AMD Claims 80% Thoroughbred Yield. UMC is Ready

[Adul]

http://www.xbitlabs.com/news/story.html?id=1043189982


Some of our friends around the industry commented on our recent news-story in regards possible low yields of AMD?s x86-64 processors (check it here). They told us a lot of interesting facts about AMD, but probably the most interesting stuff they said concerns the current yield of x86-64 and Thoroughbred processors.

Firstly, I should point out that Thoroughbred?s yield currently achieves amazing 80% what is a lot more compared to 60% we reported before based on the information announced back in may 2002. Furthermore, the cost of manufacturing one single core should be below $10, what is also less than it used to be two quarters ago. Due to astonishing yields, AMD can sell their Thoroughbred CPUs for low price, still having some space to lower them more. Unfortunately, this is not enough for the company to become profitable. In order to achieve high revenues and earn a lot, AMD should offer something a lot more powerful than Intel. It already happened in the year 2000, when the world?s second CPU maker?s sales were more than $4 billion.

As for the yield of x86-64 processors, everything should be fine in terms of yields themselves ? the number of chips produced is sufficient, but there is a problem with frequencies of the chip. At the moment there are still some issues that do not allow AMD to clock their x86-64 CPUs at high core-speeds. This is the reason why the company decided to go with the Opteron a bit earlier than with the Athlon 64 that is postponed till the second quarter.

In fact, UMC is ready to produce AMD CPUs, all AMD needs is to ask them. Due to the fact that AMD can fulfill all the orders at the moment, the company produces their chips on their own shiny Fab30. At the moment the first batch of AMD Barton CPUs should be in production in Dresden, Germany.

 

[chizow]

Yummy...I wanna see Barton before Opteron, as a new mobo and platform isn't in my immediate plans (give me 3 months and I may change my mind ).

Chiz

 

[Insane3D]

At the moment the first batch of AMD Barton CPUs should be in production in Dresden, Germany.

The yields explain why people are having such good luck with overclocking the Tbred cores...especially in the lower speeds. The 2100+ Tbred B that can sometimes hit as high as 2.4ghz and costs ~$100 is a prime example of a great "bang for the buck" CPU IMO. I hope I have good luck with mine and the 8RDA+ and Corsair PC3200 Twin pack I've got coming.

 

[dullard]

Anyone have a rough idea of how much packaging adds to the $10 core price?

Seems like AMD shot themselves in the foot with their claims of 3400+ speeds no matter what at release. They could release the Athlon 64 a lot sooner if they didn't make those claims, then have time to fix the frequency problem that isn't letting them reach those speeds.

The high price of Bartons should help AMD in becoming profitable.

Any guess on how much UMC will add to the price of manufacturing when AMD starts to use it?

 

[chizow]

Originally posted by: Insane3D

At the moment the first batch of AMD Barton CPUs should be in production in Dresden, Germany.

The yields explain why people are having such good luck with overclocking the Tbred cores...especially in the lower speeds. The 2100+ Tbred B that can sometimes hit as high as 2.4ghz and costs ~$100 is a prime example of a great "bang for the buck" CPU IMO. I hope I have good luck with mine and the 8RDA+ and Corsair PC3200 Twin pack I've got coming.

Oh man, that RAM was sooo tempting...$400 bucks though If its half as bad-ass as I think it is, I don't wanna hear about it...it'll make me spend more money

Chiz

 

[Insane3D]

Originally posted by: chizow

Originally posted by: Insane3D

At the moment the first batch of AMD Barton CPUs should be in production in Dresden, Germany.

The yields explain why people are having such good luck with overclocking the Tbred cores...especially in the lower speeds. The 2100+ Tbred B that can sometimes hit as high as 2.4ghz and costs ~$100 is a prime example of a great "bang for the buck" CPU IMO. I hope I have good luck with mine and the 8RDA+ and Corsair PC3200 Twin pack I've got coming.

Oh man, that RAM was sooo tempting...$400 bucks though If its half as bad-ass as I think it is, I don't wanna hear about it...it'll make me spend more money

Chiz

Nah, I came into some spending money, but not that much. I got the 2 x 256mb setup with the 6-2-2 settings and the silver heatspreader...$220. Icidentally, Newegg dropped their price again today on the 2100+ Tbred's...it's under $100 now.

 

[chizow]

Originally posted by: Insane3D
Nah, I came into some spending money, but not that much. I got the 2 x 256mb setup with the 6-2-2 settings and the silver heatspreader...$220. Icidentally, Newegg dropped their price again today on the 2100+ Tbred's...it's under $100 now.

Oh gotcha heh, I thought you were getting the Twin 512MBs :Q That's what I was looking at for $400 Those dimms should rock, I would get em too, but then I'd have 512MB Samsung True PC2700's sitting around, and I kinda wanna get more RAM if I upgrade.

Chiz

 

[majewski9]

512k L2 cache and 333fsb + higher core frequency! AMD is back on top again!

 

[pm]

Originally posted by: Insane3D
The yields explain why people are having such good luck with overclocking the Tbred cores...especially in the lower speeds. The 2100+ Tbred B that can sometimes hit as high as 2.4ghz and costs ~$100 is a prime example of a great "bang for the buck" CPU IMO. I hope I have good luck with mine and the 8RDA+ and Corsair PC3200 Twin pack I've got coming.

Yield is the ratio of working chips to the total number of die on a wafer (occassionally not counting chips along the edge of the wafer). It is usually determined using the lowest usable bin for that part. So if the lowest usable bin (slowest chips being shipped) is 1GHz, for example, and the highest bin (the fastest parts shipping) is 2GHz, then you would calculate the ratio of how many chips work at 1GHz vs. how many were on the wafer in total. When yield is quoted without a reference (ie. "our yield is better than 95%" without any further elaboration), then yield are numbers are open to a certain degree of intrepretation. The key bit that is missing is at which stage this yield was measured. Was this line yield? After packaging? After sort? After burn-in? And another part is what parts are counted on the wafer in the ratio.

But my main point is that speed/bin split has nothing to do with yield. You could have 100% yield on a wafer (all die on the wafer work at the minimum speed) and still have terrible bin split to your top frequency bin. In fact this situation is ofen the case, since low-Vt/low Leff wafers run faster but are more likely to have lower overall yield. In other words, faster wafers usually have lower yield.

 

[sparks]

512k L2 cache and 333fsb + higher core frequency! AMD is back on top again!

They never mentioned anything about faster core speed, at least initially. The 512K L2 cache and 333 FSB will allow AMD to ship a CPU with a lower core clock and still be higher performing. Just look at the 2600s, the 2600 that has a 266 FSB clocks in at 2.13 GHz while the 333 version runs at 2.06GHz. I wouldn't be suprised if the 3000+ clocked in at 2.25Ghz like the TBred 2800+.

 

[The Sauce]

In order to achieve high revenues and earn a lot, AMD should offer something a lot more powerful than Intel.

Yah, that does seem to be the trick though, doesn't it. Duuuuuuuuuuh.

 

[Wingznut]

Yeah, that's a pretty poor article.

$10 per chip? I suppose maybe if they were taking into consideration the materials necessary. But if you take into consideration the rest of the necessities (fab, labor, R&D, etc...), there's no way that $10 is accurate.

Same with "amazing 80%" yield. Depending on how they are defining it, that could really be a very poor statistic.

Originally posted by: Xbitlabs
In order to achieve high revenues and earn a lot, AMD should offer something a lot more powerful than Intel. It already happened in the year 2000, when the world?s second CPU maker?s sales were more than $4 billion.

Heh, I think that's my favorite part. A) This guy is a master of the obvious! Can't put anything past him! B) Sales were more than $4 billion because people were buying cpu's like they were going out of style... Not so much that they had a faster cpu than Intel. Even if they had a faster cpu today, they wouldn't be bringing in the same amount of sales. Not even close.

 

[zephyrprime]

there's no way that $10 is accurate

Yeah, $10 per chip is crazy. AMD wouldn't be losing any money at all if it only cost them anywhere near $10 per chip to make Athlons.