AMD to Polevault Zen+, Head Straight to 7nm Zen2 for EPYC

[btarunr]

AMD in its Computex 2018 address earlier today, mention that its second-generation EPYC enterprise processors will be based on its 7 nanometer "Zen 2" architecture, and not 12 nm "Zen+." The company has the 7 nm silicon ready in its labs, and will begin sampling within the second half of 2018. The first products could launch in 2019, after validations. Besides improved energy-efficiency, the 12 nm "Zen+" architecture features a minor 3-5 percent IPC uplift thanks to improved multi-core clock-speed boosting, and faster caches. "Zen 2," on the other hand, presents AMD with the opportunity to make major design changes to its silicon to achieve higher IPC uplifts. The 7 nm process introduces significant transistor density uplifts over the current process. AMD is in the process of building 4-die multi-chip modules using the 12 nm "Pinnacle Ridge" silicon for its 2nd generation Ryzen Threadripper HEDT client processor family.



View at TechPowerUp Main Site

 

[voltage]

The War begins. I don't think I need to explain...

 

[Metroid]

Good thing Lisa is not playing the Intel game, she is just making sure amd is very competitive and that is the way to think.

 

[birdie]

Intel must have shat their pants.

35% faster and 2 times more power efficient means their entire lineup becomes hugely inferior to AMD's offerings even if AMD doesn't break the 4.3GHz barrier.

And given the 8121U fiasco (a 10nm CPU without iGPU with the same power envelope), they are truly f*cked. What a time to be alive: Intel is unable to compete both on tech process and IPC. Last time and perhaps the only time it was AMD K8 over 10 years ago.

 

[dj-electric]

Good thing Lisa is not playing the Intel game, she is just making sure amd is very competitive and that is the way to think.

AMD is in danger to exhaust everything they have early. Not playing the Intel game by using all available cannons might have a severe penalty.

 

[Imsochobo]

Intel must have shat their pants.

35% faster and 2 times more power efficient means their entire lineup becomes hugely inferior to AMD's offerings even if AMD doesn't break the 4.3GHz barrier.

And given the 8121U fiasco (a 10nm CPU without iGPU with the same power envelope), they are truly f*cked. What a time to be alive: Intel is unable to compete both on tech process and IPC. Last time and perhaps the only time it was AMD K8 over 10 years ago.

Just the fact that AMD makes just as much money on epyc as intel at half the cost with same performance and More I/O alone is enough to make them shit their pants.

 

[Caring1]

"12 nm "Zen+" architecture features a minor 3-5 percent IPC uplift "
Not sure why you say "minor" as it's never called that when Intel releases a new chip, and that seems to be their average increase in IPC.

 

[TheGuruStud]

"12 nm "Zen+" architecture features a minor 3-5 percent IPC uplift "
Not sure why you say "minor" as it's never called that when Intel releases a new chip, and that seems to be their average increase in IPC.

AMD should have said 15%, then they would be on parity with intel.

 

[ShurikN]

"12 nm "Zen+" architecture features a minor 3-5 percent IPC uplift "
Not sure why you say "minor" as it's never called that when Intel releases a new chip, and that seems to be their average increase in IPC.

Intel had 0% IPC increase going from Skylake to Kaby.

 

[Vya Domus]

AMD is in danger to exhaust everything they have early. Not playing the Intel game by using all available cannons might have a severe penalty.

Except , Intel's cannon is blocked by a boulder , called 10 nm. This is the first time AMD has slight manufacturing process advantage in terms of availability , yields and maybe even performance , that's a big deal.

 

[stimpy88]

"12 nm "Zen+" architecture features a minor 3-5 percent IPC uplift "
Not sure why you say "minor" as it's never called that when Intel releases a new chip, and that seems to be their average increase in IPC.

Yeah, I have noticed this too when reading about AMDs IPC gains for this generation. I'm damn sure Intel has briefed/paid/promised to send more early access samples etc - to the tech press in return for subtly downplaying AMDs achievements.

 

[bug]

I thought this was common knowledge. Enterprises don't update as often, they actually need to see actual improvements.
Whether this was common knowledge or not, it's confirmed now.

 

[XiGMAKiD]

AMD is in danger to exhaust everything they have early. Not playing the Intel game by using all available cannons might have a severe penalty.

I don't think it's a bad idea because AMD is simply using all the time they got now to their advantage before Intel can launch their own leapfrog uArch

 

[bug]

AMD is in danger to exhaust everything they have early. Not playing the Intel game by using all available cannons might have a severe penalty.

On one hand, you don't know what they have. On the other, they're the underdog, they can't afford as many liberties as Intel.
Imho, with Zen AMD played their cards right, so instead of looking to fault them, I'm hoping they'll keep it up.

 

[Imsochobo]

Intel had 0% IPC increase going from Skylake to Kaby.

Intel had 0% from skylake->Kaby->coffee.
Only Performance by clocks or cores.

 

[bug]

Intel had 0% from skylake->Kaby->coffee.
Only Performance by clocks or cores.

It's actually a little worse than that: https://www.hardocp.com/article/2017/01/13/kaby_lake_7700k_vs_sandy_bridge_2600k_ipc_review/
We got faster PCIe lanes, faster USB ports, better IGPs, more specialized instructions, better power saving. But performance basically hasn't budged in a decade. On the other hand, this also speaks about Zen's IPC: as miraculous as it is portrayed, it only managed to get AMD to where Intel was a decade ago and still is today.

 

[Fouquin]

It's actually a little worse than that: https://www.hardocp.com/article/2017/01/13/kaby_lake_7700k_vs_sandy_bridge_2600k_ipc_review/
We got faster PCIe lanes, faster USB ports, better IGPs, more specialized instructions, better power saving. But performance basically hasn't budged in a decade. On the other hand, this also speaks about Zen's IPC: as miraculous as it is portrayed, it only managed to get AMD to where Intel was a decade ago and still is today.

A decade ago the QX9770/9775 was the fastest consumer chip from Intel, with Nehalem launching 5 months after this point.

It isn't 2021 yet, there's still time to improve before Sandy-Bridge turns ten.

 

[bug]

A decade ago the QX9770/9775 was the fastest consumer chip from Intel, with Nehalem launching 5 months after this point.

It isn't 2021 yet, there's still time to improve before Sandy-Bridge turns ten.

Fwiw, Sandy Bridge also didn't improve IPC over Nehalem. But it did bring much faster clock speeds, so overall it did bring better performance to the table.

 

[TristanX]

Zen2 -> 4.5 GHz, 5% more IPC

 

[Vya Domus]

Zen2 -> 4.5 GHz, 5% more IPC

7nm is a big step , I expect more than 4.5 Ghz , not in these server CPUs with dozens of cores though.

 

[HD64G]

Absolutely smart move from AMD imho. Until this time next year, they will be well established again with the 1st gen EPYC and will be able to offer hifher performance, lower consuption and more core cpus for somewhat higher price with higher margins. No point at all to bother with a simple refresh without drastic performance or efficiency gains. They have been on a roll lately as business at least on the cpu side.

 

[bug]

Absolutely smart move from AMD imho. Until this time next year, they will be well established again with the 1st gen EPYC and will be able to offer hifher performance, lower consuption and more core cpus for somewhat higher price with higher margins. No point at all to bother with a simple refresh without drastic performance or efficiency gains. They have been on a roll lately as business at least on the cpu side.

Nope, that's not it.
Small clock increases just don't fly in the enterprise space. Unlike home users, no enterprise user would have been interested in a small 100-200MHz clock increase. Just look at Intel and Xeons: even when they gave us small refreshes, in the enterprise space they almost always had to come up with additional cores, additional sockets support or at least lower TDP.
So this is not a smart move from AMD that no one saw coming, it's just how the enterprise market works.

 

[Fx]

"12 nm "Zen+" architecture features a minor 3-5 percent IPC uplift "
Not sure why you say "minor" as it's never called that when Intel releases a new chip, and that seems to be their average increase in IPC.

Agreed, quite a good point.

This news actually has me reconsidering of waiting for Zen on 7nm. I think the wait would be worth it.

 

[HD64G]

Nope, that's not it.
Small clock increases just don't fly in the enterprise space. Unlike home users, no enterprise user would have been interested in a small 100-200MHz clock increase. Just look at Intel and Xeons: even when they gave us small refreshes, in the enterprise space they almost always had to come up with additional cores, additional sockets support or at least lower TDP.
So this is not a smart move from AMD that no one saw coming, it's just how the enterprise market works.

So for you is a simply sensible move. That's OK for me also.

 

[bug]

So for you is a simply sensible move. That's OK for me also.

It could be more than a sensible move. If AMD doesn't feel pressured into outing half-assed products, it could be a sign they're in a comfortable position where they can release on their own terms. That is big, as I have always said that regardless of how well they sell to us, the big margins are still in the server market.