Discussion: AMD Ryzen

[jimmysmitty]

Do you have any idea what it takes and costs to research and develop a new process tech? I highly doubt it. It is not cheap and it is not going to get cheaper, consider that now companies have to look at ideas outside of Silicon with takes time to research and test.

That said, Intels R&D is not just for their process tech. Intel helps in the development of other tech such as Ethernet standards, USB standards (they just got TB3 into USB Type C ) and various others. They are heavily invested in the PC world, not just CPUs.

As mentioned, Global Foundries 14nm is from Samsung, not quite their own. Have you looked at how much Samsung spends on R&D? 13.4 Billion, 3 billion more than Intel, in 2014:

http://www.androidauthority.com/samsung-biggest-rd-spender-571711/

Either way AMD has very low R&D due to many things, one is a lack of a competitive CPU in the server market or desktop market and the change in GPUs although their last truly good GPU was the HD7970 with nothing quite trumping NVidia like it since.

They also made some good choices at the wrong time and had a few trips with K10 and Bulldozer. It might change with Zen since all their R&D can now be focused on their uArch but that means they have to wisely pick a good process tech and everything we have seen so far shows that Samsungs 14nm is inferior to TSMCs 16nm which is still not as up to par as Intels 14nm.

The next few months should be quite telling.

We can't know how much it costs to fabricate anything from Intel or AMD, because that is a trade secret. Or at least, you can strip most of the costs, until a point where you can't. Keep that in mind when calling someone out on something like that.

What we know, from third party data, is Intel's R&D is huge and several times more than AMDs. We do not know how they split it, nor how much of that R&D cost is translated as manufacturing cost via accounting magic.

For all we know, Intel is making a 99% profit on each CPU sold, or just a 1%. Anyone with some hard data to put some boundaries on that?

EDIT: Nice timing, manleysteele! So it's ~50%?

The link says that the data center group did near 50% of profit. That doesn't mean that the whole company did. The data center group is not the one that invest in foundry R&D for instance, and groups as mobile division had huge loses. Also to put things in perspective. AMD announced a 30% profit the last quarter.

The thing that struck me about the article was the 5 year lead time for producing a new design. Even though much of the design is, in fact not new, the pressure to be right must be intense. The die shrink from Haswell to Broadwell (20 nm to 14 nm) provides an interesting case. The repeated delays for Broadwell make more sense now. From all appearances, the move to Cannonlake (10 nm) is already experiencing similar problems. This also points to why AMD/GoFlo is willing to make the jump to 7 nm.

The huge cost (both money and time) of development is the reason why Globalfoundries canceled its 14XM node and licensed the 14LPP from Samsung. It is also the reason why Samsung 14LPP node used for Zen is not a pure 14nm node but a 20nm base with 14nm FINFETs.

That recently announced "7nm" is not 7nm. A jump to 7nm would increase density by 4x, but Globalfoundries claims the density will be only about 2x higher over their current node. This means that they are relabeling the 10nm node as 7nm, which implies that Globalfoundries "7nm" will be inferior to Intel "10nm" and surely also inferior to TSMC "10nm".

That is interesting.

Maybe they are doing it to try and push it out and claim an industry first? Would surely boost stock and gain some customers until Intel pushes out a true 7nm, which everything points to it not even being silicon but rather Indium Gallium Arsenide.

 

[-Fran-]

*Casually notes that said encoding/decoding is performed orders of magnitude faster using a GPU*

Please, gamerk; I feel offended by that, haha. Of course I'm not implying using the CPU is better in terms of "efficiency", more over since GPUs don't really use their computing muscle to decode 90% of content, they use the fixed pipelines, from the way decoding stages work, they don't really see much benefit in terms of quality processing.

The only exception is nVidia + CUDA, since there are some codecs that support CUDA decoding (namely madVR and LAV). Also, the integrated decoder in VLC, IIRC, also supports it. Well, even then, the CPU still takes most of the decoding processing.

All in all, my point was moving towards the more demanding streaming becomes, the more CPU resources it will consume anyway. Be it because 360° video, 4K content or new encoding matrix/codec being used that is not supported in the fixed pipelines.

Decoding is a piece of cake, my dirt old iPod can do that fine. I think it's pretty negligible especially with GPU accelerated decoding. No one is going to want a CPU with more cores because they watch Youtube videos.

Yes, fixed pipelines. Try getting H265 decoding in an iPod. Even H264 Hi10 has issues with them. I still remember how the Core2Duo generation had issues keeping up with H264 Hi8 at 1080p whereas the Core2Quads didn't have much issues. Fastforward to H264 Hi10 and 4K, current low tier i5s have issues.

This is a screenshot from... 4 years ago? XD

Spoiler

That is using LAV, with full CUDA support and my i7 @~4.5Ghz. Notice in Process explorer the GPU usage was ~35% and just look at the CPU graph.

Cheers!

 

[spat55]

Well quad core should go and die, we have had them for the past 8 years on mainstream time to upgrade them. Software needs to take advantage of more cores and more manufactures need to give me a reason to upgrade.

 

[juanrga]

I'm not hoping for AMD to best Intel I'm just hoping for AMD to kick Intel into making mainstream 6 cores maybe 8 cores. I want my next CPU to be a i7 unlocked K chip with 6 cores and 12 threads minimum for around £300-£350, so same price as a 6700k.

(1) How will AMD force Intel to do that?

(2) AMD itself is giving quad-core Zen APUs for mainstream and octo-core Zen CPUs for HEDT.

Well quad core should go and die, we have had them for the past 8 years on mainstream time to upgrade them. Software needs to take advantage of more cores and more manufactures need to give me a reason to upgrade.

And how will software take advantage of more cores? Easy to parallelize code will benefit from more cores, sure, but what happens with rest of code?

We have had the ability to make moar cores for decades. Some companies have been shipping 256-core CPUs, but they are aimed at very specific parallel workloads (accelerating multimedia codecs and so), not as general CPU for a windows/linux desktop machine.

 

[juanrga]

Spoiler

Do you have any idea what it takes and costs to research and develop a new process tech? I highly doubt it. It is not cheap and it is not going to get cheaper, consider that now companies have to look at ideas outside of Silicon with takes time to research and test.

That said, Intels R&D is not just for their process tech. Intel helps in the development of other tech such as Ethernet standards, USB standards (they just got TB3 into USB Type C ) and various others. They are heavily invested in the PC world, not just CPUs.

As mentioned, Global Foundries 14nm is from Samsung, not quite their own. Have you looked at how much Samsung spends on R&D? 13.4 Billion, 3 billion more than Intel, in 2014:

http://www.androidauthority.com/samsung-biggest-rd-spender-571711/

Either way AMD has very low R&D due to many things, one is a lack of a competitive CPU in the server market or desktop market and the change in GPUs although their last truly good GPU was the HD7970 with nothing quite trumping NVidia like it since.

They also made some good choices at the wrong time and had a few trips with K10 and Bulldozer. It might change with Zen since all their R&D can now be focused on their uArch but that means they have to wisely pick a good process tech and everything we have seen so far shows that Samsungs 14nm is inferior to TSMCs 16nm which is still not as up to par as Intels 14nm.

The next few months should be quite telling.

We can't know how much it costs to fabricate anything from Intel or AMD, because that is a trade secret. Or at least, you can strip most of the costs, until a point where you can't. Keep that in mind when calling someone out on something like that.

What we know, from third party data, is Intel's R&D is huge and several times more than AMDs. We do not know how they split it, nor how much of that R&D cost is translated as manufacturing cost via accounting magic.

For all we know, Intel is making a 99% profit on each CPU sold, or just a 1%. Anyone with some hard data to put some boundaries on that?

EDIT: Nice timing, manleysteele! So it's ~50%?

The link says that the data center group did near 50% of profit. That doesn't mean that the whole company did. The data center group is not the one that invest in foundry R&D for instance, and groups as mobile division had huge loses. Also to put things in perspective. AMD announced a 30% profit the last quarter.

The thing that struck me about the article was the 5 year lead time for producing a new design. Even though much of the design is, in fact not new, the pressure to be right must be intense. The die shrink from Haswell to Broadwell (20 nm to 14 nm) provides an interesting case. The repeated delays for Broadwell make more sense now. From all appearances, the move to Cannonlake (10 nm) is already experiencing similar problems. This also points to why AMD/GoFlo is willing to make the jump to 7 nm.

The huge cost (both money and time) of development is the reason why Globalfoundries canceled its 14XM node and licensed the 14LPP from Samsung. It is also the reason why Samsung 14LPP node used for Zen is not a pure 14nm node but a 20nm base with 14nm FINFETs.

That recently announced "7nm" is not 7nm. A jump to 7nm would increase density by 4x, but Globalfoundries claims the density will be only about 2x higher over their current node. This means that they are relabeling the 10nm node as 7nm, which implies that Globalfoundries "7nm" will be inferior to Intel "10nm" and surely also inferior to TSMC "10nm".

I had suspicions concerning that, but I couldn't substantiate them. I had assumed the move to 7nm was made as a result of buying the IBM foundries.

Globalfoundries announced the 'acquisition' of IBM foundries July 2015. Months latter the chief executive officer of GlobalFoundries, announced both 10nm and 7nm being developed by a combined team of IBM and GF engineers. They now announcing the cancellation of the 10nm and going directly for the 7nm node implies the reason is another.

 

[manleysteele]

http://arstechnica.com/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

I was thinking of this announcement.

 

[jimmysmitty]

http://arstechnica.com/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

I was thinking of this announcement.

That was an SRAM chip I think which is just a basic chip. It may or may not be valuable since SRAM is the easiest product to FAB on a new process.

It is also still silicon and I think Intel is correct to move to a non silicon based material for 7nm.

The one thing I keep in mind though is that if Intel has issues with a process tech so will most others. Most issues that arise now are just common place due to the difficulty of shrinking silicon.

 

[juanrga]

Do you have any idea what it takes and costs to research and develop a new process tech? I highly doubt it. It is not cheap and it is not going to get cheaper, consider that now companies have to look at ideas outside of Silicon with takes time to research and test.

That said, Intels R&D is not just for their process tech. Intel helps in the development of other tech such as Ethernet standards, USB standards (they just got TB3 into USB Type C ) and various others. They are heavily invested in the PC world, not just CPUs.

As mentioned, Global Foundries 14nm is from Samsung, not quite their own. Have you looked at how much Samsung spends on R&D? 13.4 Billion, 3 billion more than Intel, in 2014:

http://www.androidauthority.com/samsung-biggest-rd-spender-571711/

Either way AMD has very low R&D due to many things, one is a lack of a competitive CPU in the server market or desktop market and the change in GPUs although their last truly good GPU was the HD7970 with nothing quite trumping NVidia like it since.

They also made some good choices at the wrong time and had a few trips with K10 and Bulldozer. It might change with Zen since all their R&D can now be focused on their uArch but that means they have to wisely pick a good process tech and everything we have seen so far shows that Samsungs 14nm is inferior to TSMCs 16nm which is still not as up to par as Intels 14nm.

The next few months should be quite telling.

We can't know how much it costs to fabricate anything from Intel or AMD, because that is a trade secret. Or at least, you can strip most of the costs, until a point where you can't. Keep that in mind when calling someone out on something like that.

What we know, from third party data, is Intel's R&D is huge and several times more than AMDs. We do not know how they split it, nor how much of that R&D cost is translated as manufacturing cost via accounting magic.

For all we know, Intel is making a 99% profit on each CPU sold, or just a 1%. Anyone with some hard data to put some boundaries on that?

EDIT: Nice timing, manleysteele! So it's ~50%?

The link says that the data center group did near 50% of profit. That doesn't mean that the whole company did. The data center group is not the one that invest in foundry R&D for instance, and groups as mobile division had huge loses. Also to put things in perspective. AMD announced a 30% profit the last quarter.

The thing that struck me about the article was the 5 year lead time for producing a new design. Even though much of the design is, in fact not new, the pressure to be right must be intense. The die shrink from Haswell to Broadwell (20 nm to 14 nm) provides an interesting case. The repeated delays for Broadwell make more sense now. From all appearances, the move to Cannonlake (10 nm) is already experiencing similar problems. This also points to why AMD/GoFlo is willing to make the jump to 7 nm.

The huge cost (both money and time) of development is the reason why Globalfoundries canceled its 14XM node and licensed the 14LPP from Samsung. It is also the reason why Samsung 14LPP node used for Zen is not a pure 14nm node but a 20nm base with 14nm FINFETs.

That recently announced "7nm" is not 7nm. A jump to 7nm would increase density by 4x, but Globalfoundries claims the density will be only about 2x higher over their current node. This means that they are relabeling the 10nm node as 7nm, which implies that Globalfoundries "7nm" will be inferior to Intel "10nm" and surely also inferior to TSMC "10nm".

That is interesting.

Maybe they are doing it to try and push it out and claim an industry first? Would surely boost stock and gain some customers until Intel pushes out a true 7nm, which everything points to it not even being silicon but rather Indium Gallium Arsenide.

Globalfoundries claims risk production in 2018, which must mean volume production in 2019. Intel, TSMC, and Samsung 10nm nodes are expected for 2017. I only can guess Glofo is labeling its node as "7nm" to not look outdated compared to rest of foundries.

 

[juanrga]

http://arstechnica.com/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

I was thinking of this announcement.

What has that to do with Globalfoundries initially announcing the common Globalfoundries/IBM engineers team developing 10nm and 7nm nodes, and now announcing that the 10nm node was canceled?

 

[elbert]

AMD's 7nm code name starship is a 48 core 96 thread replacement for Naples. This CPU most likely will come late 2019 to early 2020. I think Zen+ core will be based on revised Zen still 14nm lol. My guess is late 2H 2017.

 

[manleysteele]

http://arstechnica.com/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

I was thinking of this announcement.

What has that to do with Globalfoundries initially announcing the common Globalfoundries/IBM engineers team developing 10nm and 7nm nodes, and now announcing that the 10nm node was canceled?

It's working silicon on 7nm. They have to try something to get back in the game.

 

[juanrga]

AMD's 7nm code name starship is a 48 core 96 thread replacement for Naples. This CPU most likely will come late 2019 to early 2020. I think Zen+ core will be based on revised Zen still 14nm lol. My guess is late 2H 2017.

Zen launch on early 2017. With APUs coming in late 2017, and HPC APU in 2018. No way AMD will release Zen+ at same time by economic reasons.

Now physical reasons. According to AMD Zen+ increases IPC by about 10%. This implies about 25% bigger core. We also know that AMD increases core count by 50%. Combining both this means about 1.87 bigger dies, which would imply a nonsensical 337W TDP on current 14nm. Thus Zen+ and Starship require a die shrink. Precisely Glofo "7nm" increases density by about 2x (canceling the 1.87x die increase). Therefore it is likely that Zen+ is a 2019 product made on the new node.

My guess:

Naples --> 32-core Zen; 180W; 14nm; 2017.
Starship --> 48-core Zen; ~160W; 7nm; 2019.

 

[elbert]

AMD's 7nm code name starship is a 48 core 96 thread replacement for Naples. This CPU most likely will come late 2019 to early 2020. I think Zen+ core will be based on revised Zen still 14nm lol. My guess is late 2H 2017.

Zen launch on early 2017. With APUs coming in late 2017, and HPC APU in 2018. No way AMD will release Zen+ at same time by economic reasons.

Now physical reasons. According to AMD Zen+ increases IPC by about 10%. This implies about 25% bigger core. We also know that AMD increases core count by 50%. Combining both this means about 1.87 bigger dies, which would imply a nonsensical 337W TDP on current 14nm. Thus Zen+ and Starship require a die shrink. Precisely Glofo "7nm" increases density by about 2x (canceling the 1.87x die increase). Therefore it is likely that Zen+ is a 2019 product made on the new node.

My guess:

Naples --> 32-core Zen; 180W; 14nm; 2017.
Starship --> 48-core Zen; ~160W; 7nm; 2019.

The 1.87 times is reasonable larger than the 8 core Zen. The nonsensical 337W TDP I can only guess is a belief its a Naples CPU. When it should be a 8, 12, and 16 core Zen+ on AM4+. Understand that many Intel broadwell CPU is larger than Zen. Much like Intels broadwell the Zen+ coes should have a reduced clockrate beyond the 8 core version.
Understand starship isn't a nm reduced Naples thus the Zen and zeppelin will need a major change in 4 years. IE AMD will have 3 years of revisions on the 14nm. What your suggesting is an AMD can only compete in 2017 and 2020.

 

[juanrga]

No. The nonsensical 337W is my estimation of the TDP of hypothetical Starship CPU on 14nm. This nonsensical value combined with the other arguments given is the reason why I am rather sure that Starship is a 7nm CPU.

There is no leak or confirmation for "8, 12, and 16 core Zen+ on AM4+". In fact data suggests otherwise; the 48-core leak suggests that the 7nm Zen+ die for the "AM4+" will have 12-cores (i.e. two complexes of six-core each). Therefore AMD will probably produce 6-core and 12-core Zen+ on AM4+, just like is producing now 4-core and 8-core Zen on AM4.

Edit: My computation/explanation of why Starship has to be a 7nm chip wasn't needed, it was leaked it is 7nm

http://www.eteknix.com/amd-working-48-core-7nm-processor-codenamed-starship/

 

[juanrga]

http://arstechnica.com/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

I was thinking of this announcement.

What has that to do with Globalfoundries initially announcing the common Globalfoundries/IBM engineers team developing 10nm and 7nm nodes, and now announcing that the 10nm node was canceled?

It's working silicon on 7nm. They have to try something to get back in the game.

That means nothing for their recent cancellation of the 10nm node. And they aren't "getting back in the game". As explained above their '7nm' node is not a real 7nm node and will be similar or even inferior to the "10nm" nodes will be released before by other foundries.

 

[elbert]

No. The nonsensical 337W is my estimation of the TDP of hypothetical Starship CPU on 14nm. This nonsensical value combined with the other arguments given is the reason why I am rather sure that Starship is a 7nm CPU.

There is no leak or confirmation for "8, 12, and 16 core Zen+ on AM4+". In fact data suggests otherwise; the 48-core leak suggests that the 7nm Zen+ die for the "AM4+" will have 12-cores (i.e. two complexes of six-core each). Therefore AMD will probably produce 6-core and 12-core Zen+ on AM4+, just like is producing now 4-core and 8-core Zen on AM4.

Edit: My computation/explanation of why Starship has to be a 7nm chip wasn't needed, it was leaked it is 7nm

http://www.eteknix.com/amd-working-48-core-7nm-processor-codenamed-starship/

So I was correct in pointing out your nonsensical 337W wasn't Zen+. Now there has been likes of SP4R2 with drop in 4 and 8 core compatible plus 12 and 16 core additions in 2H 2017 leaks. One example and you can find many more.
http://www.fudzilla.com/news/processors/41164-amd-2017-opteron-comes-in-three-sockets

 

[manleysteele]

http://arstechnica.com/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

I was thinking of this announcement.

What has that to do with Globalfoundries initially announcing the common Globalfoundries/IBM engineers team developing 10nm and 7nm nodes, and now announcing that the 10nm node was canceled?

It's working silicon on 7nm. They have to try something to get back in the game.

That means nothing for their recent cancellation of the 10nm node. And they aren't "getting back in the game". As explained above their '7nm' node is not a real 7nm node and will be similar or even inferior to the "10nm" nodes will be released before by other foundries.

Can you substantiate that argument?

 

[jdwii]

No. The nonsensical 337W is my estimation of the TDP of hypothetical Starship CPU on 14nm. This nonsensical value combined with the other arguments given is the reason why I am rather sure that Starship is a 7nm CPU.

There is no leak or confirmation for "8, 12, and 16 core Zen+ on AM4+". In fact data suggests otherwise; the 48-core leak suggests that the 7nm Zen+ die for the "AM4+" will have 12-cores (i.e. two complexes of six-core each). Therefore AMD will probably produce 6-core and 12-core Zen+ on AM4+, just like is producing now 4-core and 8-core Zen on AM4.

Edit: My computation/explanation of why Starship has to be a 7nm chip wasn't needed, it was leaked it is 7nm

http://www.eteknix.com/amd-working-48-core-7nm-processor-codenamed-starship/

So I was correct in pointing out your nonsensical 337W wasn't Zen+. Now there has been likes of SP4R2 with drop in 4 and 8 core compatible plus 12 and 16 core additions in 2H 2017 leaks. One example and you can find many more.
http://www.fudzilla.com/news/processors/41164-amd-2017-opteron-comes-in-three-sockets

No not really Juan was saying that if Zen+ was to be released in 2017 it would be on 14nm as full production 7nm probably won't be out until 2019 or later knowing Global foundries and like their current 14nm being inferior to TSMC 16nm i wouldn't call it to far fetched to assume their 7nm will be inferior to their competition. Most know here that 7nm, 10nm, 28nm is more and more about marketing then actual lengths or sizes.

https://www.youtube.com/watch?v=ktHciPVAZaU

 

[elbert]

Yes but Zen+ core was to release 2h 2017. Thus logic would state this must be the 12 and 16 core. Has nothing to do with starship nearly 4 years later. Zen+ core is just that. More core of zen in the workstation/enthusiast space. Look at it this way. What is AMD going to offer against Intel's 10, 12, 14 and 16 core broadwell-e/p. No way will AMD just let Intel have this. AMD can't wait 4 years to launch there next or even the next after the next and compete.

 

[jdwii]

Yes but Zen+ core was to release 2h 2017. Thus logic would state this must be the 12 and 16 core. Has nothing to do with starship nearly 4 years later. Zen+ core is just that. More core of zen in the workstation/enthusiast space. Look at it this way. What is AMD going to offer against Intel's 10, 12, 14 and 16 core broadwell-e/p. No way will AMD just let Intel have this. AMD can't wait 4 years to launch there next or even the next after the next and compete.

Amd clearly didn't make Zen to compete head on in terms of performance with Intel. If so it would be a 4+4 ALU/AGU design if so Zen would have 60% more IPC, If so they would be making a 10 core out of the gate. Amd can't do this they know this they have to make a product that can be competitive somehow. 14nm Global foundarys can not handle a 12 core Zen CPU(that has a clock rate competitive enough) without crazy power consumption. Actually currently even a 3.2Ghz clock speed is a bit much for their 8 core on 14nm with a 95 watt target.

Its not so much Amd can't design it its more Global foundry can't make it, if somehow Amd had the same fabrication process as Intel things would be different. About the only thing i actually agree with you on is it is possible for Zen+ on 14nm but with their APU businesses only where its just a 4 core. However efficacy wise Juan would be right a 10% improve in IPC doesn't mean a 10% increase in die size it means much more.

Plus even with ignoring technology for a second and thinking about marketing Amd wouldn't be in the best position if they released Zen+ so quickly after Zen you are talking less then 3 terms that would be to soon(edit heck man just look how well Intel was able to do such a thing ha ha). Amd needs contracts and deals telling HP and Dell and so on to support something completely different in just 9 months or so is crazy and not good business.

 

[-Fran-]

Plus even with ignoring technology for a second and thinking about marketing Amd wouldn't be in the best position if they released Zen+ so quickly after Zen you are talking less then 3 terms that would be to soon(edit heck man just look how well Intel was able to do such a thing ha ha). Amd needs contracts and deals telling HP and Dell and so on to support something completely different in just 9 months or so is crazy and not good business.

Regarding this bit, jdwii.

I think OEMs would like it to be the other way around. In my personal view of things, OEM would like Intel and AMD to do refreshes *faster* than 2 years. I know this might be a rough parallel, but take a look at the smartphone refresh-cycles. They try to keep them to 1 year. That is because they might have their optimal cycles with 1 year, due to fiscal year accountancy. So, if AMD can get Zen+ in 2H 2017, then OEMs would be very happy I'd say. EDIT: Forgot to say that i base this train of thought on AMD using AM4 as a single socket based platform for all Zen (and maybe Zen+?) products.

That being said, I have just a little confidence in GF getting their act together this year. Just a little.

Cheers!

 

[juanrga]

No. The nonsensical 337W is my estimation of the TDP of hypothetical Starship CPU on 14nm. This nonsensical value combined with the other arguments given is the reason why I am rather sure that Starship is a 7nm CPU.

There is no leak or confirmation for "8, 12, and 16 core Zen+ on AM4+". In fact data suggests otherwise; the 48-core leak suggests that the 7nm Zen+ die for the "AM4+" will have 12-cores (i.e. two complexes of six-core each). Therefore AMD will probably produce 6-core and 12-core Zen+ on AM4+, just like is producing now 4-core and 8-core Zen on AM4.

Edit: My computation/explanation of why Starship has to be a 7nm chip wasn't needed, it was leaked it is 7nm

http://www.eteknix.com/amd-working-48-core-7nm-processor-codenamed-starship/

So I was correct in pointing out your nonsensical 337W wasn't Zen+.

No, the nonsensical 337W was for Zen+.

Now there has been likes of SP4R2 with drop in 4 and 8 core compatible plus 12 and 16 core additions in 2H 2017 leaks. One example and you can find many more.
http://www.fudzilla.com/news/processors/41164-amd-2017-opteron-comes-in-three-sockets

Fuad has to be taken with a grain of salt. He has some legit information but then he fills on the holes of that information with his own stuff, which most of time is nonsensical, because he doesn't understand tech. An example? He got the right the leak about MCM packages for the Opteron (multi-die packages), but then the adds his own (nonsensical) stuff about three-die Opterons, which he refers to as "three Zeppelin clusters", because he doesn't even know the correct term is "dies".

I think I have also identified the source of your confusion. When he talks about the 8, 12, and 16-core Opteron being socket compatible, you incorrectly believed that those are single-die and that is the reason why you insist on that AMD will launch 12-core and 16-core die Zen for enthusiasts, when the maximum core count for AM4 (even confirmed by AMD) is 8-core. Ok, let me clarify things a bit.

AMD is releasing three sockets for Zen: AM4, SP4, and SP3. Only AM4 is for customers (both mainstream and enthusiasts). The other two sockets are for servers (therein the "S" in the socket name).

AM4 is a single-die socket. The Zeppelin die is 8-core. AMD will release 8-core and 4-core Summit Ridge CPUs for AM4. With the 4-core CPU using a Zeppelin die with one-complex disabled.

SP4 is a dual-die socket (MCM2). AMD will release 16, 12, and 8-core Snowy Owl CPUs for SP4. The 16-core Opteron is made of two 8-core dies. The 8-core Opteron is made of two 4-core dies. Here relies your confusion. You believed this 8-core Opteron was single die. The 12-core Opteron is made of one 8-core die plus one 4-core die.

SP3 is a quad-die socket (MCM4). AMD will release 32, 24, and 16-core Naples CPUs for SP3. The 32-core Opteron is made of four 8-core dies. The 16-core Opteron is made of four 4-core dies. The 24-core Opteron is made of two 8-core dies plus two 4-core dies.

 

[juanrga]

http://arstechnica.com/gadgets/2015/07/ibm-unveils-industrys-first-7nm-chip-moving-beyond-silicon/

I was thinking of this announcement.

What has that to do with Globalfoundries initially announcing the common Globalfoundries/IBM engineers team developing 10nm and 7nm nodes, and now announcing that the 10nm node was canceled?

It's working silicon on 7nm. They have to try something to get back in the game.

That means nothing for their recent cancellation of the 10nm node. And they aren't "getting back in the game". As explained above their '7nm' node is not a real 7nm node and will be similar or even inferior to the "10nm" nodes will be released before by other foundries.

Can you substantiate that argument?

Density (Intel 14nm) > Density (TSMC 16nm) ~ Density (Globalfoundries 14nm) ........................................................... [1]

Intel 10nm is a full node shrink,

Density (Intel 10nm) ~2x Density (Intel 14nm) ........................................................ [2]

TSMC has stated their 10nm node will provide about 100% higher density than 16nm.

Density (TSMC 10nm) ~2x Density (TSMC 16nm) ........................................................ [3]

Glofo has confirmed their 7nm node will provide about 2x the density of their 14nm,

Density (Globalfoundries 7nm) ~2x Density (Globalfoundries 14nm) ........................................................ [4]

Substituting [2], [3], and [4] on [1] yields

Density (Intel 10nm) > Density (TSMC 10nm) ~ Density (Globalfoundries 7nm)

This proves Globalfoundries "7nm" is not a true 7nm. Moreover their 7nm is comming late to the party. E.g., TSMC 10nm will be ready next year, whereas Globalfoundries expects risk production of their "7nm" node for 2018, which suggests volume production for 2019.

 

[juanrga]

Yes but Zen+ core was to release 2h 2017. Thus logic would state this must be the 12 and 16 core. Has nothing to do with starship nearly 4 years later. Zen+ core is just that. More core of zen in the workstation/enthusiast space. Look at it this way. What is AMD going to offer against Intel's 10, 12, 14 and 16 core broadwell-e/p. No way will AMD just let Intel have this. AMD can't wait 4 years to launch there next or even the next after the next and compete.

As explained a hundred of times. Zen is 2017--2018 product. Zen+ is 2019 product.

As explained a hundred of times, there is no 12 and 16-core Zen die. The only Zen die is 8-core. Any Zen product is obtained from combining one or more Zen dies, but the multi-die products (Naples and Snowy Owl) are SERVER CPUs.

 

[juanrga]

Plus even with ignoring technology for a second and thinking about marketing Amd wouldn't be in the best position if they released Zen+ so quickly after Zen you are talking less then 3 terms that would be to soon(edit heck man just look how well Intel was able to do such a thing ha ha). Amd needs contracts and deals telling HP and Dell and so on to support something completely different in just 9 months or so is crazy and not good business.

Regarding this bit, jdwii.

I think OEMs would like it to be the other way around. In my personal view of things, OEM would like Intel and AMD to do refreshes *faster* than 2 years. I know this might be a rough parallel, but take a look at the smartphone refresh-cycles. They try to keep them to 1 year. That is because they might have their optimal cycles with 1 year, due to fiscal year accountancy.

This year Digitimes published a report about both Intel and AMD delaying chips due to OEMs cleaning existences of older chips. Also desktop computers aren't phones. Refreshes cycles follow the trend

mobile > desktop > server