Fudzilla: New AMD Zen APU boasts up to 16 cores (plus Greenland GPU with HBM)

[Arachnotronic]

You are right Arachnotronic, they are going after intel's core market. It is a gutsy move, or maybe a desperate one, or both. The thing is, they dont seem to be trying to leverage their best advantage vs intel, and that is graphics performance. Makes me question the ATI purchase even more. Seems like they finally had a chance with Zen and HBM to make an APU a compelling purchase in some segments, but failed on both counts. And it seems like they are ceding basically the entire mobile market to intel.

Reminds me of this quote from Moby Dick ...

To the last, I grapple with thee; From Hell's heart, I stab at thee; For hate's sake, I spit my last breath at thee.

 

[Fjodor2001]

Well, 8 core/16 thread Zen is probably going to go up against SKL-E rather than BDW-E,

That's the thing though, Intel is prioritizing mobile. And down-prioritizing server and desktop, HEDT desktop in particular. The down-prioritized segments get older uArch and older nodes than mobile. So they left those flanks open for attack. I guess that's what AMD is zooming in on.

And I don't think we'll see SKL-E in 2016, but BDW-E. See this and this.

In addition, I guess it's sensible to compare BDW-E to Zen performance-wise, but not price-wise. Because I don't think AMD intends to sell 8 core Zen for $1000 like Intel does on their HEDT platform. The 8 core Zen is a replacement for the performance FX 8xxx series CPUs, so it will most likely be priced much lower.

Intel will effectively hand over the mainstream MT performance crown to AMD in 2016.

 

[CHADBOGA]

Reminds me of this quote from Moby Dick ...

I think that might have been a tweet from Abwx's to some Intel rep. :biggrin:

 

[CHADBOGA]

Well, 8 core/16 thread Zen is probably going to go up against SKL-E rather than BDW-E, so I'd imagine that if Intel really feels "threatened" by it, it will take advantage of the fact that there are some insane core count chips that will fit into the HEDT platform. Max core count by the time Skylake-EP rolls around will probably be north of 20.

Maybe we'll see ~$400-$600 8 core chips, and the Extreme Edition becomes a 10+ core chip.

How many people would be interested in a SKL-E with 12 cores clocked at 2.5Ghz and costing $1,000?

 

[Arachnotronic]

That's the thing though, Intel is prioritizing mobile. And down-prioritizing server and desktop, HEDT desktop in particular. The down-prioritized segments get older uArch and older nodes than mobile. So they left those flanks open for attack. I guess that's what AMD is zooming in on.

And I don't think we'll see SKL-E in 2016, but BDW-E. See this and this.

In addition, I guess it's sensible to compare BDW-E to Zen performance-wise, but not price-wise. Because I don't think AMD intends to sell 8 core Zen for $1000 like Intel does on their HEDT platform. The 8 core Zen is a replacement for the performance FX 8xxx series CPUs, so it will most likely be priced much lower.

Intel will effectively hand over the mainstream MT performance crown to AMD in 2016.

The only reason AMD prices its FX chips at low prices is because they won't sell at higher prices.

If AMD has an "Intel killer" as you say, then expect them to charge for it.

BTW, just curious, do you think Intel is stupid enough to just "hand over" any high-margin segment to AMD?

 

[Fjodor2001]

Yea, if "the future is Fusion" I guess it just got delayed. I honestly am blown away by this roadmap, and not in a good way. I am sure there will be a market for Zen without igp, assuming the price and performance are competitive, but that is not the chip that AMD needs to regain relevance in the mainstream market. It will appeal mainly to enthusiasts who will use a dedicated card, and that at most could be what, 5 percent of the overall cpu market?

What they needed was a Zen based APU with better cpu performance and HBM to alleviate the bandwidth bottleneck. Apparently they are getting neither, at least until 2017. I guess they are betting the farm on Zen for servers. But even this is astounding, because for years they have been touting how ARM and APUs are going to overtake the conventional x86 server market.

You are right Arachnotronic, they are going after intel's core market. It is a gutsy move, or maybe a desperate one, or both. The thing is, they dont seem to be trying to leverage their best advantage vs intel, and that is graphics performance. Makes me question the ATI purchase even more. Seems like they finally had a chance with Zen and HBM to make an APU a compelling purchase in some segments, but failed on both counts. And it seems like they are ceding basically the entire mobile market to intel.

AMD have to prioritize, just like Intel that prioritizes mobile.

AMD has chosen to go after the flanks that Intel has left open for attack due to Intel's gamble on mobile (which they are losing $4B a year on by the way). So AMD will be attacking the server and high end desktop segments first. The areas that Intel has given the least amount of love lately. Makes total sense.

Then in 2017 there'll be APUs with Zen + latest iGPU + HBM. Intel will not have improved much in that segment by then anyway, given their past record for the last couple of years (~5% yearly performance increases).

So in 2017 you will be able to get an APU with close to Intel's top end CPU performance but much better iGPU, likely at a lower price.

So good stuff ahead from AMD in 2016, and more to come in 2017 and beyond.

 

[Sweepr]

That's the thing though, Intel is prioritizing mobile. And down-prioritizing server and desktop, HEDT desktop in particular. The down-prioritized segments get older uArch and older nodes than mobile. So they left those flanks open for attack. I guess that's what AMD is zooming in on.

Desktops will get Skylake before mobile.

And I don't think we'll see SKL-E in 2016, but BDW-E. See this and this.

Broadwell-E is only coming in Q1-2016 because it was delayed, doesn't mean that they will delay Skylake-E too so it could very well arrive in late 2016.

Intel will effectively hand over the mainstream MT performance crown to AMD in 2016.

Even if this chip delivers around SB level of IPC we don't know what the clocks will be for the 8C part neither how much their first SMT implementation will improve MT performance. If they manage +20% better MT performance but Intel still delivers +25% better per core performance (compared to 4C/8T Skylake-S) then Intel will most likely not even bother to lower the price of 6C/12T Broadwell-E/Skylake-E parts, Skylake-S would do fine till Icelake arrives in 2017.

Intel will replace Core i7 6700K with a faster part as soon as Q1-2016, you can bet they will not stand still if the competition improves their position. Who knows what will be available by the time this 8-core Zen chip arrives (most likely H2-2016).

 

[Fjodor2001]

The only reason AMD prices its FX chips at low prices is because they won't sell at higher prices.

If AMD has an "Intel killer" as you say, then expect them to charge for it.

Yes, but not a ridiculous amount like $1000, because then they won't sell, like Intel's 8 core HEDT CPUs. Better to sell more units at a lower price then. The total profit for AMD will be greater then.

Remember AMD did not price the FX predecessors anywhere near $1000.

BTW, just curious, do you think Intel is stupid enough to just "hand over" any high-margin segment to AMD?

Intel likes its high margins. They usually demand 60% margin or so, much higher in the server segment. Do you think they'll be stupid enough to give up that in the server and desktop segments?

That's the choice they have to make; Give up the performance crown to AMD (i.e. performance per $), or lower their margins drastically. What do you think they'll choose?

 

[StinkyPinky]

40% increase in IPC is great, but I bet it's marketing speak. It's probably 40% on one synthetic benchmark they manipulated.

Don't get me wrong, I hope I'm wrong. An AMD CPU with Ivy Bridge or even Haswell IPC with 8 cores would attract a lot of attention, but I'm pessimistic But let's hope so because it will probably force Intel to release a 6 core mainstream CPU.

 

[mrmt]

Adding a third ALU only scales when three operations have been fully decoded and all information is present in registers. You have to feed that ALU and the code has to allow for it to be used. You will not get close to 50% in general purpose code going from 2 to 3 ALUs.

Yes, and that was a marketing slide, meaning that 40% is a best-case scenario.

 

[Arachnotronic]

Yes, but not a ridiculous amount liek $1000, because then they won't sell, like Intel's 8 core HEDT CPUs. Better to sell more units at a lower price then.

Remember AMD did not price the FX predecessors anywhere near $1000.

Back when FX chips were best-in-class, AMD did charge $1k+.

Intel likes its high margins. They usually demand 60% margin or so, much higher in the server segment. Do you think they'll be stupid enough to give up that in the server and desktop segments?

If AMD's products are super, duper, glorious like you think they are, lower margins on non-zero revenue is better than asking for high margins and getting no sales, no?

That's the choice they have to make; Give up the performance crown to AMD (i.e. performance per $), or lower their margins drastically. What do you think they'll choose?

The one that doesn't lead to Intel's revenue going to $0 because in this case, AMD would have the "superior" solution?

 

[Fjodor2001]

Broadwell-E is only coming in Q1-2016 because it was delayed, doesn't mean that they will delay Skylake-E too so it could very well arrive in late 2016.

According to the article: "Intel Will Begin Mass Production of Broadwell-E in Q1 2016". So it'll be in the shops later than that. If SKL-E would be available in 2016, that would give BDW-E a ridiculously short lifespan, and no OEMs would touch it. Not likely to happen. Past track record has shown that HEDT is given less and less priority by Intel, older uArch, older nodes, and later releases.

 

[mrmt]

That's the thing though, Intel is prioritizing mobile. And down-prioritizing server and desktop, HEDT desktop in particular. The down-prioritized segments get older uArch and older nodes than mobile. So they left those flanks open for attack. I guess that's what AMD is zooming in on.

If you think Intel is not prioritizing servers you have no clue of what Intel's business model is all about.

Intel will effectively hand over the mainstream MT performance crown to AMD in 2016.

I doubt it. If 40% is the best case scenario we can expect from Zen then AMD shouldn't reach Intel performance levels, even in mainstream SKUs. Adding more cores will only worse their performance/watt.

 

[Sweepr]

According to the article: "Intel Will Begin Mass Production of Broadwell-E in Q1 2016". So it'll be in the shops later than that. If SKL-E would be available in 2016, that would give BDW-E a ridiculously short lifespan, and no OEMs would touch it. Not likely to happen. Past track record has shown that HEDT is given less and less priority by Intel, older uArch, older nodes, and later releases.

Nope, it will be launched in Q1-2016.

You kept questioning Skylake-S's launch in 2015 (I still remmember your H2-2016 or 2017 launch prediction for desktop Skylake) till you were proven wrong and now we all know Skylake-S will arrive only one quarter after Broadwell-K.
The fact that Broadwell-E was delayed doesn't necessarily mean Skylake-E got pushed to 2017.

 

[jpiniero]

No Zen APUs in 2016 is total fail. Really, Intel is pushing desktop TDP down to 35W for the most part in Skylake and this is all they have?

 

[Arachnotronic]

According to the article: "Intel Will Begin Mass Production of Broadwell-E in Q1 2016". So it'll be in the shops later than that. If SKL-E would be available in 2016, that would give BDW-E a ridiculously short lifespan, and no OEMs would touch it. Not likely to happen. Past track record has shown that HEDT is given less and less priority by Intel, older uArch, older nodes, and later releases.

Link to this article? Roadmaps on the web say Q1 2016, and that usually means commercial availability date.

Also, I suspect BDW-E will have a shortened life-cycles because transitions to new platforms (i.e. Romley -> Grantley) tend to be better from a financial perspective than upgrades to existing platforms so I'm sure Intel wants to get it out ASAP.

BDW-E/EP will probably be made available in Q1 2016. It'll be a better perf/watt solution than HSW-E for customers that really would like to buy in early 2016 rather than late 2016, and it would also provide an upgrade path for those running the Grantley platform and want a "drop in" replacement.

SKL-E/EP brings AVX512 and other goodies, so I am sure Intel's HPC customers, among others, are clamoring to get their hands on this thing ASAP.

I might turn out to be wrong, but I'd be willing to wager a late 2016 launch for Skylake-EP.

 

[Abwx]

40% increase in IPC is great, but I bet it's marketing speak. It's probably 40% on one synthetic benchmark they manipulated. .

I would rather think that your post is not probably but most surely what one can call deffamation...

 

[Enigmoid]

Even if this chip delivers around SB level of IPC we don't know what the clocks will be for the 8C part neither how much their first SMT implementation will improve MT performance. If they manage +20% better MT performance but Intel still delivers +25% better per core performance (compared to 4C/8T Skylake-S) then Intel will most likely not even bother to lower the price of 6C/12T Broadwell-E/Skylake-E parts, Skylake-S would do fine till Icelake arrives in 2017.

Intel will replace Core i7 6700K with a faster part as soon as Q1-2016, you can bet they will not stand still if the competition improves their position. Who knows what will be available by the time this 8-core Zen chip arrives (most likely H2-2016).

What I hope is that intel respins Skylake to a 6 core version for mainstream. 14 nm should allow for a 6 core 45W mobile chips (with aggressive turbo) and the existing mainstream architecture should allow for 2 more cores to be dropped in (~20-24 mm^2 die space with L3).

 

[Fjodor2001]

Back when FX chips were best-in-class, AMD did charge $1k+.

Only the top speed binned ones. And back then all CPUs where priced higher, e.g. the desktop Intel Q6600 cost $851 at release.

 

[Fjodor2001]

Link to this article?

I already linked to it. Read the post again.

 

[Arachnotronic]

OK, listening to the Q&A...definitely coming to APUs in 2017.

 

[podspi]

Makes sense their APU is still construction-core based, especially if they are adding other magic (HBM, etc). They did a similar thing with Llano.

Assuming 40% is correct and general (granted, a large assumption), and can contain the clocks they already have, they'll be within spitting distance of Haswell.

That level of ST performance with 8-cores + HT would be VERY interesting. The only thing left to do now is wait for leaks...

 

[Maxima1]

Makes sense their APU is still construction-core based, especially if they are adding other magic (HBM, etc). They did a similar thing with Llano.

Assuming 40% is correct and general (granted, a large assumption), and can contain the clocks they already have, they'll be within spitting distance of Haswell.

That level of ST performance with 8-cores + HT would be VERY interesting. The only thing left to do now is wait for leaks...

That doesn't make sense. It means that all of the APUs are complete garbage. Who will want an APU when they're claiming ~40% increase for Zen, much less considering the already huge gap between Intel and AMD?

 

[ElFenix]

be nice if it works. not going to get myself worked up over it until i see actual benchmarks on shipping silicon in stuff that matters (video encoding, photo editing, and gaming).

That doesn't make sense. It means that all of the APUs are complete garbage. Who will want an APU when they're claiming ~40% increase for Zen, much less considering the already huge gap between Intel and AMD?

$350 desktop and notebook market, type of people who go down to wally world and just buy whatever is on sale with the shiniest plastic everything. you know, the current APU market.

 

[Arachnotronic]

Apparently K12 is still in "simulation" -- i.e. no actual silicon back.

I wonder if Zen is also in simulation as well, or if they've got actual silicon in hand?