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

[cbn]

To be honest, I wouldn't focus too much on the fact that they use CUDA. Most CUDA code is pretty close to standard C++, with some GPU-centric constructs like texture references which will have an equivalent in any other GPGPU API. I sincerely hope that OpenCL 2.1 (with it's C++ based kernel language) will make it much easier to port from CUDA.

It may be that Nvidia also makes strides in CUDA by the time OPEN CL catches up with where CUDA is today.

Then there is also differences in Linux driver quality to consider as well.

 

[Phynaz]

You said the "D" word.

 

[cbn]

Ehm the current Opterons 63xx are almost three years old, they are not able to compete today.

Summit is going live in 2018.

https://www.olcf.ornl.gov/summit/

So a future AMD CPU would have been used if AMD got the contract.

 

[JDG1980]

Not sure if this has been posted before, but the FAD 2015 Codename Decoder document on AMD's website indicates that Zen will be "sampling in 2016". Maybe I'm reading too much into this, but that doesn't sound like retail availability. Will we need to wait until 2017 for this product to actually hit the market?

Then again, this document has some clear mistakes, like the implication that Vishera has integrated graphics (which, of course, it doesn't).

 

[inf64]

Not sure if this has been posted before, but the FAD 2015 Codename Decoder document on AMD's website indicates that Zen will be "sampling in 2016". Maybe I'm reading too much into this, but that doesn't sound like retail availability. Will we need to wait until 2017 for this product to actually hit the market?

Then again, this document has some clear mistakes, like the implication that Vishera has integrated graphics (which, of course, it doesn't).

 

[Arachnotronic]

Not sure if this has been posted before, but the FAD 2015 Codename Decoder document on AMD's website indicates that Zen will be "sampling in 2016". Maybe I'm reading too much into this, but that doesn't sound like retail availability. Will we need to wait until 2017 for this product to actually hit the market?

Then again, this document has some clear mistakes, like the implication that Vishera has integrated graphics (which, of course, it doesn't).

Probably going to have to wait for 2017, yeah. And, if the next-gen Carrizo APUs are coming in mid-2016, then you can probably expect the Zen ones to come in mid-2017.

Against Cannonlake the Zen APUs go, and against Skylake-E the Zen standalone CPU goes most likely.

EDIT: If "Summit Ridge" is Zen CPU, then it seems like Q4 2016 is when the company expects to release these parts. That's better than early 2017.

 

[dark zero]

AMD entered too late.. Well it was good to see a competition, but this is ending and time to see the death of AMD. Intel won this time.

Btw, once AMD dies, Intel will triplicate their product prices the next generation after that. Celerons at 160 dollars and Pentiums at 240 dollars... Yeah.

 

[Arachnotronic]

Btw, once AMD dies, Intel will triplicate their product prices the next generation after that. Celerons at 160 dollars and Pentiums at 240 dollars... Yeah.

If they did that, people would slow down their upgrade cycles even more. If AMD went bankrupt tomorrow Intel would be absolutely stupid to try to jack up prices in the manner you describe.

 

[CHADBOGA]

Probably going to have to wait for 2017, yeah. And, if the next-gen Carrizo APUs are coming in mid-2016, then you can probably expect the Zen ones to come in mid-2017.

Against Cannonlake the Zen APUs go, and against Skylake-E the Zen standalone CPU goes most likely.

The Skylake-E is hardly going to be the Zen's main competitor.

 

[SPBHM]

If they did that, people would slow down their upgrade cycles even more. If AMD went bankrupt tomorrow Intel would be absolutely stupid to try to jack up prices in the manner you describe.

also it would force the low end to replace x86 with ARM.

 

[IntelUser2000]

also it would force the low end to replace x86 with ARM.

They won't be affected as long as Windows and applications remain firmly x86 territory.

Intel will probably increase the prices in a way most people don't notice it. Take a look at how their "Iris" lines are priced. They don't use "Iris" to make cost efficient parts, they make it to force you to buy $250+ CPUs. The 28W Iris lines are so overpriced that it was being used in $1300+ laptops. All for a 20% gain since there's no option.

Take a look at their Xeon prices:

Xeon E7 8870: $4616
Xeon E7 8890 v2: $6841
Xeon E7 8890 v3: $7174

Now:

Skylake = 15% extra performance for same price as Broadwell/Haswell

Future: 10nm Tock 5% extra performance for same price, 15% extra performance for higher price

Celerons at 160 dollars and Pentiums at 240 dollars... Yeah.

They won't make it that obvious.

http://forums.anandtech.com/showpost.php?p=37451180&postcount=3984

Broxton sits between Bay Trail and Core M in that slide. Intel also indicated Broxton taking "premium" segment. Are the results same as you are suggesting? Yes. Does it look like they aren't doing that? Yes to that too.

Against Cannonlake the Zen APUs go, and against Skylake-E the Zen standalone CPU goes most likely.

Technically if Carrizo lives up to their powerpoint slide performance in graphics, it'll be competitive to the fastest Iris Pro parts without eDRAM or HBM memory. They are about 10% slower in graphics but at 35W and cheaper. While Skylake GT4e is 50% faster than Broadwell, there's no sure confirmation that we'll see mainstream consumer parts with it since Broadwell released just now. Also, while Desktop Carrizo might be next year, Laptop is this year. Considering how 14nm brought only 20% for graphics, I doubt Cannonlake will do much better.

 

[Arachnotronic]

The Skylake-E is hardly going to be the Zen's main competitor.

Skylake-E should be the competitor to the HEDT Zen chips; Cannonlake will go up against the APUs.

That is, unless Intel does something interesting like upping the max core counts on the mainstream Cannonlake parts (which it totally should)...

 

[Arachnotronic]

Take a look at their Xeon prices:

Xeon E7 8870: $4616
Xeon E7 8890 v2: $6841
Xeon E7 8890 v3: $7174

I seriously doubt any major datacenter customer is paying anywhere close to these kinds of prices.

If you take Intel's data center group revenue of ~$14.4 billion last year and take Intel's claim that ~88% of its data center sales are from CPUs, this means the company's data center group sold $12.68 billion worth of CPUs.

Now, how many server CPUs did Intel sell in 2014? Hard to guess but let's call it 20 million.

This implies ASP of about $633. Unless you think Intel sells only a few E5s and E7s and the majority of its volumes are cheap E3s, this should tell you that ARK/price list prices are pretty much non-representative of reality

 

[SPBHM]

They won't be affected as long as Windows and applications remain firmly x86 territory.

5 or 10 years ago, sure, but now (and specially in 5 years) ARM and Google could replace MS/Intel if they go crazy with the prices for the lower end IMO

 

[IntelUser2000]

This implies ASP of about $633. Unless you think Intel sells only a few E5s and E7s and the majority of its volumes are cheap E3s, this should tell you that ARK/price list prices are pretty much non-representative of reality

You got that right, but list prices increasing will likely mean the prices customers are paying for are increasing proportionally too.

Look at the model numbers. There's "E7 8870" and "E7 8870 v2". But while E7 8870 is the highest chip, E7 8870 v2 is not, E7 8890 v2 is. While model numbers indicate E7 8870 v2 is the replacement, best-to-best should mean E7 8890 v2 is. However they charge 50% more for that. They are slyly "upselling" the best chips.

At the high-end, these are "RISC replacement" chips. People will pay for it because CPU prices take up very small portion. POWER chips for example are even more expensive, meaning Intel has more room to increased prices should they catch up to the absolute performance of POWER chips in the future. And they have no cheaper competitors.

5 or 10 years ago, sure, but now (and specially in 5 years) ARM and Google could replace MS/Intel if they go crazy with the prices for the lower end IMO

The fact that Intel uses much lower performing(compared to ARM competition) CPU/Graphics Bay Trail and Cherry Trail chips to successfully fend off ARM entering the PC market is an indication for me that at least in the near future, that's not true. Something will have to change must more drastically, like Samsung buying AMD to compete with Intel with their 14nm fabs in x86 as an example.

 

[Arachnotronic]

You got that right, but list prices increasing will likely mean the prices customers are paying for are increasing proportionally too.

Look at the model numbers. There's "E7 8870" and "E7 8870 v2". But while E7 8870 is the highest chip, E7 8870 v2 is not, E7 8890 v2 is. While model numbers indicate E7 8870 v2 is the replacement, best-to-best should mean E7 8890 v2 is. However they charge 50% more for that. They are slyly "upselling" the best chips.

At the high-end, these are "RISC replacement" chips. People will pay for it because CPU prices take up very small portion. POWER chips for example are even more expensive, meaning Intel has more room to increased prices should they catch up to the absolute performance of POWER chips in the future. And they have no cheaper competitors.

SKU to SKU, Intel delivers reasonable performance improvements (they say they target 20% SKU to SKU per gen), and if you need even more performance -- and you're willing to pay -- Intel will sell you something better and more expensive. Intel is betting on higher end SKUs leading to TCO benefits for the customer that far outweigh the increased chip prices.

Seems like a solid strategy on Intel's part.

 

[IntelUser2000]

Seems like a solid strategy on Intel's part.

I bet you they wouldn't do this if they had $/performance competition.

This would be the only way to introduce increased price on consumers without them "rebelling" because its radical, like the $160 Celeron.

For me, I think segmenting their iGPUs into GTx parts is their way of doing that. I think its ridiculous how much they charge for 20% extra graphics when 20% in graphics at this low of a performance level won't do anything.

 

[Arachnotronic]

I bet you they wouldn't do this if they had $/performance competition.

Competition obviously drives prices lower, but remember that this market is very low volume, so in order to make designing a chip -- especially a very large die chip with all of the complexities and reliability requirements of a server part -- a worthwhile venture, you need to have high ASPs to turn a reasonable profit.

For me, I think segmenting their iGPUs into GTx parts is their way of doing that. I think its ridiculous how much they charge for 20% extra graphics when 20% in graphics at this low of a performance level won't do anything.

20% extra graphics really balloons the die size. GT2 Broadwell is 82mm^2 while GT3 is 133mm^2...obviously those increased mfg costs will be passed on to the buyer...

What Intel needs to do is to make sure it is delivering good value for the extra money paid. For these big iGPU parts, Intel often pitches that you don't need the extra GDDR memory, board space, etc. associated w/ a dGPU in a mobile device. However, I don't think the performance of Intel's iGPUs are quite there to justify the kinds of premiums that Intel hopes to make with the Broadwell generation.

I think with Skylake GT4e we will see a step function increase in the value proposition of Iris Pro for high performance laptops and all-in-one PCs, and I think even in Ultrabooks GT3e + Skylake could be a pretty powerful combination. I know a MBA + Retina Display + Skylake GT3e 15W would be something I would be very hard pressed to pass up if I were in the market for a new laptop.

 

[cbn]

They won't be affected as long as Windows and applications remain firmly x86 territory.

That Continuum/Universal apps criteria MS is establishing for desktop mode for tablet and phones in Windows 10 could be the beginning of breaking that.

However, I think if Intel can provide a compelling chip then maybe MS will be forced to allow use of traditional desktop apps for phones and tablets.

But then there is the conflict with atom and Core M to consider. (ie, if Core M might be holding atom development back)

 

[cbn]

20% extra graphics really balloons the die size. GT2 Broadwell is 82mm^2 while GT3 is 133mm^2...obviously those increased mfg costs will be passed on to the buyer...

I think as the competition heats up Intel should reconsider its die size for the non GTe chips. (eg, I think GT2 is kinda of stuck in the middle where it really doesn't have a price or performance advantage).

Instead, maybe a just adequate iGPU (say GT1 specific die) and something really big like a GT3e or GT4e, but nothing in the middle. Then make four thread processors more accessible.

 

[cbn]

At the high-end, these are "RISC replacement" chips. People will pay for it because CPU prices take up very small portion. POWER chips for example are even more expensive, meaning Intel has more room to increased prices should they catch up to the absolute performance of POWER chips in the future. And they have no cheaper competitors.

The fact that IBM Power occupies the high end makes me wonder if that 165W 28C leak we got for Purely was just a smoke screen for what Intel is really developing. This especially considering what IBM is doing to open up Power via licensing.

Besides that it just makes sense to me that Intel could advance the state of technology more than what 28C would deliver.

 

[ShintaiDK]

AMD entered too late.. Well it was good to see a competition, but this is ending and time to see the death of AMD. Intel won this time.

Btw, once AMD dies, Intel will triplicate their product prices the next generation after that. Celerons at 160 dollars and Pentiums at 240 dollars... Yeah.

Intel isnt going to change prices a single bit from today if you exclude inflation. If they did they will earn less. At tripple the prices Intel would be bankrupt within a year.

CPUs is not some kind of static demand goods.

Same reason why innovation wont slow.

 

[ShintaiDK]

The Skylake-E is hardly going to be the Zen's main competitor.

In the server segment that AMD is betting everything on with Zen it is.

 

[CHADBOGA]

In the server segment that AMD is betting everything on with Zen it is.

Yeah but Arachnotronic was surely referring to desktop.

Zen will rise or fall on how it does against a i7-6770k, not the E version of Skylake which is just a niche.

 

[AtenRa]

Yeah but Arachnotronic was surely referring to desktop.

Zen will rise or fall on how it does against a i7-6770k, not the E version of Skylake which is just a niche.

ZEN will compete against socket 2011 segment, it doesnt have an iGPU and it will have more cores than i7-6770K.

ZEN APUs in 2017 will compete in the same segment as Socket 1151.

If ZEN will not be able to compete against Intel Socket 2011 then its a no go for the server from the start.