Zen ETA Q4, '16 ?

[dark zero]

Some people here seem to have too much spare time for repeating the same arguments over and over about well known facts. To follow this tradition let me quote myself again:

Read well that again... it says MAINSTREAM, but..how about the Entusiast and Server tier?

 

[Dresdenboy]

Read well that again... it says MAINSTREAM, but..how about the Entusiast and Server tier?

That's right as long as that "2017" doesn't exclude a 2017 server CPU/APU and the engineer somehow forgot, that the uarch, he's working on, will already appear in 2016.

 

[svenge]

But Sandy Bridge and Sandy Bridge-E were released 4 years ago, so the timing of their release dates vs. projections is somewhat irrelevant at this point.

Who has mentioned SB and SB-E?

I was talking about the Intel releases the last few years and the delays we've seen for those, and what we can expect going forwards (CannonLake and 10 nm delayed, KabyLake stopgap solution introduced).

You did (albeit inadvertantly) when you said that "we're discussing what Intel CPUs that AMD Zen will compete with".

Zen will end up being competitive with 2011's Sandy Bridge, and maybe 2012's Ivy Bridge on a good day with highly-optimized workloads. Haswell and the Skylake/Kaby Lake lines will be far out of reach.

 

[guskline]

Do any of us have ANY idea what ZEN will be competitive with? Seriously. We at least know the name and have some slides. Is there any silicon floating around anywhere? Hard to know what if anything it will be competitive with.

 

[jhu]

Do any of us have ANY idea what ZEN will be competitive with? Seriously. We at least know the name and have some slides. Is there any silicon floating around anywhere? Hard to know what if anything it will be competitive with.

It's unlikely we'll know until it comes out.

 

[Dresdenboy]

Zen will end up being competitive with 2011's Sandy Bridge, and maybe 2012's Ivy Bridge on a good day with highly-optimized workloads. Haswell and the Skylake/Kaby Lake lines will be far out of reach.

I agree with you while accepting your hidden assumption (or overlooking) of a 32nm bulk process for Zen.

While solely thinking about IPC don't miss advantages in production process, ISA extensions (AVX2, faster AES engines, transactional memory, and all the smaller ones), power management (Zen is not Llano), memory subsystem (DDR4), etc.

 

[tenks]

So yes, AMD's miracle chip will probably face Skylake-E for most of its life cycle.

Oh, and Broadwell-E could be launched as soon as Cebit in late Q1-2016 according to Bench-life (source of the leaks). Not sure where the Q3-2016 launch B$ is coming from.

this. thank you

This horrible logic of q1 production means q3 availability is just strait silly.

Production will begin in Jan, launching at Cebit in March. March is Q1. The end. From the same source you're basing your silly math from.


Also, why are people taking AMD's word on anything CPU related at this point? Intel's dates are ignored yet they regularly deliver and AMD's dates are the truth but they never deliver. Surprise, GloFo/TMSC is behind/not ready. I feel like that's baked into every one of their process shrinks at this point. When was the last node on time from either of those 2?

 

[jpiniero]

Production will begin in Jan, launching at Cebit in March. March is Q1. The end. From the same source you're basing your silly math from.

The latest leaked chart says it is starting production at the end of February maybe. Clearly there was an additional 6-8 week production delay likely due to how bad the 14 nm fabs are doing at something not tiny. Cebit was likely the original target for launch; there's probably not much downside from not formally announcing it then even if there isn't much/any product available for some time.

 

[DrMrLordX]

Supposedly Bristol Ridge could enter on 14 nm instead on 28 nm.. But that seems to be discarded... So AMD will hardly have some performance increase.

I would like to see 14nm Excavator as well. I just don't see AMD's budget as accomodating it, unless porting to 14nm is just as complicated/expensive as porting it to 28nm minus HDL. As it stands, Carrizo has inferior voltage scaling compared to GV-A1 Kaveri past 2.6 GHz, so there's no reason to launch a desktop chip using HDL unless GF has corrected that situation through a new stepping.

AMD DOES need a test-case for GF's 14nm process, and 14nm Excavator would fit the bill for that. But does it fit the budget?

Also, why are people taking AMD's word on anything CPU related at this point?

We just went over this very question above. Kaveri was late, but most of their other CPU launches over the last 3 years have been on time. All data points to Carrizo/Carrizo-L being ready for the June paper launch which was delayed by Win10. So AMD is actually delivering hardware. Whether or not people like or respect what is delivered is another matter . . .

 

[tenks]

We just went over this very question above. Kaveri was late, but most of their other CPU launches over the last 3 years have been on time. All data points to Carrizo/Carrizo-L being ready for the June paper launch which was delayed by Win10. So AMD is actually delivering hardware. Whether or not people like or respect what is delivered is another matter . . .

Great, a couple wins doesn't mean you shouldn't proceed without caution, especially knowing their past and the foundries they're dealing with.

 

[frozentundra123456]

Great, a couple wins doesn't mean you shouldn't proceed without caution, especially knowing their past and the foundries they're dealing with.

Yea, AMD's cpu launches the last few years have had relatively minor changes compared to a brand new architecture on a brand new node. I will be very surprised if there are not more delays, or at best a formal launch with very little availability, like intel 14nm.

 

[Guest1]

Wow, that is a total lack of foresight on AMD and GF for allowing this to happen. Intel generated 13 billion in sales last quarter. If AMD has a competitive product that can recapture even a small portion of that market share, they are losing billions for every quarter they delay.

It's not a matter of foresite but years of mismanagement. When AMD was competitive with intel they weren't paranoid enough which allowed intel to surpass them forcing them to eventually spin off their fab and leave them in the situation they are in.

 

[svenge]

I agree with you while accepting your hidden assumption (or overlooking) of a 32nm bulk process for Zen.

No, I was factoring in the "14nm" (quotes intentional) process that Global Failures is still struggling with along with AMD's general incompetence in CPU design.

And no, Jim Keller is not some sort of silicon messiah that can overcome all the problems inherent with their current situation by himself. After all, even Jesus had 12 disciples to help him, which is still probably a higher number than the headcount of AMD's CPU engineering department currently after all the waves of layoffs!

 

[dark zero]

I would like to see 14nm Excavator as well. I just don't see AMD's budget as accomodating it, unless porting to 14nm is just as complicated/expensive as porting it to 28nm minus HDL. As it stands, Carrizo has inferior voltage scaling compared to GV-A1 Kaveri past 2.6 GHz, so there's no reason to launch a desktop chip using HDL unless GF has corrected that situation through a new stepping.

AMD DOES need a test-case for GF's 14nm process, and 14nm Excavator would fit the bill for that. But does it fit the budget?

Maybe Bristol might not overclock past 4.0 Ghz and has as better a similar performance as an i5 Broadwell (theorical one) and as worse an i5 Haswell

 

[guskline]

Well the name's cool. ZEN

 

[AtenRa]

After all, even Jesus had 12 disciples to help him, which is still probably a higher number than the headcount of AMD's CPU engineering department currently after all the waves of layoffs!

I believe its Keller and one or two more engineers drawing blueprints by hand.

 

[guskline]

I believe its Keller and one or two more engineers drawing blueprints by hand.

I sure hope you are kidding. Which foundry is going to build it?

 

[guskline]

So they will be expensive or lack performance. Pick one.

Both!:\

 

[Idontcare]

I sure hope you are kidding. Which foundry is going to build it?

AMD has no choice, it has to be fabbed by GF.

 

[Dresdenboy]

I believe its Keller and one or two more engineers drawing blueprints by hand.

I heard he uses a C64 for cycle exact simulations.

 

[Dresdenboy]

Both!:\

We don't have socialism.

 

[AtenRa]

I heard he uses a C64 for cycle exact simulations.

Too expensive, they using Sinclair ZX81.

 

[svenge]

Too expensive, they using Sinclair ZX81.

Still too expensive. Probably using an Altair 8800.

 

[DrMrLordX]

Maybe Bristol might not overclock past 4.0 Ghz and has as better a similar performance as an i5 Broadwell (theorical one) and as worse an i5 Haswell

I'm still scratching my head over that last bit. Not sure what you meant there.

But if they stick to the HDL libraries, expect some pretty-wretched voltage scaling at higher clockspeeds.