New Zen microarchitecture details

[inf64]

AMD is presenting Zen today at ISSCC event:
https://submissions.mirasmart.com/ISSCC2017/PDF/ISSCC2017AdvanceProgram.pdf

Presentation starts at 2PM in SF,CA.

2:00 PM
3.2 Zen: A Next-Generation High-Performance x86 Core
T. Singh1, S. Rangarajan1, D. John1, C. Henrion2, S. Southard1, H. McIntyre3,
A. Novak1, S. Kosonocky2, R. Jotwani1, A. Schaefer1, E. Chang2, J. Bell1, M. Co1
1AMD, Austin, TX; 2AMD, Fort Collins, CO; 3AMD, Sunnyvale, CA

There should be some new info released today in the presentation (supposedly who registered early got a paper from the presentation) so keep an eye out for it.

 

[guskline]

I expect the price on the highest rated SR7 Ryzen to been overinflated until the hoopla dies down.

 

[Lovec1990]

AMD is presenting Zen today at ISSCC event:
https://submissions.mirasmart.com/ISSCC2017/PDF/ISSCC2017AdvanceProgram.pdf

Presentation starts at 2PM in SF,CA.

2:00 PM
3.2 Zen: A Next-Generation High-Performance x86 Core
T. Singh1, S. Rangarajan1, D. John1, C. Henrion2, S. Southard1, H. McIntyre3,
A. Novak1, S. Kosonocky2, R. Jotwani1, A. Schaefer1, E. Chang2, J. Bell1, M. Co1
1AMD, Austin, TX; 2AMD, Fort Collins, CO; 3AMD, Sunnyvale, CA

There should be some new info released today in the presentation (supposedly who registered early got a paper from the presentation) so keep an eye out for it.

too bad it wont be live so others from other corners of earth could watch it

 

[Insert_Nickname]

We see tons of boards. Anyone remember how it was at K7 launch?

Yes actually. It was pretty difficult to get a Slot A board for a while. This is absolutely nothing like then. Luckily.

Heh! K7 was pretty sweet, but none of the chipsets supporting it were all that good. Poor memory controllers, IDE data corruption bugs, the works.

Sure the AMD 750 and the early VIA ones weren't that good. To put it mildly. The later nForce2 and KT600/KT880 were.

Had Zero problems with the Asus A7N8X deluxe Rev 2.0 ( yes i remembered that off the top of my head )

Hey, had one of those. Was pretty good for its time.

 

[hojnikb]

Those were the times... Pretty much anyone was making chipsets

That got me thinking. Is it possible for 3rd party to make AM4 "chipsets" or are they exclusive to amd ?

 

[krumme]

Those were the times... Pretty much anyone was making chipsets

That got me thinking. Is it possible for 3rd party to make AM4 "chipsets" or are they exclusive to amd ?

Would guess not. There is not much in it these days compared to back then so no business anyway. Amd probably license a lot of the chipset ip stuff outside anyways. Zen is a Soc.

 

[Crumpet]

This is... interesting.

It's about Intel potentially responding to Ryzen, from CPC

https://www.cpchardware.com/intel-prepare-la-riposte-a-ryzen/
(link is in French, so i'll copy paste the important bit in English here)

• First there is the Core i7 7740K . It has 8 MB of L3, with a base frequency of 4.30 GHz (100 MHz more than the Core i7 7700K) at the price of a significantly higher TDP (> 100W against 91W).The Turbo frequency is not known, but should logically be 4.60 GHz.
  
  • One of the other sample was renamed Core i5 7640K . At first glance, this is also a speed bump from the Core i5 7600K: Quad Core, 6MB L3, 4.0 GHz base frequency for the 3.8 GHz against 7640K for 7600K. On the TDP side, it would also climb above the 100W. But there is much more interesting: the Core i5 7640K would have active SMT (Hyper-threading), a first for a Core i5 Desktop.

• • 
    

 

[mikk]

It makes no sense to me. If the 7700k isn't enough a 100 Mhz faster model won't be good enough either, that's just a minor 2% increase.

 

[inf64]

It makes no sense to me. If the 7700k isn't enough a 100 Mhz faster model won't be good enough either, that's just a minor 2% increase.

It's all about marketing, if they can squeeze another few % and extend the ST performance lead it would paint better picture when reviews are published. Desktop space does not have that many workloads that scale that well with more than 4 cores so having such a highly clocked part would be a plus for intel.

 

[Crumpet]

The hyperthreaded i5 however, leaves me baffled.

 

[inf64]

The hyperthreaded i5 however, leaves me baffled.

I don't think that part is going to happen. At least not under i5 moniker.

 

[DrMrLordX]

Had Zero problems with the Asus A7N8X deluxe Rev 2.0 ( yes i remembered that off the top of my head )

AMD's suppliers had growing pains early on. For example:

https://en.wikipedia.org/wiki/List_of_VIA_chipsets#Slot_A_and_Socket_A

KX133, KT133, and KT133A all had issues with IDE corruption:

https://lkml.org/lkml/2002/1/18/77

Even once that stuff was sorted out, there was still the issue of VIA's weak-ass memory controller. They finally started to get their act together when SiS fired a shot across their bow with the SiS735. VIA answered with the KT266A which was the same as the KT266 with a mem controller refresh. It might have also had the IDE corruption bug still, I don't 100% remember.

It wasn't all smooth sailing though.

I loved my SiS735-based K7S5A, but a lot of people had problems with those thanks to ECS cheaping out on components. Some revisions were apparently better than others. I could never tell . . . mine just worked.

The later nForce2 and KT600/KT880 were.

By the time nForce2 was out, things were getting a lot better in the AMD universe. But that was getting on towards the end of K7's run.

Those were the times... Pretty much anyone was making chipsets

That got me thinking. Is it possible for 3rd party to make AM4 "chipsets" or are they exclusive to amd ?

Yeah, you used to have ALi, SiS, and VIA making them. I don't know that I miss that though since Q/A was pretty rough on some of that stuff. Accountability was just not there, especially when VIA's chipset division was in play. Today it seems like AMD contracts out to have their in-house stuff done, and then they push out the result to the public, either in the form of the chips on the boards or in the form of IP embedded into their CPUs/APUs (AMD has been farming out design of their IMCs for some time now). At the very least AMD is now forced to take responsibility for failure of any of the "chipset" components.

With so much of the traditional chipset now integrated into the SoCs/APUs coming from AMD, I do not think we'll see 3rd party chipsets anymore, just 3rd party IP.

 

[Glo.]

I don't think that part is going to happen. At least not under i5 moniker.

If AMD provides 4.0 GHz, 65W 4C/8T CPU for 199$, do you think Intel can afford putting "high price" branding on the CPUs mentioned?

 

[Magic Hate Ball]

I don't think that part is going to happen. At least not under i5 moniker.

I think Intel realizes the jig is up on core/thread counts. They just released the lowest tier Pentium chips as 2c/4t directly competing with the i3's. I think we're going to see a shift upwards in the Intel line up to compete with Ryzen.

Pentium 2c/4t
i3 4c/4t
i5 4c/8t
i7 6c/12t (or more, and around the 7700k price range)

They're starting now, but the transition won't be able to be completed until they have the chips roll out over the next few years. They may have gotten too comfortable.

 

[inf64]

If AMD provides 4.0 GHz, 65W 4C/8T CPU for 199$, do you think Intel can afford putting "high price" branding on the CPUs mentioned?

What I meant was that the second SKU that CPC HW mentions would break current i5/i7 differentiation. I don't know what will intel do, they have some options, like price cuts and offering bundles with CPUs.

8T Ryzen will for sure be disruptive product, especially in price/perf. category. All parts unlocked and can be OCed on cheaper boards, no need for expensive cooling, good thermals etc.

 

[Crumpet]

That was more or less along the lines I was thinking.

But with Ryzen an estimated 4 weeks away, with these new Kaby Lake cpus in validation they're surely going to be 1-2 months behind at best?

 

[Arachnotronic]

That was more or less along the lines I was thinking.

But with Ryzen an estimated 4 weeks away, with these new Kaby Lake cpus in validation they're surely going to be 1-2 months behind at best?

The silicon itself is done, they just need to bin more aggressively to find the parts that can hit the higher frequencies. If 7740K/7640K are real, I could see Intel hitting time to market with Ryzen.

 

[Crumpet]

The silicon itself is done, they just need to bin more aggressively to find the parts that can hit the higher frequencies. If 7740K/7640K are real, I could see Intel hitting time to market with Ryzen.

Interesting. Do you think the higher TDP indicates significantly higher turbo clocks, as I can't see them needing a 9w TDP jump for 100hz.

 

[Magic Hate Ball]

The silicon itself is done, they just need to bin more aggressively to find the parts that can hit the higher frequencies. If 7740K/7640K are real, I could see Intel hitting time to market with Ryzen.

Or they could just put less garbage TIM under the IHS and reduce the gap between the chips and the IHS.

Seriously, having to delid a $350 part to fix the thermals is shameful from a company like Intel.

 

[Arachnotronic]

Interesting. Do you think the higher TDP indicates significantly higher turbo clocks, as I can't see them needing a 9w TDP jump for 100hz.

I think they might increase the TDP to allow more silicon to "pass" as the 7740K. If they restricted to 91W, they could only sell chips that hit the higher speeds at the same power consumption, and those are probably rare birds at the moment. Better to up the TDP a little more (they don't even include a HSF in the box so who cares) so that a reasonable number of chips yield than to risk a supply shortage.

 

[tamz_msc]

So Intel is preparing "Devil's Canyon" Kaby Lake?

 

[Arachnotronic]

Son Intel is preparing "Devil's Canyon" Kaby Lake?

Pretty much, but Devil's Canyon had some changes to the packaging and TIM, this new chip likely has neither enhancements.

It's a typical speed bump, which were common in the past but became rarer from Intel when they had virtually no serious competitive threats.

 

[Crumpet]

Cthulhu Lake?

 

[tamz_msc]

From the CanardPC article, it seems AMD is having trouble reaching 4GHz base clocks on the 4C parts - which means that they are failed 8C parts (since there is only one die, which is as big as 200mm2, can someone confirm its size?). If so then AMD can't price them too low or else they'll lose money. AMD will have lesser room to maneuver in the sub-250$ range if that is the case, provided Ryzen is competitive enough with 4C Intel parts.

 

[CatMerc]

On the bright side for consumers, this will force AMD to price 4C/8T lower than most of us expected. While overall performance might be lower, with games targeting 8 threads (thanks to consoles), there are occasions where 4C/4T would do worse in minimum framerates than 4C/8T, even if total throughput is lower, due to various timing issues.

I personally would take a 4C/8T at 3.5GHz over a 4C/4T at 4.5GHz.