The AMD Execution Thread [2007 - 2017]

Status
Not open for further replies.
apoppin said:
They are being "sneaky" and using the "excuses" of everything BUT how much is due to the SOI process being introduced to the fabs with all the machinery changes that it will entail.
Click to expand...
TSMC does have a SOI process, but you seem to be the only one to care: they reportedly couldn't even find a lead customer for their 40nm SOI process to test a real-world design on it! While I do think SOI is fairly attractive for CMOS Photonics and Z-RAM, don't forget it does increase cost/wafer and might lower yields, especially for a foundry that isn't generating massive volumes of it and thus has much less experience with tuning the process than, say, IBM or AMD...
 
Chatter indicates that at least one of the IBM+AMD pairing is not entirely happy with the decision to still be with SOI.

The benefits seem to be a one-time boost, while the headaches it introduces or fails to address increase at a greater rate.
 
Arun said:
TSMC does have a SOI process, but you seem to be the only one to care: they reportedly couldn't even find a lead customer for their 40nm SOI process to test a real-world design on it! While I do think SOI is fairly attractive for CMOS Photonics and Z-RAM, don't forget it does increase cost/wafer and might lower yields, especially for a foundry that isn't generating massive volumes of it and thus has much less experience with tuning the process than, say, IBM or AMD...
Click to expand...

3dilettante said:
Chatter indicates that at least one of the IBM+AMD pairing is not entirely happy with the decision to still be with SOI.

The benefits seem to be a one-time boost, while the headaches it introduces or fails to address increase at a greater rate.
Click to expand...

It makes sense. SOI becomes far less viable as gate thickness decreases. It was a great technology for the 130/90nm nodes but it's been a hindrance beyond that.
 
ShaidarHaran said:
True. How much energy is derived from petroleum then?
Click to expand...

Not sure I understand the question.

I was implying Natural gas, coal, uranium all are going up. Electricity --> more expensive --> energy intensive processes--> more expensive

Refining silicon is quite energy intensive from what I understand, that is part of the problem with pv cells, though they do not need as high of quality of silicon, they still often use cast off waste from places like TSMC.
 
Sxotty said:
Not sure I understand the question.

I was implying Natural gas, coal, uranium all are going up. Electricity --> more expensive --> energy intensive processes--> more expensive

Refining silicon is quite energy intensive from what I understand, that is part of the problem with pv cells, though they do not need as high of quality of silicon, they still often use cast off waste from places like TSMC.
Click to expand...

I misunderstood. I thought you meant that the increase in energy prices has been mostly due to the increase in the cost of energy directly derived from petroleum.
 
ShaidarHaran said:
It makes sense. SOI becomes far less viable as gate thickness decreases. It was a great technology for the 130/90nm nodes but it's been a hindrance beyond that.
Click to expand...

But IBM and AMD seem to be going full steam ahead with SOI.

http://www.edn.com/index.asp?layout=article&articleid=CA6510863

http://www.tomshardware.com/news/cpu-phenom-amd,5370.html


Also, isnt' even Intel considering fully depleted SOI for future products?

http://www.intel.com/technology/magazine/silicon/bohr-qa-0806.htm
 
There was a fair amount of speculation that SOI was on the chopping block for 32nm, and 22nm is uncertain (if IBM even stays in the business).

AMD in particular has gotten burned by SOI, going back to when it first starting using it and had to pay IBM to help get the process production-worthy.
Troubles continued with the delays in implementing SOI cores in a timely fashion.
Barcelona's poor cache density in comparison to Intel's is in part due to SOI.

SOI wafer sales haven't been setting the world on fire either, and what is a better barometer for adoption, people buying the stuff or another me-too fluff piece IBM puts out for a future process?
 
MTd2 said:
This is horrible, looks like we're in 2007:

it doesn’t look that AMD has sampled its Deneb 45nm quad-core. The word in Taiwan is that it might take some time before this happens.[...] AMD might get to 3GHz, but it looks that Intel can already do 3.2Ghz with Quad-core, eight thread Nehalem.

http://www.fudzilla.com/index.php?option=com_content&task=view&id=7676&Itemid=66

Ouch
Click to expand...

Ouch indeed, but not news really. It's generally accepted that AMD only has process headroom for 2.8-3.0GHz @ 45nm with 10h family MPUs.
 
It seems that the architecture or SOI isnt limiting the clocks if you read this on Hardware zone.

When we probed AMD if we would ever see a 3GHz Phenom in retail since clock speed is something they could really use now, John Taylor had the following comment on it:-

"That's something that we're looking at right now. As you know, you can overclock a Phenom X4 9850 stably to 3GHz or more. We've some secret things in the works that you'll here more about later this quarter; that will allow people to get significantly above 3GHz even with a Phenom X4 9850 Black Edition. These are things we're doing more at the chipset level and some updates to some chipsets coming soon. So watch that space; I've said too much already, but you'll see significantly above 3GHz performance from Phenoms for those who really understand and know what they are doing when overclocking."

So there you have it - interesting developments coming our way, direct from the source (though we've not heard much of Phenoms making it past 3GHz normally). Now, AMD being a smaller company than its competitor in both resources and finances, it has to optimize the use of its engineering resources. So while they would like to make a 3GHz Phenom right now, they've tradeoffs to make. Right now, its OEM partners/customers just want more processors, but they aren't asking for a 3GHz Phenom since they don't sell such high-speed quad-core machines and only those who deal in the boutique end of the market selling $3000 to $4000 machines to gamers really deal with this sort of needs. AMD explained that their major partners like HP, Dell, Lenovo and others aren't asking for 3GHz quad-core Phenom processors. However, they are requesting for ever more energy efficient chips, so that they can keep making machines smaller and cooler. For the moment in this quarter, AMD will be pushing out a 2.6GHz Phenom 9950 which will be 100MHz speedier than the current 9850 Black Edition. While it won't break any records from what we've seen in its performance by means of the unreleased Phenom 9900, it's nonetheless a step in the right direction for them.

Even if their OEM's aren't demanding high-speed processors, as they deal with more mainstream volume movement, AMD's upcoming transition to the 45nm process technology (also often know as the "Stars" core update on the desktop side and "Shanghai" on the server side) is another reason why they can't entirely devote themselves to cranking higher clocked Phenoms. John commented that they've a lot of engineering resources channeled in this area as they are committed to getting it out right and want a clean launch and delivery execution. Of course they are doing more than just transitioning to a 45nm process, as they have some core enhancements, 6MB of L3 cache, supports DDR3-1333 memory and comes in a new AM3 socket packaging (which is fortunately backwards compatible with older platforms like AM2 and AM2+). By AMD's estimation, a 2.6GHz Phenom today, when making the jump to the new 45nm process node, should perform 10 to 20% better. So a 2.6GHz 45nm Phenom later this year would perform about the same as a 2.8GHz Phenom of the current generation. While most of the performance gain is mainly attributed to the enlarged L3 cache (still using the same exclusive caching policy), the gains sound much more than what Intel got when making their transition to 45nm. We'll check these claims when the processor becomes available, but yet another salient point about the transition is that AMD can bring down the wattage further on some of their processors.

So right now, they are at a juncture if they should introduce higher clocked SKUs of the existing 65nm quad-core processors or stay put and shift their engineering resources to delivering their newer 45nm processors which should hopefully give a notable performance boost at even the same clocks speeds. So that's something they'll figure out along the way as they get more industry feedback on what they require. By third quarter of this year, we should be able to get a clearer picture of the situation and their direction.
Click to expand...

What kind of influence could a southbridge have on CPU clocks or is it a bios issue :?:
 
Perhaps AMD decided to update the thermal and power specs for boards with the new chipsets to allow for an even higher tier of power draw.

It should be noted that this is only for overclocks, and apparently the new speeds are to be reached by "those who really understand and know what they are doing when overclocking".

If we go by that logic, Intel's chips would maintain or increase their clock speed advantage.
 
Ouch. DTV and handheld, with handheld probably being the chief culprit.

But yeah, they've written off 1/2 the purchase price now.
 
Which is why I think the stock didn't close lower than it did; investors already knew that the merger was misguided in its original premise. AMD's market cap has been below the purchase price for some time now, essentially reflecting a 'total write-off' situation. Not really of course, but you get the idea.

Since the charge is a goodwill impairment, it doesn't effect margins or cash, but the balance sheet. Investors already expect a grim situation there - it's cash charges that would wreck the stock.
 
A possible solution: dump FAB36/FAB38 and go fabless like VIA/Centaur.
If that's not possible due to licensing agreements with Intel, then dump them both after a new FAB is ready in either New York/US or some other country with qualified engineers, like India.
That alone would chip off a lot of expenses, especially now with the strong Euro value definitively hurting exports and such.
There's also the issue with the tremendously high cost of the king-size salaries that skilled German workers often benefit of (compared to the rest of the European countries).
 
Last edited by a moderator:
Status
Not open for further replies.

Similar threads

F
Dedicated gaming GPUs bottom of the priority list?
2
Replies
34
Views
3K
Seanspeed
S
F
RDNA5 Speculation
Replies
2
Views
1K
Erinyes
E
T
Re-litigating Radeon market performance *spawn
2 3
Replies
40
Views
3K
Dictator
D
D
Intel, AMD, Microsoft & Others Team Up For Develop The “UALink”, Direct Competitor To NVIDIA NVLink
Replies
3
Views
845
pcchen
P
xpea
Datacenter GPU market supply and demand (2024)
2
Replies
28
Views
3K
Erinyes
E
Back
Top