ATI RV740 review/preview

Geez guys I can figure figurative Chinese, being a rather sleek lurker in forums and such.



Stop with the machine translations, it's killing me with lethal laughter :D




汗 - the chinese equivalent of "sweating" - aka speechlessness. Nothing about Genghis Khan LOLLLLLL
 
Last edited by a moderator:
汗 - the chinese equivalent of "sweating" - aka speechlessness. Nothing about Genghis Khan LOLLLLLL

The chinese-english dictionary i have been using gives that symbol as "hàn"(pronounced similarly to khan in english) with 2 definitions perspiration/sweat or "king/emperor"

What about the second - 死 - pronounced " sǐ "
That means "to die or impassable or rigid" doesnt it?

I saw that the first char was also sweating but the rest of the sentence didnt fit how do you get "sweating /speechlessness" together with "rigid"...suppose could mean catatonic or similar but totally didnt fit with the rest of the sentence.

So should i throw out my dictionary?
 
Last edited by a moderator:
The "sweating", is a fashionable expression originated from funny faces in Japanese Anime. You'll probably know what I am talking about if ever watched those cartoons;)
 
Well the full original quote in chinese was: 汗死 前几天啊....漏电严重,某家都不好意思说了只能小道消息散下

something quickly died, leakage, embarassment...my confidence in translation is gone...kindof sad cause the stuff he was saying kindof looked fun to try and decode.

It's cantonese. My wife helped translate it.

Goddammit, The leakage of electricity is too bad, I'm too embarrassed to say it since it's not official (street news) I can only offer this as gossip, let's "call it a day" (as in, I don't want to talk about it anymore.)
 
1% more or less good dies is a relatively minor difference in revenue for the fab, but it get amplified by one or two orders of magnitude for the design house, depending on how long it takes further down the pipeline before the defect is detected.

A more important testing tradeoff is being made wrt tester time. When going from 500ppm to 100ppm, you quickly enter the laws of diminishing returns.

But in terms of fab yield, the difference is in the noise.

Yes if you have manual workmanship then these 100ppm looks like rocket science problems. But AMD developed APM and GP willingly announced that they sold that knowledge to anyone pays enough :D

"I have to mention our APM [Automated Process Manufacturing http://www.globalfoundries.com/technology/apm ] set of technologies here. We are not a company that will jealously keep our secrets and send the invoice to the companies working with us.
We want to give [customers, Ed.] a chance to look at what's best in class and put out a truly 21st century solution that will put GlobalFoundries as the platform of choice for custom silicon"

......

"As you know, we are not afraid to license our technology to partners that are open to innovation. A couple of years ago, Microsoft, IBM and Chartered found themselves in trouble considering low yields on the Xbox 360 CPU. We were asked for help and by analyzing the manufacturing flow, we ended up implementing APM 2.0 in the Chartered Fab in Singapore. The results were instant and the Xbox CPU achieved rewarding yields in mere months after APM deployment.
All APM specifics are now our key selling points. As I mentioned before, we're not an "invoice company". As our customers learn of their own designs we can come in and help them to optimize the designs to achieve highest yields and clocks. Our policy is "Our Fab is their Fab"."

from http://www.brightsideofnews.com/print/2009/5/26/the-future-of-globalfoundries-revealed.aspx
 
Yes if you have manual workmanship then these 100ppm looks like rocket science problems. But AMD developed APM and GP willingly announced that they sold that knowledge to anyone pays enough :D
Your quote is not relevant in this context. The problem that AMD had to solve at that time, was yield. As indicated earlier, this is very coarse problem and something for a fab to solve.

When talking about ppm, it's about test coverage and not letting failed devices being declared as good. That's a different problem. The difference between 500ppm and 100ppm is in the noise wrt yield and not of any concern for the fab, but it's very important for the fabless company who's selling the devices to the downstream customer who integrates the chip into an actual product. APM wouldn't solve any of that. What you need it such a case is a very comprehensive DFT strategy and a willingness to incur the cost of much longer test times.
 
It's cantonese. My wife helped translate it.

Goddammit, The leakage of electricity is too bad, I'm too embarrassed to say it since it's not official (street news) I can only offer this as gossip, let's "call it a day" (as in, I don't want to talk about it anymore.)

Ah ok so sort of similar, "khan dies" to "god damn"...well at least i know how to swear now.

The above looked simplest to translate, There was one other more difficult sentence to translate, am very hesitant about this after difficulties above. Anyway just for laughs:
hotboy said:
昨天BOSS来了,没来得及回....据可靠线报,2家的新核又是一次飞跃,也难怪前段时间生了个畸形残疾儿
Which is something like:
Boss came yesterday, no time to turn around, according to reliable report 2 new cores to leap, also not surprising student with disabilities has birth defects.

ie Boss came by yesterday to report amd has 2 new cores. RV740 is "student with disabilities" as it was designed to learn about 40nm and is a cut down RV770 having "birth defects" ie high early failure rate.

Am presently burying my dictionary in the backyard, I can see tchock laughing now....
 
hotboy said:
昨天BOSS来了,没来得及回....据可靠线报,2家的新核又是一次飞跃,也难怪前段时间生了个畸形残疾 儿

Which is something like:
Boss came yesterday, no time to turn around, according to reliable report 2 new cores to leap, also not surprising student with disabilities has birth defects.

ie Boss came by yesterday to report amd has 2 new cores. RV740 is "student with disabilities" as it was designed to learn about 40nm and is a cut down RV770 having "birth defects" ie high early failure rate.

Am presently burying my dictionary in the backyard, I can see tchock laughing now....

Boss came buy yesterday, so I didn't have time to reply. According to a reliable source we're going to produce 2 new cores "a jump / on a new level" some time ago they produced a "handicaped" core.

Indeed rjc, as I read it here AMD willingly mutilated and basically neutered RV740 to adapt to a troublesome 40nm process. So even though yields might be low it seems to be a salvage part in general. From that point of view, getting your first revision (A11) out the door and with a good market reception while actually producing it as a test processor is quite a feat.
 
Your quote is not relevant in this context. The problem that AMD had to solve at that time, was yield. As indicated earlier, this is very coarse problem and something for a fab to solve.

When talking about ppm, it's about test coverage and not letting failed devices being declared as good. That's a different problem. The difference between 500ppm and 100ppm is in the noise wrt yield and not of any concern for the fab, but it's very important for the fabless company who's selling the devices to the downstream customer who integrates the chip into an actual product. APM wouldn't solve any of that. What you need it such a case is a very comprehensive DFT strategy and a willingness to incur the cost of much longer test times.

Well apm seemed to me as ongoing process. Maybe 1.0 was only an yield oriented but from above it more look alike whole automated binning and testing process even in this 2.0 stage. Anyway that DFT-Design for Test? you mentioned ... silicon testing ground. And the whole APM thing i mentioned cause someone above point out year old news about speculative TSMC venture into chip packing and testing, and there was much rumors what know known Globalfoundries corp will look a like before they close the deal with ATIC. GF is mor than willing to JV with TSMC for their cash inflow to build some TSMC testing ground round their APM 3.0 or more what they now blabing about beneath the carpet.
 
So, I've noticed that Newegg only carries a single 4770 model now and that it is out of stock. Previously they had a number of 4770s but they were all out of stock for weeks.
 
Boss came buy yesterday, so I didn't have time to reply. According to a reliable source we're going to produce 2 new cores "a jump / on a new level" some time ago they produced a "handicaped" core.

Indeed rjc, as I read it here AMD willingly mutilated and basically neutered RV740 to adapt to a troublesome 40nm process. So even though yields might be low it seems to be a salvage part in general. From that point of view, getting your first revision (A11) out the door and with a good market reception while actually producing it as a test processor is quite a feat.

Thankyou again neliz for checking the translation. Dont get too caught up in the RV740, the main part: "2 new cores in production of a new level" is kindof a bit important too.

Thing is that quote was less than a month ago, provided everything works smoothly(admittedly a really big assumption) amd should have production quantities of 2 new chips in a couple of months or so.

Question then becomes, what are they announcing this week? Announcing these 2 cores now will likely have a detrimental effect on sales of current products as happened to the RV770 when the RV740 was introduced.
 
2家的新核 - cores from the 2 IHVs, btw.


it's referring to "RV870"/Evergreen and GT300 being tremendous leapfrogs over their current lineups.

But gosh, you guys do have the effort. :D
I have to say sadly though, that Chiphell wasn't the place it used to be. Still fun to hang out, but not really that robust in info anymore. There are a few informants, but basically only their threadstart posts matter- the replies are mostly BS. ;)
 
Or is it lacking in stock because RV740 launched in China on the 28th of May, thus taking away stock intended for other markets?


China is crazily price sensitive (on top of REALLY rigid graphics pricing with SKUs coming in at every 100-200RMB level) compared to the other markets which would make RV740 an easier sell while replacing RV770LE/4850 would sell in the others still.
 
But we're hearing of chip supply issues, not board supply issues. So if AIB's can't even get enough chips....
 
Alternate (europe) is also getting new stock this week, all the others getting stock next week.

but then, they're only running at 20% of projected yields aren't they?
 
2家的新核 - cores from the 2 IHVs, btw.
it's referring to "RV870"/Evergreen and GT300 being tremendous leapfrogs over their current lineups.
Hmmm, i never seem to be able to get this right. When i do it by hand:
  1. 2 - 2
  2. 家 jiā - family
  3. 的 dí - really
  4. 新 xīn - new
  5. 核 hé - core
  6. 又 yòu - again
  7. 是 shì - to be
  8. 一次 yī cì - first
  9. 飞跃 fēi yuè - to leap

For some reason i thought "family" was 2 market segments(ie like performance, mainstream, entry etc). Didnt occur to me at all was 2 families "nvidia" and "amd".

The "..again to be first to leap part" seemed to suggest whoever got to the previous node first was going to do it again.

Very frustrated at the moment :???:

But gosh, you guys do have the effort. :D
Thanks - sorry to put you through all this, i wish i was better at translation, it's so hard to get the right meaning.
I have to say sadly though, that Chiphell wasn't the place it used to be. Still fun to hang out, but not really that robust in info anymore. There are a few informants, but basically only their threadstart posts matter- the replies are mostly BS. ;)
Ok will bear that in mind in future, thanks again for helping me with this.
 
Back
Top