AMD won't release Zen CPU before Q4 2016 (too late)

[CougTek]

News source (or rumor source

If true, AMD is capout IMO. There's a window that could allow them to increase their market share, but Q4 will be too late. Zen server processors will probably be performance-competitive with either the current Intel Xeon E5 V3 generation, or if AMD has especially well made the Zen core design, maybe with the Xeon E5 V4 (and only because their server CPUs will have more cores than Intel: 32 vs 22). But by the time Q4 2016 arrives, Intel will have a refreshed Xeon lineup either being released, or close to being released and those will almost certainly make the Zen-based server processors look bad.

It's not really surprising, given how often AMD has managed to miss opportunities since what, 2006? What's really saddening is that without competition, Intel's processors' prices will keep rising.

Meanwhile, the Intel Xeon E5 V4 should be released by mid-November, since it would make little sense to launch it later than that. Businesses generally don't make big purchases too close to the holiday break (no one's working to install the new stuff = frozen money).

 

[Mercutio]

I think AMD has been done for a while. It's also demonstrating mini-systems with nvidia graphics and just restored ATI to its previous state as a distinct business unit. I have no idea what the out could be for it at this point.

 

[sedrosken]

What's really saddening is that without competition, Intel's processors' prices will keep rising.

This sort of thing is why I wish Cyrix was still in business, and AMD wasn't being run by idiots.

 

[snowhiker]

[H]ard|OCP was not sent a NANO video card for testing. Because AMD needs fair reviews. Story here.

"Kyle, my position is that we need fair reviews. I doubt there is anyone in our entire industry who would disagree."

-Roy Taylor-Corporate Vice President

"Roy’s backtracking here seems humorous at best. Uh yeah....reviews need to be fair.....especially from those sites mentioned that AMD did not sample."

-Kyle Bennett-[H]ard|OCP


AMD not sending out sample Nanos for testing to all review sites is OK. Whatever. But publicly saying the review sites that didn't get samples need to be fair? Bullshit.

AMD really has a "quality" PR department. Hint to AMD, "You guys need all the help you can get." Don't publicly talk shit about one of the largest review sites out there.

 

[Stereodude]

I see some things never change...

 

[CougTek]

Despite assuring people that their Broadwell-E and Xeon EP V4 would be introduced in Q4 2015 just a few weeks ago, Intel has now apparently decided to postpone the launch of their newest flasgship processors by a quarter to Q1 2016 (source).

I see this as a direct consequence of AMD's latest screw up and delay for Zen processors' launch date. It's also yet another confirmation that marketing people shit thru their mouth.

 

[LunarMist]

Despite assuring people that their Broadwell-E and Xeon EP V4 would be introduced in Q4 2015 just a few weeks ago, Intel has now apparently decided to postpone the launch of their newest flasgship processors by a quarter to Q1 2016 (source).

I see this as a direct consequence of AMD's latest screw up and delay for Zen processors' launch date. It's also yet another confirmation that marketing people shit thru their mouth.

I'm glad to have the Haswell-E instead of waiting.

 

[CougTek]

From bad to worse : Jim Keller is leaving AMD. Rumor is that he's the best CPU architect they'e ever had. Without him to supervise and improve Zen, I don't see how this isn't going to be another major failure for the company. Perhaps their last, at this point.

 

[Buck]

From bad to worse : Jim Keller is leaving AMD. Rumor is that he's the best CPU architect they'e ever had. Without him to supervise and improve Zen, I don't see how this isn't going to be another major failure for the company. Perhaps their last, at this point.

I question what Keller has really done. Since 2012 he has been responsible for "overseeing the teams defining the roadmaps for AMD's CPU cores, systems IP, and server and client SoCs." Where has his effort taken AMD to date? So, without further detail, I cannot say that his egress is good or bad. What he had accomplished prior to being with AMD may not have translated into anything good for AMD. It could also be that AMD hindered his ability to help them be competitive.

 

[CougTek]

Keller was in charge of developping the Zen CPU architecture. Apparently, most of the development is finished, but what about further improvement to the design?

Keller out of AMD cannot be good for AMD.

 

[Mercutio]

Given that AMD is a fraction of Intel's size, fabless and stuck on badly outdated tech from its manufacturing partners, the writing really is on the wall. Intel is pushing back product releases so as not to bury it, I think.

 

[Tea]

AMD is still a thing?

 

[Stereodude]

Just barely.

 

[Handruin]

I think IBM or even Samsung should acquire AMD when Zen fails to deliver. Samsung has much better fabs that they could actually make something solid out of AMD. We need someone to carry the torch or else innovation will stifle as will the increase in costs from Intel.

 

[Mercutio]

What does Samsung want with AMD? Contracts to build APUs for Microsoft and Sony? An escalation of the competition with Intel? A reason to re-tool its 14nm fabrication process away from all the ARMs it can handle?
IBM seems to have turned its back on hardware. Yes, it still does big iron, but its management clearly wants an IP licensing-and-services corporation. Is anyone even using Power CPUs outside of i and z series systems?

I usually tell people that AMD is 1/10th the size of Intel, but as of 20 minutes ago, AMD's market cap is $1.3B and Intel's is $135B; even nVidia is worth $12.5B. That's not the whole story of who is worth what but that's not a pretty picture. AMD is worth as much as a photo sharing service people use to take pictures of food.

Just to throw out some other big names:
Apple - Has no reason. It's on Intel now. It knows Intel is better for non-mobile. Maybe it wants some better ARM engineering, but even at that it's cheaper to just buy the best ARM guys in the world or wait for Intel to finish making an Atom that makes everyone forget that ARM was ever even an option.
Google / Facebook / Amazon - the idea I've heard here is that these are companies that buy so many servers that it might on some distant day make financial sense. Because engineering CPUs is going to fix all the other costs associated with running datacenters? Let's move on.
Microsoft - To take an option away from Sony in those ever-important game console wars, maybe? Plus probably most of the countries in the Northern Hemisphere would start talking about antitrust issues, something that they really just stopped talking about maybe five years ago. Nope. Not happening.


Maybe a lateral mood to another electronics company? Merging with Asus (~US$6B market cap) or Foxconn (~US$33B) might somehow work out, but neither company actually has the know-how or wherewithal to fix AMD's fundamental problem.
There's also a dim possibility that a company like NEC or Fujitsu might be interested, but I don't think large-scale computing systems are a big enough part of either company's business to step away from what they're doing right now

Aside from screaming "Sell, Mortimer, Sell!", if I were running things at AMD, I'd be shopping for any sort of design improvement that could be adapted to the technology they do have, hopefully far, far away from places where Intel and Samsung representatives have been with their bags full of cash. I have to assume that's part of the goal of the memory designs presently being used for the new R9 GPUs. AMD got lucky enough with the NexGen purchase to escape the fate that met Cyrix, but to my recollection the circumstances now seem eerily similar.

 

[CougTek]

According to a source of absolutely unknown reliability, Zen's IPC is roughly equal to Broadwell, which is good news IMO for AMD since Intel is moving at a snail pace these days. So even with an October launch, like rumored today, with chips with twice as many cores as Intel, Zen could still end up being a very interesting deal.

AMD being AMD, they'll certainly manage to screw it up anyway, but still. They will have a fighting chance.

 

[Mercutio]

My guess is that Intel releases an Atom CPU (Kaby Lake, maybe?) with 20 better per-thread performance than Skylake Core Ms... which are themselves 20% faster per thread than Broadwell. Even if it's a paper launch and actual product doesn't ship at any time in 2016, it'll crater the possibility of anyone launching a product on Zen unless the damned things are priced at ARM-like levels.

 

[Stereodude]

According to a source of absolutely unknown reliability, Zen's IPC is roughly equal to Broadwell, which is good news IMO for AMD since Intel is moving at a snail pace these days. So even with an October launch, like rumored today, with chips with twice as many cores as Intel, Zen could still end up being a very interesting deal.

I'll believe that when I see it.

 

[CougTek]

Supposedly confirmed, but not officially by AMD itself, AMD will launch a 32-core/64-thread server CPU next year (that was already widely known). The chip will dissipate up to 180W and will integrate a 16x10GbE controller. Hopefully, it will arrive in early 2017, but no precise timeframe other than the year has been published.

Codenamed Naples, the chip uses the Zen architecture. Each Zen core has its own dedicated 512kb cache. A cluster [shurely that should be cloister.ed] of Zen cores shares a 8MB L3 cache which makes the total amount of L3 shared cache 64MB. This is a big chip and of course there will be a 16 core variant.

This will be a 14nm FinFET product manufactured in GlobalFoundries and supporting the X86 instruction set. Naples has eight independent memory channels and up to 128 lanes of gen 3 PCIe. This makes it suitable for fast NVMO memory controllers and drives. Naples also support up to 32 SATA or NVME drives.

source

 

[ddrueding]

Integrated 16x10GbE seems to be screaming for the supercomputer space. Or just really extreme routers?

 

[Handruin]

Integrated 16x10GbE seems to be screaming for the supercomputer space. Or just really extreme routers?

That and the 128 lanes of PCIe means it's really geared for NVMe and other very high bandwidth IO devices.