AMD Wants To Stop Being Known As The “Cheaper Solution”

[OverVolt]

I don't care about CPU speed anymore. I've turned into an AMD fanboy. Its weird.

 

[cbn]

focus on having more cores in the same space. Their only recent server success was with Magny Cours years back at the time having a great per-socket core count. But they went nowhere with it and lost what they had gained quickly.

I got to thinking about how that kind of strategy could affect AMD's cpu core design for high core count servers: If AMD's TDP per socket stays the same as Intel having an inferior process node will likely force AMD into a smaller core. This in order to keep efficiency up.

However, if AMD targets a higher TDP per socket (say the level of Power8 with turbo enabled ~250 watts) then they will probably keep IPC up. So that is what I expect them to do for something like a 32 core processor.

And Looking at the current Intel E5 Xeon line-up Intel does charge more money for the higher clocked, higher TDP version of the 12 core chip:

12/24T E5 Xeon: $1329 65 watt 12/24T E2650L v3, 1.8 Ghz base clock and 2.5 Ghz turbo --> http://ark.intel.com/products/81903/Intel-Xeon-Processor-E5-2650L-v3-30M-Cache-1_80-GHz

12/24T E5 Xeon: $2094 135 watt 12/24T E2690 v3, 2.6 Ghz base clock with 3.5 Ghz turbo --> http://ark.intel.com/products/81713/Intel-Xeon-Processor-E5-2690-v3-30M-Cache-2_60-GHz

(Specs according to Intel Ark on those two E5 Xeons I linked above the same. The only difference is price, clockspeed and TDP)

So higher TDP and higher clocks can result in a more expensive server chip.

 

[Headfoot]

I got to thinking about how that kind of strategy could affect their cpu core design for high core count servers: If AMD's TDP per socket stays the same as Intel having an inferior process node will likely force AMD into a smaller core. This in order to keep efficiency up.

However, if AMD targets a higher TDP per socket (say the level of Power8 with turbo enabled ~250 watts) then they will probably field a relatively big core design. So that is what I expect them to do for the really high core count server chips they sell.

And Looking at the current Intel E5 Xeon line-up Intel does charge more money for the higher clocked, higher TDP version of the 12 core chip:

12/24T E5 Xeon: $1329 65 watt 12/24T E2650L v3 --> http://ark.intel.com/products/81903/Intel-Xeon-Processor-E5-2650L-v3-30M-Cache-1_80-GHz

12/24T E5 Xeon: $2094 135 watt 12/24T E2690 v3 --> http://ark.intel.com/products/81713/Intel-Xeon-Processor-E5-2690-v3-30M-Cache-2_60-GHz

(Specs according to Intel Ark on those two E5 Xeons I linked above the same. The only difference is price, clockspeed and TDP)

TBH I think AMD's only chance is in higher core count with okay single thread to get some of those big server dollars. They haven't shown leadership in IPC design or IPC R&D in years. However, they have a ton of high-density design know how and IP from GPU design and their latest APUs. Getting more cores into the same space while not completely dropping IPC as a design goal a-la Bulldozer would provide them with a niche. And getting a niche is success for AMD, not winning the whole market. I doubt that will ever happen. But to have a $2-3 billion niche like IBM's Power does would be enough. While I highly doubt AMD could ever design a higher IPC CPU than Intel, I don't doubt they could design a much denser CPU than Intel, if they prioritize for it. Although Intel's 14nm designs have had incredibly good density improvements so this window of opportunity may already be closing...

The key is they have to hit a certain minimum single thread performance level before pushing more cores, which was bulldozers ultimate failing.

 

[Phynaz]

Q4 2014 Enterprise Embedded and Semi-Custom
Revenue = 577M
Income = 109M

Q1 2015 Enterprise Embedded and Semi-Custom
Revenue = 498M
Income = 45M

With more Semi-Custom deals in 2016 and increase of Enterprise and Embedded Revenue from ZEN servers and new SoCs APUs, i wouldn't be surprised to see the EESC group to have more than 60% of the entire AMD Revenue and largely impact the profitability of the company.

Considering how accurate you have been predicting AMD profitability, my post still stands.

 

[cbn]

Getting more cores into the same space while not completely dropping IPC as a design goal a-la Bulldozer would provide them with a niche.

With 250 watts for a 32 core design that gives AMD 7.8 watts per core.

At 150 watts for a 32 core design that gives AMD 4.7 watts per core.

So I think having 250 watts as a higher end target lets AMD design a core with a higher IPC than what they would have if forced to work with only 150 watts were available.

 

[poofyhairguy]

AMD will have a very competitive x86 CPU when Apple buys them for that licence.

 

[AtenRa]

Considering how accurate you have been predicting AMD profitability, my post still stands.

Really ???

As discussed in the video forum here http://forums.anandtech.com/showthread.php?t=2256727, AMD warned on revenue yesterday. So let's have a little fun for a week. I'll have a prize for the person that comes closest. In the event of a tie, first poster wins.

So what are your predictions for GAAP income (not revenue)?

Here's mine:

Net loss of $300M

and my prediction,

Ok lets see,

Net Income 97M

AMD Q2 2012
Net income $37 million, earnings per share $0.05, operating income $77 million

 

[TheELF]

What part of my post suggested it was 'so easy'? Getting 40% more IPC from Haswell, I think we could all agree that'd likely be quite a challenge for Intel.

Look at same frequency pentium vs i3 ,intel already boosted IPC of their cores by 10 to 40% depending on workload,and they did it years ago.

AMD gives a statement that it will incorporate SMT in every core and that they will boost IPC by 40% and every amd fanboy screams
"finally amd is going to make intel cpus" ,well no,
maybe the boost will be from smt alone,but probably they will boost single-threaded as well but only within their means.

 

[Phynaz]

Really ???



and my prediction,

I'm not going there. Anybody that is interested can search your posts. I'm not that invested.

Noticed you had to find a three year old post from me though.

 

[el etro]

Stop being called the cheaper solution is not stop being offering better value.

What AMD wants to get rid is of their recent image of cheaper and not always efficient products seller.

 

[Tsavo]

AMD will have a very competitive x86 CPU when Apple buys them for that licence.

License is non-transferable.

Intel isn't going to let anyone else have it.

 

[VirtualLarry]

License is non-transferable.

Intel isn't going to let anyone else have it.

If AMD goes, though, will the gov't force Intel to license x86 and associated patents on FRAND terms to any comer?

I believe that this has been enforced before, when certain companies controlled patents essential to certain industries, and their licensing fees were deemed to be "abusive".

 

[GreenChile]

This 40% faster IPC claim from AMD reminds me of another 40% faster claim they made.

To say I am skeptical is putting it lightly.

 

[jpiniero]

BTW, AMD's next set of bonds to pay off are at the end of 2017 ($633M). They have the cash to pay it off now, but perhaps not if they keep losing money? Even if they manage breakeven between now and then, it would eat much of the cash they have left. They could leverage more stuff or sell new bonds of course, but that may not be so easy.

If AMD goes, though, will the gov't force Intel to license x86 and associated patents on FRAND terms to any comer?

Unlikely given the competition that ARM brings.

 

[SlowSpyder]

Look at same frequency pentium vs i3 ,intel already boosted IPC of their cores by 10 to 40% depending on workload,and they did it years ago.

AMD gives a statement that it will incorporate SMT in every core and that they will boost IPC by 40% and every amd fanboy screams
"finally amd is going to make intel cpus" ,well no,
maybe the boost will be from smt alone,but probably they will boost single-threaded as well but only within their means.

I meant in true single threaded performance. But as you pointed out, IPC can be had in round about ways, not only single threaded performance per clock. But I am saying, they probably can get single threaded performance per clock up ~40% from SR / EX. Hell, had they stuck with PhII and tweaked that they'd probably have much higher single threaded IPC.

*edit - Not to mention they're going to have a much newer manufacturing process.

 

[cbn]

Looking back at my example in post #177,

If a 12C/24T chip with 135 watt TDP and 2.6 Ghz base clock/3.5 Ghz turbo sells for ~ 50% more money than a 12C/24T chip with 65 watt TDP and 1.8 Ghz base/2.5 Ghz turbo.....does that also mean a hypothetical 24C/48T with 270 watt TDP and 2.6 Ghz base clock/3.5 Ghz turbo would sell for 50% more money than a hypothetical 24C/48T with 130 watt TDP and 1.8 Ghz base/2.5 Ghz turbo?

My thinking is probably yes (although I do not know how large the niche is) and thus having a higher TDP strategy for very high core count designs might be part of AMD strategy for not wanting to be the cheaper solution.

Problem, of course, for AMD happens when Intel decides to raise TDP on their own very high core count chips. In fact we already saw this happen to small extent with the 18C Xeon E7-8890 v3 vs. 18C E5 2699 v3. The 18C E7 has a 20 watt higher TDP (165 watts vs. 145 watts) and 200 Mhz higher base clock (2.5 Ghz vs 2.3 Ghz) compared to the 18C E5.

 

[mrmt]

BTW, AMD's next set of bonds to pay off are at the end of 2017 ($633M). They have the cash to pay it off now, but perhaps not if they keep losing money? Even if they manage breakeven between now and then, it would eat much of the cash they have left. They could leverage more stuff or sell new bonds of course, but that may not be so easy.

No, they don't. AMD needs a minimum level of $600 million to keep working.

 

[CHADBOGA]

This 40% faster IPC claim from AMD reminds me of another 40% faster claim they made.

To say I am skeptical is putting it lightly.

I forgot about that.

AMD does love their 40%'s. :whiste:

 

[Phynaz]

BTW, AMD's next set of bonds to pay off are at the end of 2017 ($633M). They have the cash to pay it off now, but perhaps not if they keep losing money? Even if they manage breakeven between now and then, it would eat much of the cash they have left. They could leverage more stuff or sell new bonds of course, but that may not be so easy.



Unlikely given the competition that ARM brings.

They do not have the money to pay off that bond. At least not pay it and still have enough money to run the company.

Edit:
Mrmt beat me to it.

 

[Phynaz]

If AMD goes, though, will the gov't force Intel to license x86 and associated patents on FRAND terms to any comer?

I believe that this has been enforced before, when certain companies controlled patents essential to certain industries, and their licensing fees were deemed to be "abusive".

No.

 

[erunion]

Iirc, the only guidance AMD gave on how much cash they'd have left at the end of q2 was more than the minimum but less than the optimum. (somewhere in the 600-1billion range, I think)

But it seems that at this rate they are at risk of being at the minimum by the end of the year.

 

[cbn]

I suspect this is the beginning of the softening up process that ends with the announcement of the defunding/cancellation of K12.

If AMD wants to move HSA into Android it would probably help to have a good ARM CPU for that process.

Unfortunately some of this will depend on AMD's Linux graphics drivers as well as other projects competing for development dollars.

 

[DrMrLordX]

Currently, the only driver AMD has that supports HSA in on Linux (open-source radeon driver).

 

[TheELF]

License is non-transferable.

Intel isn't going to let anyone else have it.

There is no license on x86 up to and including the 80486,all the cpu extensions ,like amd_64 ,avx ,aes and so on are licensed and you can't make a modern cpu without those.
Anyone with enough $$$ can start manufacturing x86 CPUs they just will be junk.

 

[ShintaiDK]

This 40% faster IPC claim from AMD reminds me of another 40% faster claim they made.

To say I am skeptical is putting it lightly.

We heard plenty of big claims from AMD about performance that never arrives the last 10 years. The sole fact that everything comes from simulation and estimates at this points shows its nothing but a pivot to try and fool the last investors that havent figured it out yet.