SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Technology Stocks : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?


To: dougSF30 who wrote (148535)1/20/2005 6:06:51 PM
From: heatsinker2Read Replies (3) | Respond to of 275872
 
Doug, "It (Smithfield) probably yields *better* than high-bin Prescott"

Well that's an interesting thought, could be true...

The Smithfield thing does not make sense to me. And I don't believe Intel is totally stupid. So what's up? Maybe this is gonna be like flash. A loss leader for Intel that aims to screw up AMD's strategy. Intel has a ton of capacity, AMD doesn't. Maybe Intel wants to make dual core a standard item, forcing AMD to tie up its capacity making dual core at a low price. Since Intel will continue to have a frequency advantage in dual core, there will be some benchmarks that Intel can win, forcing AMD to match their price. I wouldn't think that AMD and the market would be stupid enough to fall into this trap, but who knows. Megahertz sells, right?



To: dougSF30 who wrote (148535)1/20/2005 6:24:39 PM
From: kpfRead Replies (1) | Respond to of 275872
 
doug

If I connect the nodes right, dies (!) for Smithfield are mezzanine yields from Intels 2BM-process. So its packaging will be more costly than fabbing its dies.

And yes, sure it's crappy. AMD folks certainly know all that. And have a dozen of 939-dualcore-parts ready to ship to reviewers by the time Smithfield is sent out for reviews. Labeled FX2 - or whatever.

K.



To: dougSF30 who wrote (148535)1/20/2005 6:25:43 PM
From: RinkRead Replies (1) | Respond to of 275872
 
Doug, re: It probably yields *better* than high-bin Prescott

Though I'm not an expert I think you're wrong about on yields.

1. The die is way bigger, meaning proportionally lower max. dies / wafer. Yield is not only good die vs. max die (e.g. 70% good), but also good die per wafer (e.g. 200/wafer). More importantly:
2. Twice the amount of core logic trannies (without much of the redundancy present in the cache subsystem) means twice the amount of failure opportunities.
3. Smithfield pushes the 90nm process to the max just like Prescott. 130W for dual core Prescott is low.
4. Lastly it makes more sense to compare Smithfield yields with Prescott yields instead of with some ambiguous high end Prescott yields.

Sure frequency is lower (that foremost means a different bin split range and has relatively little to do with yields afaik that is).

All in all this novice feels reasonably sure you're wrong.

Regards,

Rink



To: dougSF30 who wrote (148535)1/20/2005 6:38:11 PM
From: eracerRead Replies (1) | Respond to of 275872
 
Re: Who would pay more to have their games run much slower?

But that assumes the system is equipped with a video card fast enough to take advantage of higher single core CPU speeds. Most systems are still configured with slow video cards. In many cases a high end video card like a GeForce 6800GT or Radeon X800 XT isn't even an option.

I just took a look at the HP a850e series and four out of the five video options are FX5500 speed or slower. The other video card was only a Radeon 9800 non-Pro. This is for a processor which excels at gaming. On the four slower cards it wouldn't matter whether you were running a single core Athlon 64 FX-57 vs. a 1.8GHz Toledo, or a 3.8GHz Pentium 4 670J vs. 2.8GHz Smithfield. Gaming performance would be capped by the video card. Even the 9800 non-Pro will be a limiting factor in gaming performance.

This is where I think Intel has correctly positioned Smithfield in terms of pricing. It isn't targeting the serious gamer who is willing to spend $400 on a video card alone. It is priced low enough that the average Joe Blow who plans on buying a system with the default integrated graphics or FX 5200/Radeon 9200SE budget video card will consider buying one. For casual gamers a 2.8GHz Smithfield is more than enough.

How hard Intel wants to actually push the 800 series to home users remains to be seen. The benefits are questionable at best. If they do push it hard to the home user get ready to see the CD burning/SETI@Home/video encoding all-at-once multitasking benchmarks show up similar to when Intel first introduced hyperthreading.



To: dougSF30 who wrote (148535)1/20/2005 8:39:27 PM
From: UpNDownRead Replies (2) | Respond to of 275872
 
dougSF30

I think the dual-core Prescott should perform worse than single-core in single-threaded operations. According to the Prescott descriptions, single-core Prescott has pretty aggressive speculative prefetching. That sort of thing helps to make up for the high speed/low IPC/long pipeline architecture. But a dual-core product will not be able to have the same volume of prefetching at all, leading to worse performance. Doing otherwise will saturate the memory bus too much.