Spin-Off: AMD & Intel Business Practices

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Phynaz

Lifer
Mar 13, 2006
10,140
819
126
The big problem is that if Intel needs to spend 1B every Quarter or approximately 4B per year for R&D and MG&A in the mobile segment, they will have to generate more than double the amount in Revenue from that group alone in order to make a profit.

That means that, if they will sell each SoC at $25 and earn $10, then they will need to sell 400 million SoCs per year just to get even. And that is with 40% margins, they will need to lower that in order to sell that high Volume. So you see that seams far too fetched to become real even with Intel's deep pockets.
They really need to change their business plan and how they operate in order to be competitive in the Mobile market.

Who says they need to spend this every quarter? I'll bet the R&D spending is front loaded.

Your assumption is also that the knowledge and expertise gained from this research can only be used to sell socs and can't be applied to any other product, ever. This assumption is wrong.

Finally, so what if they have to sell 400M a year? There were 8B ARM devices sold last year. So they need 5% market share to break even, big deal.

Please just stop with straw man building, ok?
 

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
And it shows. I don't see AMD producing anything like what Intel has in development.

Really? I think if anyone is showing poor execution and development it's intel. ;) AMD crushes them on graphics and low power chips all while being a process node behind, without finfets. It takes 14nm finfets to match AMD's 28nm silicon? That doesn't sound too good to me.

And BTW, your slight of hand isn't really fooling anyone... "I don't see AMD producing anything like what intel has in development." How about AMD beating products that intel is producing?
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Really? I think if anyone is showing poor execution and development it's intel. ;) AMD crushes them on graphics and low power chips all while being a process node behind, without finfets. It takes 14nm finfets to match AMD's 28nm silicon? That doesn't sound too good to me.

And BTW, your slight of hand isn't really fooling anyone... "I don't see AMD producing anything like what intel has in development." How about AMD beating products that intel is producing?

You could just as well fit a Haswell-Y into a bulky 11.6'' reference platform and brag about it being faster than Bay Trail (which can be found in lots of small/light/thin 7-10'' tablets right now). There's no power numbers and no tablets for sale yet (AMD expects designs ''over the next 2 quarters''), until a recognized OEM ships an actual competitive tablet based on one of this new chips all they have is another PowerPoint winner. That's already better than last year's A4-1200 crap though. :p
 

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
You could just as well fit a Haswell-Y into a bulky 11.6'' reference platform and brag about it being faster than Bay Trail (which can be found in lots of small/light/thin 7-10'' tablets right now). There's no power numbers and no tablets for sale yet (AMD expects designs ''over the next 2 quarters''), until a recognized OEM ships an actual competitive tablet based on one of this new chips all they have is another PowerPoint winner. That's already better than last year's A4-1200 crap though. :p

Speaking of Haswell, the 2.8W Micro 6700T beats that thing, in graphics, too! That's incredible.
 

Sweepr

Diamond Member
May 12, 2006
5,148
1,142
131
Speaking of Haswell, the 2.8W Micro 6700T beats that thing, in graphics, too! That's incredible.


You mean this? Barely matches one of the slowest Core i3, and of course no CPU numbers (and no CPU/iGPU numbers from the i5 model) to hide their disadvantage. The fact that they are comparing this specific chip to Haswell-Y kinda makes you wonder about the real power numbers and market they are targetting (>11.6'' tablets). Their own slide implies that Bay Trail-T's direct competitors should be the A4-6400T and E1-6200T, with much lower CPU/iGPU clocks than the fastest model that AMD sent for testing (Dual-core 1.4GHz / Quad-core 1.6GHz with 300-350MHz iGPUs vs Quad-core 2.2GHz + 500MHz iGPU). Their single-thread CPU advantage will be gone and their graphics performance advantange will be severely reduced thanks to lower clocks.
 
Last edited:

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
And it shows. I don't see AMD producing anything like what Intel has in development.

So quoting only part of my post just to elude the question and change the goal post through some kind of rethoric.?..

Here the post again :

AMD s entire RD is at most 300m/quarter , how much do you think BT current RD need considering this number.?..

From thoses numbers AMD s small APUs line RD cost is hardly 100m/quarter so it is obvious that out of the 929m Intel lost on BT 150m at most are RD with 50m for management, 100m for the free chips and 600m subsides, that is about 700m poured in subsides for a 5m chips shippement , 140$ per chip shipped including the chip total rebate.

Now you can always insist that it s 929m in RD....:biggrin:
 

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
You mean this? Barely matches one of the slowest Core i3, and of course no CPU numbers (and no CPU/iGPU numbers from the i5 model) to hide their disadvantage. The fact that they are comparing this specific chip to Haswell-Y kinda makes you wonder about the real power numbers and market they are targetting (>11.6'' tablets). Their own slide implies that Bay Trail-T's direct competitors should be the A4-6400T and E1-6200T, with much lower CPU/iGPU clocks than the fastest model that AMD sent for testing (Dual-core 1.4GHz / Quad-core 1.6GHz with 300-350MHz iGPUs vs Quad-core 2.2GHz + 500MHz iGPU). Their single-thread CPU advantage will be gone and their graphics performance advantange will be severely reduced thanks to lower clocks.

A tablet chip with a 2.8W SDP matching a core i3 Haswell sure does put things in perspective. This came out of nowhere, blowing away expectations. That Skin Temperature Aware Power Management is brilliant, integrating ARM PSP, etc. I want one. :) Samsung and Lenova are preparing designs with much more to come.
 

bullzz

Senior member
Jul 12, 2013
404
23
81
drinking AMDs koolaid much. its obvious mullins clearly beats bay trail on GPU. but comparing it to corei3 is a bit too much. CPU perf even on a i3 is high enough that you can notice clearly the difference when compared to bay trail/mullins
 

Phynaz

Lifer
Mar 13, 2006
10,140
819
126
So quoting only part of my post just to elude the question and change the goal post through some kind of rethoric.?..

Here the post again :



From thoses numbers AMD s small APUs line RD cost is hardly 100m/quarter so it is obvious that out of the 929m Intel lost on BT 150m at most are RD with 50m for management, 100m for the free chips and 600m subsides, that is about 700m poured in subsides for a 5m chips shippement , 140$ per chip shipped including the chip total rebate.

Now you can always insist that it s 929m in RD....:biggrin:

So you are saying Intel outright lied in their investor call. You should call the SEC. The rest of us will ignore your made up numbers.
 

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
So you are saying Intel outright lied in their investor call. You should call the SEC. The rest of us will ignore your made up numbers.

They can say what they want but if AMD whole quarterly RD is 300m i guess that they ll have trouble explaining that they need 700m in the same time for a product ouperformed by a couterpart that was granted 100m at most.

That said,

The bill of materials cost for a Broxton tablet will be $20 less than for Bay Trail, Krzanich said. SoFIA, with its greater integration and smaller die size, will cost even less, he said.
http://www.pcworld.com/article/2089421/how-intel-is-buying-a-piece-of-the-tablet-market.html

Seems that the subsides business model is here to stay for a long time if 20$ is what is to be cost saved by each generation...
 

Phynaz

Lifer
Mar 13, 2006
10,140
819
126
They can say what they want but if AMD whole quarterly RD is 300m i guess that they ll have trouble explaining that they need 700m in the same time for a product ouperformed by a couterpart that was granted 100m at most.

What mobile device is AMD releasing in three years that will out perform Intel's mobile device?

Will it have an integrated LTE modem?

Seriously, you like AMD and hate Intel. We get it. But making up false numbers doesn't make any case for you, it does just the opposite.
 

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
What mobile device is AMD releasing in three years that will out perform Intel's mobile device?

Currently their linenup outperform the competition.

Will it have an integrated LTE modem?

4G is an open standard and devellopping such a modem is negligible quantity for a firm like AMD.


Seriously, you like AMD and hate Intel. We get it. But making up false numbers doesn't make any case for you, it does just the opposite.

Of course you had to end with an ad hominem attack but that s not surprising at all, it happens often when one is short of arguments, most funny is that it starts with "seriously"...
 

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
AMD is no longer even pretending they can keep pace with the other big CPU, SoC companies. They are attempting to expand into becoming a design house for integrating their technology with others IP as another revenue stream to stave off completely gutting their R&D. With the console wins being the headliners for that sales pitch.

Imo, even if they did have some funds to buy a few design wins they would be better off using that money to prep for Samsung+GF FinFet.
 
Last edited:

AtenRa

Lifer
Feb 2, 2009
14,000
3,357
136
Who says they need to spend this every quarter? I'll bet the R&D spending is front loaded.

In order to sell more you have to spend more in MG$A. Also, you dont stop to spend on R&D and even if you keep the same R&D amount you have to raise the MG&A amount. But i have assumed they will keep the same spending of R&D + MG&A in every Quarter.

Finally, so what if they have to sell 400M a year? There were 8B ARM devices sold last year. So they need 5% market share to break even, big deal.

Their target for 2014 is 40m units and 80m for 2015. Do you actually believe they will spend less than what they are already spending in R&D and MG&A in 2014 ?? So from the current data it seams they will continue to loose money in 2014 and 2015 and even in 2016 and so on.

Ohh, nobody earns 10$ per chip in that market. Just a reminder ;)
 

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
A tablet chip with a 2.8W SDP matching a core i3 Haswell sure does put things in perspective. This came out of nowhere, blowing away expectations. That Skin Temperature Aware Power Management is brilliant, integrating ARM PSP, etc. I want one. :) Samsung and Lenova are preparing designs with much more to come.

Same microarchitecture and no new process node means the same efficiency. Since Jaguar was already considerably worse than Silvermont, I don't see how Puma is going to meaningfully improve performance. Your 4.5W chip now uses some fancy boost mode to enable higher clock speeds that would otherwise require a 15W+ TDP for Jaguar SKUs, while Silvermont keeps using the same ~2.5W, which is a lot lower with higher clocks, more cores and no throttling. Nice marketing trick, that Qualcomm also uses, but that doesn't really change anything in terms of actual performance (per watt, i.e. sustained performance).
 

NTMBK

Lifer
Nov 14, 2011
10,208
4,940
136
Same microarchitecture and no new process node means the same efficiency. Since Jaguar was already considerably worse than Silvermont, I don't see how Puma is going to meaningfully improve performance. Your 4.5W chip now uses some fancy boost mode to enable higher clock speeds that would otherwise require a 15W+ TDP for Jaguar SKUs, while Silvermont keeps using the same ~2.5W, which is a lot lower with higher clocks, more cores and no throttling. Nice marketing trick, that Qualcomm also uses, but that doesn't really change anything in terms of actual performance (per watt, i.e. sustained performance).

Mullins has moved to GlobalFoundries' gate first 28nm process, while Temash was on TSMC's gate last 28nm process. Puma has been tweaked to reduce leakage- AMD claims 19% leakage reduction. And it now supports lower voltage DDR3L memory, which is more suited to tablets than the standard DDR3 memory used by Temash. And I don't understand why you're disparaging Turbo- turbo core is an essential way to balance single threaded performance, multithreaded performance and power consumption in a power constrained environment. Intel was a pioneer in turbo technology, and Temash/Kabini's awful Turbo was one of its main problems.

Anyway, as always, we should wait for some independently verified measurements before we get too carried away with the arguments.
 

Erenhardt

Diamond Member
Dec 1, 2012
3,251
105
101
Same microarchitecture and no new process node means the same efficiency. Since Jaguar was already considerably worse than Silvermont, I don't see how Puma is going to meaningfully improve performance. Your 4.5W chip now uses some fancy boost mode to enable higher clock speeds that would otherwise require a 15W+ TDP for Jaguar SKUs, while Silvermont keeps using the same ~2.5W, which is a lot lower with higher clocks, more cores and no throttling. Nice marketing trick, that Qualcomm also uses, but that doesn't really change anything in terms of actual performance (per watt, i.e. sustained performance).

Wasn't BT first to share TDP across the chip? Boost CPU when IGP is idle, and vice versa. AMD was called out for not using boost and being slower in single threaded tasks, because chip was running at stock freq.

Here comes AMD with better boost and suddenly it is cheating, faking etc.

We've seen performance and it beats intel by a wide margin with the expection of 3dmark physics benchmark. So if anyone is playing that, they can stick with intel ;)

Anyway. I think it will cost intel direly to pay for mullins design wins if you know what I mean :cool:
 

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
Mullins has moved to GlobalFoundries' gate first 28nm process, while Temash was on TSMC's gate last 28nm process. Puma has been tweaked to reduce leakage- AMD claims 19% leakage reduction. And it now supports lower voltage DDR3L memory, which is more suited to tablets than the standard DDR3 memory used by Temash.
Those are necessary changes for tablet chips, and as Anand showed, idle power still isn't as good as other SoCs like the Snapdragon 600.

And I don't understand why you're disparaging Turbo- turbo core is an essential way to balance single threaded performance, multithreaded performance and power consumption in a power constrained environment. Intel was a pioneer in turbo technology, and Temash/Kabini's awful Turbo was one of its main problems.
Turbo is great, but I'm very skeptical about it in mobile parts, where companies like Qualcomm only mention "2.5GHz clock speed!!" But in practice, a Snapdragon 800 clocks back to 0.5GHz within a ridiculously short time.

Anyway, as always, we should wait for some independently verified measurements before we get too carried away with the arguments.
Exactly, I'm waiting for a review to see if AMD really succeeded in doubling the clock speed, cache size and core count while halving TDP (A4-1250 vs A10 Micro-6700T), meaning an octupling in efficiency. I think not.
 

witeken

Diamond Member
Dec 25, 2013
3,899
193
106
Wasn't BT first to share TDP across the chip? Boost CPU when IGP is idle, and vice versa. AMD was called out for not using boost and being slower in single threaded tasks, because chip was running at stock freq.

Here comes AMD with better boost and suddenly it is cheating, faking etc.
I'm not saying that AMD is cheating. Quadcore Silvermont uses ~2.5W with all cores at 2.4GHz. Silvermont doesn't need to clock back like other SoCs, and Intel mentions both turbo and stock frequencies. So I'm wondering how much real performance increase we'll get. Since there are not a lot of improvement except for leakage, I don't think they'll be as big as AMD is promising, based on core count, frequency TDP improvements of the new SKUs.

Edit:

For example, 4.5W A10 Micro-6700T vs 15W A6-6310 only sees a 0.2GHz and 0.3GHz decrease in clock speed for the CPU and GPU respectively.
 
Last edited:

Abwx

Lifer
Apr 2, 2011
10,847
3,297
136
Those are necessary changes for tablet chips, and as Anand showed, idle power still isn't as good as other SoCs like the Snapdragon 600.

Anand quoted Beema 15W APU s idle power as being Mullins s.

Screen-Shot-2014-04-29-at-1.06.11-AM_575px.jpg


http://www.anandtech.com/show/7974/...hitecture-a10-micro-6700t-performance-preview

Quadcore Silvermont uses ~2.5W with all cores at 2.4GHz.

That s not true at all but you keep claiming it as real, power usage at 2.4 is much higher , you are quoting its SDP as being its TDP at 2.4.....
 
Last edited: