AMD Zen - Key Dates and Information

[R0H1T]

Explain what this would do exactly?

Did you miss the context wrt the post I quoted? Selling 8c/16t zen @350 would be stupid IMO, even if it were to severely dampen the sales of Kabylake. It won't though, because Intel has more $ to spend their way out of a (possible) messy situation besides there's the IGP so techinally they're not competing in the same market space or segment.

It stands to reason then that Zen should be priced appropriately in lieu of its competition i.e. HEDT so IMO 500$ is where it should debut & no less, the black or extreme edition could be 100$ more.

 

[imported_jjj]

You think Intel can't throw $ like the last time, when they were lagging AMD & most OEM's had to budge or risk getting wiped out? They did it a decade ago, they did it with Atom, trying to cram it in mobile/tablets & they'll likely do it again if push comes to shove. Don't you think AMD should make all the $ while they can, or just give away cheap octacores hoping it'd wipe Kabylake out with a single (master) stroke?

Ok so this Summit Ridge platform is CPU only and by being so, it's not really OEM. That's one of the main points and why AMD can gain share easier.
Yes i think AMD should make the most out of it and that's the main point. Achieving that is not done by shipping little volumes at insane margins but by shipping very high volumes at high margins.
We could make a parallel with the GPU market, according to JPR discrete desktop in Q3 was 13.04 million units,out of that, only some 1.5 million units were at above 250$(defined by them as enthusiast). Volumes decline very fast with price.

As for what Intel can do, Skylake 4 cores+GPU is 120+mm2, the 10 cores die is twice that, Intel doesn't have a slim many core die for consumer.They can make one but would take a bit of time and that's AMD's window of opportunity. It's a much bigger opportunity in this market than the times when AMD just had a faster core,now they have 2x the cores.
I'll say this one more time, An 8 core is just an APU where you remove the GPU and replace it with 4 extra cores. It wouldn't be cheap at 349$ by any means, it's the same math as a 4 cores APU. 349$ is where you pay a substantial premium for high end.'Look at an Intel die shot and note that the 4 cores+cache are smaller than the GPU.

It's been almost 10 years since 4 cores have become "mainstream ". http://www.anandtech.com/show/2303 If we would apply Moore's Law to that, by now we would have 128 cores LOL
Instead of that, we got bigger cores (in terms of transistor count),higher clocks but not by much as those don't scale anymore, more chipset functions on the SoC ,a pointless GPU. Ofc Moore's Law has been dead for quite a while and nobody gets 2x the transistors every 2 years at same cost. Still, there has been plenty of scaling from 65nm 10 years ago for the Core 2 Q6600 to today's 14nm. Sadly, AMD being out of it, has enabled Intel to just stay with 4 cores.
If you would compare the Core 2 Q6600 with Kaby Lake at same clocks and then look at core area and cost, it's not even funny.
A Kentsfield die (286 mm2)

 

[R0H1T]

Ok so this Summit Ridge platform is CPU only and by being so, it's not really OEM. That's one of the main points and why AMD can gain share easier.
Yes i think AMD should make the most out of it and that's the main point. Achieving that is not done by shipping little volumes at insane margins but by shipping very high volumes at high margins.
We could make a parallel with the GPU market, according to JPR discrete desktop in Q3 was 13.04 million units,out of that, only some 1.5 million units were at above 250$(defined by them as enthusiast). Volumes decline very fast with price.

The enthusiast or workstation class (hexa or octacore) CPU's haven't moved much in price but they ought to have had a steady growth in volume over the years, at least not negative. But then again you'd ask yourself do you really need an octa core CPU just because it's 350$ & the answer to that is a resounding NO, the vast majority of people don't need more than 4c/8t IMO, that the IGP in their system is just a waste of die space is another matter entirely. So by keeping octa core CPU prices relatively high, AMD could balance out their lower end offerings by pricing them appropriately but most of this depends on how Intel reacts to the upcoming competition & (dis)incentives they give their partners vis-a-vis AMD.

As for what Intel can do, Skylake 4 cores+GPU is 120+mm2, the 10 cores die is twice that, Intel doesn't have a slim many core die for consumer.They can make one but would take a bit of time and that's AMD's window of opportunity. It's a much bigger opportunity in this market than the times when AMD just had a faster core,now they have 2x the cores.
I'll say this one more time, An 8 core is just an APU where you remove the GPU and replace it with 4 extra cores. It wouldn't be cheap at 349$ by any means, it's the same math as a 4 cores APU.'Look at an Intel die shot and note that the 4 cores+cache are smaller than the GPU.

I don't see Intel making a separate die for HEDT just to counter AMD. They'll just make the next mainstream CPU a hexacore, Skylake's competition should be Zen APU & with a better IGP they might make massive inroads in the desktop & notebook space, provided Intel doesn't restart their contra revenue shens with a different name.

It's been almost 10 years since 4 cores have become "mainstream ". If we would apply Moore's Law to that, by now we would have 128 cores LOL
Instead of that, we got bigger cores (in terms of transistor count),higher clocks but not by much as those don't scale anymore, more chipset functions on the SoC ,a pointless GPU. Ofc Moore's Law has been dead for quite a while and nobody gets 2x the transistors every 2 years at same cost. Still, there has been plenty of scaling from 65nm 10 years ago for the Core 2 Q6600 to today's 14nm. Sadly, AMD being out of it, has enabled Intel to just stay with 4 cores.
If you would compare the Core 2 Q6600 with Kaby Lake at same clocks and then look at core area and cost, it's not even funny.

Well if it were all up to Intel then we'd be twiddling our thumbs on dual cores, minus HT, while some us would've left their laptops for Android powered tablets or convertibles. The AMD being "crap competition" is as much an enabler in Intel maintaining their extortionist prices as the people who replace their quad core CPU's with a 2~10% faster model each year, or two along with a motherboard upgrade.

 

[jackstar7]

https://en.wikipedia.org/wiki/Economies_of_scale

Indeed. Now also factor in the unseen non-material costs. Now adjust for investors and your other expenses. Now set retail prices.

You are the one making it sound like AMD should be all but giving these away. It was an absurd starting point, so I pushed the absurdity to its conclusion to perhaps provoke you to reconsider what you were putting forth.

 

[beginner99]

A session at GDC was spotted by VideoCardz.com ...

http://schedule.gdconf.com/session/optimizing-for-amd-ryzen-cpu-presented-by-amd

Maybe this description was submitted to GDC organizers weeks to months ago, never adjusted and since then the release schedule has changed? Is also a possibility. It's hard to believe AMD still using ES and launch should happen in around 6 weeks. To meet this timeline they already need to be shipping final product now (including boxes) and hence only thing they might be able to adjust is price but not SKUs and their specs. Given that, there should have been more leaks with final specs. So call me skeptical.

 

[imported_jjj]

The enthusiast or workstation class (hexa or octacore) CPU's haven't moved much in volume but they ought to have had a steady growth over the years, at least not negative. But then again you'd ask yourself do you really need an octa core CPU just because it's 350$ & the answer to that is a resounding NO, the vast majority of people don't need more than 4c/8t IMO, that the IGP in their system is just a waste of die space is another matter entirely. So by keeping octa core CPU prices high, AMD could balance out their lower end offerings by pricing them appropriately but most of this depends on how Intel reacts to the upcoming competition & (dis)incentives they give their partners vis-a-vis AMD.
I don't see Intel making a separate die for HEDT just to counter AMD. They'll just make the next mainstream CPU a hexacore, Skylake's competition should be Zen APU & with a better IGP they might make massive inroads in the desktop & notebook space, provided Intel doesn't restart their contra revenue shens with a different name.
Well if it were all up to Intel then we'd be twiddling our thumbs on dual cores, minus HT, while some us would've left their laptops for Android powered tablets or convertibles. The AMD being "crap competition" is as much an enabler in Intel maintaining their extortionist prices as the people who replace their CPU's with a 2% faster model each year, or two along with a motherboard upgrade.

Stating that people don't need an octa is opinion and what matters more is if they WANT one. We are not talking about the average consumer but folks that care a bit more.
Don't focus on just octa, they got 4 and 6 cores too and every price point they don't address properly ,means money going to Intel,especially before AMD has its APUs. If they price quads at 99$ to 199$, they can price hexa at 199$ to 299$ and octa from 259 to 349$ plus one or two "shiny" SKUs above that. At certain price points, the customer could chose between high clocks fewer cores or lower clocks more cores - by doing so they increase revenue a bit and give us the illusion of more options. That's more or less the pricing i was proposing, details might have gotten lost in the many posts. How many hexa SKUs they got and at what prices,would depend to some extent on yields. They do need to make it hard to buy an Intel APU for folks that don't need an integrated GPU.
I presume the 6 cores assumption is based on leaks but that doesn't mean that Intel won't change course and do an 8 core die with less L3, just 2 memory chans, less IO so something that matches Ryzen. If they start hurting badly, they can't ignore it.
For APUs and addressing the broader market, that's where gaining share is hard as Intel can apply certain business tactics. Would be very important here to strengthen the brand with Ryzen and reach consumers that way since AMD can't spend billions on marketing every quarter.

Your main concern seems to be that people won't buy 8 cores even if they can afford them ,i don't agree and AMD's sales will show us, if they price them well. Nobody needs a quad core at 350$ really when there are much cheaper quads but a great number of folks pay a premium for those anyway. The way most will think about it is, they need an upgrade soon as their box is aging, more cores is a plus and future proof, it is in their budget so they might as well upgrade and that's how AMD expands the market and starts a huge upgrade cycle. For folks that would buy something this year anyway, AMD needs to make it really hard to buy Intel. Longer term, a larger base of octa cores, pushes devs to code for many cores and that helps AMD in 2019 if they got 12 cores on 7nm then.
However, If you are right, then they would trade margins for share as folks would buy lesser SKUs. All in all they would end up with similar profits just more AMD users and better GPU and laptop sales while hurting Intel more (by grabbing more share) and making it harder for Intel to respond as Intel needs to watch their margins and ASPs or they get spanked by Wall Street.
Intel in PC makes over 30 billion per year in revenue.AMD maybe 2 and that Includes GPU- GPU could be half. AMD really can't afford to price it out of range or give Intel a break.

 

[imported_jjj]

Indeed. Now also factor in the unseen non-material costs. Now adjust for investors and your other expenses. Now set retail prices.

You are the one making it sound like AMD should be all but giving these away. It was an absurd starting point, so I pushed the absurdity to its conclusion to perhaps provoke you to reconsider what you were putting forth.

How about you do the math and show us how their costs really are and why very high margins would not be sufficient? Or how pricing it high would generate the most profits.
It's the same cost as a 4 cores APU, lower if you consider the GPU IP.
Octa cores at 249-349$ or w/e is giving it away? They could sell it at 99$ and still make a billion in gross profits this year from this platform.
You make empty claims based on nothing, you haven't provided any relevant argument about pricing,margins, volumes or any other additional relevant factors. Bring something of value to the conversation.

 

[R0H1T]

Stating that people don't need an octa is opinion and what matters more is if they WANT one. We are not talking about the average consumer but folks that care a bit more.
Don't focus on just octa, they got 4 and 6 cores too and every price point they don't address properly ,means money going to Intel,especially before AMD has its APUs. If they price quads at 99$ to 199$, they can price hexa at 199$ to 299$ and octa from 259 to 349$ plus one or two "shiny" SKUs above that. At certain price points, the customer could chose between high clocks fewer cores or lower clocks more cores - by doing so they increase revenue a bit and give us the illusion of more options. That's more or less the pricing i was proposing, details might have gotten lost in the many posts. How many hexa SKUs they got and at what prices,would depend to some extent on yields. They do need to make it hard to buy an Intel APU for folks that don't need an integrated GPU.

I'd say 90% of desktop users around the world don't need an octacore, that may not be verifiable atm but I'd fine with it being an unverifiable fact. As for non casual users, if they can pay ~1100$ for a 6900K then why not 500~600$ for something that'll likely be within 5~10% of the Intel HEDT equivalent?

I do agree that they need to price their lower end aggressively but that also depends on whether they can do something like a 4c/4t i.e. disable SMT, if not then they have a narrower price range to work with. The displacement of Intel APU's is the hard part, people buy an i7 & then game on its IGP, steam survey says there's lots of them out there.

I presume the 6 cores assumption is based on leaks but that doesn't mean that Intel won't change course and do an 8 core die with less L3, just 2 memory chans, less IO so something that matches Ryzen. If they start hurting badly, they can't ignore it.
For APUs and addressing the broader market, that's where gaining share is hard as Intel can apply certain business tactics. Would be very important here to strengthen the brand with Ryzen and reach consumers that way since AMD can't spend billions on marketing every quarter.

They might if the volumes justify that, but at that point AMD would just be better because of the single socket AM4 & the prohibitive costs when moving from Intel x70 to xx9 IMO. The thing is I don't believe most (i.e. average) people will move to Ryzen if they don't have a dGPU, even if it's cheap. The IGP is a great selling point for many, including enthusiasts. Zen should certainly bring positive momentum & word of mouth to the masses, but the thing is not many people (non enthusiasts) upgrade their desktops these days & even fewer unless they absolutely have to. These people generally go for an IGP &/or stick with cheapo OEM builds, do very little research & get the easiest option that doesn't strike their wallet hard.

The real battle will be in the APU space & notebooks at that, OEM shouldn't be ostracizing Zen APU unless there's something really bad about it or Intel tells them to. I can see AMD making a dent in the mid/upper tier of desktop market with their pricing but I do not believe that pricing an octacore CPU to 350$ will convert the mainstream (Intel) IGP loving crowd to AMD. The Zen APU would've have to do it's magic, for AMD, to claw back substantial market share from Intel.

Your main concern seems to be that people won't buy 8 cores even if they can afford them ,i don't agree and AMD's sales will show us, if they price them well. Nobody needs a quad core at 350$ really when there are much cheaper quads but a great number of folks pay a premium for those anyway. The way most will think about it is, they need an upgrade soon as their box is aging, more cores is a plus and future proof, it is in their budget so they might as well upgrade and that's how AMD expands the market and starts a huge upgrade cycle. For folks that would buy something this year anyway, AMD needs to make it really hard to buy Intel. Longer term, a larger base of octa cores, pushes devs to code for many cores and that helps AMD in 2019 if they got 12 cores on 7nm then.
However, If you are right, then they would trade margins for share as folks would buy lesser SKUs. All in all they would end up with similar profits just more AMD users and better GPU and laptop sales while hurting Intel more (by grabbing more share) and making it harder for Intel to respond as Intel needs to watch their margins and ASPs or they get spanked by Wall Street.
Intel in PC makes over 30 billion per year in revenue.AMD maybe 2 and that Includes GPU- GPU could be half. AMD really can't afford to price it out of range or give Intel a break.

More people upgrade their phones than PC parts these days, this isn't changing anytime soon & I think you're giving too much credit to people if you think they'll go to octacores just because they're cheap. Users who were eying octacores will likely dip their toes with the upcoming Zen, but the quad core crowd IMO will wait for a price war with Intel before they choose their winner.

 

[beginner99]

It stands to reason then that Zen should be priced appropriately in lieu of its competition i.e. HEDT so IMO 500$ is where it should debut & no less, the black or extreme edition could be 100$ more.

Exactly. The octo-core at least with high clocks won't sell for under $500. there is no reason it should unless it sucks like BD sucked. If it actually delivers, the server market will be all over it and hence high demand with questionable availability. AMD can then price the best 6-core vs 7700k and the best quad vs best i3 and have good margin and deliver better performance/$ (And more cores) than intel in all brackets.

 

[coercitiv]

The octo-core at least with high clocks won't sell for under $500. there is no reason it should unless it sucks like BD sucked. If it actually delivers, the server market will be all over it and hence high demand with questionable availability.

This. Even with top to bottom demand being able to scale up faster than lowering prices (in the sense of increasing/optimizing total revenue and market share), we have no indication that AMD would be able to keep up with that kind of demand at launch time. Fail to keep up and MSRP vs. retail price will be two completely different stories.

 

[imported_jjj]

Maybe this description was submitted to GDC organizers weeks to months ago, never adjusted and since then the release schedule has changed? Is also a possibility. It's hard to believe AMD still using ES and launch should happen in around 6 weeks. To meet this timeline they already need to be shipping final product now (including boxes) and hence only thing they might be able to adjust is price but not SKUs and their specs. Given that, there should have been more leaks with final specs. So call me skeptical.

It's possible that the launch is channel first and then, they just need to ship by air right before launch. It's pretty high value per volume or weight so air is ok.
And i think it's more likely they'll launch before GDC, using ES now in demos , doesn't mean they don't have final silicon and likely they started production last year.

I'd say 90% of desktop users around the world don't need an octacore, that may not be verifiable atm but I'd fine with it being an unverifiable fact. As for non casual users, if they can pay ~1100$ for a 6900K then why not 500~600$ for something that'll likely be within 5~10% of the Intel HEDT equivalent?

I do agree that they need to price their lower end aggressively but that also depends on whether they can do something like a 4c/4t i.e. disable SMT, if not then they have a narrower price range to work with. The displacement of Intel APU's is the hard part, people buy an i7 & then game on its IGP, steam survey says there's lots of them out there.
They might if the volumes justify that, but at that point AMD would just be better because of the single socket AM4 & the prohibitive costs when moving from Intel x70 to xx9 IMO. The thing is I don't believe most (i.e. average) people will move to Ryzen if they don't have a dGPU, even if it's cheap. The IGP is a great selling point for many, including enthusiasts. Zen should certainly bring positive momentum & word of mouth to the masses, but the thing is not many people (non enthusiasts) upgrade their desktops these days & even fewer unless they absolutely have to. These people generally go for an IGP &/or stick with cheapo OEM builds, do very little research & get the easiest option that doesn't strike their wallet hard.

The real battle will be in the APU space & notebooks at that, OEM shouldn't be ostracizing Zen APU unless there's something really bad about it or Intel tells them to. I can see AMD making a dent in the mid/upper tier of desktop market with their pricing but I do not believe that pricing an octacore CPU to 350$ will convert the mainstream (Intel) IGP loving crowd to AMD. The Zen APU would've have to do it's magic, for AMD, to claw back substantial market share from Intel.
More people upgrade their phones than PC parts these days, this isn't changing anytime soon & I think you're giving too much credit to people if you think they'll go to octacores just because they're cheap. Users who were eying octacores will likely dip their toes with the upcoming Zen, but the quad core crowd IMO will wait for a price war with Intel before they choose their winner.

Lets define what market Summit Ridge is addressing.
It's no integrated GPU and there is no debate there. The market is "small" but lets define how small. I think mobo sales are at about 50 million units per year ,give or take. Asus and Gigabyte shiped 16-17 million each last year ,not sure about MSI and if they have grown a lot.The other mobo players are much much smaller.
Last year discrete GPUs should have been at maybe 45-ish million units with CF or SLI as small portion.
My assumption is that Summit Ridge can address this 40-50 million units market and that it can boost it a great deal with the right price.So best case scenario ,AMD ships 40 million units at 200ASP for 8 billion revenue. Worst case 20 million and half the revenue. Both scenarios with my suggested pricing structure.
How many of these users WANT more cores? It is a niche but a sizable niche and retail gives AMD great ASPs and margins. To OEMs the price is a few times lower and lower margins too. It's not everybody that Summit Ridge is addressing, it's not mom or grandpa. it's the folks that need or care the most about the PC.
The remaining 200 million PCs sold this year, are for everybody else and AMD addresses those with APUs. Some of them might be Ryzen buyers, buying a laptop for themselves or a PC for a friend , a relative.
So remember, we are not talking about those"other 200 million" at all when it comes to Summit Ridge, just the other chunk.
The 200 millions would be in majority commercial not consumer and that's tricky as businesses buy Intel. For the consumer chunk, marketing sells Intel and there are tricks and so on.
Also very important , in laptop,when selling to OEMs, AMD might have 70$ ASPs. Intel might have towards 100$ now as they milk low power and charge a premium but if AMD is competitive, prices would compress a bit. Lets say 100$ ASPs not 70$ since some APUs would be in retail and ,lets say that AMD gains 20% share in 2018 in those 200 million -might be hard to get that without a share grab in bulk with an Apple win ( EDIT - a Macbook win would be 12-15 million units so they would have more than 20% overall in "the 200 million" with such a win. Without it,getting 20% wouldn't be easy at all). 100$x40mil= 4 billions and ofc they would have lower margins in OEM. APUs would become the most important if they gain more share but that's not gonna be easy.
In server it is hard to predict now, very nice ASPs and great margins so if they gain more than 10% share it starts to become more important than APUs too.

Summit Ridge can exploit the market it addresses now better than at any time in the future as Intel can't do much and it's likely the biggest opportunity for AMD this year and maybe the next year.At some point server or APU could become more important if they gain share.

May i ask what pricing structure you think would be more profitable for AMD? In as much detail as you want.

 

[R0H1T]

May i ask what pricing structure you think would be more profitable for AMD? In as much detail as you want.

I'd say 20 million should be doable if they don't face strong headwinds possibly even 30 million if they have a dream year, it also depends on how the competitor responds & market conditions. The price will not expand this market segment much, if at all, however a good number of SKU's & appropriate pricing is a must. In that respect ~

Zen should start at 100~200$ provided they can do 4c/4t if not then 4c/8t should initially be priced ~150$
250~400$ again with the caveat that 6c/6t is possible, or else 300$ as launch price for 6c/12t is fine IMO
500~600$ for 8c/16t prime Zen

The price tiers can be adjusted if Intel responds aggressively, conversely AMD can introduce higher clocked Zen & move the older models to a lower tier. As for Zen APU a lot will depend on when or if they can integrate HBM soon enough for it to be a viable alternative in the notebook space, with HBM I can easily see Iris Pro CPU's being made redundant. Even the regular models with dual channel DDR4 or LPDDR4x should be excellent for notebooks. The desktop APU though is what'd be really interesting, we'll see if AMD can totally decimate the 750Ti class dPU with their HBM models.

 

[dacostafilipe]

Zen should start at 100~200$ provided they can do 4c/4t if not then 4c/8t should initially be priced ~150$
250~400$ again with the caveat that 6c/6t is possible, or else 300$ as launch price for 6c/12t is fine IMO
500~600$ for 8c/16t prime Zen

Yep, this makes sense.

The price tiers can be adjusted if Intel responds aggressively, conversely AMD can introduce higher clocked Zen & move the older models to a lower tier. As for Zen APU a lot will depend on when or if they can integrate HBM soon enough for it to be a viable alternative in the notebook space, with HBM I can easily see Iris Pro CPU's being made redundant. Even the regular models with dual channel DDR4 or LPDDR4x should be excellent for notebooks. The desktop APU though is what'd be really interesting, we'll see if AMD can totally decimate the 750Ti class dPU with their HBM models.

I don't think (aka only my own opinion!) that we will see HBM on Notebooks APUs so soon as the added height from the Interposer still worries me.

It would not be a fit for Ultrabooks or Gaming Notebooks ... in what niche would such an HBM APU be used? It's not like such an APU would be cheap in the first place.

 

[imported_jjj]

I'd say 20 million should be doable if they don't face strong headwinds possibly even 30 million if they have a dream year, it also depends on how the competitor responds & market conditions. The price will not expand this market segment much, if at all, however a good number of SKU's & appropriate pricing is a must. In that respect ~

Zen should start at 100~200$ provided they can do 4c/4t if not then 4c/8t should initially be priced ~150$
250~400$ again with the caveat that 6c/6t is possible, or else 300$ as launch price for 6c/12t is fine IMO
500~600$ for 8c/16t prime Zen

The price tiers can be adjusted if Intel responds aggressively, conversely AMD can introduce higher clocked Zen & move the older models to a lower tier. As for Zen APU a lot will depend on when or if they can integrate HBM soon enough for it to be a viable alternative in the notebook space, with HBM I can easily see Iris Pro CPU's being made redundant. Even the regular models with dual channel DDR4 or LPDDR4x should be excellent for notebooks. The desktop APU though is what'd be really interesting, we'll see if AMD can totally decimate the 750Ti class dPU with their HBM models.

Those prices will be prohibitive for most. Look how few are buying 6 cores Intel today.They wouldn't ship 10 millions with those prices. No waves of upgraders at all.
They would also have a huge. gap in margins between quad and 6-8 cores.The 4 cores die might be 30% smaller only, 40% at most With heavy volumes on the 4 cores side.
And depending on die size,it would be pretty much as greedy as Intel's pricing for Broadwell-E so Intel could make an effort with their 246.3mm2 10 cores die and threaten the 8 cores and 6 cores while having the advantage in IO.
It would also make AMD a lot more vulnerable to Intel's 6 core, when it arrives.
We still assume lower IPC for Zen vs Skylake and Intel is hitting pretty high clocks on 14nm+ and that's a factor.

APUs with HBM are not viable from a cost perspective yet, I've commented on this a bunch so not gonna go in depth into that. It's perf gained vs cost ,if you get enough there and how it compares vs normal APUs and discrete GPUs..If they bring a server APU in consumer, that's possible but hard to see the value proposition and it's more likely a CPU die+ a Vega with HBM in a single package but only the Vega+HBM on the interposer. Could maybe make sense in laptop, just to simplify the mobo.
Not sure where Iris Pro lands in perf as the pricing made it irrelevant for me to waste time checking reviews but i am kinda assuming AMD can beat that with 768 cores Vega. Going from 28nm to 14nm, increasing the size by 50% and doing so with a new arch that is much more efficient, might be quite enough. Ofc they would have more memory BW with DDR4 while Vega should use a lot less.

 

[krumme]

A few extra points to consider:

1. Ceo get shortsighted bonusses and its one of the reasons our economy is dragged down. Its actually crazy what idiotic decision that can lead to. In this case eg search profit for next q2 instead of 100% 2017 to 2025 focus.
I dont know what incentive program Lisa and Pepermaster is on but there should be regulations that stock options should all be at least for 8 years plus.
In many cases the incentive programs is directly counter productive.

2. Not only is zen kind of unique that it bends the monopolys power by what seems to me like a small miracle. The new fabric enables going for niches at cost that was impossible before.
We have actually some new fabric here that can change how we get mid and highend stuff.

 

[krumme]

If you sell 8c at 500 to 600 you will not sell it above half mill. The market is miniscule. If amd tries it they will fail.

And why should you name your products sr7 and sell it for 500 or 600 when your well known competitor is 350? Bad move. Then it should be called sr9.

 

[lopri]

8C/16T @3.2/3.5 GHz for $500 will be more than reasonable assuming the IPC is comparable to Haswell. There might be more than one SKU with 8C/16T configuration. 8C/16T @3.4/3.7 GHz for $600 will be excellent since the 6900K with same config is $1,100.

Of course Intel can lower the prices of their offerings, which forces AMD to slightly adjust their prices down. That will be a welcome development in my book as well.

 

[lopri]

Or maybe 8C/16T @2.9/3.3 GHz for $400. That would be fantastic if it OC's well.

 

[ShintaiDK]

http://techfrag.com/2016/02/12/amd-zen-high-end-exascale-cpu-and-apu-specs-leaked/

https://vrworld.com/2016/02/12/cern-confirms-amd-zen-high-end-specifications/


But then, it doesn't exactly tie with the below:

But this should be borne in mind:

Yes, 2-3year old slides. You just prove my point.

 

[ShintaiDK]

And when Xeon Phi is run as a PCI-e card like a Vega 10 does, how fast can it access data from MAIN Memory. Your shifting goal posts, because a Vega based APU ( like Raven Ridge) will also have direct access to the full SOC memory bandwidth. An APU like the presented Data center APU will have 100GB/s of memory access.

So what exactly was your point all along ?

There are no PCIe versions of KNL.
http://ark.intel.com/products/family/92650/Intel-Xeon-Phi-Product-Family-x200#@Server

And my mistake, some SKUs got 115.2GB/sec main memory bandwidth and not just 102GB/sec

 

[ShintaiDK]

http://techfrag.com/2016/02/12/amd-zen-high-end-exascale-cpu-and-apu-specs-leaked/
Every CPU has 2048 bit controller. Raven Ridge APUs are the same family as Vega. Guess why? HBM2 is inherent part of the architecture. Vega appears to be designed to work ONLY with HBM2, not GDDR memory. However, some Raven Ridge APUs will not have HBM2. But every CPU has 2048 bit memory controller. And HBM2, CPU and GPU are connected together with Infinity Fabric, which you can deduct from post on forum-3dcenter.org, that I have linked above.

HBM2 adds latency. However, Infinity Fabric is for the exact reason of lowering the latency for CPU, and GPU.

Reposting old news as new? Classic.

Also another hint, do you think Naples supports 3200mhz memory? Because Ryzen doesn't

 

[itsmydamnation]

There are no PCIe versions of KNL.
http://ark.intel.com/products/family/92650/Intel-Xeon-Phi-Product-Family-x200#@Server

And my mistake, some SKUs got 115.2GB/sec main memory bandwidth and not just 102GB/sec

Not according to intel, unless they canned there own products because they had no chance at competing............
And in that case you comparing a x86 CPU to a GPU, so how pitiful is KNL with its piddly 6 channels, real big iron has 8 channels..........
did you miss AMD's answer to HPC solutions:
http://www.guru3d.com/index.php?ct=news&action=file&id=17690
http://www.guru3d.com/index.php?ct=news&action=thumb&id=17691

288 full forwarding PCI-E lanes, upto 512TB of addressable memory.

Yes, 2-3year old slides. You just prove my point.

how are these to your point:

http://abload.de/image.php?img=2017-01-09_231822rdkz6.jpg
http://abload.de/image.php?img=2017-01-09_2318247ukfb.jpg

a few notes:
Hyper transport plus enhancements
Multi socket and mutli die Ready
used on Vega
used on Summit Ridge
used on Raven ridge
.........................


But what is really funny is the massive rear guard action you have gone on, while completely missing the point of the original discussion. Which was with the changes with Vega, AMD could run a 1x 1-2gb HBM stack as a GPU cache in Raven Ridge...........

i engoy looking forward to the next 12 month where to try to recommend 1151 solutions because of miss information around PCI lanes.....lol

 

[Atari2600]

Yes, 2-3year old slides. You just prove my point.

I assume you are going to provide links that give evidence contrary to what I've shown?


If this is the best you can do, it seems these days you can't even get Trolls like you used to.

 

[ShintaiDK]

I assume you are going to provide links that give evidence contrary to what I've shown?

Since its never been a product or SKU from AMD you expect what again?

 

[ShintaiDK]

Not according to intel, unless they canned there own products because they had no chance at competing............
And in that case you comparing a x86 CPU to a GPU, so how pitiful is KNL with its piddly 6 channels, real big iron has 8 channels..........
did you miss AMD's answer to HPC solutions:
http://www.guru3d.com/index.php?ct=news&action=file&id=17690
http://www.guru3d.com/index.php?ct=news&action=thumb&id=17691

288 full forwarding PCI-E lanes, upto 512TB of addressable memory.


how are these to your point:

http://abload.de/image.php?img=2017-01-09_231822rdkz6.jpg
http://abload.de/image.php?img=2017-01-09_2318247ukfb.jpg

a few notes:
Hyper transport plus enhancements
Multi socket and mutli die Ready
used on Vega
used on Summit Ridge
used on Raven ridge
.........................


But what is really funny is the massive rear guard action you have gone on, while completely missing the point of the original discussion. Which was with the changes with Vega, AMD could run a 1x 1-2gb HBM stack as a GPU cache in Raven Ridge...........

i engoy looking forward to the next 12 month where to try to recommend 1151 solutions because of miss information around PCI lanes.....lol

You can try and change the goalpost as much as you want, there still isn't any KNL products without the 6 channel DRAM and nobody else is even close to have similar capabilities.

Vega 20 will be 2 years after Nvidia just for GPU peer to peer communications. And around 2 years behind Intel as well with KNL on Omnipath.