AMD's 2008 K8 Roadmap

[Zenoth]

In my humble opinion, the worse AMD move so far has been to suddenly cut their interests with Socket 939, which at the time was still (and still is today) found on the majority of AMD-based computers (of course it was, AM2 wasn't there, but that's the point). When AM2 arrived, AMD basically shot themselves in their on feet, because they wanted to ONLY support AM2 from then on and completely forget about S939. If they supported BOTH they would have made more money.

I myself WOULD have upgraded IF my S939 could have supported a newer AMD model (for example the 6000+). But no ... AMD had to tell to themselves "alright guys, let's lose around 90% of our own Desktop market share by moving to AM2 exclusively, and let's try to build ourselves a solid competitive economy on a 10% share, and then let's expect to see all the S939 consumers currently using our processors moving to AM2 with us". It was a stupid move at best. In my book the sudden arrival of AM2 at the price of S939's death was their biggest mistake to date.

 

[Viditor]

Originally posted by: Acanthus

Originally posted by: Viditor

Originally posted by: Acanthus

TSMC has not been fabbing any CPUs for AMD.

When they do outsource, they outsource to IBM.

IBM doesn't do outsourcing of CPUs for AMD either...only Chartered does.

IBMs Fishkill facility has been making chips for AMD for years...

http://arstechnica.com/news.ar...rication-business.html

Currently, Chartered Semiconductor handles some of AMD's manufacturing, and AMD told Ars Technica last fall that its plans called for Chartered to eventually manufacture CPUs on a 65nm process. AMD also has a long-standing partnership with IBM under which AMD gets to use Big Blue's East Fishkill, NY, plant for R&D and manufacturing.

I didnt know they also outsourced to Chartered.

East Fishkill is only for R&D. That's where the 65nm, 45nm, and 32nm is developed.

 

[coldpower27]

Originally posted by: Zenoth
In my humble opinion, the worse AMD move so far has been to suddenly cut their interests with Socket 939, which at the time was still (and still is today) found on the majority of AMD-based computers (of course it was, AM2 wasn't there, but that's the point). When AM2 arrived, AMD basically shot themselves in their on feet, because they wanted to ONLY support AM2 from then on and completely forget about S939. If they supported BOTH they would have made more money.

I myself WOULD have upgraded IF my S939 could have supported a newer AMD model (for example the 6000+). But no ... AMD had to tell to themselves "alright guys, let's lose around 90% of our own Desktop market share by moving to AM2 exclusively, and let's try to build ourselves a solid competitive economy on a 10% share, and then let's expect to see all the S939 consumers currently using our processors moving to AM2 with us". It was a stupid move at best. In my book the sudden arrival of AM2 at the price of S939's death was their biggest mistake to date.

That is one of the drawbacks of the IMC, it required a whole platform change to support a new memory type. AMD didn't plan far enough ahead for that, it would have been better if the memory controller for the Dual Channel Athlon 64's was designed with both DDR and DDR2 circuitry, that way it could support either memory type, and would make a more substantial transition then just Dual Channel from Socket 754. It would also make supporting both DDR and DDR2 an option at least.

Does supporting 939 make sense from the OEM perspective though, I know it makes sense of the DIY crowd, but if AMD wants to break into the OEM space it had to make things simpler and that meant DDR2 only.

It's hard to say if the sales from DIY would compensate for the development costs of having both a DDR and DDR2 based version of the same SKU.

 

[taltamir]

The sales of "white box" computers.. that is, by small stores who assemble it themselves from off the shelf components, are several times the sales of all OEM combined...

And those small stores are perfectly capable of providing drop in upgrade services cheap. There will defintely be a huge demand for it. Actually I think AMD did GOOD by leaving the 939 behind... it made sense at the time... however now with the flop of their phenoms they need to sell where they are the strongest... that is, a more bang for the buck then AMD for owners of existing AM2 or 939 motherboards. Which means a brisbane 939 would be a perfect AMD product right now.

 

[SexyK]

Originally posted by: taltamir
The sales of "white box" computers.. that is, by small stores who assemble it themselves from off the shelf components, are several times the sales of all OEM combined...

Dude, I would love to see some stats on that. No chance small shops building from off the shelf components outsells Dell, let alone all OEM's combined.

 

[Viditor]

Originally posted by: SexyK

Originally posted by: taltamir
The sales of "white box" computers.. that is, by small stores who assemble it themselves from off the shelf components, are several times the sales of all OEM combined...

Dude, I would love to see some stats on that. No chance small shops building from off the shelf components outsells Dell, let alone all OEM's combined.

Agreed...the Channel (source for all white box and DIY CPUs) is only a small fraction of OEM sales. Dell's sales alone are significantly larger than the entire channel...

 

[taltamir]

not from what I have read.

Anyways... I took the liberty of translating AMDs roadmaps from PR to english:

2006: Hope that intel will continue to use the useless netburst architecture (worse performance per clock on P4 then the P3) instead of something more competitive.
2007: Hope that for some reason the aging X2 cores will still stay competative with intels new offering, assuming that intel will keep on sucking like it has in the past.
2008: Hope that people will buy the monolithic monster called phenom even though it fails in every aspect. Suggest that they couple it with an inferior AMD video card and AMD chipset for a total suckage experience. Try to survive till 2009.
2009: Hope that Fusion will beat the pants off of everything else. Hope that nvidia will die as it is locked out by AMD and Intel as each fortifies its own platform. Hope that the intel video card suck.

AMDs plans seem to hinge on the opponents sucking and messing up every single time while they would get lucky. While in fact it is AMD who keeps on making mistakes.
A proper plan should accommodate for both bad or good luck.

 

[Viditor]

Originally posted by: taltamir
not from what I have read.

Anyways... I took the liberty of translating AMDs roadmaps from PR to english:

2006: Hope that intel will continue to use the useless netburst architecture (worse performance per clock on P4 then the P3) instead of something more competitive.
2007: Hope that for some reason the aging X2 cores will still stay competative with intels new offering, assuming that intel will keep on sucking like it has in the past.
2008: Hope that people will buy the monolithic monster called phenom even though it fails in every aspect. Suggest that they couple it with an inferior AMD video card and AMD chipset for a total suckage experience. Try to survive till 2009.
2009: Hope that Fusion will beat the pants off of everything else. Hope that nvidia will die as it is locked out by AMD and Intel as each fortifies its own platform. Hope that the intel video card suck.

AMDs plans seem to hinge on the opponents sucking and messing up every single time while they would get lucky. While in fact it is AMD who keeps on making mistakes.
A proper plan should accommodate for both bad or good luck.

Where did you read that?

As to your "translation", are you familiar with the term "FUD"?

 

[taltamir]

where did you read that dell sells more then the whiteboxes?

I recall reading specific data on the issue. I don't recall where exactly.

You on the other hand assume the opposite with absolutely no data, recalled or otherwise, to corroborate your point.

So unless you claim to have actually READ somewhere that I am wrong (meaning we are both working off of memory and written info is required) or actually have sometimes substantial I am gonna keep assuming I am right on this.

 

[CTho9305]

Originally posted by: jiffylube1024
If only they released Brisbane 65nm cores for 939 all along!!!!! But no, they wanted to force AM2 down our throats, and once Core2 was released it was all over...

FWIW, it's a lot of work to port a processor across process nodes nowadays, and some of the uglier things are parts like memory interfaces because they're very "analogy" - they're more sensitive to the things that get worse on smaller process nodes (e.g. variation). I'm not speaking for any companies.

 

[Viditor]

Originally posted by: taltamir
where did you read that dell sells more then the whiteboxes?

I recall reading specific data on the issue. I don't recall where exactly.

You on the other hand assume the opposite with absolutely no data, recalled or otherwise, to corroborate your point.

So unless you claim to have actually READ somewhere that I am wrong (meaning we are both working off of memory and written info is required) or actually have sometimes substantial I am gonna keep assuming I am right on this.

I have read it, but can't post it...I subscribe to both iSuppli and IDG, but the data requires you to sign an NDA so publishing without permission isn't allowed.

I can certainly point you to hints of the truth howoever...
link

"In the second quarter of 2005, whitebox or local brand PC makers like SVOA shipped only 560,000 notebooks worldwide vs. 4.5 million name-brand notebooks shipped during the same period. Whitebox builders realized only 9 percent growth, compared with 27.5 percent growth in sales for tier-one OEMs (global name brands)"

 

[jiffylube1024]

Originally posted by: CTho9305

Originally posted by: jiffylube1024
If only they released Brisbane 65nm cores for 939 all along!!!!! But no, they wanted to force AM2 down our throats, and once Core2 was released it was all over...

FWIW, it's a lot of work to port a processor across process nodes nowadays, and some of the uglier things are parts like memory interfaces because they're very "analogy" - they're more sensitive to the things that get worse on smaller process nodes (e.g. variation). I'm not speaking for any companies.

You are right.

Basically, I guess, it all comes down to AMD switching to 65nm too slowly (Intel was on 65nm by the time P965/975 was out, while AMD launched AM2 at 90nm and had a defecit in cache and clockspeed from day one).

Now, history is repeating itself in even more stunning fashion, as the 65nm Phenoms have been halted, and the month that was supposed to see their wide distribution (January 08) will instead see them working on ironing out bugs in the architecture, while Intel trots out a ready 45nm update, with new features, slightly higher IPC and more cache...

And AMD's only possible response to Intel's agressive node change is to bring back AM2 Brisbane chips, migrate the entire architecture to 65nm, and oh yeah to kill off all 2MB cache models because they take more die space and are more expensive to make, especially when your transistors are 45% larger...

Seriously, how cheap are AMD CPU's and motherboards going to have to be in order to get sales in 2008?

Phenom sounded interesting for awhile - "true" 4 core architecture, onboard memory controller, etc. However, we saw that Intel's Core2 with 4MB of cache easily offset AMD's latency advantage with the on-die memory controller. In 2008 Intel's got Quad Cores with 12MB of cache... AMD's Phenom doesn't even win in clock-per-clock performance, let alone the fact that they're 600 MHz slower at the high end (2.4 GHz vs 3 GHz) and oh yeah, they have a TLB bug that is preventing their release now. And picking nits, AMD is using a L3 cache that's fixed at 2 GHz in Phenom (2x the HT bus speed, right?), rather than just making it all full-speed L2 cache.

It's just a perfect shit storm in AMD's face and I'm worried for the CPU industry. We need a serious challenger to Intel to keep prices down!