AMD Graphics Chip Shortage Hits PC Vendors 97
CWmike writes "An offshore AMD foundry is having trouble ramping up production of a new 40-nanometer GPU, forcing PC makers to delay shipments of desktop and laptop computers, AMD confirmed today. TSMC is struggling to get up to speed manufacturing AMD's 5800 series, 40-nm GPUs, according to Jim McGregor, an analyst at In-Stat. He added that the foundry is in full production, but so far yields are below expectation. Matt Davis, a spokesman for AMD, confirmed that TSMC is having issues with production of the chips. He added that it's not clear how far behind the foundry is on production expectations. 'The design is sound. It's just a matter of trying to get TSMC to a point where they can yield. They're feeling the manufacturing crunch,' said Davis. 'We're a little bit under yield but we're working back into a manufacturing schedule we want for these parts. TSMC can only kick them out so fast at this point.' He said that PC vendors are being affected but declined to say how many vendors are feeling the pinch or which ones. 'It's the end of the whip,' he added. '[The vendors] are going to have a hard time.'"
A post at Anandtech suggests we'll see price hikes for the 5800-series Radeons until this situation sorts itself out.
Still has a lead on nVidia (Score:2, Interesting)
Bad Financial News for AMD (Score:3, Interesting)
---
Graphics Cards [feeddistiller.com] Feed @ Feed Distiller [feeddistiller.com]
Re:Still has a lead on nVidia (Score:3, Interesting)
Re:Still has a lead on nVidia (Score:5, Interesting)
I don't know why, but people always assume that nVidia parts are at the least equal, and for the most part better than ATi. Granted they have been in the past, but anyone savvy enough to know about graphics cards should also know how much things can change with every next generation.
I've heard people actually say "It's safe to say that the HD 3800 was pretty much a failure". That had to be one of the dumbest comments I've ever heard from a so-called "true gamer".
Re:This is where Intel rules (Score:5, Interesting)
Part of the problem in particular with this one seems to be the process. TSMC has decided to blaze their own trail as it were and is going outside the ITRS roadmap. You'll note it says 40nm chips and that's not a typo. They have a 40nm process, whereas pretty much everything else (like Intel and AMD CPUs) are 45nm currently and working on moving to 32nm.
Ok well this roadmap with set nodes isn't for nothing. You don't semiconductor manufacturing in a vacuum, the foundries buy hardware from a number of companies to be able to make their fab work. As such it is useful if everyone has a common goal to work on. If machines for one step are for one process and machines for another are for a different process, you have problems.
Well TSMC has decided to go ahead and make their own process, not something part of the ITRS standard. Ok well that means they are buying some custom equipment or modifying the procedure or the like.
The result? Well it seems to be poor yields. They had a lot of trouble bringing it online, took longer than they planned, and now it doesn't work as well as they'd hoped.
This isn't isn't entirely surprising. How well it works out for them in the long run remains to be seen. They do have the smallest process on the market now as far as I'm aware and both nVidia and ATi are placing orders using it. However I wonder if they'll be shopping elsewhere for future cards, given the problems this is having. They can't change what they've got now (a design for one process doesn't work on another as is) but they can change what they do in the future.
You are also correct, Intel rocks at fabs. They generally beat just about everyone to market with on a new node and they seem to be able to keep yields high enough to meet demand and keep prices at whatever level they like.
Re:This is where Intel rules (Score:3, Interesting)
True, but there was an article about why they (knowingly) do it this way. (Somewhere at AnandTech, afaicr; perhaps the GlobalFoundries article?)
I believe the gist of it was that the pace of GPU refreshes is much shorter than CPU's, and consequently it makes economic sense to both design-for and migrate to so-called half-node production steps. Both AMD (ATI) and NVIDIA been doing it this way for a while now, and I believe it has burned them in the past as well.
ButJudging by the fact that they continue down the same path, though, means it must make some kind of economic sense for them.
Re:Still has a lead on nVidia (Score:3, Interesting)
They always manage to stay on top because they are a monopolist in the gfx industry. They are the Inte£/Micro$oft of their respective industry.
Remember the partial precision era (5800)? They just happened to continue using PP well up to the 8 series...
3Dmark? They threatened to leave the sponsors group when things didn't go their way, a few years back.
They have PhysX in 3Dmark, when no one else has it in hardware to artificially boost benchmark scores (which basically sells hardware to 99% of non-enthusiasts).
Remember when 3Dmark ran on rails? The biggest cheat that the public found out about....
They have a very long history of dirty tricks, anti-competitive and anti-consumer behavior. The latest one is the disabling of physx when not paired with an $vidia card as the renderer. The customers already bought the right to use physx with their ati cards but $vidia disabled it and then gave a complete bull$hit answer as to why.
That and lots more over the years.
They are a scum company, which is why i have been $vidia-free for 7 years.
Re:Bad Financial News for AMD (Score:3, Interesting)
I believe global foundaries can do 40nm standard silicon either now or soon, so AMD should perphaps switch to there part owned foundary.
No, they can't. Global Foundries can do 45nm, and soon 32nm, but not 40. Also, Global Foundries uses SOI while TSMC is bulk.
I'm sure AMD will use GF eventually for their graphics chips, but for right now, I'm also sure it will take less time for TSMC to sort themselves out than it would to modify the design for a very different process.
Also, don't expect graphics by itself to make or break AMD. It helps being on top their, but it's a small portion of their overall revenue. To stay afloat, AMD has to compete with Intel and that's all there is to it.
Re:Still has a lead on nVidia (Score:3, Interesting)
nVidia manages to stay in my systems on the assumption that it will work better in Linux. So far, I have never been wrong about this; ATI has always been an abject nightmare for me, while nVidia has usually worked. Note that I am not the fanboy who will say it "just works" which would be a lie. But, it can be made to work. I've been flip-flopping between ATI and nVidia and back in the day had 3dfx and even Permedia and PowerVR at times and I've spent most of my time with nVidia and never regretted it. I've regretted every moment I spent with ATI, most especially on Linux. I've heard that some people with just the right-generation GPU have great results with the free ati driver; everyone else suffers.
I don't care if I'm off by five or ten FPS, I care that output is pretty and that I can use my video card.
This is why AMD/Foundry needs more fabs (Score:3, Interesting)
AMD has needed new fabs to increase capacity for a long time now. After AMD purchased ATI, I always found it odd that there wasn't more of a push to build more fabs and bring their GPU production in-house. At the least, NVIDIA should also be suffering from TSMC having problems, even though they may not be feeling the crunch at the moment.
Re:Bad Financial News for AMD (Score:4, Interesting)
i'd hate to be on a one cpu maker planet
You mean one x86 CPU manufacturer. TI, Samsung, Qualcomm, and a dozen other companies all make ARM chips and these outsell x86 by a large margin.
Re:This is where Intel rules (Score:3, Interesting)
Well they've only done it once before, that was with TSMC's 55nm process. Prior to that, all GPUs I am aware of were on ITRS nodes. In terms of the 55nm chips they did get a bit burned on supply but it worked pretty well, more or less. Of course while 55nm was non standard, it wasn't blazing new ground. When TSMC was bringing 55nm online, Intel already had 45nm products for sale. Also from my understanding their 55nm process was more or less a shrink on the 65nm process they have. Not much changes in terms of design and implementation. Not so on the 40nm process, it might be a shrink from a 45nm process, but they don't have one of those. They go from 55nm to 40nm, nothing in the middle.
That nVidia and ATi with with it doesn't surprise me. At the time they were designing these chips, nobody else really had the capacity to take their orders (Global Foundries wasn't up and running when these were starting design) and in graphics there is a heavy push for smaller. GPUs use too much power and give off too much heat. Anything you can do to lower that is good, because it means you can have more power. GPUs also can scale to pretty ridiculous levels since they are so parallel so there's always the possibility of adding more transistors if you've got the power/silicon budget. So makes sense to me they'd try TSMC's 40nm process.
Also, when they did it, they didn't know how bad it'd be. If they could have seen to now, I think perhaps it would have been different. Like I said, there were some problems on the 55nm process, but not big ones. The 40nm process has been much more problematic.
We'll see what they do next. TSMC hasn't said, publicly at least, what the next process is they are bringing online or when. Global foundries is saying they can take 32nm orders now for delivery in early 2010. So perhaps TSMC will lose out on the next gen of cards.