New Zen microarchitecture details

[Doom2pro]

I'm planning on building a premium VR rig, but I'm torn if I should get the top notch chipset so I can crossfire two Vega cards, or if I should get the chipset below that and just get one Vega card.

I guess prices will determine that, so hurry up and wait I guess :\

 

[The Stilt]

Despite the product isn't launched yet and the performance still isn't officially confirmed, yet I have to say that I'm simply stunned. Not only by Ryzen, but the sudden change AMD has gone through. In reality of course the change hasn't happened over night however it still feels like AMD has suddenly been waken from a bad spell, which has lasted for almost seven years. Most of you are familiar with the things I've very openly criticized during the past couple years, in regards of the products themselves and the marketing / general policies AMD has practiced and executed. I'm fully aware that many of the things I said were harsh, but that was mostly intentional and I don't regret any of it as I never had to lie about any of it. In those cases where the harsh comments weren't intentional, they became more harsh than intended out of plain frustration.

If a year ago I would have been asked to bet my life either on AMD or against AMD, I wouldn't have had any other choice but to bet against them.

- Hitting competitive frequencies on a design manufactured on a low power manufacturing process, targeted for handheld devices - (allegedly)
- Meeting, let alone exceeding the advertized marketing figures (in this case the 40% IPC improvement over XV). Let alone by a figure, which apparently can be nearly 40% higher than originally quoted (up to 55% absolute over XV) - (allegedly)
- Be able to even remotely keep up with the original promises, let alone vastly exceeding them.

For the record, I was wrong about Zen.
For the Fmax part due having a prejudice against low power processes in general and against GlobalFoundries in particular, and for the IPC part because I couldn't even dream that AMD could meet, let alone exceed their advertized IPC figures.

If I wore a hat, I would tip it to AMD right now.

 

[F-Rex]

but but but.... but..... but....... FO4.....

Process experts told us on this forum that:
1 - 14lpp was low power/low performance and frequencies higher than 2.8/3.2 turbo should not be expected.
2- A0 was first and probably last ryzen spin. Don't expect new revision.
3- ryzen ipc Will be lower than Ivy.

And when first demo showed up, we were told that 20% error margin was good.

So let me give you my Guess with the same 20% error margin: 3.8ghz/4.8ghz 1 core turbo.

 

[AtenRa]

B350 is all I need!

Im looking at the X300 😉

 

[F-Rex]

[SARCASM]
But... But... But... INTEL process is better and FO4 is a meaningless measure!!!!11!!!1!one!eleven! How can this be possible?
[/SARCASM]

14lpp is low power. Samsung and gf don't know how to develop a process. Only Intel know how to achieve 1 GHz+ frequencies.
Could our process experts help us?
Perhaps amd has rebadged some i7 6900k.

 

[raghu78]

I think everybody here now agrees that AMD and GF has done exceedingly well on the clock frequencies. I think this blows away the allegations that Zen is a low frequency design and that 14LPP is a low power process and not suitable for high frequencies. When stock turbo on a 8C/16T hits 4 Ghz you can damn well bet its a high performance CPU design on a high performance process. btw I think 3.6 Ghz base and 4.0 Ghz turbo will be the launch frequencies. I think 4.5-4.6 Ghz max OC on air for 24x7 use looks realistic now. The future looks bright for AMD with Zen and Vega.

 

[inf64]

I have to say that I also underestimated Zen 🙂. Apparently AMD not only out-clocked my estimated Zen, they out-IPCed it too 😀. Six months ago I wouldn't even dare dreaming of 3.3Ghz base clocks, let alone 3.6Ghz. I also wrote this :

I think the SRx branding is real and the XP is fake. AMD MUST hit at least ~3.2Ghz base and ~3.6Ghz ST Turbo clocks if they want to be semi-competitive against Broadwell-E parts. They likely won't match top 8 core part but could land close, especially if the unlocked top 8C/16T part can clock to 4-4.2Ghz on good cooling.

As all of you can see, I ALMOST nailed it.

 

[Glo.]

I will not be that radical, as you guys are.

If you want my opinion, Highest end Ryzen CPU will top at 4.0 GHz base clock, and 4.2 GHz all core turbo and 95W TDP.

I think I have to quote myself 😉

 

[dahorns]

This is all very exciting, but, my translation from http://www.hardwareluxx.de/index.ph...-doom-auf-ryzen-und-vega-cpu-bei-3-6-ghz.html says:

At CES 2017, however, we were able to take a brief look at a RYZEN system and open the system information of Windows 10. This indicated a clock of 3.6 GHz. The term also suggests that the processor reaches a base clock of 3.6 GHz, but individual cores can operate at up to 3.9 GHz.

Doesn't that suggest single core turbo of 3.9, not all cores? Is it mistranslated?

 

[inf64]

This is all very exciting, but, my translation from http://www.hardwareluxx.de/index.ph...-doom-auf-ryzen-und-vega-cpu-bei-3-6-ghz.html says:



Doesn't that suggest single core turbo of 3.9, not all cores? Is it mistranslated?

It doesn't matter. ST Turbo of 3.9Ghz is way higher than any of us expected. Base is important as it is minimum guaranteed frequency for any load on maximum loaded cores(8). All cores and half cores Turbo is also important because it will surely go above base 3.6Ghz in that case. We are now talking 5% higher than all core Turbo of 6900K which is 1000+ dollars SKU.

Also CanardPC already said via twitter they have 3.6/4Ghz part. Things are moving really fast now.

 

[bjt2]

This is all very exciting, but, my translation from http://www.hardwareluxx.de/index.ph...-doom-auf-ryzen-und-vega-cpu-bei-3-6-ghz.html says:



Doesn't that suggest single core turbo of 3.9, not all cores? Is it mistranslated?

Yes, but is still an ES. There is no point to have much higher max core turbo if the purpose of an ES is just to verify that the clock changes when you have low load... +300 or +1000 MHz what is the difference, considering the purpose of the ES?

 

[Tup3x]

It's pretty obvious now that AMD has a real winner here. They are definitely back in competition. Now if the price is also better than expected....

 

[dahorns]

It's pretty obvious now that AMD has a real winner here. They are definitely back in competition. Now if the price is also better than expected....

That's the part I doubt. People are expecting $200 or $300 for the 8-core. That is never going to happen. I'd still wager $600-$800 for the 8c/16t part. Still a great deal for an enthusiast chip.

 

[AtenRa]

I want a mini-itx X300 board with 65W TDP 6C 12T ZEN + Vega NANO 8GB HBM2 yesterday

That's the part I doubt. People are expecting $200 or $300 for the 8-core. That is never going to happen. I'd still wager $600-$800 for the 8c/16t part. Still a great deal for an enthusiast chip.

Im waiting $400 to $600 for the 8C 16T SKUs

 

[Abwx]

Obviously also the process and the final actual design counts. But if on a worse process you see way higher frequencies, can't be other than the FO4. Then the other details gives you +-200MHz... But here there is not doubt: +400MHz base with 45W less is only possible with FO4. Or the INTEL process that is shitty...

14nm LPP has 20% better transconductance than Intel s 14nm, so at equal input capacitance this mechanicaly get you 17% lower gate delay (at same VDD), or FO4 if you prefer.

Dunno for said input capacitance but there s no reason that it would be higher for GF s 14nm.

 

[Glo.]

That's the part I doubt. People are expecting $200 or $300 for the 8-core. That is never going to happen. I'd still wager $600-$800 for the 8c/16t part. Still a great deal for an enthusiast chip.

This is NOT enthusiast chip. SR3, SR5 and SR7 are meant to be Socket 1151 CPU competitors. Putting it on the 500$ price range is outrageous, despite the core counts.

 

[inf64]

This is NOT enthusiast chip. SR3, SR5 and SR7 are meant to be Socket 1151 CPU competitors. Putting it on the 500$ price range is outrageous, despite the core counts.

Hmm if it beats 6900K, why would you even think about comparing it to Socket 1151 SKUs? AMD is not a charity, they will price it above i7 s1151 mainstream range for sure. For sure they will undercut every single intel SKUs in similar performance bracket, they want both higher margins and higher market share.

 

[Tup3x]

This is NOT enthusiast chip. SR3, SR5 and SR7 are meant to be Socket 1151 CPU competitors. Putting it on the 500$ price range is outrageous, despite the core counts.

Yeah. This should be more or less the same size as mainstream i7 chips (since no GPU), right? So shouldn't the manufacturing costs be the same? Also this lacks the things that make the Intel HEDT platform expensive so why this should cost 600 €?

Personally I'm expecting them to pretty much match the mainstream i7 pricing with 8 core Zen.

 

[laamanaator]

This is NOT enthusiast chip. SR3, SR5 and SR7 are meant to be Socket 1151 CPU competitors. Putting it on the 500$ price range is outrageous, despite the core counts.

But if ryzen performs like an "enthusiast chip", why should it cost any less than said enthusiast chip even if the platform it fits in is touted as a mainstream platform? If ryzen performs like a 6900K, it's going to be priced relatively similarly.

 

[lolfail9001]

Yeah. This should be more or less the same size as mainstream i7 chips (since no GPU), right? So shouldn't the manufacturing costs be the same? Also this lacks the things that make the Intel HEDT platform expensive so why this should cost 600 €?

Yeah, why not 800 GBP in line with 6900k? Platform costs are lower anyways.

 

[Glo.]

Hmm if it beats 6900K, why would you even think about comparing it to Socket 1151 SKUs? AMD is not a charity, they will price it above i7 s1151 mainstream range for sure.

Of course they will. However... IMO if Fott is right and the die for 8C/16T CPU costs AMD 18-20$ to make, base 8C/16T chip will start at around 350-400$.

Highest end Ryzen chip with my prediction of 4 GHz/4.2 GHz boost(Black Edition), would be 500$. If you consider it as a i7 1151 competitor do you think this is charity? Or just upping the prices by AMD? Think about it this way.

And best part: https://twitter.com/BitsAndChipsEng/status/814837806933692416

 

[riggnix]

According to canard PC, they have step F4 and is 3.6/4.0 https://twitter.com/CPCHardware/status/817044837358780416 (I suppose still ES)...

I'm seriously starting to freak out about these clocks 🙂 Can't wait to get my hands on this chip. At this point I don't even care about the price anymore 😀
What I don't really get is why they insist on max 95W TDP. It seems that those chips clock pretty high, why not push it to 125W? Or 140W like they did with AM3+! Just imagine this baby with 4+ GHz base!

I'm assuming no one mentioned anything about a release date, right?

 

[riggnix]

Yeah, why not 800 GBP in line with 6900k? Platform costs are lower anyways.

AMD desperately needs to get market share. So I'm guessing (hoping) they try to be noticeably cheaper then Intel. But that still would mean 400 $ at the VERY least i guess.

 

[AtenRa]

Ok, i believe im aiming for a mini-itx X300 mobo with 65W TDP 6C 12T ZEN and Vega Nano some time in summer to replaice my current Core i7 3770K system.

 

[Glo.]

I'm seriously starting to freak out about these clocks 🙂 Can't wait to get my hands on this chip. At this point I don't even care about the price anymore 😀
What I don't really get is why they insist on max 95W TDP. It seems that those chips clock pretty high, why not push it to 125W? Or 140W like they did with AM3+! Just imagine this baby with 4+ GHz base!

I'm assuming no one mentioned anything about a release date, right?

Because it is not 6900K competitor, but its Socket 1151 competitor.