Slashdot Mirror
China Releases 2nd generation MIPS Chip
eldawg writes writes with news of the launch of a second-generation Chinese 64-bit MIPS CPU. "The Godson-2 or 'Dragon' went into production last week. News reports indicate that, 'The CPU is 95% MIPS compatible using an unauthorized and unlicensed variation of the MIPS architecture, which is owned by the American company MIPS Technologies...The Godson-2 is pretty much a copy of the MIPS R10000 which makes it on par with 1995 technology.' The Chinese plan on using these chips in consumer electronics for the local market, but one can assume that they will eventually end up in exported electronic goods. I wonder if MIPS Technology will sit idly by when this happens?"
A patent granted in USA is not automatically valid elsewhere, and you cannot infringe on a patent where it's not valid. The Chinese will infringe on MIPS patents if they try to export their chip to countries where the MIPS patens are valid.
It does not implement the bits that are patented. IIRC there are patents MIPS equivalent of SIMD instructions and a few others. The chinese were wise enough to skip these so they in fact can export this and MIPS technologies will have to sit and watch.
Do you have any facts about this, or is it your intuition?.
It was one of the design criteria. There was plenty of information about it 1-2 years ago. It was carefully and deliberately designed around MIPS patents. The rest of the architecture and the instruction set is an industry standard and in the public domain.
If the Godson-2 is "pretty much a copy of the MIPS R10000".
It is as far as instruction set is concerned. It is not as far as technology and implementation. While R10000 was not a bad CPU, I would expect "Godson" to be considerably better. It should consume less and scale to higher frequencies. China has manufacturing capability on 150nm (and possibly less) which was not available to anyone in 1995
Baker's Law: Misery no longer loves company. Nowadays it insists on it
http://www.sigsegv.cx/
"The Register" has a better write up on this story (sorry guys). Apparently they've managed to get Windows CE, Linux, and VxWorks up and running on the CPU.
As for the patented instruction sets, apparently they aren't used in the chip. (Supposedly that's why it's 95% compatible).
Currently the chip clocks in at 400-500Mhz, but the next generation is going to be around the 1Ghz mark - by which point China is going to be spitting out all manner of sub $200 computers I imagine.
READY.
PRINT ""+-0
See AMD vs Intel ca. 1991
You do realize, that AMD had a license to second-source Intel parts, right? That litigation was over the terms of that license.
-jcr
The only title of honor that a tyrant can grant is "Enemy of the State."
http://www.china.org.cn/english/2004/Jan/85390.ht
An agency of the Chinese government announced that economic growth reached 9.1% for 2003.
http://www.theglobeandmail.com/servlet/ArticleNew
The truth is very different and much more compelling. The International Olympic Committee told the Chinese to slow down construction due to fears that the Olympic venues would become white elephants (read the link).
http://en.wikipedia.org/wiki/Three_Gorges_Dam
Probably not very up-to-date but wikipedia says that one generator was online in 2003 and all 26 are expected to come online by 2009. So the dam being operational now doesn't mean much if it's producing less than 10% of its full capacity.
It doesn't implement the unaligned memory access instructions. This is one of the only useful patents held by MIPS.
.
They could not implement the unaligned memory access instructions of the MIPS architecture. MIPS have a patent for this.
.
The Godson-2 apparently is a clone of the R10000, customized for Chinese manufacturing plants. They probably shrunk the die and put it in a different box, but it doesn't change the processor.
No, they recreated it from scratch from the instruction set.
You can't just take an existing die and somehow create a processor design from it (well, you could strip each layer off, then reverse engineer each layer, then spend ages working out what bit does what - easier to just design it from scratch), and then shrink it. They probably even have a license for the design and layout software they're using.
Godson 1 was the first implementation from scratch. Godson 2 is clearly the first step towards modernisation and optimisation of that design - 150nm or 180nm instead of what? 350nm? Maybe alternative non-patent infringing implementations of the missing functionality?
It is a different processor, as different as an Opteron is from a Xeon anyway.
The Chinese know that this processor will be used in products abroad. They know it'll be looked at in detail. Their capitalist side will ensure that there won't be any problems in this regard.
A brief description with picture of the chip:u /0312/258718.html/
u /0312/258719.html/
http://www.pconline.com.cn/pchardware/foreline/cp
A 13-page write-up documenting the tough work and challenges faced by one of the chip scientists (e.g. pipelines/branch-prediction/cache design, packaging, etc...):
http://www.pconline.com.cn/pchardware/foreline/cp
Interesting bits from those Chinese sites:
- (back in 2003) they're already running Linux on it, with applications such as MP3 audio/mpeg movie playing, Mozilla, OpenOffice, games...
- (back in 2003) Max clock 300MHz, 1-2W power consumption, 1% CPU load for playing MP3, 23% for mpeg movie, SPEC_CPU2000 score of 300
- will reach 1GHz by early 2006
- it will be used in low-cost PC with price RMB1,000
- the 3rd gen of the chip will incorporate multi-core design
The US International Trade Commission can issue an exclusion order which allows US Customs to seize products that violate US patents, trademarks, and copyrights.
Mea navis aericumbens anguillis abundat
You can't just take an existing die and somehow create a processor design from it (well, you could strip each layer off, then reverse engineer each layer, then spend ages working out what bit does what - easier to just design it from scratch)
Regarding deprocessing then reverse engineering and copying each layer, I know it sounds absurd, but I have a taiwanese friend of mine who said this happens all the time. In fact this person had done it before when working back in Taiwan. She was particularly proud of having copied an RF design and having only a 3dB sensitivity difference between the copy and the original (which IMO is pretty good for having only a die to work off of).
Our group laughed about this on one RF IC we did because the top layer was copper. As soon as one tried to deprocess the die the copper would corrode and turn all green. If you looked at it under a scope it was just a big green blob, so we called it our 'IP protection layer'.
He didn't say that the rate of literacy went up by 1000%, but that literacy went up by 1000%.