is far more important.
 
What makes you think they are held back by the process? I don't think so. What makes you think, that the next available process at TSMC or GF is going to target high clocks? They won't, they never have, at least not when you compare them with Intel fabs across all of history. The only CPU made by AMD that came close to the same clockspeed as Intel was Buldozer, and that because of the long pipeline, which of course came at the cost of reduced IPC, and wasn't worth it at all.

I'm sure it's possible, that next gen Ryzens will be able to achieve higher clockspeeds, but the opposite is also possible. Clockspeed is nice to have, but overall design is far more important.

First off, strange answer. You just... don't think so.. because?

Intel's process, and AMD never having 'similar' clock speeds is irrelevant to the conversation of Zen 2 having higher clock speeds than Zen & Zen+.

7nm node, that Zen 2 is going to be built on, has been specifically designed for higher clocks, you can see it in any GF marketing materials on the process. This is reason alone to believe that Zen 2 will have higher clocks than Zen & Zen+. How much higher? Who knows, but I imagine if we see clocks held back due to architecture, we may see it on 7nm.
 
First off, strange answer. You just... don't think so.. because?

Intel's process, and AMD never having 'similar' clock speeds is irrelevant to the conversation of Zen 2 having higher clock speeds than Zen & Zen+.

7nm node, that Zen 2 is going to be built on, has been specifically designed for higher clocks, you can see it in any GF marketing materials on the process. This is reason alone to believe that Zen 2 will have higher clocks than Zen & Zen+. How much higher? Who knows, but I imagine if we see clocks held back due to architecture, we may see it on 7nm.

I clearly explained why I don't think so, but if you want to take your facts from GF marketing material, go ahead.
 
What makes you think they are held back by the process? I don't think so. What makes you think, that the next available process at TSMC or GF is going to target high clocks?

Because Zen uses 14LPP and 12LP (= 14LPP+) and both are nodes optimized for efficiency not performance. The higher performance node at Glofo is 14HP (HP = High Performance) 14HP is the node used by IBM to get 5GHz.

TSMC and Glofo have different flavors for 7nm, the SOC nodes target mobile applications but the HPC nodes target high performance. TSMC already demoed an ARM core clocked at 4.2GHz on 7HPC. And Glofo 7HPC targets 5GHz.
 
Because Zen uses 14LPP and 12LP (= 14LPP+) and both are nodes optimized for efficiency not performance. The higher performance node at Glofo is 14HP (HP = High Performance) 14HP is the node used by IBM to get 5GHz.

TSMC and Glofo have different flavors for 7nm, the SOC nodes target mobile applications but the HPC nodes target high performance. TSMC already demoed an ARM core clocked at 4.2GHz on 7HPC. And Glofo 7HPC targets 5GHz.

Zen uses 14LPP and 12LP over HP process, because it's better for the overall design and characteristics, I'm sure there are tradeoffs to using the HP process, otherwise I'm sure AMD would have been using it.
 
Zen uses 14LPP and 12LP over HP process, because it's better for the overall design and characteristics, I'm sure there are tradeoffs to using the HP process, otherwise I'm sure AMD would have been using it.

Allow me to speculate a moment. AMD designed Zen to be ridiculously easy to scale cores with the CCX and multiple die concepts. They wanted to essentially have two dies (one APU die w/1 CCX, one straight CPU only die, 2 CCXs). This kept costs low. They could then fill their whole product stack, from desktop to HEDT to server, and waste very few of the dies, relatively speaking. So it needed to be power efficient for the server market. But it needed to scale to higher clock speeds with the desktop, but stay within in a reasonable-ish TDP envelope at higher clocks with multiple dies.

So I speculate that AMD compromised a bit and used the 14LPP process to check as many boxes as they could. The tradeoff in clock speed would be made up with more cores in every product segment. Servers still get high efficiency. Then push Zen+ on refined process (whatever you want to call it) to push clocks a little more.

With AMD doing better, due to Zen sales (plus unloading GPUs by the boatload due to crypto), and with the 7nm process available soon... I figure AMD has more room to maneuver with Zen 2. The product stack might not be as monolithic.
 
Allow me to speculate a moment. AMD designed Zen to be ridiculously easy to scale cores with the CCX and multiple die concepts.
With AMD doing better, due to Zen sales (plus unloading GPUs by the boatload due to crypto), and with the 7nm process available soon... I figure AMD has more room to maneuver with Zen 2. The product stack might not be as monolithic.

I follow your speculation and believe very similar. AMD had to make a simple design, repeatable yet flexible with the innovative Infinity Fabric. They just don't have the resources of their competitors - not even close. Interestingly,they appear to be following the "tick - tock" cadence pioneered by we all know who - and so far appears to be working for them. With actual cash flow now that ZEN is selling, hopefully AMD can dump some real R and D into the Radeon Group and come up with something better on the GPU end.
 
I follow your speculation and believe very similar. AMD had to make a simple design, repeatable yet flexible with the innovative Infinity Fabric. They just don't have the resources of their competitors - not even close. Interestingly,they appear to be following the "tick - tock" cadence pioneered by we all know who - and so far appears to be working for them. With actual cash flow now that ZEN is selling, hopefully AMD can dump some real R and D into the Radeon Group and come up with something better on the GPU end.

Crypto sure bailed them out of a lackluster (but not entirely useless) Vega product.
 
But yeah, I want to see a "ZEN" type success for the GPU side, but its probably years out if it ever happens. Nvidia needs to screw up like Intel and I'm not sure that's even possible.
 
But yeah, I want to see a "ZEN" type success for the GPU side, but its probably years out if it ever happens. Nvidia needs to screw up like Intel and I'm not sure that's even possible.
FX5800, 480 and 580 would like to have a word with you...
And yes they all have differing reasons IMO for being bleh products, some worse than others.
 
FX5800, 480 and 580 would like to have a word with you...
And yes they all have differing reasons IMO for being bleh products, some worse than others.

Understand.. AMD is about 2 Gens behind. So Nv would have to mess up several times in a row - comparable to Intel's latest set of Fiascoes. Nv just seems to be executing pretty flawlessly lately and the only good part is it appears to be inflating Jensen's Ego. The more, the better so maybe AMD to catch Nvidia sleeping down the road. Not really sure what a 7nm Vega will give us other than a slightly cooler card and maybe a few MHz. Obviously, not enough.
 
Understand.. AMD is about 2 Gens behind. So Nv would have to mess up several times in a row - comparable to Intel's latest set of Fiascoes. Nv just seems to be executing pretty flawlessly lately and the only good part is it appears to be inflating Jensen's Ego. The more, the better so maybe AMD to catch Nvidia sleeping down the road. Not really sure what a 7nm Vega will give us other than a slightly cooler card and maybe a few MHz. Obviously, not enough.
I don't disagree, just ribbing you a bit lol.
 
  • Like
Reactions: Mega6
like this
You won't be able to touch anything from AMD on high end until January it looks like. Vega 20s are targeted at AI and Datacenter and will require deep pockets. The next wave of Cloud, AI and datacenter is just outside. Catch it!
 
Last edited:
Zen uses 14LPP and 12LP over HP process, because it's better for the overall design and characteristics, I'm sure there are tradeoffs to using the HP process, otherwise I'm sure AMD would have been using it.

AMD uses 14LPP because it is a node optimized for efficiency and favors AMD plans in the server market. EPYC competes in the cheap server market instead in the top heavy market as IBM and ORACLE do.

Top EPYC SKU is 32 core at 2.2GHz. The whole EPYC line span between 2.0GHz and 2.4GHz so a node optimized for 2.5GHz as 14LPP is the natural choice for AMD. In the other hand IBM want the 14HP node (optimized for 4GHz) because IBM is selling 24 cores at 3.8GHz.

Let us wait and see what happens with Zen2 and the 7nm node, with 7SOC (optimized for efficiency) and 7HPC (optimized for performance). There are two hypothesis on the table:

  • Those that claim that AMD will use 7SOC for all Zen2 dies.
  • Those that claim that AMD will use 7SOC for server and 7HPC for desktop dies.
 
Last edited:
Latest rumor is TSMC will make Epyc chips and Ryzen will be GF. Which could mean two different processes may be in use, which will make things interesting.
 
EPYC 2 on TSMC isn't a rumor. Su confirmed it.

AMD server uses a MCM approach just to reuse the same dies on desktop/HEDT. The Zeppelin dies that don't pass server-grade qualification are reused for desktop/HEDT.

If TSMC is only used for EPYC, what happens to all the dies that didn't pass qualification?

Why not CPU dies on TSMC and APU dies on Glofo? Something similar happened in the pass, but with TSMC doing the APUs and Glofo the CPUs.
 
  • Like
Reactions: Nobu
like this
EPYC 2 on TSMC isn't a rumor. Su confirmed it.

AMD server uses a MCM approach just to reuse the same dies on desktop/HEDT. The Zeppelin dies that don't pass server-grade qualification are reused for desktop/HEDT.

If TSMC is only used for EPYC, what happens to all the dies that didn't pass qualification?

Why not CPU dies on TSMC and APU dies on Glofo? Something similar happened in the pass, but with TSMC doing the APUs and Glofo the CPUs.

Ok HEDT get the best Epyc dies for the higher clocks it will allow since servers dont get overclocked and AMD has even said this. Secondly they can always repurpose them for some lower power stuff if they have a large rash of poor yields even tho that issue is not present with them currently. TSMC makes sense for this since they will make far less Epyc chips then they will Ryzen chips and AMD has to buy so many wafers from GF. As for your theory of a APU only at GF makes no sense since they would never sell enough of them to make contract requirements, unless they suddenly make a bigger impact in Notebooks. Lower quantity will always go to TSMC not GF.
 
Ok HEDT get the best Epyc dies for the higher clocks it will allow since servers dont get overclocked and AMD has even said this.

AMD said hat Threadripper has the top binned dies compared to mainstream Ryzen, not that ThreadRipper gets the best dies of the whole silicon production.

Getting the best dies for EPYC, i.e. the ones that can hit the highest frequencies at a given voltage, doesn't imply you have to push the clocks to the limit. Or said in another way the server die has to use less voltage to get 3.0GHz compared to a deskop die. Efficiency is key in servers. Getting the more efficient dies for Theadripepr and the worse for EPYC doesn't make any sense.
 
Allow me to speculate a moment. AMD designed Zen to be ridiculously easy to scale cores with the CCX and multiple die concepts. They wanted to essentially have two dies (one APU die w/1 CCX, one straight CPU only die, 2 CCXs). This kept costs low. They could then fill their whole product stack, from desktop to HEDT to server, and waste very few of the dies, relatively speaking. So it needed to be power efficient for the server market. But it needed to scale to higher clock speeds with the desktop, but stay within in a reasonable-ish TDP envelope at higher clocks with multiple dies.

So I speculate that AMD compromised a bit and used the 14LPP process to check as many boxes as they could. The tradeoff in clock speed would be made up with more cores in every product segment. Servers still get high efficiency. Then push Zen+ on refined process (whatever you want to call it) to push clocks a little more.

With AMD doing better, due to Zen sales (plus unloading GPUs by the boatload due to crypto), and with the 7nm process available soon... I figure AMD has more room to maneuver with Zen 2. The product stack might not be as monolithic.
The reason why there are no high clock speeds is because that was never in the plan. When AMD saw the server market billions were up for grabs the only way for AMD to move forward was to ignore the desktop attributes and get the ball rolling for them to keep making money on a process which allowed to do both and what we saw with Precision Boost and XFR were more or less introduced for the desktop to offset the lack of frequency.
 
These rumors just seem very unlikely. This just seems to be the general over-hyping that happens before major products are released.

They are just rumors, where there is usually some truth to them. Up to you to decide how much truth there is to them. I always find it fun to speculate a bit on what might happen and then compare them to release. I mean after all many thought Ryzen would be less powerful then it ended up actually being. Clock speed was a bit less then hoped for tho, but that is the fun part about speculating.
 
They are just rumors, where there is usually some truth to them. Up to you to decide how much truth there is to them.

Like the older rumors that Zen achieved 5GHz on air, was faster than Kabylake on gaming, surpassed Intel Broadwell on IPC, top SKU was only $300, quad-channel AM4, HBM on the APUs, and rest of hype and fabricated nonsense?
 
Like the older rumors that Zen achieved 5GHz on air, was faster than Kabylake on gaming, surpassed Intel Broadwell on IPC, top SKU was only $300, quad-channel AM4, HBM on the APUs, and rest of hype and fabricated nonsense?

Yeah whatever Juanrga, it's rumors and at least AMD is progressing forward rather then backwards like Intel with their exploits and patches these days. Still waiting on that 5GHz 28 core chip from Intel.....
 
Like the older rumors that Zen achieved 5GHz on air, was faster than Kabylake on gaming, surpassed Intel Broadwell on IPC, top SKU was only $300, quad-channel AM4, HBM on the APUs, and rest of hype and fabricated nonsense?
Rather than rumors, most of that sounds like speculation, or ideals, that people came up with as what they thought would be best case scenario. Except for the IPC/gaming rumors, I would take any of those claims with a huge grain of salt (even not knowing what I do now), and even those two are pretty farfetched.
 
Yeah whatever Juanrga, it's rumors and at least AMD is progressing forward rather then backwards like Intel with their exploits and patches these days. Still waiting on that 5GHz 28 core chip from Intel.....

No one negated that AMD is progressing. Just mentioning that those leaks about Zen2 are so suspicious as former leaks about Zen.

Rather than rumors, most of that sounds like speculation, or ideals, that people came up with as what they thought would be best case scenario. Except for the IPC/gaming rumors, I would take any of those claims with a huge grain of salt (even not knowing what I do now), and even those two are pretty farfetched.

They were leaks. The 5GHz on air come from an easter egg publication. The faster than Kabylake on gaming from AMD slides. The IPC from leaks about CPU-Z. the HBM on Zen APUs from bitsandchips,...

All fake leaks.
 
Like I said, put me down for a ~5% IPC increase, a 6 core CCX (or, one more 4 core CCX), and ~4.7GHz boost clocks for Zen 2. So a 12 core with almost-Skylake IPC on AM4, and only a few hundred MHz behind Intel's lineup.

And I'd buy the hell out of that if true, too.
 
They were leaks. The 5GHz on air come from an easter egg publication. The faster than Kabylake on gaming from AMD slides. The IPC from leaks about CPU-Z. the HBM on Zen APUs from bitsandchips,...

All fake leaks.
Right, and the ones with half a brain (who weren't also trolling) might have given one or two of them some credence, given their source, with a good bit of skepticism to keep things in perspective. Aside from the reasonable speculation of ~4GHz boost, I haven't seen a good source of any other rumors going on about TR2 here yet. IPC comparisons are dubious at best for now (except those using zen2 Epyc or current TR as a base, though that'd still be only speculation), because we know nothing about the test systems, and both the TR2 and (supposed) new intel part are both very suspect, imho, with no other leaks to corroborate or reinforce the information we have.
 
Right, and the ones with half a brain (who weren't also trolling) might have given one or two of them some credence, given their source, with a good bit of skepticism to keep things in perspective. Aside from the reasonable speculation of ~4GHz boost, I haven't seen a good source of any other rumors going on about TR2 here yet. IPC comparisons are dubious at best for now (except those using zen2 Epyc or current TR as a base, though that'd still be only speculation), because we know nothing about the test systems, and both the TR2 and (supposed) new intel part are both very suspect, imho, with no other leaks to corroborate or reinforce the information we have.

I think most people are skeptical until we see more of the same thing from other sources, then you have to figure there is likely some truth to it then. This is also why I called the thread Zen 2 rumors and not Zen 2 facts or OMG Zen 2 is Awesome. I think Juanrga spends to much time on Reddit and applies that logic here on this forum when most of us are more skeptical. I just noticed some news started to trickle out about Zen 2 and figured to give it a thread where others can bring news about it and discuss it. Also that 1 sample leak on TR2 makes it impossible to really know for sure but it's not outside the realm of possibility as well. Thus just a item to discuss. This whole Hype thing gets way overblown as most any enthusiast gets excited about new tech about to release otherwise they wouldn't be on these forums. Sometimes it leads to wild speculation but I think very few actually believe them and of course sometimes things end up as a dud and deserve the bashing they get.
 
I think most people are skeptical until we see more of the same thing from other sources, then you have to figure there is likely some truth to it then. This is also why I called the thread Zen 2 rumors and not Zen 2 facts or OMG Zen 2 is Awesome. I think Juanrga spends to much time on Reddit and applies that logic here on this forum when most of us are more skeptical. I just noticed some news started to trickle out about Zen 2 and figured to give it a thread where others can bring news about it and discuss it. Also that 1 sample leak on TR2 makes it impossible to really know for sure but it's not outside the realm of possibility as well. Thus just a item to discuss. This whole Hype thing gets way overblown as most any enthusiast gets excited about new tech about to release otherwise they wouldn't be on these forums. Sometimes it leads to wild speculation but I think very few actually believe them and of course sometimes things end up as a dud and deserve the bashing they get.

You are wrong. I barely visit Reddit and when I visit I do as external guy without account.

FYI people here claimed that the 5GHz on air 'leak' for Zen was real and that IPC above Kabylake was granted. Just as some people here is now taking that 15% IPC increase 'leak' as granted telling me that AMD will surpass Intel on single thread.

You seem to not read your own threads, or other threads, or even your own posts: "So I think a 15% IPC increase may be possible"
 
You are wrong. I barely visit Reddit and when I visit I do as external guy without account.

FYI people here claimed that the 5GHz on air 'leak' for Zen was real and that IPC above Kabylake was granted. Just as some people here is now taking that 15% IPC increase 'leak' as granted telling me that AMD will surpass Intel on single thread.

You seem to not read your own threads, or other threads, or even your own posts: "So I think a 15% IPC increase may be possible"

Look get out of my thread if you want to talk about Zen, this is Zen 2 cause to be honest not one soul in the AMD sub forum cares what you think and yes I think a 15% gain in IPC is possible, will get a better idea when benchmarks start to leak. You are not a engineer, you dont work in the industry so you dont know anything better then the rest of us so stop asserting your correct and no one else can be. Id read your posts but they are always edited.. wonder why. So do us a favor and stay on topic if your going to post on here, I know Intel is boring lately as they are stuck on what 14++++ now but this thread isnt about that either.
 
Look get out of my thread if you want to talk about Zen, this is Zen 2 cause to be honest not one soul in the AMD sub forum cares what you think and yes I think a 15% gain in IPC is possible, will get a better idea when benchmarks start to leak. You are not a engineer, you dont work in the industry so you dont know anything better then the rest of us so stop asserting your correct and no one else can be. Id read your posts but they are always edited.. wonder why. So do us a favor and stay on topic if your going to post on here, I know Intel is boring lately as they are stuck on what 14++++ now but this thread isnt about that either.

I am talking about Zen2, in case you didn't notice. I am mentioning why the Zen2 rumors that you bring here to us (8-core CCX, 16 core 32 threads for AM4, 15% IPC gains,...) seem irrational hype rather than real leaks. You may believe that all that is "possible" for Zen2, but many here are skeptics. I have mentioned Zen leaks, because, as I said to you, "those leaks about Zen2 are so suspicious as former leaks about Zen."

P.S.: Amazing that you talk about staying on-topic, when you are repeatedly trying to move the discussion towards Intel chips as you did in #72.
 
Back
Top