Discussion: AMD Ryzen

[-Fran-]

I see zen is floating on a lot of rumors. Including estimates and rumors on performance.
If the flagship Zen 8core-16thread performs 50% or more of 6900K for $500-$600, it should be a good deal.

I suppose Zen will come with the wrath stock cooler?

Well, given (recent) past history with Bulldozer (Vishera), they will price it as high as they can given the performance. If, by some miraculous reason, they are actually on par or, cows be flying, faster than Intel they would price them right next to them. The hope would be for them to price them inclined to win market share (personal hope, that is) and force Intel to lower prices. I don't know if AMD is in a good position to start a price war with Intel, TBH, but there's that.

And that is a great question and logic would tell me they will indeed ship Zen CPUs with the Wraith HSF. I have not seen nor remember anything pointing in that direction, but I would imagine they will.

Cheers!

 

[TechyInAZ]

I'd hope AMD would put Wraith coolers in their Zen platforms, as the Wraith cooler knocks the socks off of it's previous stock coolers and Intel's stock cooler.

However, my gut is telling me that AMD will most likely make a new Wraith cooler for Zen. That would be awesome if that's true.

 

[-Fran-]

Well, that raises another question... I know there was this talk at some point, but is AM4 going to be socket-lock compatible with current AM3 and FM2 sockets?

I really wouldn't want to spend more money replacing my good old TT Frio if that were the case 😛

Cheers!

 

[TechyInAZ]

Well, that raises another question... I know there was this talk at some point, but is AM4 going to be socket-lock compatible with current AM3 and FM2 sockets?

I really wouldn't want to spend more money replacing my good old TT Frio if that were the case 😛

Cheers!

No, AM4 is going to be completely independent of all older sockets. So no backwards compatibility.

 

[TehPenguin]

I get the feeling Zen will not be the one to win the market for AMD but the one which decides if their next generation will be successful.
If Zen succeeds, AMD wins attention and breaks the infamous chain of disappointing releases.

 

[juanrga]

AMD didn't dismissed it. They didn't comment on boards and costs. From the above link: "AMD stated that the "Zen" processor platform is on-track, and the company will not comment on "customer specific board-level solutions."

The company that commented on the rumor was ASMedia, but their answer to the Zen issue was weak. Their designs could have passed formal certification and still have the long-wire USB issue requiring the extra board circuity that will increase costs.

Moreover, it is evident that there is a delay. The AM4 mobos were supposed to appear time ago, much before Zen. In fact ASMedia has claimed that Zen chipset has not still started production: "The AMD Zen chipset will be in production by the end of the year and will go on sale in 2017."

http://techfrag.com/2016/06/22/amd-zen-chipset-price-issues/

Now, it would be glad to know how Zen could be released in Q4 2016 if the chipset will be for sale in 2017. I claim again that Zen has been delayed to 2017.

Bolded mine.

AMD has no issues with their chipset (according to what I have read) and it seems the problems stem from some partners and not all.

I still classify it as a rumor and salt is required, since AMD has stated otherwise.

Cost is something of another conversation.

"Is on-track" is not saying much really. It can be on-track with the issue within and the extra cost from extra circuity needed to be added to the board.

AMD is not making the chipset. They outsourced to another company: ASMedia. The rumors are about ASMedia having issues with the Zen chipset and as quoted above AMD didn't comment on "customer specific board-level solutions." Therefore they are not confirming/denying the rumor. They are not confirming if the boards will be more expensive or not.

 

[-Fran-]

No, AM4 is going to be completely independent of all older sockets. So no backwards compatibility.

Ah, ok; too bad then.

"Is on-track" is not saying much really. It can be on-track with the issue within and the extra cost from extra circuity needed to be added to the board.

AMD is not making the chipset. They outsourced to another company: ASMedia. The rumors are about ASMedia having issues with the Zen chipset and as quoted above AMD didn't comment on "customer specific board-level solutions." Therefore they are not confirming/denying the rumor. They are not confirming if the boards will be more expensive or not.

I really think you're reading it wrong, but so be it.

Cheers!

 

[TechyInAZ]

Careful...(a bit edgy there on being civil).

At least right now, Zen isn't slated until 2017. So that should give ASMedia enough time to fix any bugs their having.

 

[Cazalan]

"Is on-track" is not saying much really. It can be on-track with the issue within and the extra cost from extra circuity needed to be added to the board.

AMD is not making the chipset. They outsourced to another company: ASMedia. The rumors are about ASMedia having issues with the Zen chipset and as quoted above AMD didn't comment on "customer specific board-level solutions." Therefore they are not confirming/denying the rumor. They are not confirming if the boards will be more expensive or not.

I really think you're reading it wrong, but so be it.

Cheers!

Indeed. Both AMD and ASMedia are saying it is not the chip. Meaning it is a layout issue with a specific PCB maker.

It is a new part with new DRC (Design Rule Checks). They flubbed the first PCB spin. No big deal. Those are quick to turn.

 

[juanrga]

Both AMD and ASMedia are saying it is not the chip. Meaning it is a layout issue with a specific PCB maker.

It is a new part with new DRC (Design Rule Checks). They flubbed the first PCB spin. No big deal. Those are quick to turn.

AMD didn't confirm/deny the rumor. ASMedia denied the rumor and said its chip passed all qualifications. On the other hand Digitimes source claim the source of the problem is on the ASMedia chip. If the chipset has issues with long wires, some mobo makers could solve it by placing the USB ports the more close possible to the chipset to reduce the signal loses. But this doesn't work with frontal case USB ports, those mobo maker will have to add extra circuity on the mobo to increase USB signal, and this increases costs.

 

[jaymc]

Hey people,

Just read the tread from start to finish... (took some time by the way, and one thing is clear to me toms hardware is still where the book stops when it comes to tech). Which is great as I'm a long time fan of the site. I didn't understand it all, but there's some really knowledgeable people posting here thats for sure.

Anyway It looks like the Zen will not be an amazing chip but hopefully it gets AMD back some market share and some much needed revenue... I was also reading up about the company's debt and from what I could gather they have managed to put most of their major debt repayment off until 2019 !!

This would make it sound plausible that they are going to go all out for 2018 (zen+).... As it's actually that which will be do or die for them I think, because if they are not making money by 2019 it's gonna be a sad time for the CPU industry as intel will likely gain a monopoly when AMD goes bankrupt.

I also (excuse my ignorance here) seen this: samsung is to develope 7nm chips in their FAB in 2017 http://vrworld.com/2016/05/09/samsung-manufacture-ultra-violet-7nm-chips-2017/

"Samsung is deploying a process originally developed by a joined task force between AMD, GlobalFoundries, IBMand Samsung, thus we estimate GlobalFoundries will be the second foundry to offer EUV to mass market, followed by Intel and TSMC."

EUV is whats is required to produce 7nm and it looks like intel is gonna be left to last (great news for the CPU market) with this technology. Samsung own's it now patents an all as far as I can see.

So does this give weight to the theory that AMD could skip 10nm and go straight to 7nm... As far as I know they have sold off global foundries to IBM anyway so why couldn't they just contract samsung to produce the 7nm version of their chip... Or as it states global foudries will be second to do it... I wonder if this is what AMD has had in mind all along.???

Also can I ask, if this is the case would it put them ahead of intel... ?? Or can intel just do the same or would they need to redesign their hole chip or what exactly is their roadmap at the moment ???

It raises a lot of questions.. If this was to happen where would it put AMD ?

Regards,
Jay

 

[-Fran-]

Hey people,

Just read the tread from start to finish... (took some time by the way, and one thing is clear to me toms hardware is still where the book stops when it comes to tech). Which is great as I'm a long time fan of the site. I didn't understand it all, but there's some really knowledgeable people posting here thats for sure.

Anyway It looks like the Zen will not be an amazing chip but hopefully it gets AMD back some market share and some much needed revenue... I was also reading up about the company's debt and from what I could gather they have managed to put most of their major debt repayment off until 2019 !!

This would make it sound plausible that they are going to go all out for 2018 (zen+).... As it's actually that which will be do or die for them I think, because if they are not making money by 2019 it's gonna be a sad time for the CPU industry as intel will likely gain a monopoly when AMD goes bankrupt.

I also (excuse my ignorance here) seen this: samsung is to develope 7nm chips in their FAB in 2017 http://vrworld.com/2016/05/09/samsung-manufacture-ultra-violet-7nm-chips-2017/

So does this give weight to the theory that AMD could skip 10nm and go straight to 7nm... As far as I know they have sold off global foundries to IBM anyway so why couldn't they just contract samsung to produce the 7nm version of their chip... I wonder if this is what AMD has had in mind all along.???

Also can I ask, if this is the case would it put them ahead of intel... ?? Or can intel just do the same or would they need to redesign their hole chip or what exactly is their roadmap at the moment ???

It raises a lot of questions.. If this was to happen where would it put AMD ?

Regards,
Jay

Given how 20nm was avoided like the plague by some big players, I would imagine 10nm could be a repeat if the numbers don't show promise in early prototypes and testing. Now, I haven't personally read or found anything pointing in that direction. Plus, foundries are always "prototyping" the next node early on, so there is a long road between "prototype", "testing" and "sampling". Also, even if Samsung is ready to sample 7nm, I'd say they would want to use it with their own products first. They still have to combat Apple in the mobile sector.

I can't deny that AMD would be better of going outside of GF for their CPU "fabbing" needs, but they are in bed with GF still. AMD still has shares and some deals when they sold it off to the external fellas (it wasn't IBM IIRC, it was some middle eastern investors); the usual "contractual obligations". I'm not 100% sure they got rid of them yet.

And even if they skip 10nm, there is no guarantee they'd be ahead of Intel. The "Nanometer" race is just one side of the whole picture. Plus, I'm pretty darn sure Intel is also ahead of just "prototyping" 7nm. Don't forget Intel has been ahead of the Fabbing game for a long while now and it doesn't seem to change this time either.

What happened currently, according to information laying around, is that GF licensed/partnered the manufacturing from Samsung in 14nm and is using it for Zen. If things remain the same going forward, they could repeat for 10/7 nm when they arrive in volume.

Cheers!

 

[jaymc]

It states that "GlobalFoundries will be the second foundry to offer EUV to mass market" ... An it looks like intel will be last to get it.. (thats by design i'm sure, they wont outsource to intel until their finished with it)

EUV is whats required to produce 7nm.. "Still, there is one major technology barrier we need to pass in order to get to the next generation of semiconductors: Extreme Ultra Violet Lithography or UEV."

Samsung have purchased it patents an all from a dutch company... It also states that "Samsung Electronics is pushing to be the first manufacturer in 7nm with their own memory and logic processors."

Could it be that intel has fell behind in the FAB industry ??

"We do not know if any of the cores/cells worked during this trial run but 7nm and Extreme Ultra Violet are getting closer by the day."

Jay

 

[-Fran-]

It states that "GlobalFoundries will be the second foundry to offer EUV to mass market" ... An it looks like intel will be last.

EUV is whats required to produce 7nm.. Samsung have purchased it patents an all from a dutch company... It also states that "Samsung Electronics is pushing to be the first manufacturer in 7nm with their own memory and logic processors."

Could it be that intel has fell behind in the FAB industry ??

Jay

Just be mindful of the fine print in those announcements/publications.

Like I said, "nanometers" are just one part of the whole picture. When you have simpler chips to build it isn't so hard to tweak and actually produce something. CPUs and, more importantly, full complex CPUs (not SoCs components) are a massive difference in terms of building complexity. The die size of each chip you make increases the complexity and all the associated fabbing process.

Cheers!

 

[jaymc]

check this out http://wccftech.com/intel-abandoning-silicon-7nm/

Intel have definately fell behind, infact they have given up on 7nm....

Samsung more or less have it sorted... which means AMD and global foundries will have it before intel.

I think this claim that AMD will go straight to 7nm may have more weight to it than previously realized. It would be great to see the underdog getting one up on intel !!

Jay

 

[-Fran-]

Ugh... I don't really believe or trust most of the things reported in WTFBBQTech. Always take them with a pinch of salt, no matter how favorable or reasonable they sound. Plus, that article is from 1 year ago. Intel themselves said they would not use Silicon, but no one really know what they are doing. Or at least, I haven't read up on anything in that direction.

For all we know, they haven't found anything to go under 10nm, or they might have something in the works and ready for testing.

Cheers!

 

[Inkiad]

The intel CEO sait in 2010 they have silicon alternative prototype processors running in lab. This does not mean we will see consumer version of them anytime soon.
Intel already has issues getting satisfactory yield in 14nm. 7nm will be tough. As it is said we can go till 5nm, then we have to look for silicon alternatives.

 

[jaymc]

Found this intel are now saying they are gonna use EUV..(they werent going to but have done a uturn). They will have to licence it from samsung... which means they wont get it until last !!! after IBM and AMD...

Check this out https://www.linkedin.com/pulse/intel-wants-use-euv-lithography-7nm-process-sri-karthik-kadapa

Interesting times in the CPU industry intel have definately fell behind in FAB process...

"EUV is also expected to bring significant benefits in terms of yield and cycle time."...

Jay

 

[juanrga]

Hey people,

Just read the tread from start to finish... (took some time by the way, and one thing is clear to me toms hardware is still where the book stops when it comes to tech). Which is great as I'm a long time fan of the site. I didn't understand it all, but there's some really knowledgeable people posting here thats for sure.

Anyway It looks like the Zen will not be an amazing chip but hopefully it gets AMD back some market share and some much needed revenue... I was also reading up about the company's debt and from what I could gather they have managed to put most of their major debt repayment off until 2019 !!

This would make it sound plausible that they are going to go all out for 2018 (zen+).... As it's actually that which will be do or die for them I think, because if they are not making money by 2019 it's gonna be a sad time for the CPU industry as intel will likely gain a monopoly when AMD goes bankrupt.

I also (excuse my ignorance here) seen this: samsung is to develope 7nm chips in their FAB in 2017 http://vrworld.com/2016/05/09/samsung-manufacture-ultra-violet-7nm-chips-2017/

"Samsung is deploying a process originally developed by a joined task force between AMD, GlobalFoundries, IBMand Samsung, thus we estimate GlobalFoundries will be the second foundry to offer EUV to mass market, followed by Intel and TSMC."

EUV is whats is required to produce 7nm and it looks like intel is gonna be left to last (great news for the CPU market) with this technology. Samsung own's it now patents an all as far as I can see.

So does this give weight to the theory that AMD could skip 10nm and go straight to 7nm... As far as I know they have sold off global foundries to IBM anyway so why couldn't they just contract samsung to produce the 7nm version of their chip... Or as it states global foudries will be second to do it... I wonder if this is what AMD has had in mind all along.???

Also can I ask, if this is the case would it put them ahead of intel... ?? Or can intel just do the same or would they need to redesign their hole chip or what exactly is their roadmap at the moment ???

It raises a lot of questions.. If this was to happen where would it put AMD ?

Regards,
Jay

The vrworld article has been updated. If you check it now you can see that the quote you give has changed to:

Samsung is deploying a process originally developed by a joined task force between [strike]AMD[/strike], GlobalFoundries, [strike]IBM[/strike] and Samsung, thus we estimate GlobalFoundries will be the second foundry to offer EUV to mass market, followed by Intel and TSMC.

Apart from that, the author is making up stuff. Don't forget this author is the same guy that claimed recently that AMD had launched K12, when AMD only launched the A57-based Seattle Opteron.

Samsung purchasing machinery for the foundries in the second half of 2017 doesn't mean that will be producing commercial chips in the same date. Samsung is purchasing the machinery for R&D. Let us check Samsung foundry strategy:

Samsung Foundry 10nm will be in production by the end of 2016
Samsung is approaching 10nm differently than TSMC. Rather than doing a quick node transition from 10nm to 7nm, Samsung will focus on 10nm as a full node by building out different versions targeted at multiple markets. According to Samsung a “true” 10nm can be done using double patterning thus saving the cost of triple or quad patterning. Samsung does use triple patterning on one of the metal layers but still allows bidirectional routing which is easier to design to.

Samsung Foundry 7nm will use EUV for cost reduction
As I was told at SPIE, Samsung will use EUV for 7nm logic before using EUV for memory. An executive from ASML EUV (Dr. Hans Meiling) even presented at the Samsung event to bring everyone up to date. Given that Samsung 10nm will be a full node, delaying 7nm until 2020 (EUV ETA) should not be a problem.

I.e. 10nm for late 2016 early 2017. And 7nm somewhat between 2019 and 2020. But this is Samsung not Globalfoundries.

Moreover, didn't Lisa Su say that AMD will remain in the 14nm node for years?

Final remarks. AMD didn't sold off global foundries to IBM. First, Global Foundries is not property of AMD. Second, it is IBM who 'sold' its own foundries to Globalfoundries.

 

[juanrga]

Given how 20nm was avoided like the plague by some big players, I would imagine 10nm could be a repeat if the numbers don't show promise in early prototypes and testing.

It was avoided for consumer products because of costs. Otherwise 20nm was used in high performance (expensive) chips.

I can't deny that AMD would be better of going outside of GF for their CPU "fabbing" needs, but they are in bed with GF still. AMD still has shares and some deals when they sold it off to the external fellas (it wasn't IBM IIRC, it was some middle eastern investors); the usual "contractual obligations". I'm not 100% sure they got rid of them yet.

AMD is tied to GF up to 2025.

And even if they skip 10nm, there is no guarantee they'd be ahead of Intel. The "Nanometer" race is just one side of the whole picture. Plus, I'm pretty darn sure Intel is also ahead of just "prototyping" 7nm. Don't forget Intel has been ahead of the Fabbing game for a long while now and it doesn't seem to change this time either.

Samsung 7nm is scheduled for about same time than Intel 7nm. The only way for Glofo to be ahead of both is if relabel the 10nm node as 7nm, somewhat as the 20nm with FinFET was labeled "14nm".

What happened currently, according to information laying around, is that GF licensed/partnered the manufacturing from Samsung in 14nm and is using it for Zen. If things remain the same going forward, they could repeat for 10/7 nm when they arrive in volume.

Samsung and Globalfoundries break apart after the fiasco of the 14nm collaboration. Samsung is no more licensing nodes to GF. The GF 10nm node and ahead are in-house designs.

 

[8350rocks]

It states that "GlobalFoundries will be the second foundry to offer EUV to mass market" ... An it looks like intel will be last to get it.. (thats by design i'm sure, they wont outsource to intel until their finished with it)

EUV is whats required to produce 7nm.. "Still, there is one major technology barrier we need to pass in order to get to the next generation of semiconductors: Extreme Ultra Violet Lithography or UEV."

Samsung have purchased it patents an all from a dutch company... It also states that "Samsung Electronics is pushing to be the first manufacturer in 7nm with their own memory and logic processors."

Could it be that intel has fell behind in the FAB industry ??

"We do not know if any of the cores/cells worked during this trial run but 7nm and Extreme Ultra Violet are getting closer by the day."

Jay

Bear in mind, in some cases, such as 14nm XM-FF, you have scenarios where the foundry is using a hybrid process that uses a larger BEOL and a significantly smaller FEOL. 14nm XM is a process using 20nm BEOL and 14nm FEOL, so it is not a "true" 14nm process. Samsung's process, however, is a 14nm process to my knowledge. The FEOL and BEOL on their process is not significantly different.

7nm could potentially be a scenario where 10nm BEOL is being used, but a 7nm FEOL is being used as well. We do not have enough information to truly glean what we are looking at, and it seems that Samsung and GF are content to leave it as such until we get some actual white papers on the process node.

 

[juanrga]

Bear in mind, in some cases, such as 14nm XM-FF, you have scenarios where the foundry is using a hybrid process that uses a larger BEOL and a significantly smaller FEOL. 14nm XM is a process using 20nm BEOL and 14nm FEOL, so it is not a "true" 14nm process. Samsung's process, however, is a 14nm process to my knowledge. The FEOL and BEOL on their process is not significantly different.

14XM doesn't exist. It was a process node once in the R&D pipeline of Globalfoundries but was canceled because performed very bad. It is worth recalling that XM did mean eXtreme Mobility. After cancellation Globalfoundries got a collaboration agreement with Samsung to use Samsung 14nm tech.

Globalfoundries has licensed 14LPE and 14LPP tech from Samsung. None are true 14nm. Both are hybrid nodes with a '20nm' BEOL and '14nm' FEOL.

GloFo/Samsung "14nm":
48-nm Fin Pitch
78-nm Gate Pitch
64-nm 2D Interconnect Pitch
0.064μm² HD SRAM

Intel "14nm":
42-nm Fin Pitch
70-nm Gate Pitch
52-nm 1D Interconnect Pitch
0.0499μm² HD SRAM

 

[jaymc]

What do yis make of this....

http://wccftech.com/amd-naples-32-core-zen/

Jay

 

[TehPenguin]

What do yis make of this....

http://wccftech.com/amd-naples-32-core-zen/

Jay

Clocked at 1GHz? 😀

Hopefully this development will make more sense than just creating the marketing "power" of having more cores.

 

[juanrga]

What do yis make of this....

http://wccftech.com/amd-naples-32-core-zen/

Jay

I will copy/paste my comments from their site:

There is no "Zeppelin platform which will include a 32 Zen Core/Vega based APU as well." The HPC/server APU has only 16 Zen cores.

64 weaker threads is not any advantage. The real advantage will be the 8-channel DDR4 memory subsystem.

I started posting that the AMD would use a multidie approach with 8-core dies much before Fudzilla 'exclusives'. However, I said that AMD would use a quad-die configuration on interposer, instead a traditional MCM package.

http://semiaccurate.com/forums/showpost.php?p=233677&postcount=422

I don't buy Fudzilla's claim that Naples will come in 4 different variants: 2 cores, 4 cores, 16 cores and 32 cores. Since the unit is an 8-core die, a 2-core chip would mean that 75% of the die is damaged or fussed-off. Yields cannot be that bad. I would expect 12-core and 24-core versions as well.