AMD will launch AM4 platform in March 2016 says industry source

[AtenRa]

Who gona pair a G3258 with a high end card? at most 250X/360/GTX750/GTX750TI/GTX950 maybe, its stupid to pair a
$300 card with a $70 cpu.

Open Youtube, search for g3258 + GTX970 or R9 290/390

Also, G3258 stutters with slower cards like 960, 760, 770, 280, 280X, 285, 380 etc

 

[cbn]

Open Youtube, search for g3258 + GTX970 or R9 290/390

Do you have an example that is obviously not someone benchmarking or testing?

 

[AtenRa]

Do you have an example that is obviously not someone benchmarking or testing?

What do you mean ?

 

[cbn]

Here are the AMD APU vs Core i3 desktop prices from January 2014:

http://forums.anandtech.com/showpost.php?p=35947650&postcount=746

Using Newegg prices are a comparison for pre-built desktops:

http://www.newegg.com/Product/Produc...el Core i3 (Core i3 starts @ $399)

http://www.newegg.com/Product/Produc...8-Series APU (A8-6500 APU starts @ $479)

http://www.newegg.com/Product/Produc...0-Series APU (A10-6700 APU starts @ $499)

Using the Current Fry's sale for Pre-builts here is what I am coming up with:

http://www.frys-electronics-ads.com/...4130-Processor (Core i3 for $348)

http://www.frys-electronics-ads.com/...6500-Processor (A8-6500 for $448)

Now granted the Core i3 systems come with 4GB RAM vs. 8GB for the AMD systems, but that still a pretty big disparity in pricing IMO.

^^^^ Notice how Core i3 desktops (despite a rather high processor list price) are still cheaper than both the A8 and A10 desktops.

Now here are the current price trends (2 years later):

Here are the listings for new (not refurbished) A8 desktops (some of these are A8-6410 cat core processors):

http://www.newegg.com/Product/Produ...14&IsNodeId=1&bop=And&Order=PRICE&PageSize=30

(Kaveri starts @ $399.99)

Here are the listings for new (not refurbished) A10 desktops

http://www.newegg.com/Product/Produ...14&IsNodeId=1&bop=And&Order=PRICE&PageSize=30

(Kaveri starts @ $419.99)

Here are the listings for new (not refurbished) Core i3 desktops

http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=100019096 4814 600014733&IsNodeId=1

(Core i3 4170 starts @ $329.99)

Notice how the Core i3 desktops are still significantly cheaper than the A10 and A8 desktops.

With that info noted, what do you think should be done?

1.) Should AMD keep on fighting on Desktop with A10 and A8 Bristol Ridge APUs?

2.) Or do you think AMD would be better off focusing all 4C/512sp and 4C/384sp Bristol Ridge into Mobile (particularly the 35W category)?

3.) Or perhaps a middle of the road approach where all 512sp dies are allocated to mobile, but 4C/384sp is allowed on desktop?

Personally, I am thinking #2 is the best option. Then maybe at some later time (perhaps when Zen APU launches) there could be some left over 4C/384sp and 4C/512sp Bristol Ridge made available on AM4?

The pricing isn't up to AMD though, is it? This smells like the OEMs are (at Intel's behest?) screwing AMD over.

I build computers for a living. The A10 is the same price as an i3 and the A8 is cheaper, and our prices for customers reflect this. Sounds like a more subtle market-corruption tactic of Intel's...

I've been wondering how much OEM's fixed costs are affecting the final price desktop price because AMD has lower volume?

And same goes for the laptops. Intel sells more chips so the cost an OEM incurs for developing and maintaining single laptop design gets spread over more units.

With that mentioned, BGA socket FP4 (that Bristol Ridge will use) also uses Stoney Ridge (2C Excavator, 192sp GCN 1.2, single channel DDR4 (and DDR3?)). The question is whether or not AMD (or OEMs) would offer Stoney Ridge as 35W or not?

If AMD had more 35W laptop processors in the form of Stoney Ridge (@ 35W) adding to the 4C + 512sp Bristol Ridge and 4C + 384sp Bristol Ridge 35W totals I figure that would amount to a pretty decent volume for 35W FP4.

Another question is how would people react to a 35W Stoney Ridge? (My opinion is that if I only have two cores, I want them to be fast ones.....and certainly there is room in that 35W TDP to clock both the cpu cores and 192sp GCN 1.2 iGPU fairly high.) Cost should be low as well considering 1M Excavator 192sp/single channel should be around the same size as the quad cat core 128sp/single channel chips.

P.S. For the 15W BGA FP4, AMD could just keep on using Carrizo. With the low TDP there is no reason to go beyond dual channel DDR3 1600 anyway.

So with that scheme I outlined above:

Carrizo would be high cost 15W FP4
Carrizo-L would be low cost 15W FP4 (remember this is quad cat core/128sp/single channel)

Bristol Ridge would be high cost 35W FP4
Stoney Ridge would be low cost 35W FP4 (dual core Excavator/192sp/single channel that I estimate to have roughly the same die size as Carrizo-L)

 

[Azuma Hazuki]

Then you should make AMD aware of this, so they can get some free billions in a lawsuit.

Else its time to take the tinfoil hat off.

And just how in bloody buggery would anyone who isn't hacking around in Intel's and/or their OEM partners' finances (or the bigwigs' personal emails...) be able to prove this?

Given Intel's past behavior, I don't think this is at all tinfoil to suspect. And you're a jerk for immediately jumping to that conclusion. If you wonder why you are sometimes accused of being an Intel shill, this kind of corrosive idiocy is why.


Name calling is not allowed in the technical forums.
Markfw900 Anandtech moderator

 

[Arachnotronic]

And just how in bloody buggery would anyone who isn't hacking around in Intel's and/or their OEM partners' finances (or the bigwigs' personal emails...) be able to prove this?

You don't think that Intel or one of its OEM partners will at some point piss somebody off, who then goes ahead and spills the beans on it, sending many executives to prison all the while collecting a big fat "whistleblower" check from the government? You're funny

 

[AtenRa]

The pricing isn't up to AMD though, is it? This smells like the OEMs are (at Intel's behest?) screwing AMD over.

I build computers for a living. The A10 is the same price as an i3 and the A8 is cheaper, and our prices for customers reflect this. Sounds like a more subtle market-corruption tactic of Intel's...

Reasons that may attribute to this,

1. OEM fierce competition on Intel products with low margins.
OEMs continuously battle for selling Intel products, they sell as cheap as possible to gain market share that they solely relying on Intel rebate programs to gain profit.
On AMD products they solely relying on selling at higher margins, thus making the product more expensive to the consumer.

2. Higher volumes on Intel products equals lower BOM and faster ROI.
Higher volumes means lower cost not only of the CPU but for many other components, like motherboards etc. Also since they sell higher volumes of Intel products (PCs, Laptops, AIOs etc) they have a faster ROI at lower margins. Since most products have a 12 month cycle, selling less AMD products mean they have to sell at higher margins in order to get a ROI at the same 12 months period, making the selling price of the product higher for the Consumer.

3. Better iGPU performance as an added value.
OEMs always added higher margins for their products featuring a dGPU over Intels iGPU. An AMD APU having far better iGPU performance and software ecosystem (for games and professional software), gives them an excuse to sell their AMD product as a premium device and thus raise the selling price (higher margins) making the product even more expensive to the consumer but offering higher profit and faster ROI for the OEM.

I believe those are the top 3 reasons why pre-build OEM PCs/Laptops and AIOs are not cheaper vs Intel counterparts although AMD APU retail prices are cheaper or equal to Intels counterparts.

 

[Azuma Hazuki]

You don't think that Intel or one of its OEM partners will at some point piss somebody off, who then goes ahead and spills the beans on it, sending many executives to prison all the while collecting a big fat "whistleblower" check from the government? You're funny

What, and derail the giant incestuous gravy train that is the USA's idea of a free market? Surely you jest. Anyone who tried that would end up committing suicide by stabbing himself in the back about 80 times with a knife that isn't found at the scene...

 

[mrmt]

And just how in bloody buggery would anyone who isn't hacking around in Intel's and/or their OEM partners' finances (or the bigwigs' personal emails...) be able to prove this?

An i3 die is half the die size of AMD APUs and Intel yields are above the average in the industry, so for a given return rate Intel needs this ASP to be much lower than AMD can think of. Given that Intel SKUs will have far greater numbers than anything AMD and as a consequence some costs (marketing material, supporting personnel) benefits from the economies of scale kicking in, and as a last feature channel distributors should be having to compensate the lower AMD volumes with higher margin per unit.

Given these features I would be surprised if Intel prices weren't lower than AMD's.

 

[AtenRa]

An i3 die is half the die size of AMD APUs and Intel yields are above the average in the industry, so for a given return rate Intel needs this ASP to be much lower than AMD can think of. Given that Intel SKUs will have far greater numbers than anything AMD and as a consequence some costs (marketing material, supporting personnel) benefits from the economies of scale kicking in, and as a last feature channel distributors should be having to compensate the lower AMD volumes with higher margin per unit.

Given these features I would be surprised if Intel prices weren't lower than AMD's.

I know you like to compare die sizes but a 14nm Core i3 even at half the die size vs 28nm APU is not cheaper.

 

[ShintaiDK]

I know you like to compare die sizes but a 14nm Core i3 even at half the die size vs 28nm APU is not cheaper.

Keep telling yourself that

Intel got lower cost at 14nm.

 

[Puffnstuff]

I know you like to compare die sizes but a 14nm Core i3 even at half the die size vs 28nm APU is not cheaper.

Apparently someone doesn't understand the relationship between die size and dies per wafer. The more dies you can place on a given wafer the lower the cost will be and that is just reality not to mention good business.

 

[Fjodor2001]

Apparently someone doesn't understand the relationship between die size and dies per wafer. The more dies you can place on a given wafer the lower the cost will be and that is just reality not to mention good business.

The cost of the wafer also comes into play. Wafers on later process nodes are more expensive.

What determines the price is the total of all this, wafer price and die area on that wafer.

Is there anyone that has all the data to prove that a chip on 14 nm is cheaper than on 28 nm, or vice versa? And doesn't it differ depending on the chip's die area (e.g. since the yield for large die area chips potentially is lower on later nodes, until yields have improved)?

 

[IntelUser2000]

Keep telling yourself that

Intel got lower cost at 14nm.

No they don't, for all intents and purposes.

In theory they do. Not in actuality.

Reason #1: Their yields suck now, which is why you see $400-500 6700K chips. Also if we take PCWatch article as true(and they are pretty credible), Intel knew that the yield was going to be underwhelming so they had to architect Skylake for longer pipeline stages. Hence the menial IPC gain despite a wide uArch.

Low yields mean higher cost, and not being able to max out process features. For example, density.

#2: Transistor vs. Transistor Intel might be denser than TSMC, GF, Samsung, etc. None of Intel designs reflect that. Atom was barely competitive when on 22nm and others were on 28nm. Now ARM is closer to Core than Atom, leaving the latter behind a trail of dust.

Atom vs ARM core might be pretty close in size now, but the mediocre Airmont is at 80mm2 die while ARM chips are at 100mm2 with 2x the performance.

As long as Intel keeps making chips that are uncompetitive in price/perf and mm/perf things like "process density" can be ignored. It doesn't matter. Theory = good, Practice = suck

 

[ShintaiDK]

6700K is in broad supply again. And it was only retailer markup.

But do you know if its yield or capacity? 14nm requires a lot more fab time.

You seem to confuse transistor density with different libraries and layouts. High speed designs will never be very dense.

 

[IntelUser2000]

By the way some details about next APU platform based on Zen is out. It says it features HBM memory with 128GB/s bandwidth. If it becomes a cheap alternative to Iris Pro of that generation, there might be a good uptake there.

And honestly, AMD drivers are way better than Intel for gaming. Regardless of how Intel wants to have people view them, for graphics all they care about is making GPU the next "HD Audio".

I don't have fantasies about Zen beating Intel as a CPU. Make a truly $150 discrete GPU replacement APU and they'd do find as a company and make enough people happy for few years.

 

[mrmt]

No they don't, for all intents and purposes.

In theory they do. Not in actuality.

Reason #1: Their yields suck now, which is why you see $400-500 6700K chips. Also if we take PCWatch article as true(and they are pretty credible), Intel knew that the yield was going to be underwhelming so they had to architect Skylake for longer pipeline stages. Hence the menial IPC gain despite a wide uArch.

Skylake was architected since 2010, do you mean that Intel knew that their yields on a process that would be deployed in 2015 would suck? Do you think that's credible?

 

[AtenRa]

Keep telling yourself that

Intel got lower cost at 14nm.

1: 14nm R&D more than 2X the cost of 28nm

2: 14nm design IC cost is more than 2 or 3 times that of 28nm

3: 14nm needs more masks than 28nm,

4: 14nm yields are lower than what 28nm are today,

1+2+3+4 makes Core i3 not cheaper than 28nm APUs today.

 

[AtenRa]

Apparently someone doesn't understand the relationship between die size and dies per wafer. The more dies you can place on a given wafer the lower the cost will be and that is just reality not to mention good business.

see my post above

 

[frozentundra123456]

By the way some details about next APU platform based on Zen is out. It says it features HBM memory with 128GB/s bandwidth. If it becomes a cheap alternative to Iris Pro of that generation, there might be a good uptake there.

And honestly, AMD drivers are way better than Intel for gaming. Regardless of how Intel wants to have people view them, for graphics all they care about is making GPU the next "HD Audio".

I don't have fantasies about Zen beating Intel as a CPU. Make a truly $150 discrete GPU replacement APU and they'd do find as a company and make enough people happy for few years.

Source for any of these "details"?

 

[ShintaiDK]

1: 14nm R&D more than 2X the cost of 28nm

2: 14nm design IC cost is more than 2 or 3 times that of 28nm

3: 14nm needs more masks than 28nm,

4: 14nm yields are lower than what 28nm are today,

1+2+3+4 makes Core i3 not cheaper than 28nm APUs today.

Feel free to provide the documentation.

because you have some flawed estimates about cost on both nodes. And its been ever so since AMD got a disadvantage. However with 14nm AMD products you tend to forget all those. Specially since they cant afford the designs to remove most of the penalties.

 

[cbn]

By the way some details about next APU platform based on Zen is out. It says it features HBM memory with 128GB/s bandwidth. If it becomes a cheap alternative to Iris Pro of that generation, there might be a good uptake there.

And honestly, AMD drivers are way better than Intel for gaming. Regardless of how Intel wants to have people view them, for graphics all they care about is making GPU the next "HD Audio".

I don't have fantasies about Zen beating Intel as a CPU. Make a truly $150 discrete GPU replacement APU and they'd do find as a company and make enough people happy for few years.

Source for any of these "details"?

http://wccftech.com/xbox-one-may-be-getting-a-new-apu-based-on-amds-polaris-architecture/

 

[cbn]

Reasons that may attribute to this,

1. OEM fierce competition on Intel products with low margins.
OEMs continuously battle for selling Intel products, they sell as cheap as possible to gain market share that they solely relying on Intel rebate programs to gain profit.
On AMD products they solely relying on selling at higher margins, thus making the product more expensive to the consumer.

2. Higher volumes on Intel products equals lower BOM and faster ROI.
Higher volumes means lower cost not only of the CPU but for many other components, like motherboards etc. Also since they sell higher volumes of Intel products (PCs, Laptops, AIOs etc) they have a faster ROI at lower margins. Since most products have a 12 month cycle, selling less AMD products mean they have to sell at higher margins in order to get a ROI at the same 12 months period, making the selling price of the product higher for the Consumer.

3. Better iGPU performance as an added value.
OEMs always added higher margins for their products featuring a dGPU over Intels iGPU. An AMD APU having far better iGPU performance and software ecosystem (for games and professional software), gives them an excuse to sell their AMD product as a premium device and thus raise the selling price (higher margins) making the product even more expensive to the consumer but offering higher profit and faster ROI for the OEM.

I believe those are the top 3 reasons why pre-build OEM PCs/Laptops and AIOs are not cheaper vs Intel counterparts although AMD APU retail prices are cheaper or equal to Intels counterparts.

For #3, I see this as advantage for laptops (and AIO desktops), but not for conventional desktops. (re: With a Core i3 desktop a person can easily add a video card like the 38W PNY GT 730 GDDR5. And the price is definitely low enough that it would still be cheaper than A8 APU desktop. Performance wise Core i3 + GT 730 GDDR5 would be faster as well).

 

[cbn]

Even with A10, Core i3 + GT 730 GDDR5 would be faster.

 

[dark zero]

Skylake was architected since 2010, do you mean that Intel knew that their yields on a process that would be deployed in 2015 would suck? Do you think that's credible?

It was similar than Pentium 4 fiasco. They have years in advantage on their experimental nodes, so they knew that they won't make it. However at difference of the current case, AMD was there. Now there is no competition near Intel.

The 14nm Atom was a big example on how a badly implemented node shrink leads on a massive dissaster and was only saved by contravenue, otherwise it would be easy prey for ARM makers.

If it weren't for DDR4, the True Quad Core i5 for mobile, the HT Pentium M and BLCK OC, Skylake would be less impactful than Sandy Bridge