Wednesday, April 2, 2008

Reader Comments

at 4/2/2008 2:55:22 PM, bk said:
Brian, being able to sell a defective chip for some revenue instead of trashing it is definitely worth doing to help improve yields. You know that IBM does this with their CELL processors used in the PS3 where only 7 of the 8 SPEs on each chip need to work. However, you are mistaken if you think this can be accomplished as a last-minute design fix. It requires a lot of forethought and hard work to ensure the rest of the chip will work properly with one of the cores being defective. This does not come for free - you have to work at it. Trust me.

at 4/2/2008 3:04:17 PM, Brian Dipert said:
Dear bk, I daresay that one of the four x86 cores on a Phenom die represents a much larger percentage of the total die area than one of the eight SPEs on a Cell die. Therefore, the quad-core yield loss must be SUBSTANTIAL to make a triple-core recovery at all economically viable. My general rule of thumb: if you can see a structure on a die with the naked eye, it's fab-expensive (or said another way, not cost-insignificant) ;-)

at 4/2/2008 3:49:41 PM, Hank said:
No one has 100% yield, so no one has 100% working cores. The choice is to scrap the chip or sell it as partially-good. Remember the 386SX, a 386 with a defective floating-point unit. With increasing number of cores and process variability, product will increasingly ship with some bad cores and varying core frequencies.

at 4/2/2008 5:27:54 PM, T Rex said:
AMD now has healthy quad-cores to compete against Intel, and triple-core and dual-core to compete againt Intel dual-core. Why on Earth do you see that as a bad thing?

at 4/2/2008 5:34:32 PM, Brian Dipert said:
Dear T Rex, have you seen any benchmark results, either absolute performance or performance-per-dollar, on the AMD vs Intel alternatives you quote?

at 4/3/2008 7:38:59 AM, Joseph said:
Brian, you got the information about Intel's Dunnington 6-core wrong. It isn't (as you say) a processor with its cache in the missing "hole". Dunnington is just a 3-die version of their Penryn generation. As you may or may not know, Intel's current quad-cores are just two dual-cores strapped together in a single package. They just did the same thing one more time with Dunnington by strapping together 3 dies instead of two.

at 4/3/2008 7:47:42 AM, T Rex said:
Thanks - a triple-core will beat a dual-core on multi-threaded applications (digital media like iTunes and Nero, for instance). I just keep seeing this false argument that it’s somehow troubling that AMD has a problem because...it has a unique product to compete against Intel dual-core, especially in retail and business desktops? AMD just launched four new quad-cores, so there is a good quad-to-quad competition. AMD clearly positions the triple-core against Intel dual-core. Search “AMD Phenom X3 + AMD 780” in YouTube.

at 4/3/2008 8:07:11 AM, Brian Dipert said:
Dear Joseph, sorry but you're the one who's incorrect. Dunnington is a 45 nm-based, Penryn-derived single die six-core design with L3 cache in the 'hole' where the 7th and 8th cores might otherwise go. Hit Google for die shots.

at 4/3/2008 9:30:06 AM, Stiggle said:
Scrapping less silicon is also very green! By using the 3-core and maybe 2-core functional chips instead of scrapping processed silicon surely makes good economical sense as well as reducing waste. In our through away society we need to follow this example to reduce our waste products and wasted energy producing scrap silicon.

at 4/3/2008 12:27:26 PM, bk said:
Brian, these chips with 4 cores on them have about half the die devoted to Cache RAM and half is logic. If you have a defect that affects the RAM, most of the time it can be bypassed by using space word or bit lines. If the defect happens to land in logic, the core is dead. DRAM/SRAM yields are much better the logic chip yields because of the repair aspect. With a partial good die approach it may be possible to double the yields on such large area chips, although you will get less $ for a device that is not all good, it is better than just tossing it. Note that you still get all of the L3 cache on the 3-core chips, so single threads might actually run faster having to share the L3 with fewer processors. It is interesting to consider whether this partial good AMD strategy is better or worse than Intel''s dual core times 2 module approach. When Intel has a defective (logic) core, they toss out a dual core chip yet can still put 2 chips on a module for 4 cores per module. The total silicon area for 4 cores for Intel is more, but it is a higher percentage of cache that is repairable, so they can probably repair most chip defects, yet a logic failure for intel is likely a total loss. It would be interesting to know who is tossing a higher percentage of their chips. It could well be Intel, but they only toss out two cores when they do that whereas AMD tosses four cores. In any case, a partial good strategy provides a meaningful way to deal with logic side defects to help get your effective yield up and improve profit margins.

at 4/3/2008 4:47:54 PM, CYI said:
<< Dear Joseph, sorry but you''re the one who''s incorrect. Dunnington is a 45 nm-based, Penryn-derived single die six-core design with L3 cache in the ''hole'' where the 7th and 8th cores might otherwise go. Hit Google for die shots. >> When I first looked at the photo of the die, it looked as if there are only three cores on a single die. There are three very clearly identifiable core blocks with the remainder consisting of cache. However, after downloading the full-sized image, you can see that each of the three core blocks contains two symmetric cores for a total of six on a single die.

at 4/10/2008 8:49:21 AM, WCM said:
Hey, I dont really care about Intel vs AMD. I have 3 sons, 6 computers and a Gbit ethernet network. As a consumer, the PC just has to met is goals, to play games smoothly. Five of the PC''''s are using the AMD 64bit 3GHz CPU''''s and one laptop has the intel 2.2GHz celeron CPU. All the PC''''s do their jobs very well so I havent updated a single PC in 2 years. Consumer PC''''s is where AMD made its name and as long as AMD remembers consumers just want to play their games, AMD will just be fine. What i dont see is AMD advertizing its ability to play games. The ideo card companies did it and AMD needs to do it or face trash articles like this. BTW, all the PC''''s use Nvidia GPU''''s. We have never been fans of ATI GPU''''s. They have alway seemd to be the cause of game crash''''s. AMD made one mistake, buying a faied GPU company ATI. If they had bought Nvidia, AMD would havebeen the king of the gamers PC''''s. If Nvidia ends up buying AMD, gamers would be in heaven and Intel would be out in the cold. the moral is we have had what we need in single and dual cores, why would we slow game playing down with slower clock speed tripple ad quad core CPU that only manage to eat up main memory needed by the game? Windows Vista sucks and Windows XP can only see 3.25GB of RAM. So right now, why would anyone want to share 3.25GB of ram between 4 slower CPUs while games can''''t use 3 of the CPU''''s anyway. Dual cores will rock till games can uses all the cores and that is still a long way off.

at 4/11/2008 6:29:22 PM, Alan R. Weiss said:
Design wins matter. Cost matters. Quality matters. Yield matters. Performance has become so amazing (we at Synchromesh Computing have benchmarked Barcelonas) that they are <gasp!> practically Microsoft-proof for everyday and even power user consumers (certainly that not the case with servers, but for consumers ....). Now it comes down to design wins - how many computer makers can AMD and Intel line up to buy their parts? And at what price points? In other words, a technical analysis devoid of marketing analysis is simply silly.

Post a comment

Display Name
Before submitting this form, please type the characters displayed above: