Full Skylake reveal result? Waiting for Zen.

[guskline]

Define "affordable" 8 core/16 thread. I realize the 5960x is not inexpensive. However, a Zen with 8 cores and 16 threads? Unlikely to be in the $200-$400 range. More likely $700-$800. Since we are all speculating, myself included, and I am correct on my price assumption, would you opt for a $700 Zen or a $1000 5960x?

 

[ShintaiDK]

​

They're just licensing it from Samsung, they can't screw it up this time, right?

...right?

They already lost the 30% A9 for Apple because they screwed up.

 

[AtenRa]

An 8-core 16x threads ZEN at 14nm could be very small, 150-180mm2. Selling the High-end SKUs for $500-$600 will bring amazing Margins for AMD in the desktop. 6-Core Zen at $300 could be a nice alternative to Core i7 Socket 1151 and will also bring very high margins.

So i dont see the need to price them at $800-1000 but certainly not at sub $300.

 

[pepone1234]

They're just licensing it from Samsung, they can't screw it up this time, right?

...right?

they can't even do a smaller SoC design apple gave them so i think they will screw up.

 

[ShintaiDK]

An 8-core 16x threads ZEN at 14nm could be very small, 150-180mm2. Selling the High-end SKUs for $500-$600 will bring amazing Margins for AMD in the desktop. 6-Core Zen at $300 could be a nice alternative to Core i7 Socket 1151 and will also bring very high margins.

So i dont see the need to price them at $800-1000 but certainly not at sub $300.

Lets see how it performs. Marketing slides is one thing, reality another.

AMD havent delivered to its promise for 10 years now.

 

[witeken]

An 8-core 16x threads ZEN at 14nm could be very small, 150-180mm2. Selling the High-end SKUs for $500-$600 will bring amazing Margins for AMD in the desktop. 6-Core Zen at $300 could be a nice alternative to Core i7 Socket 1151 and will also bring very high margins.

So i dont see the need to price them at $800-1000 but certainly not at sub $300.

Price per transistor isn't much lower, if it isn't higher, so margins will not be amazing. Impossible anyway if you don't have nice market share.

 

[AtenRa]

Price per transistor isn't much lower, if it isn't higher, so margins will not be amazing. Impossible anyway if you don't have nice market share.

Even if 14nm wafer production cost is 2x higher than current 28nm HDL, a 160mm2 8x core 16x die sold at $300 will get you 2x the profit than a 250mm2 Kaveri sold at $150.
So even with the same market share, you still get 2x more profits thus increasing margins.

 

[guskline]

An 8-core 16x threads ZEN at 14nm could be very small, 150-180mm2. Selling the High-end SKUs for $500-$600 will bring amazing Margins for AMD in the desktop. 6-Core Zen at $300 could be a nice alternative to Core i7 Socket 1151 and will also bring very high margins.

So i dont see the need to price them at $800-1000 but certainly not at sub $300.

AtenRa: I made my comments concerning price based upon the comments of AMD's CEO that they want to up price on their premium products and that a Zen 8 core/16 thread will likely be their top premium product.

I also doubt that the Zen 8c/16t will be as small as you mention IF it also includes the gpu. I suspect an 8c/16t cpu with igpu that performs to premium standards will not be "cheap" to make.

 

[ShintaiDK]

AtenRa: I made my comments concerning price based upon the comments of AMD's CEO that they want to up price on their premium products and that a Zen 8 core/16 thread will likely be their top premium product.

I also doubt that the Zen 8c/16t will be as small as you mention IF it also includes the gpu. I suspect an 8c/16t cpu with igpu that performs to premium standards will not be "cheap" to make.

The "highend" Zen is without IGP.

 

[SPBHM]

I remember the wait for Phenom, the wait for Bulldozer... so... I don't think it's a good idea.

we know to little about Zen, if you need a new CPU, Skylake and Haswell-e are extremely impressive right now, do you think Zen OC in 2-3 years will easily beat a 5820K OC? I wouldn't bet on that

 

[guskline]

ShintaiDK, I did not realize the Zen 8c/16t will be without igpu. That being said, the new Skylake 6700k is @122MM on a 14nm wafer. Going to 8c/16t probably won't double it, but I really wonder if a 150-180mm2 size is possible.
http://www.anandtech.com/show/9505/skylake-cpu-package-analysis

 

[StrangerGuy]

Remember the Bulldozer, Steamroller, Piledriver days? :whiste:

You forgot Phenom and all the crappy AMD/VIA chipsets before nForce 2.

Athlon XP was already losing steam to 2-3GHz Northwood P4s. The HT enabled P4s outran it. It was still a great budget OC chip, but budget nonetheless. Doesn't exactly helped AMD margins.

A64 only won when Intel went full Netburst retard with Prescott instead of with a desktop Pentium M derivative. Then AMD just sat there patting themselves on the back happily with dreams that Netburst will never die.

Phenom II was price competitive against Core 2/Nehalem only if you wanted to keep your AM2 mobo and DDR2 RAM.

And that's only for the desktop segment, Intel had an unassailable lead in portables since 2003 with Pentium M/Centrino. How many years before AMD took that higher margin and eventually much bigger segment seriously enough? So I don't get the AMD romanticism.

 

[AtenRa]

ShintaiDK, I did not realize the Zen 8c/16t will be without igpu. That being said, the new Skylake 6700k is @122MM on a 14nm wafer. Going to 8c/16t probably won't double it, but I really wonder if a 150-180mm2 size is possible.
http://www.anandtech.com/show/9505/skylake-cpu-package-analysis

8-core 16x threads, 20MB L3 cache Haswell-E HEDT CPU die at 22nm is 355mm2, that also includes quad memory controller and 40 PCI-e lanes.
The same die ported to 14nm would be close to 160mm2.
Since i dont believe ZEN will have quad Channel Memory Controller and since we already know that each ZEN Core is bellow 10mm2 at 14nm, we can roughly conclude the 8Core 16x Threads ZEN die to be close to 160-180mm2.

 

[sandorski]

Time will tell how it performs.

 

[mrmt]

ShintaiDK, I did not realize the Zen 8c/16t will be without igpu. That being said, the new Skylake 6700k is @122MM on a 14nm wafer. Going to 8c/16t probably won't double it, but I really wonder if a 150-180mm2 size is possible.
http://www.anandtech.com/show/9505/skylake-cpu-package-analysis

In fact it's usually more than double. For instance, 4C Haswell with iGPU is 177mm^2, while 8C Haswell-E is 356mm^2. Given that Samsung 14nm will be less dense than Intel 14nm, 300mm+^2 doesn't seem much off the mark.

I think a few big questions is how high this Samsung process will be able to clock Zen and how much clock they are going to lose from going with this high number of cores. Intel 8C chips lose almost 25% of the base clock compared to its 4C siblings, on top of the extra 40+% power consumption, I don't think AMD will fare much better with their 8C CPUs.

AMD might be trapping itself into a corner again, catering for small niches of the consumer market but failing badly with everyone else.

 

[Fjodor2001]

In fact it's usually more than double. For instance, 4C Haswell with iGPU is 177mm^2, while 8C Haswell-E is 356mm^2. Given that Samsung 14nm will be less dense than Intel 14nm, 300mm+^2 doesn't seem much off the mark.

Haswell is on 22 nm. Zen will be on Samsung/GF 14 nm, which is more dense than Intel 22 nm. So 300 mm² is way too much.

 

[sm625]

Anyone NOT waiting for Zen?

I'm not waiting for it. I'm trying to get to it but it just keeps slipping away from me.

 

[IndyColtsFan]

I'm pretty pessimistic about Zen. I'm hoping to be pleasantly surprised, but Sandy Bridge level IPC, when compared with Skylake, is almost as far behind as Piledriver was behind Ivy Bridge. And we don't know that 8c/16t will be affordable, only that it will exist.

I was also pretty pessimistic about Skylake, and it didn't turn out to be much.

Agreed.

I waited a couple of months extra when I built my current system in 2011, as Bulldozer (IIRC; could have the codename wrong but it was Phenom II's successor) was just around the corner and I wanted to see what it offered. Well, the rest is history and I ended up with an i7-2600K instead.

I'm itching to build a new system soon but I can't see myself waiting to see what Zen offers unless we're looking at a reveal in January at the very latest. I could probably wait that long if I knew for sure it was coming out since I probably won't be building a new system in the October/November timeframe anyway. At any rate, I am very skeptical of AMD's ability to deliver. I wish them the best and hope they DO deliver, but I won't count on it.

 

[frozentundra123456]

Yea, the OP is certainly entitled to not like Skylake and expect the world of Zen. Otherwise, not really sure the point of this thread, except to start the usual AMD/Intel bickering. There are plenty of Zen threads already, and they always end up with AMD fans fueling the hype train and Intel fans (and everyone else really) saying, lets wait and see.

 

[nismotigerwvu]

You forgot Phenom and all the crappy AMD/VIA chipsets before nForce 2.

Athlon XP was already losing steam to 2-3GHz Northwood P4s. The HT enabled P4s outran it. It was still a great budget OC chip, but budget nonetheless. Doesn't exactly helped AMD margins.

A64 only won when Intel went full Netburst retard with Prescott instead of with a desktop Pentium M derivative. Then AMD just sat there patting themselves on the back happily with dreams that Netburst will never die.

Phenom II was price competitive against Core 2/Nehalem only if you wanted to keep your AM2 mobo and DDR2 RAM.

And that's only for the desktop segment, Intel had an unassailable lead in portables since 2003 with Pentium M/Centrino. How many years before AMD took that higher margin and eventually much bigger segment seriously enough? So I don't get the AMD romanticism.

You seem to be remembering the Pentium 4 through rose tinted glasses. Northwood-C was the only time the P4 looked like a good buy, during the rest of it's run it was looking very much akin to Bulldozer. Have a look through some old articles:

http://www.anandtech.com/show/661/17
http://www.anandtech.com/show/818

The P4 did well with Quake 3 Arena, but struggled with about every other benchmark (other games included), just like how you'll find specific tasks that Bulldozer took to. AMD was super competitive and more often the outright performance leader from the launch of the original Athlon through Conroe (you could argue for the K6(2/3) era as well, but they were just competitive budget chips). It may have been a decade since they really matter, but it's odd to see so many people on these forums taking what has happened in recent history and applying it back further. Also, the Phenom II stacked up well outside of the very high end. Anand even called it "A True Return to Competition", have a look:

http://www.anandtech.com/show/2702

 

[Fjodor2001]

There are plenty of Zen threads already, and they always end up with Intel fans fueling the doom&gloom train and AMD fans (and everyone else really) saying, lets wait and see.

Fixed that for you.

 

[ShintaiDK]

Fixed that for you.

Right, because Zen is already close to Haswell/Skylake with 8 cores and 4Ghz and 95W all selling for 200-300$. And everyone questioning that is just full of doom and gloom.

We already know where you stand thanks to your own quote.

 

[TheELF]

Oh?Everybody claiming sandy level performance isn't hype?

 

[Enigmoid]

8-core 16x threads, 20MB L3 cache Haswell-E HEDT CPU die at 22nm is 355mm2, that also includes quad memory controller and 40 PCI-e lanes.
The same die ported to 14nm would be close to 160mm2.
Since i dont believe ZEN will have quad Channel Memory Controller and since we already know that each ZEN Core is bellow 10mm2 at 14nm, we can roughly conclude the 8Core 16x Threads ZEN die to be close to 160-180mm2.

Its going to be limited if it doesn't. Since the first Zen chips will be server derivatives it is quite likely 8C/16T Zen will be more than double channel, especially if the performance is there.

 

[AtenRa]

Its going to be limited if it doesn't. Since the first Zen chips will be server derivatives it is quite likely 8C/16T Zen will be more than double channel, especially if the performance is there.

Well we have to wait and see, it may have a quad memory channel after all. But that will only add another 10-12mm2 to the die size.