Fabless chip shop AMD was set to launch its "Cayman" GPU (aka AMD Radeon HD 6970) in the week of November 22nd - just in time for the tech-Xmas season – but reliable word from the Taiwanese street is that the product has been delayed due to yield problems at apparently not-so-fab Fab TSMC.
The high-end HD 6970 Cayman GPU-based board is reported to have a whopping 2GB of GDDR5 memory and a 256-bit controller which – at least in terms of RAM quantity- is more powerful than Nvidia’s upcoming GeForce GTX 580 card which only sports a sniff-worthy 1.5GB.
Of course, as most people with common sense know, 1GB is more than enough for even the most high end cards, but GPUs are an ego based game and bigger is always better – at least according to the mad men of tech marketing.
It doesn’t matter how big you are, however, if you don’t show up and put some skin in the game which, according to our talkative Taiwanese buddies is likely the case for AMD which simply cannot yield the product in any significant volume at manufacturer TSMC on time.
According to our sources, yields for Cayman are still in the single digits and there's minimal chance AMD will have any shot of achieving its schedule or production/ship targets for Q4. This could also really hurt if Nvidia successfully launches its GTX 580 in volume – despite its smaller RAM offering.
No doubt this will all come to light when samples slated for press testing get held up in the holiday post. We’ll bring you more as we hear it.
There is very little chance that nVidia will be able to successfully launch GTX580 if AMD is having Cayman yield problems; unless nVidia counts having one single barely operating chip as a success.
She's making your site look bad, cause she has less ram than the bros.
I'm not saying the yeilds are not bad, but AMD has a solid foundation here, so what has changed?
Frequency? Power curve? Something seems very wrong here. I'd be more inclined to say TSMC has another broken process that was previously working fine.
let me get this clear... your objection to this article is that it's written by a woman? are you a caveman? What the heck is wrong with you? You don't even level any rational or logical criticism of the article. Instead, you just make yourself look rude, sexist and backward.
The article is based on information from trusted sources. A reporter's job is to bring that kind of information to the public. How does it make TechEye look bad to have its journalists actually producing original content from sources in the field? You, sir, are a nincompoop.
unfortunately, solid reputation at 40nm means very little in this case as it all comes down to performance targeting and how many die cut the mustard for the frequency you need for a high end product.
I'm sure AMD will have no problem pulling hundreds of samples out of its hat, as fabs produce a fair amount when a product is at single digit yields, but it doesn't mean much if you can't get into high volume production, as there's no way to sustain reasonable demand otherwise.
Folks, it doesn't matter if 40nm yields on previous products were pristine or crap. What it comes down to is performance targeting of the die or lot - which is something that Intel, AMD and NV all do with their fabs/foundries today (and have for some time). You can have a wafer with 80 per cent yield for a product at a particular frequency. As soon as you change the production mix to target a higher performance variant the yields can go to hell in a handbasket. You still have usable die in many cases but not at the level you need for a particular performance target (this case high end). So as much as the T 40nm process should be refined by now (??) this can all mean nothing if they don't have the mix right when they do their performance targeting during the production ramp. Whether you've sampled or not means nothing because you can just throw mass at that problem because the numbers needed are low.
No, this woman just doesn't know anything. She is making herself look bad. despite her smaller RAM offering, she also lacks citing any real source. It could be an NVIDIA fanboy on the streets of Taiwan, spreading FUD.
The only spec she used to compare the cards was the size of their RAM. In the enthusiast community, that is known as a joke. They way she used RAM size to tie the whole article together makes her seem like a huge n00b. She really should have a male editor proof-read and fix her articles before publication.
Folks, it doesn't matter if 40nm yields on previous products were pristine or crap. What it comes down to is performance targeting of the die or lot - which is something that Intel, AMD and NV all do with their fabs/foundries today (and have for some time). You can have a wafer with 80 per cent yield for a product at a particular frequency. As soon as you change the production mix to target a higher performance variant the yields can go to hell in a handbasket. You still have usable die in many cases but not at the level you need for a particular performance target (this case high end). So as much as the T 40nm process should be refined by now (??) this can all mean nothing if they don't have the mix right when they do their performance targeting during the production ramp. Whether you've sampled or not means nothing because you can just throw mass silicon at that problem because the numbers needed are low.
We do our best to get the balance right but unfortunately due to the volume of spam we received previously we've had to adopt this method - even if we did allow OAuth logins this would have to remain in place. We welcome feedback and suggestions on this though!
James
CTO, TechEye
a) you seem like an angry, sad person. That must suck for you.
b) Not too many other specs are currently known about EITHER offering (except the RAM which is quite well publicized), unless you have any you'd like to enlighten us with.... you being such a male guru on all things unimportant and all....
c) Lydia actually makes a point of saying that RAM has very little importance over 1GB... so... you would appear to be a little dense when it comes to picking upon irony here.
what is it, are you feeling emasculated or something, honey? there, there now....
But on to stuff that really matters... I wouldn't be so quick about that RAM doesn't matter, at least, not in this case. Generally speaking Lydia is right, but don't forget that Caymans are able to power six screens at the same time. That's when 2 Gb comes in very handy. Now, few will have such a set up, but with the large expected performance and little required performance (mostly due to the next generation consoles taking their time, and few game titles are developed specifically for PCs these days), three monitor setups will make sense to more people. It is a high end card too, after all. For such setups too only one Gb will be very little, borderline really.
And, of course knowledge of 40 nm matters. Engineers new to a production will not manage the same as those with some two years experience on it. With more experience you can achieve higher targets - but perhaps AMD was a little too bold?
If there are indeed issues, I'd wager that it's mostly the new architecture (with 4D et al) didn't yet work out too well. Rumors say it's the drivers that aren't ready yet, which sounds like something that could have been avoided...
But, as a last point, weren't there leaked benchmarks already? And leaked specs and all of that? We'll see how things turn out. If the yields are really that bad I guess they're in trouble anyway, seeing as there's so little time left to build up stocks.
I posted a comment, it was approved. Then it was responded to by others. When I tried to reply to those responses, my comments had to be approved. For my second and third posts, I used the exact same name/email/website as the first post. Same browser, just a few hours later. Firefox/3.6.12 and Win7 64-bit.
@kbobisdunce
learn2sarcasm. plz.
@SB
I followed a link to this site hoping for a good article. What I got were a poorly written article only containing unfounded, so-bad-it's-unbelievable rumors. A badly coded site with a broken comment section. Dumb comments from readers who have a 2nd grader's reading comprehension and clearly missed the point of my remarks. And a nice fanboy flamewar in the comments.
@kbobisdunce errr SB,
I visit Semiaccurate regularly. If you ignore their, then they are a great hardware site with REAL exclusive stories and sources from within companies, not just people found on the streets of Taiwan.
@SB
The specs of the GTX 580 have been known for several days, since before this article was published. This article also promises updates with "more as we hear it," well, here's a news story citing hard specs: http://fudzilla.com/graphics/item/20777-evga-gtx-580-pictured-listed-at-$499. New news stories about Cayman have been slow to come out , but if you look back at semiaccurate, they have some articles listing with some pretty realistic specs which could be cited, and are more informative than just "2GB RAM OMG"
FYI, I got this after posting comment #4 (and I'm betting I'll also get it after posting this, cmment #5) from the same computer/browser/session was also added to the approval queue
Thanks for your comment! In order to fight spam, the first comment you make on this site is moderated. We will email you once the comment has been approved. Thanks for your patience! If you have any problems don't hestitate to contact us.
gj.
If thats about to move, AMD could be waiting to set final specs. After all, if you have room to push a slightly higher clock in bios at the expense of a reduction of X chips per wafer you might make that call for the halo value,
And perhaps thats where the yield issue does lie. If they're pushing for say 1GHz just to be the first there and to hit performance level X, they might be finding the 6970 or 6990 is hard to come by but they plenty of the generation below available.
Still, I do wonder what the source is and does it hold up. We'll know more soon.