Article: AMD's Mobile Strategy
By: Vincent Diepeveen (diep.delete@this.xs4all.nl), December 30, 2011 11:17 am
Room: Moderated Discussions
David Kanter (dkanter@realworldtech.com) on 12/30/11 wrote:
---------------------------
>Doug Siebert on 12/20/11 wrote:
>>> Basically your option is a total gamble,
>>> and unless everything went exactly
>>> according to plan, and Intel was not able
>>> to use their process generation and cash
>>> advantage to take back the performance
>>> lead, AMD would be bankrupt before they
>>> could get another design out the door.
>>
>>Is making low-profit bargain bin chips a better strategy? >AMD's products need
>>differentiation from Intel's products that some group of >customers will pay extra
>>for. For mobile processors, a differentiator could be a high-quality low-power
>>H.264 encoder.
>
>Intel already has that, and it's probably better.
>
>>For desktops and workstations, a differentiator could be higher
>>single thread performance rather than more cores per socket (e.g. a 6 GHz dual-core
>>water-cooled processor).
>
>Water cooling costs too much money and most people aren't interested at all. So
>if you had something like this, it might sell to perhaps 10K people per year.
>
>>Another differentiator could be larger caches (e.g. 256
>>MBytes of L4 cache on the processor socket).
>
>Nobody can afford that much cache except for IBM mainframes. Perhaps with 3D integration it's feasible.
>
>>Another differentiator could be enabling
>>3rd-party innovation by making the cache-coherency >protocol on the HyperTransport
>>bus publically available and royalty-free.
>
>How would that help?
>
>>Intel is adding more cores per socket so I think AMD should focus on single thread
>>performance. If Intel was focused on single thread performance, then AMD should
>>focus on more threads per socket. This is more profitable than trying to make a
>>cheaper version of whatever Intel makes, especially when >AMD's version just has a lower price, not a lower cost.
>
>Actually, Intel is much more focused on single threaded performance than AMD. AMD
>is going down the path of more cores...and it doesn't seem to help too much.
I slightly disagree - AMD seems to go down the path of producing a CPU that fulfills all "specifications", yet just isn't a good performer.
It doesn't have 12 logical cores like 2010's gulftown already had, it has just 8 minicores.
It's not having a bigger bandwidth than Sandy Bridge, and it sure has to compete against Sandy Bridge, which it can't do.
It does have on paper however have everything. It has small L1's, it has slow caches, allowing it to clock high, it has 8 minicores, it has it has it has everything except performance.
It's a third world design.
The real question with respect to Bulldozer is of course: how many years delay did they have with it in total, everything counted?
>David
---------------------------
>Doug Siebert on 12/20/11 wrote:
>>> Basically your option is a total gamble,
>>> and unless everything went exactly
>>> according to plan, and Intel was not able
>>> to use their process generation and cash
>>> advantage to take back the performance
>>> lead, AMD would be bankrupt before they
>>> could get another design out the door.
>>
>>Is making low-profit bargain bin chips a better strategy? >AMD's products need
>>differentiation from Intel's products that some group of >customers will pay extra
>>for. For mobile processors, a differentiator could be a high-quality low-power
>>H.264 encoder.
>
>Intel already has that, and it's probably better.
>
>>For desktops and workstations, a differentiator could be higher
>>single thread performance rather than more cores per socket (e.g. a 6 GHz dual-core
>>water-cooled processor).
>
>Water cooling costs too much money and most people aren't interested at all. So
>if you had something like this, it might sell to perhaps 10K people per year.
>
>>Another differentiator could be larger caches (e.g. 256
>>MBytes of L4 cache on the processor socket).
>
>Nobody can afford that much cache except for IBM mainframes. Perhaps with 3D integration it's feasible.
>
>>Another differentiator could be enabling
>>3rd-party innovation by making the cache-coherency >protocol on the HyperTransport
>>bus publically available and royalty-free.
>
>How would that help?
>
>>Intel is adding more cores per socket so I think AMD should focus on single thread
>>performance. If Intel was focused on single thread performance, then AMD should
>>focus on more threads per socket. This is more profitable than trying to make a
>>cheaper version of whatever Intel makes, especially when >AMD's version just has a lower price, not a lower cost.
>
>Actually, Intel is much more focused on single threaded performance than AMD. AMD
>is going down the path of more cores...and it doesn't seem to help too much.
I slightly disagree - AMD seems to go down the path of producing a CPU that fulfills all "specifications", yet just isn't a good performer.
It doesn't have 12 logical cores like 2010's gulftown already had, it has just 8 minicores.
It's not having a bigger bandwidth than Sandy Bridge, and it sure has to compete against Sandy Bridge, which it can't do.
It does have on paper however have everything. It has small L1's, it has slow caches, allowing it to clock high, it has 8 minicores, it has it has it has everything except performance.
It's a third world design.
The real question with respect to Bulldozer is of course: how many years delay did they have with it in total, everything counted?
>David